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
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.
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.
21 #include "jconfigint.h"
25 * In the current system design, the main buffer need never be a full-image
26 * buffer; any full-height buffers will be found inside the coefficient or
27 * postprocessing controllers. Nonetheless, the main controller is not
28 * trivial. Its responsibility is to provide context rows for upsampling/
29 * rescaling, and doing this in an efficient fashion is a bit tricky.
31 * Postprocessor input data is counted in "row groups". A row group
32 * is defined to be (v_samp_factor * DCT_scaled_size / min_DCT_scaled_size)
33 * sample rows of each component. (We require DCT_scaled_size values to be
34 * chosen such that these numbers are integers. In practice DCT_scaled_size
35 * values will likely be powers of two, so we actually have the stronger
36 * condition that DCT_scaled_size / min_DCT_scaled_size is an integer.)
37 * Upsampling will typically produce max_v_samp_factor pixel rows from each
38 * row group (times any additional scale factor that the upsampler is
41 * The coefficient controller will deliver data to us one iMCU row at a time;
42 * each iMCU row contains v_samp_factor * DCT_scaled_size sample rows, or
43 * exactly min_DCT_scaled_size row groups. (This amount of data corresponds
44 * to one row of MCUs when the image is fully interleaved.) Note that the
45 * number of sample rows varies across components, but the number of row
46 * groups does not. Some garbage sample rows may be included in the last iMCU
47 * row at the bottom of the image.
49 * Depending on the vertical scaling algorithm used, the upsampler may need
50 * access to the sample row(s) above and below its current input row group.
51 * The upsampler is required to set need_context_rows TRUE at global selection
52 * time if so. When need_context_rows is FALSE, this controller can simply
53 * obtain one iMCU row at a time from the coefficient controller and dole it
54 * out as row groups to the postprocessor.
56 * When need_context_rows is TRUE, this controller guarantees that the buffer
57 * passed to postprocessing contains at least one row group's worth of samples
58 * above and below the row group(s) being processed. Note that the context
59 * rows "above" the first passed row group appear at negative row offsets in
60 * the passed buffer. At the top and bottom of the image, the required
61 * context rows are manufactured by duplicating the first or last real sample
62 * row; this avoids having special cases in the upsampling inner loops.
64 * The amount of context is fixed at one row group just because that's a
65 * convenient number for this controller to work with. The existing
66 * upsamplers really only need one sample row of context. An upsampler
67 * supporting arbitrary output rescaling might wish for more than one row
68 * group of context when shrinking the image; tough, we don't handle that.
69 * (This is justified by the assumption that downsizing will be handled mostly
70 * by adjusting the DCT_scaled_size values, so that the actual scale factor at
71 * the upsample step needn't be much less than one.)
73 * To provide the desired context, we have to retain the last two row groups
74 * of one iMCU row while reading in the next iMCU row. (The last row group
75 * can't be processed until we have another row group for its below-context,
76 * and so we have to save the next-to-last group too for its above-context.)
77 * We could do this most simply by copying data around in our buffer, but
78 * that'd be very slow. We can avoid copying any data by creating a rather
79 * strange pointer structure. Here's how it works. We allocate a workspace
80 * consisting of M+2 row groups (where M = min_DCT_scaled_size is the number
81 * of row groups per iMCU row). We create two sets of redundant pointers to
82 * the workspace. Labeling the physical row groups 0 to M+1, the synthesized
83 * pointer lists look like this:
85 * master pointer --> 0 master pointer --> 0
94 * We read alternate iMCU rows using each master pointer; thus the last two
95 * row groups of the previous iMCU row remain un-overwritten in the workspace.
96 * The pointer lists are set up so that the required context rows appear to
97 * be adjacent to the proper places when we pass the pointer lists to the
100 * The above pictures describe the normal state of the pointer lists.
101 * At top and bottom of the image, we diddle the pointer lists to duplicate
102 * the first or last sample row as necessary (this is cheaper than copying
103 * sample rows around).
