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
116 (j_decompress_ptr cinfo, JSAMPARRAY output_buf,
117 JDIMENSION *out_row_ctr, JDIMENSION out_rows_avail);
118METHODDEF(void) process_data_context_main
119 (j_decompress_ptr cinfo, JSAMPARRAY output_buf,
120 JDIMENSION *out_row_ctr, JDIMENSION out_rows_avail);
121#ifdef QUANT_2PASS_SUPPORTED
122METHODDEF(void) process_data_crank_post
123 (j_decompress_ptr cinfo, JSAMPARRAY output_buf,
124 JDIMENSION *out_row_ctr, JDIMENSION out_rows_avail);
125#endif
126
127
128LOCAL(void)
129alloc_funny_pointers (j_decompress_ptr cinfo)
130/* Allocate space for the funny pointer lists.
131 * This is done only once, not once per pass.
132 */
133{
134 my_main_ptr main_ptr = (my_main_ptr) cinfo->main;
135 int ci, rgroup;
136 int M = cinfo->_min_DCT_scaled_size;
137 jpeg_component_info *compptr;
138 JSAMPARRAY xbuf;
139
140 /* Get top-level space for component array pointers.
141 * We alloc both arrays with one call to save a few cycles.
142 */
143 main_ptr->xbuffer[0] = (JSAMPIMAGE)
144 (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
145 cinfo->num_components * 2 * sizeof(JSAMPARRAY));
146 main_ptr->xbuffer[1] = main_ptr->xbuffer[0] + cinfo->num_components;
147
148 for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
149 ci++, compptr++) {
150 rgroup = (compptr->v_samp_factor * compptr->_DCT_scaled_size) /
151 cinfo->_min_DCT_scaled_size; /* height of a row group of component */
152 /* Get space for pointer lists --- M+4 row groups in each list.
153 * We alloc both pointer lists with one call to save a few cycles.
154 */
155 xbuf = (JSAMPARRAY)
156 (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
157 2 * (rgroup * (M + 4)) * sizeof(JSAMPROW));
158 xbuf += rgroup; /* want one row group at negative offsets */
159 main_ptr->xbuffer[0][ci] = xbuf;
160 xbuf += rgroup * (M + 4);
161 main_ptr->xbuffer[1][ci] = xbuf;
162 }
163}
164
165
166LOCAL(void)
167make_funny_pointers (j_decompress_ptr cinfo)
168/* Create the funny pointer lists discussed in the comments above.
169 * The actual workspace is already allocated (in main_ptr->buffer),
170 * and the space for the pointer lists is allocated too.
171 * This routine just fills in the curiously ordered lists.
172 * This will be repeated at the beginning of each pass.
173 */
174{
175 my_main_ptr main_ptr = (my_main_ptr) cinfo->main;
176 int ci, i, rgroup;
177 int M = cinfo->_min_DCT_scaled_size;
178 jpeg_component_info *compptr;
179 JSAMPARRAY buf, xbuf0, xbuf1;
180
181 for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
182 ci++, compptr++) {
183 rgroup = (compptr->v_samp_factor * compptr->_DCT_scaled_size) /
184 cinfo->_min_DCT_scaled_size; /* height of a row group of component */
185 xbuf0 = main_ptr->xbuffer[0][ci];
186 xbuf1 = main_ptr->xbuffer[1][ci];
187 /* First copy the workspace pointers as-is */
188 buf = main_ptr->buffer[ci];
189 for (i = 0; i < rgroup * (M + 2); i++) {
190 xbuf0[i] = xbuf1[i] = buf[i];
191 }
192 /* In the second list, put the last four row groups in swapped order */
193 for (i = 0; i < rgroup * 2; i++) {
194 xbuf1[rgroup*(M-2) + i] = buf[rgroup*M + i];
195 xbuf1[rgroup*M + i] = buf[rgroup*(M-2) + i];
196 }
197 /* The wraparound pointers at top and bottom will be filled later
198 * (see set_wraparound_pointers, below). Initially we want the "above"
199 * pointers to duplicate the first actual data line. This only needs
200 * to happen in xbuffer[0].
