pngdec: Add support for PIX_FMT_Y400A
[libav.git] / libavcodec / pngdec.c
1 /*
2 * PNG image format
3 * Copyright (c) 2003 Fabrice Bellard
4 *
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
11 *
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21 #include "avcodec.h"
22 #include "bytestream.h"
23 #include "png.h"
24 #include "dsputil.h"
25
26 /* TODO:
27 * - add 2, 4 and 16 bit depth support
28 */
29
30 #include <zlib.h>
31
32 //#define DEBUG
33
34 typedef struct PNGDecContext {
35 DSPContext dsp;
36
37 const uint8_t *bytestream;
38 const uint8_t *bytestream_start;
39 const uint8_t *bytestream_end;
40 AVFrame picture1, picture2;
41 AVFrame *current_picture, *last_picture;
42
43 int state;
44 int width, height;
45 int bit_depth;
46 int color_type;
47 int compression_type;
48 int interlace_type;
49 int filter_type;
50 int channels;
51 int bits_per_pixel;
52 int bpp;
53
54 uint8_t *image_buf;
55 int image_linesize;
56 uint32_t palette[256];
57 uint8_t *crow_buf;
58 uint8_t *last_row;
59 uint8_t *tmp_row;
60 int pass;
61 int crow_size; /* compressed row size (include filter type) */
62 int row_size; /* decompressed row size */
63 int pass_row_size; /* decompress row size of the current pass */
64 int y;
65 z_stream zstream;
66 } PNGDecContext;
67
68 /* Mask to determine which y pixels can be written in a pass */
69 static const uint8_t png_pass_dsp_ymask[NB_PASSES] = {
70 0xff, 0xff, 0x0f, 0xcc, 0x33, 0xff, 0x55,
71 };
72
73 /* Mask to determine which pixels to overwrite while displaying */
74 static const uint8_t png_pass_dsp_mask[NB_PASSES] = {
75 0xff, 0x0f, 0xff, 0x33, 0xff, 0x55, 0xff
76 };
77
78 /* NOTE: we try to construct a good looking image at each pass. width
79 is the original image width. We also do pixel format conversion at
80 this stage */
81 static void png_put_interlaced_row(uint8_t *dst, int width,
82 int bits_per_pixel, int pass,
83 int color_type, const uint8_t *src)
84 {
85 int x, mask, dsp_mask, j, src_x, b, bpp;
86 uint8_t *d;
87 const uint8_t *s;
88
89 mask = ff_png_pass_mask[pass];
90 dsp_mask = png_pass_dsp_mask[pass];
91 switch(bits_per_pixel) {
92 case 1:
93 /* we must initialize the line to zero before writing to it */
94 if (pass == 0)
95 memset(dst, 0, (width + 7) >> 3);
96 src_x = 0;
97 for(x = 0; x < width; x++) {
98 j = (x & 7);
99 if ((dsp_mask << j) & 0x80) {
100 b = (src[src_x >> 3] >> (7 - (src_x & 7))) & 1;
101 dst[x >> 3] |= b << (7 - j);
102 }
103 if ((mask << j) & 0x80)
104 src_x++;
105 }
106 break;
107 default:
108 bpp = bits_per_pixel >> 3;
109 d = dst;
110 s = src;
111 if (color_type == PNG_COLOR_TYPE_RGB_ALPHA) {
112 for(x = 0; x < width; x++) {
113 j = x & 7;
114 if ((dsp_mask << j) & 0x80) {
115 *(uint32_t *)d = (s[3] << 24) | (s[0] << 16) | (s[1] << 8) | s[2];
116 }
117 d += bpp;
118 if ((mask << j) & 0x80)
119 s += bpp;
120 }
121 } else {
122 for(x = 0; x < width; x++) {
123 j = x & 7;
124 if ((dsp_mask << j) & 0x80) {
125 memcpy(d, s, bpp);
126 }
127 d += bpp;