105 * This scheme breaks down if M < 2, ie, min_DCT_scaled_size is 1. In that
106 * situation each iMCU row provides only one row group so the buffering logic
107 * must be different (eg, we must read two iMCU rows before we can emit the
108 * first row group). For now, we simply do not support providing context
109 * rows when min_DCT_scaled_size is 1. That combination seems unlikely to
110 * be worth providing --- if someone wants a 1/8th-size preview, they probably
111 * want it quick and dirty, so a context-free upsampler is sufficient.
115 /* Forward declarations */
116 METHODDEF(void) process_data_simple_main(j_decompress_ptr cinfo
,
117 JSAMPARRAY output_buf
,
118 JDIMENSION
*out_row_ctr
,
119 JDIMENSION out_rows_avail
);
120 METHODDEF(void) process_data_context_main(j_decompress_ptr cinfo
,
121 JSAMPARRAY output_buf
,
122 JDIMENSION
*out_row_ctr
,
123 JDIMENSION out_rows_avail
);
124 #ifdef QUANT_2PASS_SUPPORTED
125 METHODDEF(void) process_data_crank_post(j_decompress_ptr cinfo
,
126 JSAMPARRAY output_buf
,
127 JDIMENSION
*out_row_ctr
,
128 JDIMENSION out_rows_avail
);
133 alloc_funny_pointers(j_decompress_ptr cinfo
)
134 /* Allocate space for the funny pointer lists.
135 * This is done only once, not once per pass.
138 my_main_ptr main_ptr
= (my_main_ptr
)cinfo
->main
;
140 int M
= cinfo
->_min_DCT_scaled_size
;
141 jpeg_component_info
*compptr
;
144 /* Get top-level space for component array pointers.
145 * We alloc both arrays with one call to save a few cycles.
147 main_ptr
->xbuffer
[0] = (JSAMPIMAGE
)
148 (*cinfo
->mem
->alloc_small
) ((j_common_ptr
)cinfo
, JPOOL_IMAGE
,
149 cinfo
->num_components
* 2 * sizeof(JSAMPARRAY
));
150 main_ptr
->xbuffer
[1] = main_ptr
->xbuffer
[0] + cinfo
->num_components
;
152 for (ci
= 0, compptr
= cinfo
->comp_info
; ci
< cinfo
->num_components
;
154 rgroup
= (compptr
->v_samp_factor
* compptr
->_DCT_scaled_size
) /
155 cinfo
->_min_DCT_scaled_size
; /* height of a row group of component */
156 /* Get space for pointer lists --- M+4 row groups in each list.
157 * We alloc both pointer lists with one call to save a few cycles.
160 (*cinfo
->mem
->alloc_small
) ((j_common_ptr
)cinfo
, JPOOL_IMAGE
,
161 2 * (rgroup
* (M
+ 4)) * sizeof(JSAMPROW
));
162 xbuf
+= rgroup
; /* want one row group at negative offsets */
163 main_ptr
->xbuffer
[0][ci
] = xbuf
;
164 xbuf
+= rgroup
* (M
+ 4);
165 main_ptr
->xbuffer
[1][ci
] = xbuf
;
171 make_funny_pointers(j_decompress_ptr cinfo
)
172 /* Create the funny pointer lists discussed in the comments above.
173 * The actual workspace is already allocated (in main_ptr->buffer),
174 * and the space for the pointer lists is allocated too.
175 * This routine just fills in the curiously ordered lists.
176 * This will be repeated at the beginning of each pass.
179 my_main_ptr main_ptr
= (my_main_ptr
)cinfo
->main
;
181 int M
= cinfo
->_min_DCT_scaled_size
;
182 jpeg_component_info
*compptr
;
183 JSAMPARRAY buf
, xbuf0
, xbuf1
;
185 for (ci
= 0, compptr
= cinfo
->comp_info
; ci
< cinfo
->num_components
;
187 rgroup
= (compptr
->v_samp_factor
* compptr
->_DCT_scaled_size
) /
188 cinfo
->_min_DCT_scaled_size
; /* height of a row group of component */
189 xbuf0
= main_ptr
->xbuffer
[0][ci
];
190 xbuf1
= main_ptr
->xbuffer
[1][ci
];
191 /* First copy the workspace pointers as-is */
192 buf
= main_ptr
->buffer
[ci
];
193 for (i
= 0; i
< rgroup
* (M
+ 2); i
++) {
194 xbuf0
[i
] = xbuf1
[i
] = buf
[i
];
196 /* In the second list, put the last four row groups in swapped order */
197 for (i
= 0; i
< rgroup
* 2; i
++) {
198 xbuf1
[rgroup
* (M
- 2) + i
] = buf
[rgroup
* M
+ i
];
199 xbuf1
[rgroup
* M
+ i
] = buf
[rgroup
* (M
- 2) + i
];
201 /* The wraparound pointers at top and bottom will be filled later
202 * (see set_wraparound_pointers, below). Initially we want the "above"
203 * pointers to duplicate the first actual data line. This only needs
204 * to happen in xbuffer[0].
