1 #if !defined(_FX_JPEG_TURBO_)
2 /*
3 * jccoefct.c
4 *
5 * Copyright (C) 1994-1997, Thomas G. Lane.
6 * This file is part of the Independent JPEG Group's software.
7 * For conditions of distribution and use, see the accompanying README file.
8 *
9 * This file contains the coefficient buffer controller for compression.
10 * This controller is the top level of the JPEG compressor proper.
11 * The coefficient buffer lies between forward-DCT and entropy encoding steps.
12 */
13
14 #define JPEG_INTERNALS
15 #include "jinclude.h"
16 #include "jpeglib.h"
17
18
19 /* We use a full-image coefficient buffer when doing Huffman optimization,
20 * and also for writing multiple-scan JPEG files. In all cases, the DCT
21 * step is run during the first pass, and subsequent passes need only read
22 * the buffered coefficients.
23 */
24 #ifdef ENTROPY_OPT_SUPPORTED
25 #define FULL_COEF_BUFFER_SUPPORTED
26 #else
27 #ifdef C_MULTISCAN_FILES_SUPPORTED
28 #define FULL_COEF_BUFFER_SUPPORTED
29 #endif
30 #endif
31
32
33 /* Private buffer controller object */
34
35 typedef struct {
36 struct jpeg_c_coef_controller pub; /* public fields */
37
38 JDIMENSION iMCU_row_num; /* iMCU row # within image */
39 JDIMENSION mcu_ctr; /* counts MCUs processed in current row */
40 int MCU_vert_offset; /* counts MCU rows within iMCU row */
41 int MCU_rows_per_iMCU_row; /* number of such rows needed */
42
43 /* For single-pass compression, it's sufficient to buffer just one MCU
44 * (although this may prove a bit slow in practice). We allocate a
45 * workspace of C_MAX_BLOCKS_IN_MCU coefficient blocks, and reuse it for each
46 * MCU constructed and sent. (On 80x86, the workspace is FAR even though
47 * it's not really very big; this is to keep the module interfaces unchanged
48 * when a large coefficient buffer is necessary.)
49 * In multi-pass modes, this array points to the current MCU's blocks
50 * within the virtual arrays.
51 */
52 JBLOCKROW MCU_buffer[C_MAX_BLOCKS_IN_MCU];
53
54 /* In multi-pass modes, we need a virtual block array for each component. */
55 jvirt_barray_ptr whole_image[MAX_COMPONENTS];
56 } my_coef_controller;
57
58 typedef my_coef_controller * my_coef_ptr;
59
60
61 /* Forward declarations */
62 METHODDEF(boolean) compress_data
63 JPP((j_compress_ptr cinfo, JSAMPIMAGE input_buf));
64 #ifdef FULL_COEF_BUFFER_SUPPORTED
65 METHODDEF(boolean) compress_first_pass
66 JPP((j_compress_ptr cinfo, JSAMPIMAGE input_buf));
67 METHODDEF(boolean) compress_output
68 JPP((j_compress_ptr cinfo, JSAMPIMAGE input_buf));
69 #endif
70
71
72 LOCAL(void)
start_iMCU_row(j_compress_ptr cinfo)73 start_iMCU_row (j_compress_ptr cinfo)
74 /* Reset within-iMCU-row counters for a new row */
75 {
76 my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
77
78 /* In an interleaved scan, an MCU row is the same as an iMCU row.
79 * In a noninterleaved scan, an iMCU row has v_samp_factor MCU rows.
80 * But at the bottom of the image, process only what's left.
81 */
82 if (cinfo->comps_in_scan > 1) {
83 coef->MCU_rows_per_iMCU_row = 1;
84 } else {
85 if (coef->iMCU_row_num < (cinfo->total_iMCU_rows-1))
86 coef->MCU_rows_per_iMCU_row = cinfo->cur_comp_info[0]->v_samp_factor;
87 else
88 coef->MCU_rows_per_iMCU_row = cinfo->cur_comp_info[0]->last_row_height;
89 }
90
91 coef->mcu_ctr = 0;
92 coef->MCU_vert_offset = 0;
93 }
94
95
96 /*
97 * Initialize for a processing pass.
