1/*
2 * jdmainct.c
3 *
4 * This file was part of the Independent JPEG Group's software:
5 * Copyright (C) 1994-1996, Thomas G. Lane.
6 * libjpeg-turbo Modifications:
7 * Copyright (C) 2010, 2016, D. R. Commander.
8 * For conditions of distribution and use, see the accompanying README.ijg
9 * file.
10 *
11 * This file contains the main buffer controller for decompression.
12 * The main buffer lies between the JPEG decompressor proper and the
13 * post-processor; it holds downsampled data in the JPEG colorspace.
14 *
15 * Note that this code is bypassed in raw-data mode, since the application
16 * supplies the equivalent of the main buffer in that case.
17 */
18
19#include "jinclude.h"
20#include "jdmainct.h"
21
22
23/*
24 * In the current system design, the main buffer need never be a full-image
25 * buffer; any full-height buffers will be found inside the coefficient or
26 * postprocessing controllers. Nonetheless, the main controller is not
27 * trivial. Its responsibility is to provide context rows for upsampling/
28 * rescaling, and doing this in an efficient fashion is a bit tricky.
29 *
30 * Postprocessor input data is counted in "row groups". A row group
31 * is defined to be (v_samp_factor * DCT_scaled_size / min_DCT_scaled_size)
32 * sample rows of each component. (We require DCT_scaled_size values to be
33 * chosen such that these numbers are integers. In practice DCT_scaled_size
34 * values will likely be powers of two, so we actually have the stronger
35 * condition that DCT_scaled_size / min_DCT_scaled_size is an integer.)
36 * Upsampling will typically produce max_v_samp_factor pixel rows from each
37 * row group (times any additional scale factor that the upsampler is
38 * applying).
39 *
40 * The coefficient controller will deliver data to us one iMCU row at a time;
41 * each iMCU row contains v_samp_factor * DCT_scaled_size sample rows, or
42 * exactly min_DCT_scaled_size row groups. (This amount of data corresponds
43 * to one row of MCUs when the image is fully interleaved.) Note that the
44 * number of sample rows varies across components, but the number of row
45 * groups does not. Some garbage sample rows may be included in the last iMCU
46 * row at the bottom of the image.
47 *
48 * Depending on the vertical scaling algorithm used, the upsampler may need
49 * access to the sample row(s) above and below its current input row group.
50 * The upsampler is required to set need_context_rows TRUE at global selection
51 * time if so. When need_context_rows is FALSE, this controller can simply
52 * obtain one iMCU row at a time from the coefficient controller and dole it
53 * out as row groups to the postprocessor.
54 *
55 * When need_context_rows is TRUE, this controller guarantees that the buffer
56 * passed to postprocessing contains at least one row group's worth of samples
57 * above and below the row group(s) being processed. Note that the context
58 * rows "above" the first passed row group appear at negative row offsets in
59 * the passed buffer. At the top and bottom of the image, the required
60 * context rows are manufactured by duplicating the first or last real sample
61 * row; this avoids having special cases in the upsampling inner loops.
62 *
63 * The amount of context is fixed at one row group just because that's a
64 * convenient number for this controller to work with. The existing
65 * upsamplers really only need one sample row of context. An upsampler
66 * supporting arbitrary output rescaling might wish for more than one row
67 * group of context when shrinking the image; tough, we don't handle that.
68 * (This is justified by the assumption that downsizing will be handled mostly
69 * by adjusting the DCT_scaled_size values, so that the actual scale factor at
70 * the upsample step needn't be much less than one.)
71 *
72 * To provide the desired context, we have to retain the last two row groups
73 * of one iMCU row while reading in the next iMCU row. (The last row group
74 * can't be processed until we have another row group for its below-context,
75 * and so we have to save the next-to-last group too for its above-context.)
76 * We could do this most simply by copying data around in our buffer, but
77 * that'd be very slow. We can avoid copying any data by creating a rather
78 * strange pointer structure. Here's how it works. We allocate a workspace
79 * consisting of M+2 row groups (where M = min_DCT_scaled_size is the number
80 * of row groups per iMCU row). We create two sets of redundant pointers to
81 * the workspace. Labeling the physical row groups 0 to M+1, the synthesized
82 * pointer lists look like this:
83 * M+1 M-1
84 * master pointer --> 0 master pointer --> 0
85 * 1 1
86 * ... ...
