Sync with trunk (r48042), except win32k/ntuser/cursoricon.c
[reactos.git] / dll / 3rdparty / libjpeg / jccoefct.c
1 /*
2 * jccoefct.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 compression.
9 * This controller is the top level of the JPEG compressor proper.
10 * The coefficient buffer lies between forward-DCT and entropy encoding steps.
11 */
12
13 #define JPEG_INTERNALS
14 #include "jinclude.h"
15 #include "jpeglib.h"
16
17
18 /* We use a full-image coefficient buffer when doing Huffman optimization,
19 * and also for writing multiple-scan JPEG files. In all cases, the DCT
20 * step is run during the first pass, and subsequent passes need only read
21 * the buffered coefficients.
22 */
23 #ifdef ENTROPY_OPT_SUPPORTED
24 #define FULL_COEF_BUFFER_SUPPORTED
25 #else
26 #ifdef C_MULTISCAN_FILES_SUPPORTED
27 #define FULL_COEF_BUFFER_SUPPORTED
28 #endif
29 #endif
30
31
32 /* Private buffer controller object */
33
34 typedef struct {
35 struct jpeg_c_coef_controller pub; /* public fields */
36
37 JDIMENSION iMCU_row_num; /* iMCU row # within image */
38 JDIMENSION mcu_ctr; /* counts MCUs processed in current row */
39 int MCU_vert_offset; /* counts MCU rows within iMCU row */
40 int MCU_rows_per_iMCU_row; /* number of such rows needed */
41
42 /* For single-pass compression, it's sufficient to buffer just one MCU
43 * (although this may prove a bit slow in practice). We allocate a
44 * workspace of C_MAX_BLOCKS_IN_MCU coefficient blocks, and reuse it for each
45 * MCU constructed and sent. (On 80x86, the workspace is FAR even though
46 * it's not really very big; this is to keep the module interfaces unchanged
47 * when a large coefficient buffer is necessary.)
48 * In multi-pass modes, this array points to the current MCU's blocks
49 * 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
62 JPP((j_compress_ptr cinfo, JSAMPIMAGE input_buf));
63 #ifdef FULL_COEF_BUFFER_SUPPORTED
64 METHODDEF(boolean) compress_first_pass
65 JPP((j_compress_ptr cinfo, JSAMPIMAGE input_buf));
66 METHODDEF(boolean) compress_output
67 JPP((j_compress_ptr cinfo, JSAMPIMAGE input_buf));
68 #endif
69
70
71 LOCAL(void)
72 start_iMCU_row (j_compress_ptr cinfo)
73 /* Reset within-iMCU-row counters for a new row */
74 {
75 my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
76
77 /* In an interleaved scan, an MCU row is the same as an iMCU row.
78 * In a noninterleaved scan, an iMCU row has v_samp_factor MCU rows.
79 * But at the bottom of the image, process only what's left.
80 */
81 if (cinfo->comps_in_scan > 1) {
82 coef->MCU_rows_per_iMCU_row = 1;
83 } else {
84 if (coef->iMCU_row_num < (cinfo->total_iMCU_rows-1))
85 coef->MCU_rows_per_iMCU_row = cinfo->cur_comp_info[0]->v_samp_factor;
86 else
87 coef->MCU_rows_per_iMCU_row = cinfo->cur_comp_info[0]->last_row_height;
88 }
89
90 coef->mcu_ctr = 0;
91 coef->MCU_vert_offset = 0;
92 }
93
94
95 /*
96 * Initialize for a processing pass.
97 */
98
99 METHODDEF(void)
100 start_pass_coef (j_compress_ptr cinfo, J_BUF_MODE pass_mode)
101 {
102 my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
103
104 coef->iMCU_row_num = 0;
105 start_iMCU_row(cinfo);
106
107 switch (pass_mode) {
108 case JBUF_PASS_THRU:
109 if (coef->whole_image[0] != NULL)
110 ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
111 coef->pub.compress_data = compress_data;
112 break;
113 #ifdef FULL_COEF_BUFFER_SUPPORTED
114 case JBUF_SAVE_AND_PASS:
115 if (coef->whole_image[0] == NULL)
116 ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
117 coef->pub.compress_data = compress_first_pass;
118 break;
119 case JBUF_CRANK_DEST:
120 if (coef->whole_image[0] == NULL)
121 ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
122 coef->pub.compress_data = compress_output;
123 break;
124 #endif
125 default:
126 ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
127 break;
128 }
129 }
130
131
132 /*
133 * Process some data in the single-pass case.
134 * We process the equivalent of one fully interleaved MCU row ("iMCU" row)
135 * per call, ie, v_samp_factor block rows for each component in the image.
136 * Returns TRUE if the iMCU row is completed, FALSE if suspended.
137 *
138 * NB: input_buf contains a plane for each component in image,
139 * which we index according to the component's SOF position.