128 if ((mask << j) & 0x80)
129 s += bpp;
130 }
131 }
132 break;
133 }
134 }
135
136 void ff_add_png_paeth_prediction(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp)
137 {
138 int i;
139 for(i = 0; i < w; i++) {
140 int a, b, c, p, pa, pb, pc;
141
142 a = dst[i - bpp];
143 b = top[i];
144 c = top[i - bpp];
145
146 p = b - c;
147 pc = a - c;
148
149 pa = abs(p);
150 pb = abs(pc);
151 pc = abs(p + pc);
152
153 if (pa <= pb && pa <= pc)
154 p = a;
155 else if (pb <= pc)
156 p = b;
157 else
158 p = c;
159 dst[i] = p + src[i];
160 }
161 }
162
163 #define UNROLL1(bpp, op) {\
164 r = dst[0];\
165 if(bpp >= 2) g = dst[1];\
166 if(bpp >= 3) b = dst[2];\
167 if(bpp >= 4) a = dst[3];\
168 for(; i < size; i+=bpp) {\
169 dst[i+0] = r = op(r, src[i+0], last[i+0]);\
170 if(bpp == 1) continue;\
171 dst[i+1] = g = op(g, src[i+1], last[i+1]);\
172 if(bpp == 2) continue;\
173 dst[i+2] = b = op(b, src[i+2], last[i+2]);\
174 if(bpp == 3) continue;\
175 dst[i+3] = a = op(a, src[i+3], last[i+3]);\
176 }\
177 }
178
179 #define UNROLL_FILTER(op)\
180 if(bpp == 1) UNROLL1(1, op)\
181 else if(bpp == 2) UNROLL1(2, op)\
182 else if(bpp == 3) UNROLL1(3, op)\
183 else if(bpp == 4) UNROLL1(4, op)\
184 else {\
185 for (; i < size; i += bpp) {\
186 int j;\
187 for (j = 0; j < bpp; j++)\
188 dst[i+j] = op(dst[i+j-bpp], src[i+j], last[i+j]);\
189 }\
190 }
191
192 /* NOTE: 'dst' can be equal to 'last' */
193 static void png_filter_row(DSPContext *dsp, uint8_t *dst, int filter_type,
194 uint8_t *src, uint8_t *last, int size, int bpp)
195 {
196 int i, p, r, g, b, a;
197
198 switch(filter_type) {
199 case PNG_FILTER_VALUE_NONE:
200 memcpy(dst, src, size);
201 break;
202 case PNG_FILTER_VALUE_SUB:
203 for(i = 0; i < bpp; i++) {
204 dst[i] = src[i];
205 }
206 if(bpp == 4) {
207 p = *(int*)dst;
208 for(; i < size; i+=bpp) {
209 int s = *(int*)(src+i);
210 p = ((s&0x7f7f7f7f) + (p&0x7f7f7f7f)) ^ ((s^p)&0x80808080);
211 *(int*)(dst+i) = p;
212 }
213 } else {
214 #define OP_SUB(x,s,l) x+s
215 UNROLL_FILTER(OP_SUB);
216 }
217 break;
218 case PNG_FILTER_VALUE_UP:
219 dsp->add_bytes_l2(dst, src, last, size);
220 break;
221 case PNG_FILTER_VALUE_AVG:
222 for(i = 0; i < bpp; i++) {
223 p = (last[i] >> 1);
224 dst[i] = p + src[i];
225 }
226 #define OP_AVG(x,s,l) (((x + l) >> 1) + s) & 0xff
227 UNROLL_FILTER(OP_AVG);
228 break;
229 case PNG_FILTER_VALUE_PAETH:
230 for(i = 0; i < bpp; i++) {
231 p = last[i];
232 dst[i] = p + src[i];
233 }
234 if(bpp > 1 && size > 4) {
235 // would write off the end of the array if we let it process the last pixel with bpp=3
236 int w = bpp==4 ? size : size-3;
237 dsp->add_png_paeth_prediction(dst+i, src+i, last+i, w-i, bpp);
238 i = w;
239 }
240 ff_add_png_paeth_prediction(dst+i, src+i, last+i, size-i, bpp);
241 break;
242 }
243 }
244
245 static av_always_inline void convert_to_rgb32_loco(uint8_t *dst, const uint8_t *src, int width, int loco)
246 {
247 int j;
248 unsigned int r, g, b, a;
249
250 for(j = 0;j < width; j++) {