98 */
99
100 METHODDEF(void)
start_pass_coef(j_compress_ptr cinfo,J_BUF_MODE pass_mode)101 start_pass_coef (j_compress_ptr cinfo, J_BUF_MODE pass_mode)
102 {
103 my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
104
105 coef->iMCU_row_num = 0;
106 start_iMCU_row(cinfo);
107
108 switch (pass_mode) {
109 case JBUF_PASS_THRU:
110 if (coef->whole_image[0] != NULL)
111 ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
112 coef->pub.compress_data = compress_data;
113 break;
114 #ifdef FULL_COEF_BUFFER_SUPPORTED
115 case JBUF_SAVE_AND_PASS:
116 if (coef->whole_image[0] == NULL)
117 ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
118 coef->pub.compress_data = compress_first_pass;
119 break;
120 case JBUF_CRANK_DEST:
121 if (coef->whole_image[0] == NULL)
122 ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
123 coef->pub.compress_data = compress_output;
124 break;
125 #endif
126 default:
127 ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
128 break;
129 }
130 }
131
132
133 /*
134 * Process some data in the single-pass case.
135 * We process the equivalent of one fully interleaved MCU row ("iMCU" row)
136 * per call, ie, v_samp_factor block rows for each component in the image.
137 * Returns TRUE if the iMCU row is completed, FALSE if suspended.
138 *
139 * NB: input_buf contains a plane for each component in image,
140 * which we index according to the component's SOF position.
141 */
142
143 METHODDEF(boolean)
compress_data(j_compress_ptr cinfo,JSAMPIMAGE input_buf)144 compress_data (j_compress_ptr cinfo, JSAMPIMAGE input_buf)
145 {
146 my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
147 JDIMENSION MCU_col_num; /* index of current MCU within row */
148 JDIMENSION last_MCU_col = cinfo->MCUs_per_row - 1;
149 JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1;
150 int blkn, bi, ci, yindex, yoffset, blockcnt;
151 JDIMENSION ypos, xpos;
152 jpeg_component_info *compptr;
153
154 /* Loop to write as much as one whole iMCU row */
155 for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row;
156 yoffset++) {
157 for (MCU_col_num = coef->mcu_ctr; MCU_col_num <= last_MCU_col;
158 MCU_col_num++) {
159 /* Determine where data comes from in input_buf and do the DCT thing.
160 * Each call on forward_DCT processes a horizontal row of DCT blocks
161 * as wide as an MCU; we rely on having allocated the MCU_buffer[] blocks
162 * sequentially. Dummy blocks at the right or bottom edge are filled in
163 * specially. The data in them does not matter for image reconstruction,
164 * so we fill them with values that will encode to the smallest amount of
165 * data, viz: all zeroes in the AC entries, DC entries equal to previous
166 * block's DC value. (Thanks to Thomas Kinsman for this idea.)
167 */
168 blkn = 0;
169 for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
170 compptr = cinfo->cur_comp_info[ci];
171 blockcnt = (MCU_col_num < last_MCU_col) ? compptr->MCU_width
172 : compptr->last_col_width;
173 xpos = MCU_col_num * compptr->MCU_sample_width;
174 ypos = yoffset * DCTSIZE; /* ypos == (yoffset+yindex) * DCTSIZE */
175 for (yindex = 0; yindex < compptr->MCU_height; yindex++) {
176 if (coef->iMCU_row_num < last_iMCU_row ||
177 yoffset+yindex < compptr->last_row_height) {
178 (*cinfo->fdct->forward_DCT) (cinfo, compptr,
179 input_buf[compptr->component_index],
180 coef->MCU_buffer[blkn],
181 ypos, xpos, (JDIMENSION) blockcnt);
182 if (blockcnt < compptr->MCU_width) {
183 /* Create some dummy blocks at the right edge of the image. */
184 jzero_far((void FAR *) coef->MCU_buffer[blkn + blockcnt],
185 (compptr->MCU_width - blockcnt) * SIZEOF(JBLOCK));
186 for (bi = blockcnt; bi < compptr->MCU_width; bi++) {
187 coef->MCU_buffer[blkn+bi][0][0] = coef->MCU_buffer[blkn+bi-1][0][0];
188 }
189 }
190 } else {
191 /* Create a row of dummy blocks at the bottom of the image. */
192 jzero_far((void FAR *) coef->MCU_buffer[blkn],
193 compptr->MCU_width * SIZEOF(JBLOCK));
194 for (bi = 0; bi < compptr->MCU_width; bi++) {
195 coef->MCU_buffer[blkn+bi][0][0] = coef->MCU_buffer[blkn-1][0][0];
196 }
197 }
198 blkn += compptr->MCU_width;
199 ypos += DCTSIZE;
200 }
201 }
202 /* Try to write the MCU. In event of a suspension failure, we will
203 * re-DCT the MCU on restart (a bit inefficient, could be fixed...)