201 */
202 for (i = 0; i < rgroup; i++) {
203 xbuf0[i - rgroup] = xbuf0[0];
204 }
205 }
206}
207
208
209LOCAL(void)
210set_bottom_pointers (j_decompress_ptr cinfo)
211/* Change the pointer lists to duplicate the last sample row at the bottom
212 * of the image. whichptr indicates which xbuffer holds the final iMCU row.
213 * Also sets rowgroups_avail to indicate number of nondummy row groups in row.
214 */
215{
216 my_main_ptr main_ptr = (my_main_ptr) cinfo->main;
217 int ci, i, rgroup, iMCUheight, rows_left;
218 jpeg_component_info *compptr;
219 JSAMPARRAY xbuf;
220
221 for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
222 ci++, compptr++) {
223 /* Count sample rows in one iMCU row and in one row group */
224 iMCUheight = compptr->v_samp_factor * compptr->_DCT_scaled_size;
225 rgroup = iMCUheight / cinfo->_min_DCT_scaled_size;
226 /* Count nondummy sample rows remaining for this component */
227 rows_left = (int) (compptr->downsampled_height % (JDIMENSION) iMCUheight);
228 if (rows_left == 0) rows_left = iMCUheight;
229 /* Count nondummy row groups. Should get same answer for each component,
230 * so we need only do it once.
231 */
232 if (ci == 0) {
233 main_ptr->rowgroups_avail = (JDIMENSION) ((rows_left-1) / rgroup + 1);
234 }
235 /* Duplicate the last real sample row rgroup*2 times; this pads out the
236 * last partial rowgroup and ensures at least one full rowgroup of context.
237 */
238 xbuf = main_ptr->xbuffer[main_ptr->whichptr][ci];
239 for (i = 0; i < rgroup * 2; i++) {
240 xbuf[rows_left + i] = xbuf[rows_left-1];
241 }
242 }
243}
244
245
246/*
247 * Initialize for a processing pass.
248 */
249
250METHODDEF(void)
251start_pass_main (j_decompress_ptr cinfo, J_BUF_MODE pass_mode)
252{
253 my_main_ptr main_ptr = (my_main_ptr) cinfo->main;
254
255 switch (pass_mode) {
256 case JBUF_PASS_THRU:
257 if (cinfo->upsample->need_context_rows) {
258 main_ptr->pub.process_data = process_data_context_main;
259 make_funny_pointers(cinfo); /* Create the xbuffer[] lists */
260 main_ptr->whichptr = 0; /* Read first iMCU row into xbuffer[0] */
261 main_ptr->context_state = CTX_PREPARE_FOR_IMCU;
262 main_ptr->iMCU_row_ctr = 0;
263 } else {
264 /* Simple case with no context needed */
265 main_ptr->pub.process_data = process_data_simple_main;
266 }
267 main_ptr->buffer_full = FALSE; /* Mark buffer empty */
268 main_ptr->rowgroup_ctr = 0;
269 break;
270#ifdef QUANT_2PASS_SUPPORTED
271 case JBUF_CRANK_DEST:
272 /* For last pass of 2-pass quantization, just crank the postprocessor */
273 main_ptr->pub.process_data = process_data_crank_post;
274 break;
275#endif
276 default:
277 ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
278 break;
279 }
280}
281
282
283/*
284 * Process some data.
285 * This handles the simple case where no context is required.