251 r = src[0];
252 g = src[1];
253 b = src[2];
254 a = src[3];
255 if(loco) {
256 r = (r+g)&0xff;
257 b = (b+g)&0xff;
258 }
259 *(uint32_t *)dst = (a << 24) | (r << 16) | (g << 8) | b;
260 dst += 4;
261 src += 4;
262 }
263 }
264
265 static void convert_to_rgb32(uint8_t *dst, const uint8_t *src, int width, int loco)
266 {
267 if(loco)
268 convert_to_rgb32_loco(dst, src, width, 1);
269 else
270 convert_to_rgb32_loco(dst, src, width, 0);
271 }
272
273 static void deloco_rgb24(uint8_t *dst, int size)
274 {
275 int i;
276 for(i=0; i<size; i+=3) {
277 int g = dst[i+1];
278 dst[i+0] += g;
279 dst[i+2] += g;
280 }
281 }
282
283 /* process exactly one decompressed row */
284 static void png_handle_row(PNGDecContext *s)
285 {
286 uint8_t *ptr, *last_row;
287 int got_line;
288
289 if (!s->interlace_type) {
290 ptr = s->image_buf + s->image_linesize * s->y;
291 /* need to swap bytes correctly for RGB_ALPHA */
292 if (s->color_type == PNG_COLOR_TYPE_RGB_ALPHA) {
293 png_filter_row(&s->dsp, s->tmp_row, s->crow_buf[0], s->crow_buf + 1,
294 s->last_row, s->row_size, s->bpp);
295 convert_to_rgb32(ptr, s->tmp_row, s->width, s->filter_type == PNG_FILTER_TYPE_LOCO);
296 FFSWAP(uint8_t*, s->last_row, s->tmp_row);
297 } else {
298 /* in normal case, we avoid one copy */
299 if (s->y == 0)
300 last_row = s->last_row;
301 else
302 last_row = ptr - s->image_linesize;
303
304 png_filter_row(&s->dsp, ptr, s->crow_buf[0], s->crow_buf + 1,
305 last_row, s->row_size, s->bpp);
306 }
307 /* loco lags by 1 row so that it doesn't interfere with top prediction */
308 if (s->filter_type == PNG_FILTER_TYPE_LOCO &&
309 s->color_type == PNG_COLOR_TYPE_RGB && s->y > 0)
310 deloco_rgb24(ptr - s->image_linesize, s->row_size);
311 s->y++;
312 if (s->y == s->height) {
313 s->state |= PNG_ALLIMAGE;
314 if (s->filter_type == PNG_FILTER_TYPE_LOCO &&
315 s->color_type == PNG_COLOR_TYPE_RGB)
316 deloco_rgb24(ptr, s->row_size);
317 }
318 } else {
319 got_line = 0;
320 for(;;) {
321 ptr = s->image_buf + s->image_linesize * s->y;
322 if ((ff_png_pass_ymask[s->pass] << (s->y & 7)) & 0x80) {
323 /* if we already read one row, it is time to stop to
324 wait for the next one */
325 if (got_line)
326 break;
327 png_filter_row(&s->dsp, s->tmp_row, s->crow_buf[0], s->crow_buf + 1,
328 s->last_row, s->pass_row_size, s->bpp);
329 FFSWAP(uint8_t*, s->last_row, s->tmp_row);
330 got_line = 1;
331 }
332 if ((png_pass_dsp_ymask[s->pass] << (s->y & 7)) & 0x80) {
333 /* NOTE: RGB32 is handled directly in png_put_interlaced_row */
334 png_put_interlaced_row(ptr, s->width, s->bits_per_pixel, s->pass,
335 s->color_type, s->last_row);
336 }
337 s->y++;
338 if (s->y == s->height) {
339 for(;;) {
340 if (s->pass == NB_PASSES - 1) {
341 s->state |= PNG_ALLIMAGE;
342 goto the_end;
343 } else {
344 s->pass++;
345 s->y = 0;
346 s->pass_row_size = ff_png_pass_row_size(s->pass,
347 s->bits_per_pixel,
348 s->width);
349 s->crow_size = s->pass_row_size + 1;
350 if (s->pass_row_size != 0)
351 break;
