1#if !defined(_FX_JPEG_TURBO_)
2/*
3 * jdmainct.c
4 *
5 * Copyright (C) 1994-1996, 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 main buffer controller for decompression.
10 * The main buffer lies between the JPEG decompressor proper and the
11 * post-processor; it holds downsampled data in the JPEG colorspace.
12 *
13 * Note that this code is bypassed in raw-data mode, since the application
14 * supplies the equivalent of the main buffer in that case.
15 */
16
17#define JPEG_INTERNALS
18#include "jinclude.h"
19#include "jpeglib.h"
20
21
22/*
23 * In the current system design, the main buffer need never be a full-image
24 * buffer; any full-height buffers will be found inside the coefficient or
25 * postprocessing controllers.  Nonetheless, the main controller is not
26 * trivial.  Its responsibility is to provide context rows for upsampling/
27 * rescaling, and doing this in an efficient fashion is a bit tricky.
28 *
29 * Postprocessor input data is counted in "row groups".  A row group
30 * is defined to be (v_samp_factor * DCT_scaled_size / min_DCT_scaled_size)
31 * sample rows of each component.  (We require DCT_scaled_size values to be
32 * chosen such that these numbers are integers.  In practice DCT_scaled_size
33 * values will likely be powers of two, so we actually have the stronger
34 * condition that DCT_scaled_size / min_DCT_scaled_size is an integer.)
35 * Upsampling will typically produce max_v_samp_factor pixel rows from each
36 * row group (times any additional scale factor that the upsampler is
37 * applying).
38 *
39 * The coefficient controller will deliver data to us one iMCU row at a time;
40 * each iMCU row contains v_samp_factor * DCT_scaled_size sample rows, or
41 * exactly min_DCT_scaled_size row groups.  (This amount of data corresponds
42 * to one row of MCUs when the image is fully interleaved.)  Note that the
43 * number of sample rows varies across components, but the number of row
44 * groups does not.  Some garbage sample rows may be included in the last iMCU
45 * row at the bottom of the image.
46 *
47 * Depending on the vertical scaling algorithm used, the upsampler may need
48 * access to the sample row(s) above and below its current input row group.
49 * The upsampler is required to set need_context_rows TRUE at global selection
50 * time if so.  When need_context_rows is FALSE, this controller can simply
51 * obtain one iMCU row at a time from the coefficient controller and dole it
52 * out as row groups to the postprocessor.
53 *
54 * When need_context_rows is TRUE, this controller guarantees that the buffer
55 * passed to postprocessing contains at least one row group's worth of samples
56 * above and below the row group(s) being processed.  Note that the context
57 * rows "above" the first passed row group appear at negative row offsets in
58 * the passed buffer.  At the top and bottom of the image, the required
59 * context rows are manufactured by duplicating the first or last real sample
60 * row; this avoids having special cases in the upsampling inner loops.
61 *
62 * The amount of context is fixed at one row group just because that's a
63 * convenient number for this controller to work with.  The existing
64 * upsamplers really only need one sample row of context.  An upsampler
65 * supporting arbitrary output rescaling might wish for more than one row
66 * group of context when shrinking the image; tough, we don't handle that.
67 * (This is justified by the assumption that downsizing will be handled mostly
68 * by adjusting the DCT_scaled_size values, so that the actual scale factor at
69 * the upsample step needn't be much less than one.)
70 *
71 * To provide the desired context, we have to retain the last two row groups
72 * of one iMCU row while reading in the next iMCU row.  (The last row group
73 * can't be processed until we have another row group for its below-context,
74 * and so we have to save the next-to-last group too for its above-context.)