286 */
287
288METHODDEF(void)
289process_data_simple_main (j_decompress_ptr cinfo,
290 JSAMPARRAY output_buf, JDIMENSION *out_row_ctr,
291 JDIMENSION out_rows_avail)
292{
293 my_main_ptr main_ptr = (my_main_ptr) cinfo->main;
294 JDIMENSION rowgroups_avail;
295
296 /* Read input data if we haven't filled the main buffer yet */
297 if (! main_ptr->buffer_full) {
298 if (! (*cinfo->coef->decompress_data) (cinfo, main_ptr->buffer))
299 return; /* suspension forced, can do nothing more */
300 main_ptr->buffer_full = TRUE; /* OK, we have an iMCU row to work with */
301 }
302
303 /* There are always min_DCT_scaled_size row groups in an iMCU row. */
304 rowgroups_avail = (JDIMENSION) cinfo->_min_DCT_scaled_size;
305 /* Note: at the bottom of the image, we may pass extra garbage row groups
306 * to the postprocessor. The postprocessor has to check for bottom
307 * of image anyway (at row resolution), so no point in us doing it too.
308 */
309
310 /* Feed the postprocessor */
311 (*cinfo->post->post_process_data) (cinfo, main_ptr->buffer,
312 &main_ptr->rowgroup_ctr, rowgroups_avail,
313 output_buf, out_row_ctr, out_rows_avail);
314
315 /* Has postprocessor consumed all the data yet? If so, mark buffer empty */
316 if (main_ptr->rowgroup_ctr >= rowgroups_avail) {
317 main_ptr->buffer_full = FALSE;
318 main_ptr->rowgroup_ctr = 0;
319 }
320}
321
322
323/*
324 * Process some data.
325 * This handles the case where context rows must be provided.
326 */
327
328METHODDEF(void)
329process_data_context_main (j_decompress_ptr cinfo,
330 JSAMPARRAY output_buf, JDIMENSION *out_row_ctr,
331 JDIMENSION out_rows_avail)
332{
333 my_main_ptr main_ptr = (my_main_ptr) cinfo->main;
334
335 /* Read input data if we haven't filled the main buffer yet */
336 if (! main_ptr->buffer_full) {
337 if (! (*cinfo->coef->decompress_data) (cinfo,
338 main_ptr->xbuffer[main_ptr->whichptr]))
339 return; /* suspension forced, can do nothing more */
340 main_ptr->buffer_full = TRUE; /* OK, we have an iMCU row to work with */
341 main_ptr->iMCU_row_ctr++; /* count rows received */
342 }
343
344 /* Postprocessor typically will not swallow all the input data it is handed
345 * in one call (due to filling the output buffer first). Must be prepared
346 * to exit and restart. This switch lets us keep track of how far we got.
347 * Note that each case falls through to the next on successful completion.
348 */
349 switch (main_ptr->context_state) {
350 case CTX_POSTPONED_ROW:
351 /* Call postprocessor using previously set pointers for postponed row */
352 (*cinfo->post->post_process_data) (cinfo, main_ptr->xbuffer[main_ptr->whichptr],
353 &main_ptr->rowgroup_ctr, main_ptr->rowgroups_avail,
354 output_buf, out_row_ctr, out_rows_avail);
355 if (main_ptr->rowgroup_ctr < main_ptr->rowgroups_avail)
356 return; /* Need to suspend */
357 main_ptr->context_state = CTX_PREPARE_FOR_IMCU;
358 if (*out_row_ctr >= out_rows_avail)
359 return; /* Postprocessor exactly filled output buf */
360 /*FALLTHROUGH*/
361 case CTX_PREPARE_FOR_IMCU:
362 /* Prepare to process first M-1 row groups of this iMCU row */
363 main_ptr->rowgroup_ctr = 0;
364 main_ptr->rowgroups_avail = (JDIMENSION) (cinfo->_min_DCT_scaled_size - 1);
365 /* Check for bottom of image: if so, tweak pointers to "duplicate"
366 * the last sample row, and adjust rowgroups_avail to ignore padding rows.