352 /* skip pass if empty row */
353 }
354 }
355 }
356 }
357 the_end: ;
358 }
359 }
360
361 static int png_decode_idat(PNGDecContext *s, int length)
362 {
363 int ret;
364 s->zstream.avail_in = length;
365 s->zstream.next_in = s->bytestream;
366 s->bytestream += length;
367
368 if(s->bytestream > s->bytestream_end)
369 return -1;
370
371 /* decode one line if possible */
372 while (s->zstream.avail_in > 0) {
373 ret = inflate(&s->zstream, Z_PARTIAL_FLUSH);
374 if (ret != Z_OK && ret != Z_STREAM_END) {
375 return -1;
376 }
377 if (s->zstream.avail_out == 0) {
378 if (!(s->state & PNG_ALLIMAGE)) {
379 png_handle_row(s);
380 }
381 s->zstream.avail_out = s->crow_size;
382 s->zstream.next_out = s->crow_buf;
383 }
384 }
385 return 0;
386 }
387
388 static int decode_frame(AVCodecContext *avctx,
389 void *data, int *data_size,
390 AVPacket *avpkt)
391 {
392 const uint8_t *buf = avpkt->data;
393 int buf_size = avpkt->size;
394 PNGDecContext * const s = avctx->priv_data;
395 AVFrame *picture = data;
396 AVFrame *p;
397 uint8_t *crow_buf_base = NULL;
398 uint32_t tag, length;
399 int ret, crc;
400
401 FFSWAP(AVFrame *, s->current_picture, s->last_picture);
402 avctx->coded_frame= s->current_picture;
403 p = s->current_picture;
404
405 s->bytestream_start=
406 s->bytestream= buf;
407 s->bytestream_end= buf + buf_size;
408
409 /* check signature */
410 if (memcmp(s->bytestream, ff_pngsig, 8) != 0 &&
411 memcmp(s->bytestream, ff_mngsig, 8) != 0)
412 return -1;
413 s->bytestream+= 8;
414 s->y=
415 s->state=0;
416 // memset(s, 0, sizeof(PNGDecContext));
417 /* init the zlib */
418 s->zstream.zalloc = ff_png_zalloc;
419 s->zstream.zfree = ff_png_zfree;
420 s->zstream.opaque = NULL;
421 ret = inflateInit(&s->zstream);
422 if (ret != Z_OK)
423 return -1;
424 for(;;) {
425 int tag32;
426 if (s->bytestream >= s->bytestream_end)
427 goto fail;
428 length = bytestream_get_be32(&s->bytestream);
429 if (length > 0x7fffffff)
430 goto fail;
431 tag32 = bytestream_get_be32(&s->bytestream);
432 tag = bswap_32(tag32);
433 dprintf(avctx, "png: tag=%c%c%c%c length=%u\n",
434 (tag & 0xff),
435 ((tag >> 8) & 0xff),
436 ((tag >> 16) & 0xff),
437 ((tag >> 24) & 0xff), length);
438 switch(tag) {
439 case MKTAG('I', 'H', 'D', 'R'):
440 if (length != 13)
441 goto fail;
442 s->width = bytestream_get_be32(&s->bytestream);
443 s->height = bytestream_get_be32(&s->bytestream);
444 if(avcodec_check_dimensions(avctx, s->width, s->height)){
445 s->width= s->height= 0;
446 goto fail;
447 }
448 s->bit_depth = *s->bytestream++;
449 s->color_type = *s->bytestream++;
450 s->compression_type = *s->bytestream++;
451 s->filter_type = *s->bytestream++;
452 s->interlace_type = *s->bytestream++;
453 crc = bytestream_get_be32(&s->bytestream);
454 s->state |= PNG_IHDR;
455 dprintf(avctx, "width=%d height=%d depth=%d color_type=%d compression_type=%d filter_type=%d interlace_type=%d\n",
456 s->width, s->height, s->bit_depth, s->color_type,
457 s->compression_type, s->filter_type, s->interlace_type);