140 */
141
142 METHODDEF(boolean)
143 compress_data (j_compress_ptr cinfo, JSAMPIMAGE input_buf)
144 {
145 my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
146 JDIMENSION MCU_col_num; /* index of current MCU within row */
147 JDIMENSION last_MCU_col = cinfo->MCUs_per_row - 1;
148 JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1;
149 int blkn, bi, ci, yindex, yoffset, blockcnt;
150 JDIMENSION ypos, xpos;
151 jpeg_component_info *compptr;
152 forward_DCT_ptr forward_DCT;
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 forward_DCT = cinfo->fdct->forward_DCT[compptr->component_index];
172 blockcnt = (MCU_col_num < last_MCU_col) ? compptr->MCU_width
173 : compptr->last_col_width;
174 xpos = MCU_col_num * compptr->MCU_sample_width;
175 ypos = yoffset * compptr->DCT_v_scaled_size;
176 /* ypos == (yoffset+yindex) * DCTSIZE */
177 for (yindex = 0; yindex < compptr->MCU_height; yindex++) {
178 if (coef->iMCU_row_num < last_iMCU_row ||
179 yoffset+yindex < compptr->last_row_height) {
180 (*forward_DCT) (cinfo, compptr,
181 input_buf[compptr->component_index],
182 coef->MCU_buffer[blkn],
183 ypos, xpos, (JDIMENSION) blockcnt);
184 if (blockcnt < compptr->MCU_width) {
185 /* Create some dummy blocks at the right edge of the image. */
186 jzero_far((void FAR *) coef->MCU_buffer[blkn + blockcnt],
187 (compptr->MCU_width - blockcnt) * SIZEOF(JBLOCK));
188 for (bi = blockcnt; bi < compptr->MCU_width; bi++) {
189 coef->MCU_buffer[blkn+bi][0][0] = coef->MCU_buffer[blkn+bi-1][0][0];
190 }
191 }
192 } else {
193 /* Create a row of dummy blocks at the bottom of the image. */
194 jzero_far((void FAR *) coef->MCU_buffer[blkn],
195 compptr->MCU_width * SIZEOF(JBLOCK));
196 for (bi = 0; bi < compptr->MCU_width; bi++) {
197 coef->MCU_buffer[blkn+bi][0][0] = coef->MCU_buffer[blkn-1][0][0];
198 }
199 }
200 blkn += compptr->MCU_width;
201 ypos += compptr->DCT_v_scaled_size;
202 }
203 }
204 /* Try to write the MCU. In event of a suspension failure, we will
205 * re-DCT the MCU on restart (a bit inefficient, could be fixed...)
206 */
207 if (! (*cinfo->entropy->encode_mcu) (cinfo, coef->MCU_buffer)) {
208 /* Suspension forced; update state counters and exit */
209 coef->MCU_vert_offset = yoffset;
210 coef->mcu_ctr = MCU_col_num;
211 return FALSE;
212 }
213 }
214 /* Completed an MCU row, but perhaps not an iMCU row */
215 coef->mcu_ctr = 0;
216 }
217 /* Completed the iMCU row, advance counters for next one */
218 coef->iMCU_row_num++;
219 start_iMCU_row(cinfo);
220 return TRUE;
221 }
222
223
224 #ifdef FULL_COEF_BUFFER_SUPPORTED
225
226 /*
227 * Process some data in the first pass of a multi-pass case.
228 * We process the equivalent of one fully interleaved MCU row ("iMCU" row)
229 * per call, ie, v_samp_factor block rows for each component in the image.
230 * This amount of data is read from the source buffer, DCT'd and quantized,
231 * and saved into the virtual arrays. We also generate suitable dummy blocks
232 * as needed at the right and lower edges. (The dummy blocks are constructed
233 * in the virtual arrays, which have been padded appropriately.) This makes
234 * it possible for subsequent passes not to worry about real vs. dummy blocks.
235 *
236 * We must also emit the data to the entropy encoder. This is conveniently
237 * done by calling compress_output() after we've loaded the current strip
238 * of the virtual arrays.
239 *
240 * NB: input_buf contains a plane for each component in image. All
241 * components are DCT'd and loaded into the virtual arrays in this pass.
242 * However, it may be that only a subset of the components are emitted to
243 * the entropy encoder during this first pass; be careful about looking
244 * at the scan-dependent variables (MCU dimensions, etc).
245 */
246
247 METHODDEF(boolean)
248 compress_first_pass (j_compress_ptr cinfo, JSAMPIMAGE input_buf)
249 {
250 my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
251 JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1;
252 JDIMENSION blocks_across, MCUs_across, MCUindex;
253 int bi, ci, h_samp_factor, block_row, block_rows, ndummy;
254 JCOEF lastDC;
255 jpeg_component_info *compptr;
256 JBLOCKARRAY buffer;
257 JBLOCKROW thisblockrow, lastblockrow;
258 forward_DCT_ptr forward_DCT;
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 forward_DCT = cinfo->fdct->forward_DCT[ci];
282 /* Perform DCT for all non-dummy blocks in this iMCU row. Each call
283 * on forward_DCT processes a complete horizontal row of DCT blocks.
284 */
285 for (block_row = 0; block_row < block_rows; block_row++) {
286 thisblockrow = buffer[block_row];
287 (*forward_DCT) (cinfo, compptr, input_buf[ci], thisblockrow,
288 (JDIMENSION) (block_row * compptr->DCT_v_scaled_size),
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 }