87 * M-3 M-3
88 * M-2 M
89 * M-1 M+1
90 * M M-2
91 * M+1 M-1
92 * 0 0
93 * We read alternate iMCU rows using each master pointer; thus the last two
94 * row groups of the previous iMCU row remain un-overwritten in the workspace.
95 * The pointer lists are set up so that the required context rows appear to
96 * be adjacent to the proper places when we pass the pointer lists to the
97 * upsampler.
98 *
99 * The above pictures describe the normal state of the pointer lists.
100 * At top and bottom of the image, we diddle the pointer lists to duplicate
101 * the first or last sample row as necessary (this is cheaper than copying
102 * sample rows around).
103 *
104 * This scheme breaks down if M < 2, ie, min_DCT_scaled_size is 1. In that
105 * situation each iMCU row provides only one row group so the buffering logic
106 * must be different (eg, we must read two iMCU rows before we can emit the
107 * first row group). For now, we simply do not support providing context
108 * rows when min_DCT_scaled_size is 1. That combination seems unlikely to
109 * be worth providing --- if someone wants a 1/8th-size preview, they probably
110 * want it quick and dirty, so a context-free upsampler is sufficient.
111 */
112
113
114/* Forward declarations */
115METHODDEF(void) process_data_simple_main(j_decompress_ptr cinfo,
116 JSAMPARRAY output_buf,
117 JDIMENSION *out_row_ctr,
118 JDIMENSION out_rows_avail);
119METHODDEF(void) process_data_context_main(j_decompress_ptr cinfo,
120 JSAMPARRAY output_buf,
121 JDIMENSION *out_row_ctr,
122 JDIMENSION out_rows_avail);
123#ifdef QUANT_2PASS_SUPPORTED
124METHODDEF(void) process_data_crank_post(j_decompress_ptr cinfo,
125 JSAMPARRAY output_buf,
126 JDIMENSION *out_row_ctr,
127 JDIMENSION out_rows_avail);
128#endif
129
130
131LOCAL(void)
132alloc_funny_pointers(j_decompress_ptr cinfo)
133/* Allocate space for the funny pointer lists.
134 * This is done only once, not once per pass.
135 */
136{
137 my_main_ptr main_ptr = (my_main_ptr)cinfo->main;
138 int ci, rgroup;
139 int M = cinfo->_min_DCT_scaled_size;
140 jpeg_component_info *compptr;
141 JSAMPARRAY xbuf;
142
143 /* Get top-level space for component array pointers.
144 * We alloc both arrays with one call to save a few cycles.
145 */
146 main_ptr->xbuffer[0] = (JSAMPIMAGE)
147 (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE,
148 cinfo->num_components * 2 * sizeof(JSAMPARRAY));
149 main_ptr->xbuffer[1] = main_ptr->xbuffer[0] + cinfo->num_components;
150
151 for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
152 ci++, compptr++) {
153 rgroup = (compptr->v_samp_factor * compptr->_DCT_scaled_size) /
154 cinfo->_min_DCT_scaled_size; /* height of a row group of component */
155 /* Get space for pointer lists --- M+4 row groups in each list.
156 * We alloc both pointer lists with one call to save a few cycles.
157 */
158 xbuf = (JSAMPARRAY)
159 (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE,
160 2 * (rgroup * (M + 4)) * sizeof(JSAMPROW));
161 xbuf += rgroup; /* want one row group at negative offsets */
162 main_ptr->xbuffer[0][ci] = xbuf;
163 xbuf += rgroup * (M + 4);
164 main_ptr->xbuffer[1][ci] = xbuf;
165 }
166}
167
168
169LOCAL(void)
170make_funny_pointers(j_decompress_ptr cinfo)
171/* Create the funny pointer lists discussed in the comments above.
172 * The actual workspace is already allocated (in main_ptr->buffer),
173 * and the space for the pointer lists is allocated too.
174 * This routine just fills in the curiously ordered lists.
175 * This will be repeated at the beginning of each pass.