458 break;
459 case MKTAG('I', 'D', 'A', 'T'):
460 if (!(s->state & PNG_IHDR))
461 goto fail;
462 if (!(s->state & PNG_IDAT)) {
463 /* init image info */
464 avctx->width = s->width;
465 avctx->height = s->height;
466
467 s->channels = ff_png_get_nb_channels(s->color_type);
468 s->bits_per_pixel = s->bit_depth * s->channels;
469 s->bpp = (s->bits_per_pixel + 7) >> 3;
470 s->row_size = (avctx->width * s->bits_per_pixel + 7) >> 3;
471
472 if (s->bit_depth == 8 &&
473 s->color_type == PNG_COLOR_TYPE_RGB) {
474 avctx->pix_fmt = PIX_FMT_RGB24;
475 } else if (s->bit_depth == 8 &&
476 s->color_type == PNG_COLOR_TYPE_RGB_ALPHA) {
477 avctx->pix_fmt = PIX_FMT_RGB32;
478 } else if (s->bit_depth == 8 &&
479 s->color_type == PNG_COLOR_TYPE_GRAY) {
480 avctx->pix_fmt = PIX_FMT_GRAY8;
481 } else if (s->bit_depth == 16 &&
482 s->color_type == PNG_COLOR_TYPE_GRAY) {
483 avctx->pix_fmt = PIX_FMT_GRAY16BE;
484 } else if (s->bit_depth == 16 &&
485 s->color_type == PNG_COLOR_TYPE_RGB) {
486 avctx->pix_fmt = PIX_FMT_RGB48BE;
487 } else if (s->bit_depth == 1 &&
488 s->color_type == PNG_COLOR_TYPE_GRAY) {
489 avctx->pix_fmt = PIX_FMT_MONOBLACK;
490 } else if (s->color_type == PNG_COLOR_TYPE_PALETTE) {
491 avctx->pix_fmt = PIX_FMT_PAL8;
492 } else if (s->color_type == PNG_COLOR_TYPE_GRAY_ALPHA) {
493 avctx->pix_fmt = PIX_FMT_Y400A;
494 } else {
495 goto fail;
496 }
497 if(p->data[0])
498 avctx->release_buffer(avctx, p);
499
500 p->reference= 0;
501 if(avctx->get_buffer(avctx, p) < 0){
502 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
503 goto fail;
504 }
505 p->pict_type= FF_I_TYPE;
506 p->key_frame= 1;
507 p->interlaced_frame = !!s->interlace_type;
508
509 /* compute the compressed row size */
510 if (!s->interlace_type) {
511 s->crow_size = s->row_size + 1;
512 } else {
513 s->pass = 0;
514 s->pass_row_size = ff_png_pass_row_size(s->pass,
515 s->bits_per_pixel,
516 s->width);
517 s->crow_size = s->pass_row_size + 1;
518 }
519 dprintf(avctx, "row_size=%d crow_size =%d\n",
520 s->row_size, s->crow_size);
521 s->image_buf = p->data[0];
522 s->image_linesize = p->linesize[0];
523 /* copy the palette if needed */
524 if (s->color_type == PNG_COLOR_TYPE_PALETTE)
525 memcpy(p->data[1], s->palette, 256 * sizeof(uint32_t));
526 /* empty row is used if differencing to the first row */
527 s->last_row = av_mallocz(s->row_size);
528 if (!s->last_row)
529 goto fail;
530 if (s->interlace_type ||
531 s->color_type == PNG_COLOR_TYPE_RGB_ALPHA) {
532 s->tmp_row = av_malloc(s->row_size);
533 if (!s->tmp_row)
534 goto fail;
535 }
536 /* compressed row */
537 crow_buf_base = av_malloc(s->row_size + 16);
538 if (!crow_buf_base)
539 goto fail;
540
541 /* we want crow_buf+1 to be 16-byte aligned */
542 s->crow_buf = crow_buf_base + 15;
543 s->zstream.avail_out = s->crow_size;
544 s->zstream.next_out = s->crow_buf;
545 }
546 s->state |= PNG_IDAT;
547 if (png_decode_idat(s, length) < 0)
548 goto fail;
549 /* skip crc */