204 */
205 if (! (*cinfo->entropy->encode_mcu) (cinfo, coef->MCU_buffer)) {
206 /* Suspension forced; update state counters and exit */
207 coef->MCU_vert_offset = yoffset;
208 coef->mcu_ctr = MCU_col_num;
209 return FALSE;
210 }
211 }
212 /* Completed an MCU row, but perhaps not an iMCU row */
213 coef->mcu_ctr = 0;
214 }
215 /* Completed the iMCU row, advance counters for next one */
216 coef->iMCU_row_num++;
217 start_iMCU_row(cinfo);
218 return TRUE;
219 }
220
221
222 #ifdef FULL_COEF_BUFFER_SUPPORTED
223
224 /*
225 * Process some data in the first pass of a multi-pass case.
226 * We process the equivalent of one fully interleaved MCU row ("iMCU" row)
227 * per call, ie, v_samp_factor block rows for each component in the image.
228 * This amount of data is read from the source buffer, DCT'd and quantized,
229 * and saved into the virtual arrays. We also generate suitable dummy blocks
230 * as needed at the right and lower edges. (The dummy blocks are constructed
231 * in the virtual arrays, which have been padded appropriately.) This makes
232 * it possible for subsequent passes not to worry about real vs. dummy blocks.
233 *
234 * We must also emit the data to the entropy encoder. This is conveniently
235 * done by calling compress_output() after we've loaded the current strip
236 * of the virtual arrays.
237 *
238 * NB: input_buf contains a plane for each component in image. All
239 * components are DCT'd and loaded into the virtual arrays in this pass.
240 * However, it may be that only a subset of the components are emitted to
241 * the entropy encoder during this first pass; be careful about looking
242 * at the scan-dependent variables (MCU dimensions, etc).
243 */
244
245 METHODDEF(boolean)
compress_first_pass(j_compress_ptr cinfo,JSAMPIMAGE input_buf)246 compress_first_pass (j_compress_ptr cinfo, JSAMPIMAGE input_buf)
247 {
248 my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
249 JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1;
250 JDIMENSION blocks_across, MCUs_across, MCUindex;
251 int bi, ci, h_samp_factor, block_row, block_rows, ndummy;
252 JCOEF lastDC;
253 jpeg_component_info *compptr;
254 JBLOCKARRAY buffer;
255 JBLOCKROW thisblockrow, lastblockrow;
256
257 for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
258 ci++, compptr++) {
259 /* Align the virtual buffer for this component. */
260 buffer = (*cinfo->mem->access_virt_barray)
261 ((j_common_ptr) cinfo, coef->whole_image[ci],
262 coef->iMCU_row_num * compptr->v_samp_factor,
263 (JDIMENSION) compptr->v_samp_factor, TRUE);
264 /* Count non-dummy DCT block rows in this iMCU row. */
265 if (coef->iMCU_row_num < last_iMCU_row)
266 block_rows = compptr->v_samp_factor;
267 else {
268 /* NB: can't use last_row_height here, since may not be set! */
269 block_rows = (int) (compptr->height_in_blocks % compptr->v_samp_factor);
270 if (block_rows == 0) block_rows = compptr->v_samp_factor;
271 }
272 blocks_across = compptr->width_in_blocks;
273 h_samp_factor = compptr->h_samp_factor;
274 /* Count number of dummy blocks to be added at the right margin. */
275 ndummy = (int) (blocks_across % h_samp_factor);
276 if (ndummy > 0)
277 ndummy = h_samp_factor - ndummy;
278 /* Perform DCT for all non-dummy blocks in this iMCU row. Each call
279 * on forward_DCT processes a complete horizontal row of DCT blocks.