176 */
177{
178 my_main_ptr main_ptr = (my_main_ptr)cinfo->main;
179 int ci, i, rgroup;
180 int M = cinfo->_min_DCT_scaled_size;
181 jpeg_component_info *compptr;
182 JSAMPARRAY buf, xbuf0, xbuf1;
183
184 for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
185 ci++, compptr++) {
186 rgroup = (compptr->v_samp_factor * compptr->_DCT_scaled_size) /
187 cinfo->_min_DCT_scaled_size; /* height of a row group of component */
188 xbuf0 = main_ptr->xbuffer[0][ci];
189 xbuf1 = main_ptr->xbuffer[1][ci];
190 /* First copy the workspace pointers as-is */
191 buf = main_ptr->buffer[ci];
192 for (i = 0; i < rgroup * (M + 2); i++) {
193 xbuf0[i] = xbuf1[i] = buf[i];
194 }
195 /* In the second list, put the last four row groups in swapped order */
196 for (i = 0; i < rgroup * 2; i++) {
197 xbuf1[rgroup * (M - 2) + i] = buf[rgroup * M + i];
198 xbuf1[rgroup * M + i] = buf[rgroup * (M - 2) + i];
199 }
200 /* The wraparound pointers at top and bottom will be filled later
201 * (see set_wraparound_pointers, below). Initially we want the "above"
202 * pointers to duplicate the first actual data line. This only needs
203 * to happen in xbuffer[0].
204 */
205 for (i = 0; i < rgroup; i++) {
206 xbuf0[i - rgroup] = xbuf0[0];
207 }
208 }
209}
210
211
212LOCAL(void)
213set_bottom_pointers(j_decompress_ptr cinfo)
214/* Change the pointer lists to duplicate the last sample row at the bottom
215 * of the image. whichptr indicates which xbuffer holds the final iMCU row.
216 * Also sets rowgroups_avail to indicate number of nondummy row groups in row.
217 */
218{
219 my_main_ptr main_ptr = (my_main_ptr)cinfo->main;
220 int ci, i, rgroup, iMCUheight, rows_left;
221 jpeg_component_info *compptr;
222 JSAMPARRAY xbuf;
223
224 for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
225 ci++, compptr++) {
226 /* Count sample rows in one iMCU row and in one row group */
227 iMCUheight = compptr->v_samp_factor * compptr->_DCT_scaled_size;
228 rgroup = iMCUheight / cinfo->_min_DCT_scaled_size;
229 /* Count nondummy sample rows remaining for this component */
230 rows_left = (int)(compptr->downsampled_height % (JDIMENSION)iMCUheight);
231 if (rows_left == 0) rows_left = iMCUheight;
232 /* Count nondummy row groups. Should get same answer for each component,
233 * so we need only do it once.
234 */
235 if (ci == 0) {
236 main_ptr->rowgroups_avail = (JDIMENSION)((rows_left - 1) / rgroup + 1);
237 }
238 /* Duplicate the last real sample row rgroup*2 times; this pads out the
239 * last partial rowgroup and ensures at least one full rowgroup of context.
240 */
241 xbuf = main_ptr->xbuffer[main_ptr->whichptr][ci];
242 for (i = 0; i < rgroup * 2; i++) {
243 xbuf[rows_left + i] = xbuf[rows_left - 1];
244 }
245 }
246}
247
248
249/*
250 * Initialize for a processing pass.
251 */
252
253METHODDEF(void)
254start_pass_main(j_decompress_ptr cinfo, J_BUF_MODE pass_mode)
255{
256 my_main_ptr main_ptr = (my_main_ptr)cinfo->main;
257
258 switch (pass_mode) {
259 case JBUF_PASS_THRU:
260 if (cinfo->upsample->need_context_rows) {
261 main_ptr->pub.process_data = process_data_context_main;
262 make_funny_pointers(cinfo); /* Create the xbuffer[] lists */
263 main_ptr->whichptr = 0; /* Read first iMCU row into xbuffer[0] */
264 main_ptr->context_state = CTX_PREPARE_FOR_IMCU;
265 main_ptr->iMCU_row_ctr = 0;
266 } else {
267 /* Simple case with no context needed */
268 main_ptr->pub.process_data = process_data_simple_main;
269 }
270 main_ptr->buffer_full = FALSE; /* Mark buffer empty */
271 main_ptr->rowgroup_ctr = 0;
272 break;
273#ifdef QUANT_2PASS_SUPPORTED
274 case JBUF_CRANK_DEST:
275 /* For last pass of 2-pass quantization, just crank the postprocessor */
276 main_ptr->pub.process_data = process_data_crank_post;
277 break;
278#endif
279 default:
280 ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
281 break;
282 }
283}
284
285
286/*
287 * Process some data.
