0.1.29
by Andreas Cadhalpun
Import upstream version 2.8 |
1 |
/*
|
2 |
* Vidvox Hap decoder
|
|
3 |
* Copyright (C) 2015 Vittorio Giovara <vittorio.giovara@gmail.com>
|
|
4 |
* Copyright (C) 2015 Tom Butterworth <bangnoise@gmail.com>
|
|
5 |
*
|
|
6 |
* This file is part of FFmpeg.
|
|
7 |
*
|
|
8 |
* FFmpeg is free software; you can redistribute it and/or
|
|
9 |
* modify it under the terms of the GNU Lesser General Public
|
|
10 |
* License as published by the Free Software Foundation; either
|
|
11 |
* version 2.1 of the License, or (at your option) any later version.
|
|
12 |
*
|
|
13 |
* FFmpeg is distributed in the hope that it will be useful,
|
|
14 |
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
15 |
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
16 |
* Lesser General Public License for more details.
|
|
17 |
*
|
|
18 |
* You should have received a copy of the GNU Lesser General Public
|
|
19 |
* License along with FFmpeg; if not, write to the Free Software
|
|
20 |
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
|
21 |
*/
|
|
22 |
||
23 |
/**
|
|
24 |
* @file
|
|
25 |
* Hap decoder
|
|
26 |
*
|
|
27 |
* Fourcc: Hap1, Hap5, HapY
|
|
28 |
*
|
|
29 |
* https://github.com/Vidvox/hap/blob/master/documentation/HapVideoDRAFT.md
|
|
30 |
*/
|
|
31 |
||
32 |
#include <stdint.h> |
|
33 |
||
34 |
#include "libavutil/imgutils.h" |
|
35 |
||
36 |
#include "avcodec.h" |
|
37 |
#include "bytestream.h" |
|
38 |
#include "hap.h" |
|
39 |
#include "internal.h" |
|
40 |
#include "memory.h" |
|
41 |
#include "snappy.h" |
|
42 |
#include "texturedsp.h" |
|
43 |
#include "thread.h" |
|
44 |
||
45 |
/* The first three bytes are the size of the section past the header, or zero
|
|
46 |
* if the length is stored in the next long word. The fourth byte in the first
|
|
47 |
* long word indicates the type of the current section. */
|
|
48 |
static int parse_section_header(GetByteContext *gbc, int *section_size, |
|
49 |
enum HapSectionType *section_type) |
|
50 |
{
|
|
51 |
if (bytestream2_get_bytes_left(gbc) < 4) |
|
52 |
return AVERROR_INVALIDDATA; |
|
53 |
||
54 |
*section_size = bytestream2_get_le24(gbc); |
|
55 |
*section_type = bytestream2_get_byte(gbc); |
|
56 |
||
57 |
if (*section_size == 0) { |
|
58 |
if (bytestream2_get_bytes_left(gbc) < 4) |
|
59 |
return AVERROR_INVALIDDATA; |
|
60 |
||
61 |
*section_size = bytestream2_get_le32(gbc); |
|
62 |
}
|
|
63 |
||
64 |
if (*section_size > bytestream2_get_bytes_left(gbc) || *section_size < 0) |
|
65 |
return AVERROR_INVALIDDATA; |
|
66 |
else
|
|
67 |
return 0; |
|
68 |
}
|
|
69 |
||
70 |
static int hap_parse_decode_instructions(HapContext *ctx, int size) |
|
71 |
{
|
|
72 |
GetByteContext *gbc = &ctx->gbc; |
|
73 |
int section_size; |
|
74 |
enum HapSectionType section_type; |
|
75 |
int is_first_table = 1, had_offsets = 0, had_compressors = 0, had_sizes = 0; |
|
76 |
int i, ret; |
|
77 |
||
78 |
while (size > 0) { |
|
79 |
int stream_remaining = bytestream2_get_bytes_left(gbc); |
|
80 |
ret = parse_section_header(gbc, §ion_size, §ion_type); |
|
81 |
if (ret != 0) |
|
82 |
return ret; |
|
83 |
||
84 |