280 */
281 for (block_row = 0; block_row < block_rows; block_row++) {
282 thisblockrow = buffer[block_row];
283 (*cinfo->fdct->forward_DCT) (cinfo, compptr,
284 input_buf[ci], thisblockrow,
285 (JDIMENSION) (block_row * DCTSIZE),
286 (JDIMENSION) 0, blocks_across);
287 if (ndummy > 0) {
288 /* Create dummy blocks at the right edge of the image. */
289 thisblockrow += blocks_across; /* => first dummy block */
290 jzero_far((void FAR *) thisblockrow, ndummy * SIZEOF(JBLOCK));
291 lastDC = thisblockrow[-1][0];
292 for (bi = 0; bi < ndummy; bi++) {
293 thisblockrow[bi][0] = lastDC;
294 }
295 }
296 }
297 /* If at end of image, create dummy block rows as needed.
298 * The tricky part here is that within each MCU, we want the DC values
299 * of the dummy blocks to match the last real block's DC value.
300 * This squeezes a few more bytes out of the resulting file...
301 */
302 if (coef->iMCU_row_num == last_iMCU_row) {
303 blocks_across += ndummy; /* include lower right corner */
304 MCUs_across = blocks_across / h_samp_factor;
305 for (block_row = block_rows; block_row < compptr->v_samp_factor;
306 block_row++) {
307 thisblockrow = buffer[block_row];
308 lastblockrow = buffer[block_row-1];
309 jzero_far((void FAR *) thisblockrow,
310 (size_t) (blocks_across * SIZEOF(JBLOCK)));
311 for (MCUindex = 0; MCUindex < MCUs_across; MCUindex++) {
312 lastDC = lastblockrow[h_samp_factor-1][0];
313 for (bi = 0; bi < h_samp_factor; bi++) {
314 thisblockrow[bi][0] = lastDC;
315 }
316 thisblockrow += h_samp_factor; /* advance to next MCU in row */
317 lastblockrow += h_samp_factor;
318 }
319 }
320 }
321 }
322 /* NB: compress_output will increment iMCU_row_num if successful.
323 * A suspension return will result in redoing all the work above next time.
324 */
325
326 /* Emit data to the entropy encoder, sharing code with subsequent passes */
327 return compress_output(cinfo, input_buf);
328 }
329
330
331 /*
332 * Process some data in subsequent passes of a multi-pass case.
333 * We process the equivalent of one fully interleaved MCU row ("iMCU" row)
334 * per call, ie, v_samp_factor block rows for each component in the scan.
335 * The data is obtained from the virtual arrays and fed to the entropy coder.
336 * Returns TRUE if the iMCU row is completed, FALSE if suspended.
337 *
338 * NB: input_buf is ignored; it is likely to be a NULL pointer.
339 */
340
341 METHODDEF(boolean)
compress_output(j_compress_ptr cinfo,JSAMPIMAGE input_buf)342 compress_output (j_compress_ptr cinfo, JSAMPIMAGE input_buf)
343 {
344 my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
345 JDIMENSION MCU_col_num; /* index of current MCU within row */
346 int blkn, ci, xindex, yindex, yoffset;
347 JDIMENSION start_col;
348 JBLOCKARRAY buffer[MAX_COMPS_IN_SCAN];
349 JBLOCKROW buffer_ptr;
350 jpeg_component_info *compptr;
351
352 /* Align the virtual buffers for the components used in this scan.
353 * NB: during first pass, this is safe only because the buffers will
354 * already be aligned properly, so jmemmgr.c won't need to do any I/O.