288 * This handles the simple case where no context is required.
289 */
290
291METHODDEF(void)
292process_data_simple_main(j_decompress_ptr cinfo, JSAMPARRAY output_buf,
293 JDIMENSION *out_row_ctr, JDIMENSION out_rows_avail)
294{
295 my_main_ptr main_ptr = (my_main_ptr)cinfo->main;
296 JDIMENSION rowgroups_avail;
297
298 /* Read input data if we haven't filled the main buffer yet */
299 if (!main_ptr->buffer_full) {
300 if (!(*cinfo->coef->decompress_data) (cinfo, main_ptr->buffer))
301 return; /* suspension forced, can do nothing more */
302 main_ptr->buffer_full = TRUE; /* OK, we have an iMCU row to work with */
303 }
304
305 /* There are always min_DCT_scaled_size row groups in an iMCU row. */
306 rowgroups_avail = (JDIMENSION)cinfo->_min_DCT_scaled_size;
307 /* Note: at the bottom of the image, we may pass extra garbage row groups
308 * to the postprocessor. The postprocessor has to check for bottom
309 * of image anyway (at row resolution), so no point in us doing it too.
310 */
311
312 /* Feed the postprocessor */
313 (*cinfo->post->post_process_data) (cinfo, main_ptr->buffer,
314 &main_ptr->rowgroup_ctr, rowgroups_avail,
315 output_buf, out_row_ctr, out_rows_avail);
316
317 /* Has postprocessor consumed all the data yet? If so, mark buffer empty */
318 if (main_ptr->rowgroup_ctr >= rowgroups_avail) {
319 main_ptr->buffer_full = FALSE;
320 main_ptr->rowgroup_ctr = 0;
321 }
322}
323
324
325/*
326 * Process some data.
327 * This handles the case where context rows must be provided.
328 */
329
330METHODDEF(void)
331process_data_context_main(j_decompress_ptr cinfo, JSAMPARRAY output_buf,
332 JDIMENSION *out_row_ctr, JDIMENSION out_rows_avail)
333{
334 my_main_ptr main_ptr = (my_main_ptr)cinfo->main;
335
336 /* Read input data if we haven't filled the main buffer yet */
337 if (!main_ptr->buffer_full) {
338 if (!(*cinfo->coef->decompress_data) (cinfo,
339 main_ptr->xbuffer[main_ptr->whichptr]))
340 return; /* suspension forced, can do nothing more */
341 main_ptr->buffer_full = TRUE; /* OK, we have an iMCU row to work with */
342 main_ptr->iMCU_row_ctr++; /* count rows received */
343 }
344
345 /* Postprocessor typically will not swallow all the input data it is handed
346 * in one call (due to filling the output buffer first). Must be prepared
347 * to exit and restart. This switch lets us keep track of how far we got.
348 * Note that each case falls through to the next on successful completion.
349 */
350 switch (main_ptr->context_state) {
351 case CTX_POSTPONED_ROW:
352 /* Call postprocessor using previously set pointers for postponed row */
353 (*cinfo->post->post_process_data) (cinfo,
354 main_ptr->xbuffer[main_ptr->whichptr],
355 &main_ptr->rowgroup_ctr,
356 main_ptr->rowgroups_avail, output_buf,
357 out_row_ctr, out_rows_avail);
358 if (main_ptr->rowgroup_ctr < main_ptr->rowgroups_avail)
359 return; /* Need to suspend */
360 main_ptr->context_state = CTX_PREPARE_FOR_IMCU;
361 if (*out_row_ctr >= out_rows_avail)
362 return; /* Postprocessor exactly filled output buf */
363 /*FALLTHROUGH*/
364 case CTX_PREPARE_FOR_IMCU:
365 /* Prepare to process first M-1 row groups of this iMCU row */
366 main_ptr->rowgroup_ctr = 0;
367 main_ptr->rowgroups_avail = (JDIMENSION)(cinfo->_min_DCT_scaled_size - 1);
368 /* Check for bottom of image: if so, tweak pointers to "duplicate"
369 * the last sample row, and adjust rowgroups_avail to ignore padding rows.