206 for (i
= 0; i
< rgroup
; i
++) {
207 xbuf0
[i
- rgroup
] = xbuf0
[0];
214 set_bottom_pointers(j_decompress_ptr cinfo
)
215 /* Change the pointer lists to duplicate the last sample row at the bottom
216 * of the image. whichptr indicates which xbuffer holds the final iMCU row.
217 * Also sets rowgroups_avail to indicate number of nondummy row groups in row.
220 my_main_ptr main_ptr
= (my_main_ptr
)cinfo
->main
;
221 int ci
, i
, rgroup
, iMCUheight
, rows_left
;
222 jpeg_component_info
*compptr
;
225 for (ci
= 0, compptr
= cinfo
->comp_info
; ci
< cinfo
->num_components
;
227 /* Count sample rows in one iMCU row and in one row group */
228 iMCUheight
= compptr
->v_samp_factor
* compptr
->_DCT_scaled_size
;
229 rgroup
= iMCUheight
/ cinfo
->_min_DCT_scaled_size
;
230 /* Count nondummy sample rows remaining for this component */
231 rows_left
= (int)(compptr
->downsampled_height
% (JDIMENSION
)iMCUheight
);
232 if (rows_left
== 0) rows_left
= iMCUheight
;
233 /* Count nondummy row groups. Should get same answer for each component,
234 * so we need only do it once.
237 main_ptr
->rowgroups_avail
= (JDIMENSION
)((rows_left
- 1) / rgroup
+ 1);
239 /* Duplicate the last real sample row rgroup*2 times; this pads out the
240 * last partial rowgroup and ensures at least one full rowgroup of context.
242 xbuf
= main_ptr
->xbuffer
[main_ptr
->whichptr
][ci
];
243 for (i
= 0; i
< rgroup
* 2; i
++) {
244 xbuf
[rows_left
+ i
] = xbuf
[rows_left
- 1];
251 * Initialize for a processing pass.
255 start_pass_main(j_decompress_ptr cinfo
, J_BUF_MODE pass_mode
)
257 my_main_ptr main_ptr
= (my_main_ptr
)cinfo
->main
;
261 if (cinfo
->upsample
->need_context_rows
) {
262 main_ptr
->pub
.process_data
= process_data_context_main
;
263 make_funny_pointers(cinfo
); /* Create the xbuffer[] lists */
264 main_ptr
->whichptr
= 0; /* Read first iMCU row into xbuffer[0] */
265 main_ptr
->context_state
= CTX_PREPARE_FOR_IMCU
;
266 main_ptr
->iMCU_row_ctr
= 0;
268 /* Simple case with no context needed */
269 main_ptr
->pub
.process_data
= process_data_simple_main
;
271 main_ptr
->buffer_full
= FALSE
; /* Mark buffer empty */
272 main_ptr
->rowgroup_ctr
= 0;
274 #ifdef QUANT_2PASS_SUPPORTED
275 case JBUF_CRANK_DEST
:
276 /* For last pass of 2-pass quantization, just crank the postprocessor */
277 main_ptr
->pub
.process_data
= process_data_crank_post
;
281 ERREXIT(cinfo
, JERR_BAD_BUFFER_MODE
);
289 * This handles the simple case where no context is required.