550 crc = bytestream_get_be32(&s->bytestream);
551 break;
552 case MKTAG('P', 'L', 'T', 'E'):
553 {
554 int n, i, r, g, b;
555
556 if ((length % 3) != 0 || length > 256 * 3)
557 goto skip_tag;
558 /* read the palette */
559 n = length / 3;
560 for(i=0;i<n;i++) {
561 r = *s->bytestream++;
562 g = *s->bytestream++;
563 b = *s->bytestream++;
564 s->palette[i] = (0xff << 24) | (r << 16) | (g << 8) | b;
565 }
566 for(;i<256;i++) {
567 s->palette[i] = (0xff << 24);
568 }
569 s->state |= PNG_PLTE;
570 crc = bytestream_get_be32(&s->bytestream);
571 }
572 break;
573 case MKTAG('t', 'R', 'N', 'S'):
574 {
575 int v, i;
576
577 /* read the transparency. XXX: Only palette mode supported */
578 if (s->color_type != PNG_COLOR_TYPE_PALETTE ||
579 length > 256 ||
580 !(s->state & PNG_PLTE))
581 goto skip_tag;
582 for(i=0;i<length;i++) {
583 v = *s->bytestream++;
584 s->palette[i] = (s->palette[i] & 0x00ffffff) | (v << 24);
585 }
586 crc = bytestream_get_be32(&s->bytestream);
587 }
588 break;
589 case MKTAG('I', 'E', 'N', 'D'):
590 if (!(s->state & PNG_ALLIMAGE))
591 goto fail;
592 crc = bytestream_get_be32(&s->bytestream);
593 goto exit_loop;
594 default:
595 /* skip tag */
596 skip_tag:
597 s->bytestream += length + 4;
598 break;
599 }
600 }
601 exit_loop:
602 /* handle p-frames only if a predecessor frame is available */
603 if(s->last_picture->data[0] != NULL) {
604 if(!(avpkt->flags & AV_PKT_FLAG_KEY)) {
605 int i, j;
606 uint8_t *pd = s->current_picture->data[0];
607 uint8_t *pd_last = s->last_picture->data[0];
608
609 for(j=0; j < s->height; j++) {
610 for(i=0; i < s->width * s->bpp; i++) {
611 pd[i] += pd_last[i];
612 }
613 pd += s->image_linesize;
614 pd_last += s->image_linesize;
615 }
616 }
617 }
618
619 *picture= *s->current_picture;
620 *data_size = sizeof(AVFrame);
621
622 ret = s->bytestream - s->bytestream_start;
623 the_end:
624 inflateEnd(&s->zstream);
625 av_free(crow_buf_base);
626 s->crow_buf = NULL;
627 av_freep(&s->last_row);
628 av_freep(&s->tmp_row);
629 return ret;
630 fail:
631 ret = -1;
632 goto the_end;
633 }
634
635 static av_cold int png_dec_init(AVCodecContext *avctx){
636 PNGDecContext *s = avctx->priv_data;
637
638 s->current_picture = &s->picture1;
639 s->last_picture = &s->picture2;
640 avcodec_get_frame_defaults(&s->picture1);
641 avcodec_get_frame_defaults(&s->picture2);
642 dsputil_init(&s->dsp, avctx);
643
644 return 0;
645 }
646
647 static av_cold int png_dec_end(AVCodecContext *avctx)
648 {
649 PNGDecContext *s = avctx->priv_data;
650
651 if (s->picture1.data[0])
652 avctx->release_buffer(avctx, &s->picture1);
653 if (s->picture2.data[0])
654 avctx->release_buffer(avctx, &s->picture2);
655
656 return 0;
657 }
658
659 AVCodec png_decoder = {
660 "png",
661 AVMEDIA_TYPE_VIDEO,
662 CODEC_ID_PNG,
663 sizeof(PNGDecContext),
664 png_dec_init,
665 NULL,
666 png_dec_end,
667 decode_frame,
668 CODEC_CAP_DR1 /*| CODEC_CAP_DRAW_HORIZ_BAND*/,
669 NULL,
670 .long_name = NULL_IF_CONFIG_SMALL("PNG image"),
671 };