370 */
371 if (main_ptr->iMCU_row_ctr == cinfo->total_iMCU_rows)
372 set_bottom_pointers(cinfo);
373 main_ptr->context_state = CTX_PROCESS_IMCU;
374 /*FALLTHROUGH*/
375 case CTX_PROCESS_IMCU:
376 /* Call postprocessor using previously set pointers */
377 (*cinfo->post->post_process_data) (cinfo,
378 main_ptr->xbuffer[main_ptr->whichptr],
379 &main_ptr->rowgroup_ctr,
380 main_ptr->rowgroups_avail, output_buf,
381 out_row_ctr, out_rows_avail);
382 if (main_ptr->rowgroup_ctr < main_ptr->rowgroups_avail)
383 return; /* Need to suspend */
384 /* After the first iMCU, change wraparound pointers to normal state */
385 if (main_ptr->iMCU_row_ctr == 1)
386 set_wraparound_pointers(cinfo);
387 /* Prepare to load new iMCU row using other xbuffer list */
388 main_ptr->whichptr ^= 1; /* 0=>1 or 1=>0 */
389 main_ptr->buffer_full = FALSE;
390 /* Still need to process last row group of this iMCU row, */
391 /* which is saved at index M+1 of the other xbuffer */
392 main_ptr->rowgroup_ctr = (JDIMENSION)(cinfo->_min_DCT_scaled_size + 1);
393 main_ptr->rowgroups_avail = (JDIMENSION)(cinfo->_min_DCT_scaled_size + 2);
394 main_ptr->context_state = CTX_POSTPONED_ROW;
395 }
396}
397
398
399/*
400 * Process some data.
401 * Final pass of two-pass quantization: just call the postprocessor.
402 * Source data will be the postprocessor controller's internal buffer.
403 */
404
405#ifdef QUANT_2PASS_SUPPORTED
406
407METHODDEF(void)
408process_data_crank_post(j_decompress_ptr cinfo, JSAMPARRAY output_buf,
409 JDIMENSION *out_row_ctr, JDIMENSION out_rows_avail)
410{
411 (*cinfo->post->post_process_data) (cinfo, (JSAMPIMAGE)NULL,
412 (JDIMENSION *)NULL, (JDIMENSION)0,
413 output_buf, out_row_ctr, out_rows_avail);
414}
415
416#endif /* QUANT_2PASS_SUPPORTED */
417
418
419/*
420 * Initialize main buffer controller.
421 */
422
423GLOBAL(void)
424jinit_d_main_controller(j_decompress_ptr cinfo, boolean need_full_buffer)
425{
426 my_main_ptr main_ptr;
427 int ci, rgroup, ngroups;
428 jpeg_component_info *compptr;
429
430 main_ptr = (my_main_ptr)
431 (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE,
432 sizeof(my_main_controller));
433 cinfo->main = (struct jpeg_d_main_controller *)main_ptr;
434 main_ptr->pub.start_pass = start_pass_main;
435
436 if (need_full_buffer) /* shouldn't happen */
437 ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
438
439 /* Allocate the workspace.
440 * ngroups is the number of row groups we need.
441 */
442 if (cinfo->upsample->need_context_rows) {
443 if (cinfo->_min_DCT_scaled_size < 2) /* unsupported, see comments above */
444 ERREXIT(cinfo, JERR_NOTIMPL);
445 alloc_funny_pointers(cinfo); /* Alloc space for xbuffer[] lists */
446 ngroups = cinfo->_min_DCT_scaled_size + 2;
447 } else {
448 ngroups = cinfo->_min_DCT_scaled_size;
449 }
450
451 for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
452 ci++, compptr++) {
453 rgroup = (compptr->v_samp_factor * compptr->_DCT_scaled_size) /
454 cinfo->_min_DCT_scaled_size; /* height of a row group of component */
455 main_ptr->buffer[ci] = (*cinfo->mem->alloc_sarray)
456 ((j_common_ptr)cinfo, JPOOL_IMAGE,
457 compptr->width_in_blocks * compptr->_DCT_scaled_size,
458 (JDIMENSION)(rgroup * ngroups));
459 }
460}
461