293 process_data_simple_main(j_decompress_ptr cinfo
, JSAMPARRAY output_buf
,
294 JDIMENSION
*out_row_ctr
, JDIMENSION out_rows_avail
)
296 my_main_ptr main_ptr
= (my_main_ptr
)cinfo
->main
;
297 JDIMENSION rowgroups_avail
;
299 /* Read input data if we haven't filled the main buffer yet */
300 if (!main_ptr
->buffer_full
) {
301 if (!(*cinfo
->coef
->decompress_data
) (cinfo
, main_ptr
->buffer
))
302 return; /* suspension forced, can do nothing more */
303 main_ptr
->buffer_full
= TRUE
; /* OK, we have an iMCU row to work with */
306 /* There are always min_DCT_scaled_size row groups in an iMCU row. */
307 rowgroups_avail
= (JDIMENSION
)cinfo
->_min_DCT_scaled_size
;
308 /* Note: at the bottom of the image, we may pass extra garbage row groups
309 * to the postprocessor. The postprocessor has to check for bottom
310 * of image anyway (at row resolution), so no point in us doing it too.
313 /* Feed the postprocessor */
314 (*cinfo
->post
->post_process_data
) (cinfo
, main_ptr
->buffer
,
315 &main_ptr
->rowgroup_ctr
, rowgroups_avail
,
316 output_buf
, out_row_ctr
, out_rows_avail
);
318 /* Has postprocessor consumed all the data yet? If so, mark buffer empty */
319 if (main_ptr
->rowgroup_ctr
>= rowgroups_avail
) {
320 main_ptr
->buffer_full
= FALSE
;
321 main_ptr
->rowgroup_ctr
= 0;
328 * This handles the case where context rows must be provided.
332 process_data_context_main(j_decompress_ptr cinfo
, JSAMPARRAY output_buf
,
333 JDIMENSION
*out_row_ctr
, JDIMENSION out_rows_avail
)
335 my_main_ptr main_ptr
= (my_main_ptr
)cinfo
->main
;
337 /* Read input data if we haven't filled the main buffer yet */
338 if (!main_ptr
->buffer_full
) {
339 if (!(*cinfo
->coef
->decompress_data
) (cinfo
,
340 main_ptr
->xbuffer
[main_ptr
->whichptr
]))
341 return; /* suspension forced, can do nothing more */
342 main_ptr
->buffer_full
= TRUE
; /* OK, we have an iMCU row to work with */
343 main_ptr
->iMCU_row_ctr
++; /* count rows received */
346 /* Postprocessor typically will not swallow all the input data it is handed
347 * in one call (due to filling the output buffer first). Must be prepared
348 * to exit and restart. This switch lets us keep track of how far we got.
349 * Note that each case falls through to the next on successful completion.
351 switch (main_ptr
->context_state
) {
352 case CTX_POSTPONED_ROW
:
353 /* Call postprocessor using previously set pointers for postponed row */
354 (*cinfo
->post
->post_process_data
) (cinfo
,
355 main_ptr
->xbuffer
[main_ptr
->whichptr
],
356 &main_ptr
->rowgroup_ctr
,
357 main_ptr
->rowgroups_avail
, output_buf
,
358 out_row_ctr
, out_rows_avail
);
359 if (main_ptr
->rowgroup_ctr
< main_ptr
->rowgroups_avail
)
360 return; /* Need to suspend */
361 main_ptr
->context_state
= CTX_PREPARE_FOR_IMCU
;
362 if (*out_row_ctr
>= out_rows_avail
)
363 return; /* Postprocessor exactly filled output buf */
364 FALLTHROUGH
/*FALLTHROUGH*/
365 case CTX_PREPARE_FOR_IMCU
:
366 /* Prepare to process first M-1 row groups of this iMCU row */
367 main_ptr
->rowgroup_ctr
= 0;
368 main_ptr
->rowgroups_avail
= (JDIMENSION
)(cinfo
->_min_DCT_scaled_size
- 1);
369 /* Check for bottom of image: if so, tweak pointers to "duplicate"
370 * the last sample row, and adjust rowgroups_avail to ignore padding rows.