355 */
356 for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
357 compptr = cinfo->cur_comp_info[ci];
358 buffer[ci] = (*cinfo->mem->access_virt_barray)
359 ((j_common_ptr) cinfo, coef->whole_image[compptr->component_index],
360 coef->iMCU_row_num * compptr->v_samp_factor,
361 (JDIMENSION) compptr->v_samp_factor, FALSE);
362 }
363
364 /* Loop to process one whole iMCU row */
365 for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row;
366 yoffset++) {
367 for (MCU_col_num = coef->mcu_ctr; MCU_col_num < cinfo->MCUs_per_row;
368 MCU_col_num++) {
369 /* Construct list of pointers to DCT blocks belonging to this MCU */
370 blkn = 0; /* index of current DCT block within MCU */
371 for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
372 compptr = cinfo->cur_comp_info[ci];
373 start_col = MCU_col_num * compptr->MCU_width;
374 for (yindex = 0; yindex < compptr->MCU_height; yindex++) {
375 buffer_ptr = buffer[ci][yindex+yoffset] + start_col;
376 for (xindex = 0; xindex < compptr->MCU_width; xindex++) {
377 coef->MCU_buffer[blkn++] = buffer_ptr++;
378 }
379 }
380 }
381 /* Try to write the MCU. */
382 if (! (*cinfo->entropy->encode_mcu) (cinfo, coef->MCU_buffer)) {
383 /* Suspension forced; update state counters and exit */
384 coef->MCU_vert_offset = yoffset;
385 coef->mcu_ctr = MCU_col_num;
386 return FALSE;
387 }
388 }
389 /* Completed an MCU row, but perhaps not an iMCU row */
390 coef->mcu_ctr = 0;
391 }
392 /* Completed the iMCU row, advance counters for next one */
393 coef->iMCU_row_num++;
394 start_iMCU_row(cinfo);
395 return TRUE;
396 }
397
398 #endif /* FULL_COEF_BUFFER_SUPPORTED */
399
400
401 /*
402 * Initialize coefficient buffer controller.
403 */
404
405 GLOBAL(void)
jinit_c_coef_controller(j_compress_ptr cinfo,boolean need_full_buffer)406 jinit_c_coef_controller (j_compress_ptr cinfo, boolean need_full_buffer)
407 {
408 my_coef_ptr coef;
409
410 coef = (my_coef_ptr)
411 (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
412 SIZEOF(my_coef_controller));
413 cinfo->coef = (struct jpeg_c_coef_controller *) coef;
414 coef->pub.start_pass = start_pass_coef;
415
416 /* Create the coefficient buffer. */
417 if (need_full_buffer) {
418 #ifdef FULL_COEF_BUFFER_SUPPORTED
419 /* Allocate a full-image virtual array for each component, */
420 /* padded to a multiple of samp_factor DCT blocks in each direction. */
421 int ci;
422 jpeg_component_info *compptr;
423
424 for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
425 ci++, compptr++) {
426 coef->whole_image[ci] = (*cinfo->mem->request_virt_barray)
427 ((j_common_ptr) cinfo, JPOOL_IMAGE, FALSE,
428 (JDIMENSION) jround_up((long) compptr->width_in_blocks,
429 (long) compptr->h_samp_factor),
430 (JDIMENSION) jround_up((long) compptr->height_in_blocks,
431 (long) compptr->v_samp_factor),
432 (JDIMENSION) compptr->v_samp_factor);
433 }
434 #else
435 ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
436 #endif
437 } else {
438 /* We only need a single-MCU buffer. */
439 JBLOCKROW buffer;
440 int i;
441
442 buffer = (JBLOCKROW)
443 (*cinfo->mem->alloc_large) ((j_common_ptr) cinfo, JPOOL_IMAGE,
444 C_MAX_BLOCKS_IN_MCU * SIZEOF(JBLOCK));
445 for (i = 0; i < C_MAX_BLOCKS_IN_MCU; i++) {
446 coef->MCU_buffer[i] = buffer + i;
447 }
448 coef->whole_image[0] = NULL; /* flag for no virtual arrays */
449 }
450 }
451
452 #endif //_FX_JPEG_TURBO_
453