75 * We could do this most simply by copying data around in our buffer, but
76 * that'd be very slow.  We can avoid copying any data by creating a rather
77 * strange pointer structure.  Here's how it works.  We allocate a workspace
78 * consisting of M+2 row groups (where M = min_DCT_scaled_size is the number
79 * of row groups per iMCU row).  We create two sets of redundant pointers to
80 * the workspace.  Labeling the physical row groups 0 to M+1, the synthesized
81 * pointer lists look like this:
82 *                   M+1                          M-1
83 * master pointer --> 0         master pointer --> 0
84 *                    1                            1
85 *                   ...                          ...
86 *                   M-3                          M-3
87 *                   M-2                           M
88 *                   M-1                          M+1
89 *                    M                           M-2
90 *                   M+1                          M-1
91 *                    0                            0
92 * We read alternate iMCU rows using each master pointer; thus the last two
93 * row groups of the previous iMCU row remain un-overwritten in the workspace.
94 * The pointer lists are set up so that the required context rows appear to
95 * be adjacent to the proper places when we pass the pointer lists to the
96 * upsampler.
97 *
98 * The above pictures describe the normal state of the pointer lists.
99 * At top and bottom of the image, we diddle the pointer lists to duplicate
100 * the first or last sample row as necessary (this is cheaper than copying
101 * sample rows around).
102 *
103 * This scheme breaks down if M < 2, ie, min_DCT_scaled_size is 1.  In that
104 * situation each iMCU row provides only one row group so the buffering logic
105 * must be different (eg, we must read two iMCU rows before we can emit the
106 * first row group).  For now, we simply do not support providing context
107 * rows when min_DCT_scaled_size is 1.  That combination seems unlikely to
108 * be worth providing --- if someone wants a 1/8th-size preview, they probably
109 * want it quick and dirty, so a context-free upsampler is sufficient.
110 */
111
112
113/* Private buffer controller object */
114
115typedef struct {
116  struct jpeg_d_main_controller pub; /* public fields */
117
118  /* Pointer to allocated workspace (M or M+2 row groups). */
119  JSAMPARRAY buffer[MAX_COMPONENTS];
120
121  boolean buffer_full;		/* Have we gotten an iMCU row from decoder? */
122  JDIMENSION rowgroup_ctr;	/* counts row groups output to postprocessor */
123
124  /* Remaining fields are only used in the context case. */
125
126  /* These are the master pointers to the funny-order pointer lists. */
127  JSAMPIMAGE xbuffer[2];	/* pointers to weird pointer lists */
128
129  int whichptr;			/* indicates which pointer set is now in use */
130  int context_state;		/* process_data state machine status */
131  JDIMENSION rowgroups_avail;	/* row groups available to postprocessor */
132  JDIMENSION iMCU_row_ctr;	/* counts iMCU rows to detect image top/bot */
133} my_main_controller;
134
135typedef my_main_controller * my_main_ptr;
136
137/* context_state values: */
138#define CTX_PREPARE_FOR_IMCU	0	/* need to prepare for MCU row */
139#define CTX_PROCESS_IMCU	1	/* feeding iMCU to postprocessor */
140#define CTX_POSTPONED_ROW	2	/* feeding postponed row group */
141
142
143/* Forward declarations */
144METHODDEF(void) process_data_simple_main
145	JPP((j_decompress_ptr cinfo, JSAMPARRAY output_buf,
146	     JDIMENSION *out_row_ctr, JDIMENSION out_rows_avail));
147METHODDEF(void) process_data_context_main
148	JPP((j_decompress_ptr cinfo, JSAMPARRAY output_buf,
149	     JDIMENSION *out_row_ctr, JDIMENSION out_rows_avail));
150#ifdef QUANT_2PASS_SUPPORTED
151METHODDEF(void) process_data_crank_post
152	JPP((j_decompress_ptr cinfo, JSAMPARRAY output_buf,
153	     JDIMENSION *out_row_ctr, JDIMENSION out_rows_avail));
154#endif
155
156
157LOCAL(void)
158alloc_funny_pointers (j_decompress_ptr cinfo)
159/* Allocate space for the funny pointer lists.