size -= stream_remaining - bytestream2_get_bytes_left(gbc); |
|
85 |
||
86 |
switch (section_type) { |
|
87 |
case HAP_ST_COMPRESSOR_TABLE: |
|
88 |
ret = ff_hap_set_chunk_count(ctx, section_size, is_first_table); |
|
89 |
if (ret != 0) |
|
90 |
return ret; |
|
91 |
for (i = 0; i < section_size; i++) { |
|
92 |
ctx->chunks[i].compressor = bytestream2_get_byte(gbc) << 4; |
|
93 |
}
|
|
94 |
had_compressors = 1; |
|
95 |
is_first_table = 0; |
|
96 |
break; |
|
97 |
case HAP_ST_SIZE_TABLE: |
|
98 |
ret = ff_hap_set_chunk_count(ctx, section_size / 4, is_first_table); |
|
99 |
if (ret != 0) |
|
100 |
return ret; |
|
101 |
for (i = 0; i < section_size / 4; i++) { |
|
102 |
ctx->chunks[i].compressed_size = bytestream2_get_le32(gbc); |
|
103 |
}
|
|
104 |
had_sizes = 1; |
|
105 |
is_first_table = 0; |
|
106 |
break; |
|
107 |
case HAP_ST_OFFSET_TABLE: |
|
108 |
ret = ff_hap_set_chunk_count(ctx, section_size / 4, is_first_table); |
|
109 |
if (ret != 0) |
|
110 |
return ret; |
|
111 |
for (i = 0; i < section_size / 4; i++) { |
|
112 |
ctx->chunks[i].compressed_offset = bytestream2_get_le32(gbc); |
|
113 |
}
|
|
114 |
had_offsets = 1; |
|
115 |
is_first_table = 0; |
|
116 |
break; |
|
117 |
default: |
|
118 |
break; |
|
119 |
}
|
|
120 |
size -= section_size; |
|
121 |
}
|
|
122 |
||
123 |
if (!had_sizes || !had_compressors) |
|
124 |
return AVERROR_INVALIDDATA; |
|
125 |
||
126 |
/* The offsets table is optional. If not present than calculate offsets by
|
|
127 |
* summing the sizes of preceding chunks. */
|
|
128 |
if (!had_offsets) { |
|
129 |
size_t running_size = 0; |
|
130 |
for (i = 0; i < ctx->chunk_count; i++) { |
|
131 |
ctx->chunks[i].compressed_offset = running_size; |
|
132 |
running_size += ctx->chunks[i].compressed_size; |
|
133 |
}
|
|
134 |
}
|
|
135 |
||
136 |
return 0; |
|
137 |
}
|
|
138 |
||
139 |
static int hap_can_use_tex_in_place(HapContext *ctx) |
|
140 |
{
|
|
141 |
int i; |
|
142 |
size_t running_offset = 0; |
|
143 |
for (i = 0; i < ctx->chunk_count; i++) { |
|
144 |
if (ctx->chunks[i].compressed_offset != running_offset |
|
145 |
|| ctx->chunks[i].compressor != HAP_COMP_NONE) |
|
146 |
return 0; |
|
147 |
running_offset += ctx->chunks[i].compressed_size; |
|
148 |
}
|
|
149 |
return 1; |
|
150 |
}
|
|
151 |
||
152 |
static int hap_parse_frame_header(AVCodecContext *avctx) |
|
153 |
{
|
|
154 |
HapContext *ctx = avctx->priv_data; |
|
155 |
GetByteContext *gbc = &ctx->gbc; |
|
156 |
int section_size; |
|
157 |
enum HapSectionType section_type; |
|
158 |
const char *compressorstr; |
|
159 |
int i, ret; |
|
160 |
||
161 |
ret = parse_section_header(gbc, §ion_size, §ion_type); |
|
162 |
if (ret != 0) |
|
163 |
return ret; |
|
164 |
||
165 |
if ((avctx->codec_tag == MKTAG('H','a','p','1') && (section_type & 0x0F) != HAP_FMT_RGBDXT1) || |
|
166 |
(avctx->codec_tag == MKTAG('H','a','p','5') && (section_type & 0x0F) != HAP_FMT_RGBADXT5) || |
|
167 |
(avctx->codec_tag == MKTAG('H','a','p','Y') && (section_type & 0x0F) != HAP_FMT_YCOCGDXT5)) { |
|
168 |
av_log(avctx, AV_LOG_ERROR, |
|
169 |
"Invalid texture format %#04x.\n", section_type & 0x0F); |
|
170 |