367 */
368 if (main_ptr->iMCU_row_ctr == cinfo->total_iMCU_rows)
369 set_bottom_pointers(cinfo);
370 main_ptr->context_state = CTX_PROCESS_IMCU;
371 /*FALLTHROUGH*/
372 case CTX_PROCESS_IMCU:
373 /* Call postprocessor using previously set pointers */
374 (*cinfo->post->post_process_data) (cinfo, main_ptr->xbuffer[main_ptr->whichptr],
375 &main_ptr->rowgroup_ctr, main_ptr->rowgroups_avail,
376 output_buf, out_row_ctr, out_rows_avail);
377 if (main_ptr->rowgroup_ctr < main_ptr->rowgroups_avail)
378 return; /* Need to suspend */
379 /* After the first iMCU, change wraparound pointers to normal state */
380 if (main_ptr->iMCU_row_ctr == 1)
381 set_wraparound_pointers(cinfo);
382 /* Prepare to load new iMCU row using other xbuffer list */
383 main_ptr->whichptr ^= 1; /* 0=>1 or 1=>0 */
384 main_ptr->buffer_full = FALSE;
385 /* Still need to process last row group of this iMCU row, */
386 /* which is saved at index M+1 of the other xbuffer */
387 main_ptr->rowgroup_ctr = (JDIMENSION) (cinfo->_min_DCT_scaled_size + 1);
388 main_ptr->rowgroups_avail = (JDIMENSION) (cinfo->_min_DCT_scaled_size + 2);
389 main_ptr->context_state = CTX_POSTPONED_ROW;
390 }
391}
392
393
394/*
395 * Process some data.
396 * Final pass of two-pass quantization: just call the postprocessor.
397 * Source data will be the postprocessor controller's internal buffer.
398 */
399
400#ifdef QUANT_2PASS_SUPPORTED
401
402METHODDEF(void)
403process_data_crank_post (j_decompress_ptr cinfo,
404 JSAMPARRAY output_buf, JDIMENSION *out_row_ctr,
405 JDIMENSION out_rows_avail)
406{
407 (*cinfo->post->post_process_data) (cinfo, (JSAMPIMAGE) NULL,
408 (JDIMENSION *) NULL, (JDIMENSION) 0,
409 output_buf, out_row_ctr, out_rows_avail);
410}
411
412#endif /* QUANT_2PASS_SUPPORTED */
413
414
415/*
416 * Initialize main buffer controller.
417 */
418
419GLOBAL(void)
420jinit_d_main_controller (j_decompress_ptr cinfo, boolean need_full_buffer)
421{
422 my_main_ptr main_ptr;
423 int ci, rgroup, ngroups;
424 jpeg_component_info *compptr;
425
426 main_ptr = (my_main_ptr)
427 (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
428 sizeof(my_main_controller));
429 cinfo->main = (struct jpeg_d_main_controller *) main_ptr;
430 main_ptr->pub.start_pass = start_pass_main;
431
432 if (need_full_buffer) /* shouldn't happen */
433 ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
434
435 /* Allocate the workspace.
436 * ngroups is the number of row groups we need.
437 */
438 if (cinfo->upsample->need_context_rows) {
439 if (cinfo->_min_DCT_scaled_size < 2) /* unsupported, see comments above */
440 ERREXIT(cinfo, JERR_NOTIMPL);
441 alloc_funny_pointers(cinfo); /* Alloc space for xbuffer[] lists */
442 ngroups = cinfo->_min_DCT_scaled_size + 2;
443 } else {
444 ngroups = cinfo->_min_DCT_scaled_size;
445 }
446
447 for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
448 ci++, compptr++) {
449 rgroup = (compptr->v_samp_factor * compptr->_DCT_scaled_size) /
450 cinfo->_min_DCT_scaled_size; /* height of a row group of component */
451 main_ptr->buffer[ci] = (*cinfo->mem->alloc_sarray)
452 ((j_common_ptr) cinfo, JPOOL_IMAGE,
453 compptr->width_in_blocks * compptr->_DCT_scaled_size,
454 (JDIMENSION) (rgroup * ngroups));
455 }
456}
457