160 * This is done only once, not once per pass.
161 */
162{
163  my_main_ptr main = (my_main_ptr) cinfo->main;
164  int ci, rgroup;
165  int M = cinfo->min_DCT_scaled_size;
166  jpeg_component_info *compptr;
167  JSAMPARRAY xbuf;
168
169  /* Get top-level space for component array pointers.
170   * We alloc both arrays with one call to save a few cycles.
171   */
172  main->xbuffer[0] = (JSAMPIMAGE)
173    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
174				cinfo->num_components * 2 * SIZEOF(JSAMPARRAY));
175  main->xbuffer[1] = main->xbuffer[0] + cinfo->num_components;
176
177  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
178       ci++, compptr++) {
179    rgroup = (compptr->v_samp_factor * compptr->DCT_scaled_size) /
180      cinfo->min_DCT_scaled_size; /* height of a row group of component */
181    /* Get space for pointer lists --- M+4 row groups in each list.
182     * We alloc both pointer lists with one call to save a few cycles.
183     */
184    xbuf = (JSAMPARRAY)
185      (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
186				  2 * (rgroup * (M + 4)) * SIZEOF(JSAMPROW));
187    xbuf += rgroup;		/* want one row group at negative offsets */
188    main->xbuffer[0][ci] = xbuf;
189    xbuf += rgroup * (M + 4);
190    main->xbuffer[1][ci] = xbuf;
191  }
192}
193
194
195LOCAL(void)
196make_funny_pointers (j_decompress_ptr cinfo)
197/* Create the funny pointer lists discussed in the comments above.
198 * The actual workspace is already allocated (in main->buffer),
199 * and the space for the pointer lists is allocated too.
200 * This routine just fills in the curiously ordered lists.
201 * This will be repeated at the beginning of each pass.
202 */
203{
204  my_main_ptr main = (my_main_ptr) cinfo->main;
205  int ci, i, rgroup;
206  int M = cinfo->min_DCT_scaled_size;
207  jpeg_component_info *compptr;
208  JSAMPARRAY buf, xbuf0, xbuf1;
209
210  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
211       ci++, compptr++) {
212    rgroup = (compptr->v_samp_factor * compptr->DCT_scaled_size) /
213      cinfo->min_DCT_scaled_size; /* height of a row group of component */
214    xbuf0 = main->xbuffer[0][ci];
215    xbuf1 = main->xbuffer[1][ci];
216    /* First copy the workspace pointers as-is */
217    buf = main->buffer[ci];
218    for (i = 0; i < rgroup * (M + 2); i++) {
219      xbuf0[i] = xbuf1[i] = buf[i];
220    }
221    /* In the second list, put the last four row groups in swapped order */
222    for (i = 0; i < rgroup * 2; i++) {
223      xbuf1[rgroup*(M-2) + i] = buf[rgroup*M + i];
224      xbuf1[rgroup*M + i] = buf[rgroup*(M-2) + i];
225    }
226    /* The wraparound pointers at top and bottom will be filled later
227     * (see set_wraparound_pointers, below).  Initially we want the "above"
228     * pointers to duplicate the first actual data line.  This only needs
229     * to happen in xbuffer[0].
230     */
231    for (i = 0; i < rgroup; i++) {
232      xbuf0[i - rgroup] = xbuf0[0];
233    }
234  }
235}
236
237
238LOCAL(void)
239set_wraparound_pointers (j_decompress_ptr cinfo)
240/* Set up the "wraparound" pointers at top and bottom of the pointer lists.
241 * This changes the pointer list state from top-of-image to the normal state.