return AVERROR_INVALIDDATA; |
|
171 |
}
|
|
172 |
||
173 |
switch (section_type & 0xF0) { |
|
174 |
case HAP_COMP_NONE: |
|
175 |
case HAP_COMP_SNAPPY: |
|
176 |
ret = ff_hap_set_chunk_count(ctx, 1, 1); |
|
177 |
if (ret == 0) { |
|
178 |
ctx->chunks[0].compressor = section_type & 0xF0; |
|
179 |
ctx->chunks[0].compressed_offset = 0; |
|
180 |
ctx->chunks[0].compressed_size = section_size; |
|
181 |
}
|
|
182 |
if (ctx->chunks[0].compressor == HAP_COMP_NONE) { |
|
183 |
compressorstr = "none"; |
|
184 |
} else { |
|
185 |
compressorstr = "snappy"; |
|
186 |
}
|
|
187 |
break; |
|
188 |
case HAP_COMP_COMPLEX: |
|
189 |
ret = parse_section_header(gbc, §ion_size, §ion_type); |
|
190 |
if (ret == 0 && section_type != HAP_ST_DECODE_INSTRUCTIONS) |
|
191 |
ret = AVERROR_INVALIDDATA; |
|
192 |
if (ret == 0) |
|
193 |
ret = hap_parse_decode_instructions(ctx, section_size); |
|
194 |
compressorstr = "complex"; |
|
195 |
break; |
|
196 |
default: |
|
197 |
ret = AVERROR_INVALIDDATA; |
|
198 |
break; |
|
199 |
}
|
|
200 |
||
201 |
if (ret != 0) |
|
202 |
return ret; |
|
203 |
||
204 |
/* Check the frame is valid and read the uncompressed chunk sizes */
|
|
205 |
ctx->tex_size = 0; |
|
206 |
for (i = 0; i < ctx->chunk_count; i++) { |
|
207 |
HapChunk *chunk = &ctx->chunks[i]; |
|
208 |
||
209 |
/* Check the compressed buffer is valid */
|
|
210 |
if (chunk->compressed_offset + chunk->compressed_size > bytestream2_get_bytes_left(gbc)) |
|
211 |
return AVERROR_INVALIDDATA; |
|
212 |
||
213 |
/* Chunks are unpacked sequentially, ctx->tex_size is the uncompressed
|
|
214 |
* size thus far */
|
|
215 |
chunk->uncompressed_offset = ctx->tex_size; |
|
216 |
||
217 |
/* Fill out uncompressed size */
|
|
218 |
if (chunk->compressor == HAP_COMP_SNAPPY) { |
|
219 |
GetByteContext gbc_tmp; |
|
220 |
int64_t uncompressed_size; |
|
221 |
bytestream2_init(&gbc_tmp, gbc->buffer + chunk->compressed_offset, |
|
222 |
chunk->compressed_size); |
|
223 |
uncompressed_size = ff_snappy_peek_uncompressed_length(&gbc_tmp); |
|
224 |
if (uncompressed_size < 0) { |
|
225 |
return uncompressed_size; |
|
226 |
}
|
|
227 |
chunk->uncompressed_size = uncompressed_size; |
|
228 |
} else if (chunk->compressor == HAP_COMP_NONE) { |
|
229 |
chunk->uncompressed_size = chunk->compressed_size; |
|
230 |
} else { |
|
231 |
return AVERROR_INVALIDDATA; |
|
232 |
}
|
|
233 |
ctx->tex_size += chunk->uncompressed_size; |
|
234 |
}
|
|
235 |
||
236 |
av_log(avctx, AV_LOG_DEBUG, "%s compressor\n", compressorstr); |
|
237 |
||
238 |
return ret; |
|
239 |
}
|
|
240 |
||
241 |
static int decompress_chunks_thread(AVCodecContext *avctx, void *arg, |
|
242 |
int chunk_nb, int thread_nb) |
|
243 |
{
|
|
244 |
HapContext *ctx = avctx->priv_data; |
|
245 |
||
246 |
HapChunk *chunk = &ctx->chunks[chunk_nb]; |
|
247 |
GetByteContext gbc; |
|
248 |
uint8_t *dst = ctx->tex_buf + chunk->uncompressed_offset; |
|
249 |
||
250 |
bytestream2_init(&gbc, ctx->gbc.buffer + chunk->compressed_offset, chunk->compressed_size); |
|
251 |
||
252 |
if (chunk->compressor == HAP_COMP_SNAPPY) { |
|
253 |
int ret; |
|
254 |
int64_t uncompressed_size = ctx->tex_size; |