372 if (main_ptr
->iMCU_row_ctr
== cinfo
->total_iMCU_rows
)
373 set_bottom_pointers(cinfo
);
374 main_ptr
->context_state
= CTX_PROCESS_IMCU
;
375 FALLTHROUGH
/*FALLTHROUGH*/
376 case CTX_PROCESS_IMCU
:
377 /* Call postprocessor using previously set pointers */
378 (*cinfo
->post
->post_process_data
) (cinfo
,
379 main_ptr
->xbuffer
[main_ptr
->whichptr
],
380 &main_ptr
->rowgroup_ctr
,
381 main_ptr
->rowgroups_avail
, output_buf
,
382 out_row_ctr
, out_rows_avail
);
383 if (main_ptr
->rowgroup_ctr
< main_ptr
->rowgroups_avail
)
384 return; /* Need to suspend */
385 /* After the first iMCU, change wraparound pointers to normal state */
386 if (main_ptr
->iMCU_row_ctr
== 1)
387 set_wraparound_pointers(cinfo
);
388 /* Prepare to load new iMCU row using other xbuffer list */
389 main_ptr
->whichptr
^= 1; /* 0=>1 or 1=>0 */
390 main_ptr
->buffer_full
= FALSE
;
391 /* Still need to process last row group of this iMCU row, */
392 /* which is saved at index M+1 of the other xbuffer */
393 main_ptr
->rowgroup_ctr
= (JDIMENSION
)(cinfo
->_min_DCT_scaled_size
+ 1);
394 main_ptr
->rowgroups_avail
= (JDIMENSION
)(cinfo
->_min_DCT_scaled_size
+ 2);
395 main_ptr
->context_state
= CTX_POSTPONED_ROW
;
402 * Final pass of two-pass quantization: just call the postprocessor.
403 * Source data will be the postprocessor controller's internal buffer.
406 #ifdef QUANT_2PASS_SUPPORTED
409 process_data_crank_post(j_decompress_ptr cinfo
, JSAMPARRAY output_buf
,
410 JDIMENSION
*out_row_ctr
, JDIMENSION out_rows_avail
)
412 (*cinfo
->post
->post_process_data
) (cinfo
, (JSAMPIMAGE
)NULL
,
413 (JDIMENSION
*)NULL
, (JDIMENSION
)0,
414 output_buf
, out_row_ctr
, out_rows_avail
);
417 #endif /* QUANT_2PASS_SUPPORTED */
421 * Initialize main buffer controller.
425 jinit_d_main_controller(j_decompress_ptr cinfo
, boolean need_full_buffer
)
427 my_main_ptr main_ptr
;
428 int ci
, rgroup
, ngroups
;
429 jpeg_component_info
*compptr
;
431 main_ptr
= (my_main_ptr
)
432 (*cinfo
->mem
->alloc_small
) ((j_common_ptr
)cinfo
, JPOOL_IMAGE
,
433 sizeof(my_main_controller
));
434 cinfo
->main
= (struct jpeg_d_main_controller
*)main_ptr
;
435 main_ptr
->pub
.start_pass
= start_pass_main
;
437 if (need_full_buffer
) /* shouldn't happen */
438 ERREXIT(cinfo
, JERR_BAD_BUFFER_MODE
);
440 /* Allocate the workspace.
441 * ngroups is the number of row groups we need.
443 if (cinfo
->upsample
->need_context_rows
) {
444 if (cinfo
->_min_DCT_scaled_size
< 2) /* unsupported, see comments above */
445 ERREXIT(cinfo
, JERR_NOTIMPL
);
446 alloc_funny_pointers(cinfo
); /* Alloc space for xbuffer[] lists */
447 ngroups
= cinfo
->_min_DCT_scaled_size
+ 2;
449 ngroups
= cinfo
->_min_DCT_scaled_size
;
452 for (ci
= 0, compptr
= cinfo
->comp_info
; ci
< cinfo
->num_components
;
454 rgroup
= (compptr
->v_samp_factor
* compptr
->_DCT_scaled_size
) /
455 cinfo
->_min_DCT_scaled_size
; /* height of a row group of component */
456 main_ptr
->buffer
[ci
] = (*cinfo
->mem
->alloc_sarray
)
457 ((j_common_ptr
)cinfo
, JPOOL_IMAGE
,
458 compptr
->width_in_blocks
* compptr
->_DCT_scaled_size
,
459 (JDIMENSION
)(rgroup
* ngroups
));