242 */
243{
244  my_main_ptr main = (my_main_ptr) cinfo->main;
245  int ci, i, rgroup;
246  int M = cinfo->min_DCT_scaled_size;
247  jpeg_component_info *compptr;
248  JSAMPARRAY xbuf0, xbuf1;
249
250  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
251       ci++, compptr++) {
252    rgroup = (compptr->v_samp_factor * compptr->DCT_scaled_size) /
253      cinfo->min_DCT_scaled_size; /* height of a row group of component */
254    xbuf0 = main->xbuffer[0][ci];
255    xbuf1 = main->xbuffer[1][ci];
256    for (i = 0; i < rgroup; i++) {
257      xbuf0[i - rgroup] = xbuf0[rgroup*(M+1) + i];
258      xbuf1[i - rgroup] = xbuf1[rgroup*(M+1) + i];
259      xbuf0[rgroup*(M+2) + i] = xbuf0[i];
260      xbuf1[rgroup*(M+2) + i] = xbuf1[i];
261    }
262  }
263}
264
265
266LOCAL(void)
267set_bottom_pointers (j_decompress_ptr cinfo)
268/* Change the pointer lists to duplicate the last sample row at the bottom
269 * of the image.  whichptr indicates which xbuffer holds the final iMCU row.
270 * Also sets rowgroups_avail to indicate number of nondummy row groups in row.
271 */
272{
273  my_main_ptr main = (my_main_ptr) cinfo->main;
274  int ci, i, rgroup, iMCUheight, rows_left;
275  jpeg_component_info *compptr;
276  JSAMPARRAY xbuf;
277
278  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
279       ci++, compptr++) {
280    /* Count sample rows in one iMCU row and in one row group */
281    iMCUheight = compptr->v_samp_factor * compptr->DCT_scaled_size;
282    rgroup = iMCUheight / cinfo->min_DCT_scaled_size;
283    /* Count nondummy sample rows remaining for this component */
284    rows_left = (int) (compptr->downsampled_height % (JDIMENSION) iMCUheight);
285    if (rows_left == 0) rows_left = iMCUheight;
286    /* Count nondummy row groups.  Should get same answer for each component,
287     * so we need only do it once.
288     */
289    if (ci == 0) {
290      main->rowgroups_avail = (JDIMENSION) ((rows_left-1) / rgroup + 1);
291    }
292    /* Duplicate the last real sample row rgroup*2 times; this pads out the
293     * last partial rowgroup and ensures at least one full rowgroup of context.
294     */
295    xbuf = main->xbuffer[main->whichptr][ci];
296    for (i = 0; i < rgroup * 2; i++) {
297      xbuf[rows_left + i] = xbuf[rows_left-1];
298    }
299  }
300}
301
302
303/*
304 * Initialize for a processing pass.
305 */
306
307METHODDEF(void)
308start_pass_main (j_decompress_ptr cinfo, J_BUF_MODE pass_mode)
309{
310  my_main_ptr main = (my_main_ptr) cinfo->main;
311
312  switch (pass_mode) {
313  case JBUF_PASS_THRU:
314    if (cinfo->upsample->need_context_rows) {
315      main->pub.process_data = process_data_context_main;
316      make_funny_pointers(cinfo); /* Create the xbuffer[] lists */
317      main->whichptr = 0;	/* Read first iMCU row into xbuffer[0] */
318      main->context_state = CTX_PREPARE_FOR_IMCU;
319      main->iMCU_row_ctr = 0;
320    } else {
321      /* Simple case with no context needed */
322      main->pub.process_data = process_data_simple_main;
323    }
324    main->buffer_full = FALSE;	/* Mark buffer empty */
325    main->rowgroup_ctr = 0;
326    break;
327#ifdef QUANT_2PASS_SUPPORTED
328  case JBUF_CRANK_DEST:
329    /* For last pass of 2-pass quantization, just crank the postprocessor */
330    main->pub.process_data = process_data_crank_post;
331    break;
332#endif
333  default:
334    ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
335    break;
336  }
337}
338
339
340/*
341 * Process some data.
342 * This handles the simple case where no context is required.
343 */
344
345METHODDEF(void)
346process_data_simple_main (j_decompress_ptr cinfo,
347			  JSAMPARRAY output_buf, JDIMENSION *out_row_ctr,
348			  JDIMENSION out_rows_avail)
349{
350  my_main_ptr main = (my_main_ptr) cinfo->main;
351  JDIMENSION rowgroups_avail;
352
353  /* Read input data if we haven't filled the main buffer yet */
354  if (! main->buffer_full) {
355    if (! (*cinfo->coef->decompress_data) (cinfo, main->buffer))
356      return;			/* suspension forced, can do nothing more */
357    main->buffer_full = TRUE;	/* OK, we have an iMCU row to work with */
358  }
359
360  /* There are always min_DCT_scaled_size row groups in an iMCU row. */
361  rowgroups_avail = (JDIMENSION) cinfo->min_DCT_scaled_size;
362  /* Note: at the bottom of the image, we may pass extra garbage row groups
363   * to the postprocessor.  The postprocessor has to check for bottom
364   * of image anyway (at row resolution), so no point in us doing it too.
365   */
366
367  /* Feed the postprocessor */
368  (*cinfo->post->post_process_data) (cinfo, main->buffer,
369				     &main->rowgroup_ctr, rowgroups_avail,
370				     output_buf, out_row_ctr, out_rows_avail);
371
372  /* Has postprocessor consumed all the data yet? If so, mark buffer empty */
373  if (main->rowgroup_ctr >= rowgroups_avail) {
374    main->buffer_full = FALSE;
375    main->rowgroup_ctr = 0;
376  }
377}
378
379
380/*
381 * Process some data.
382 * This handles the case where context rows must be provided.
383 */
384
385METHODDEF(void)
386process_data_context_main (j_decompress_ptr cinfo,
387			   JSAMPARRAY output_buf, JDIMENSION *out_row_ctr,
388			   JDIMENSION out_rows_avail)
389{
390  my_main_ptr main = (my_main_ptr) cinfo->main;
391
392  /* Read input data if we haven't filled the main buffer yet */
393  if (! main->buffer_full) {
394    if (! (*cinfo->coef->decompress_data) (cinfo,
395					   main->xbuffer[main->whichptr]))
396      return;			/* suspension forced, can do nothing more */
397    main->buffer_full = TRUE;	/* OK, we have an iMCU row to work with */
398    main->iMCU_row_ctr++;	/* count rows received */
399  }
400
401  /* Postprocessor typically will not swallow all the input data it is handed
402   * in one call (due to filling the output buffer first).  Must be prepared
403   * to exit and restart.  This switch lets us keep track of how far we got.
404   * Note that each case falls through to the next on successful completion.
405   */
406  switch (main->context_state) {
407  case CTX_POSTPONED_ROW:
408    /* Call postprocessor using previously set pointers for postponed row */
409    (*cinfo->post->post_process_data) (cinfo, main->xbuffer[main->whichptr],
410			&main->rowgroup_ctr, main->rowgroups_avail,
411			output_buf, out_row_ctr, out_rows_avail);
412    if (main->rowgroup_ctr < main->rowgroups_avail)
413      return;			/* Need to suspend */
414    main->context_state = CTX_PREPARE_FOR_IMCU;
415    if (*out_row_ctr >= out_rows_avail)
416      return;			/* Postprocessor exactly filled output buf */
417    /*FALLTHROUGH*/
418  case CTX_PREPARE_FOR_IMCU:
419    /* Prepare to process first M-1 row groups of this iMCU row */
420    main->rowgroup_ctr = 0;
421    main->rowgroups_avail = (JDIMENSION) (cinfo->min_DCT_scaled_size - 1);
422    /* Check for bottom of image: if so, tweak pointers to "duplicate"
423     * the last sample row, and adjust rowgroups_avail to ignore padding rows.
424     */
425    if (main->iMCU_row_ctr == cinfo->total_iMCU_rows)
426      set_bottom_pointers(cinfo);
427    main->context_state = CTX_PROCESS_IMCU;
428    /*FALLTHROUGH*/
429  case CTX_PROCESS_IMCU:
430    /* Call postprocessor using previously set pointers */
431    (*cinfo->post->post_process_data) (cinfo, main->xbuffer[main->whichptr],
432			&main->rowgroup_ctr, main->rowgroups_avail,
433			output_buf, out_row_ctr, out_rows_avail);
434    if (main->rowgroup_ctr < main->rowgroups_avail)
435      return;			/* Need to suspend */
436    /* After the first iMCU, change wraparound pointers to normal state */
437    if (main->iMCU_row_ctr == 1)
438      set_wraparound_pointers(cinfo);
439    /* Prepare to load new iMCU row using other xbuffer list */
440    main->whichptr ^= 1;	/* 0=>1 or 1=>0 */
441    main->buffer_full = FALSE;
442    /* Still need to process last row group of this iMCU row, */
443    /* which is saved at index M+1 of the other xbuffer */
444    main->rowgroup_ctr = (JDIMENSION) (cinfo->min_DCT_scaled_size + 1);
445    main->rowgroups_avail = (JDIMENSION) (cinfo->min_DCT_scaled_size + 2);
446    main->context_state = CTX_POSTPONED_ROW;
447  }
448}
449
450
451/*
452 * Process some data.
453 * Final pass of two-pass quantization: just call the postprocessor.
454 * Source data will be the postprocessor controller's internal buffer.
455 */
456
457#ifdef QUANT_2PASS_SUPPORTED
458
459METHODDEF(void)
460process_data_crank_post (j_decompress_ptr cinfo,
461			 JSAMPARRAY output_buf, JDIMENSION *out_row_ctr,
462			 JDIMENSION out_rows_avail)
463{
464  (*cinfo->post->post_process_data) (cinfo, (JSAMPIMAGE) NULL,
465				     (JDIMENSION *) NULL, (JDIMENSION) 0,
466				     output_buf, out_row_ctr, out_rows_avail);
467}
468
469#endif /* QUANT_2PASS_SUPPORTED */
470
471
472/*
473 * Initialize main buffer controller.
474 */
475
476GLOBAL(void)
477jinit_d_main_controller (j_decompress_ptr cinfo, boolean need_full_buffer)
478{
479  my_main_ptr main;
480  int ci, rgroup, ngroups;
481  jpeg_component_info *compptr;
482
483  main = (my_main_ptr)
484    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
485				SIZEOF(my_main_controller));
486  cinfo->main = (struct jpeg_d_main_controller *) main;
487  main->pub.start_pass = start_pass_main;
488
489  if (need_full_buffer)		/* shouldn't happen */
490    ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
491
492  /* Allocate the workspace.
493   * ngroups is the number of row groups we need.
494   */
495  if (cinfo->upsample->need_context_rows) {
496    if (cinfo->min_DCT_scaled_size < 2) /* unsupported, see comments above */
497      ERREXIT(cinfo, JERR_NOTIMPL);
498    alloc_funny_pointers(cinfo); /* Alloc space for xbuffer[] lists */
499    ngroups = cinfo->min_DCT_scaled_size + 2;
500  } else {
501    ngroups = cinfo->min_DCT_scaled_size;
502  }
503
504  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
505       ci++, compptr++) {
506    rgroup = (compptr->v_samp_factor * compptr->DCT_scaled_size) /
507      cinfo->min_DCT_scaled_size; /* height of a row group of component */
508    main->buffer[ci] = (*cinfo->mem->alloc_sarray)
509			((j_common_ptr) cinfo, JPOOL_IMAGE,
510			 compptr->width_in_blocks * compptr->DCT_scaled_size,
511			 (JDIMENSION) (rgroup * ngroups));
512  }
513}
514
515#endif //_FX_JPEG_TURBO_
516