|
255 |
||
256 |
/* Uncompress the frame */
|
|
257 |
ret = ff_snappy_uncompress(&gbc, dst, &uncompressed_size); |
|
258 |
if (ret < 0) { |
|
259 |
av_log(avctx, AV_LOG_ERROR, "Snappy uncompress error\n"); |
|
260 |
return ret; |
|
261 |
}
|
|
262 |
} else if (chunk->compressor == HAP_COMP_NONE) { |
|
263 |
bytestream2_get_buffer(&gbc, dst, chunk->compressed_size); |
|
264 |
}
|
|
265 |
||
266 |
return 0; |
|
267 |
}
|
|
268 |
||
269 |
static int decompress_texture_thread(AVCodecContext *avctx, void *arg, |
|
270 |
int slice, int thread_nb) |
|
271 |
{
|
|
272 |
HapContext *ctx = avctx->priv_data; |
|
273 |
AVFrame *frame = arg; |
|
274 |
const uint8_t *d = ctx->tex_data; |
|
275 |
int w_block = avctx->coded_width / TEXTURE_BLOCK_W; |
|
276 |
int h_block = avctx->coded_height / TEXTURE_BLOCK_H; |
|
277 |
int x, y; |
|
278 |
int start_slice, end_slice; |
|
279 |
int base_blocks_per_slice = h_block / ctx->slice_count; |
|
280 |
int remainder_blocks = h_block % ctx->slice_count; |
|
281 |
||
282 |
/* When the frame height (in blocks) doesn't divide evenly between the
|
|
283 |
* number of slices, spread the remaining blocks evenly between the first
|
|
284 |
* operations */
|
|
285 |
start_slice = slice * base_blocks_per_slice; |
|
286 |
/* Add any extra blocks (one per slice) that have been added before this slice */
|
|
287 |
start_slice += FFMIN(slice, remainder_blocks); |
|
288 |
||
289 |
end_slice = start_slice + base_blocks_per_slice; |
|
290 |
/* Add an extra block if there are still remainder blocks to be accounted for */
|
|
291 |
if (slice < remainder_blocks) |
|
292 |
end_slice++; |
|
293 |
||
294 |
for (y = start_slice; y < end_slice; y++) { |
|
295 |
uint8_t *p = frame->data[0] + y * frame->linesize[0] * TEXTURE_BLOCK_H; |
|
296 |
int off = y * w_block; |
|
297 |
for (x = 0; x < w_block; x++) { |
|
298 |
ctx->tex_fun(p + x * 16, frame->linesize[0], |
|
299 |
d + (off + x) * ctx->tex_rat); |
|
300 |
}
|
|
301 |
}
|
|
302 |
||
303 |
return 0; |
|
304 |
}
|
|
305 |
||
306 |
static int hap_decode(AVCodecContext *avctx, void *data, |
|
307 |
int *got_frame, AVPacket *avpkt) |
|
308 |
{
|
|
309 |
HapContext *ctx = avctx->priv_data; |
|
310 |
ThreadFrame tframe; |
|
311 |
int ret, i; |
|
312 |
int tex_size; |
|
313 |
||
314 |
bytestream2_init(&ctx->gbc, avpkt->data, avpkt->size); |
|
315 |
||
316 |
/* Check for section header */
|
|
317 |
ret = hap_parse_frame_header(avctx); |
|
318 |
if (ret < 0) |
|
319 |
return ret; |
|
320 |
||
321 |
/* Get the output frame ready to receive data */
|
|
322 |
tframe.f = data; |
|
323 |
ret = ff_thread_get_buffer(avctx, &tframe, 0); |
|
324 |
if (ret < 0) |
|
325 |
return ret; |
|
326 |
if (avctx->codec->update_thread_context) |
|
327 |
ff_thread_finish_setup(avctx); |
|
328 |
||
329 |
/* Unpack the DXT texture */
|
|
330 |
if (hap_can_use_tex_in_place(ctx)) { |
|
331 |
/* Only DXTC texture compression in a contiguous block */
|
|
332 |
ctx->tex_data = ctx->gbc.buffer; |
|
333 |
tex_size = bytestream2_get_bytes_left(&ctx->gbc); |
|
334 |
} else { |
|
335 |
/* Perform the second-stage decompression */
|
|
336 |
ret = av_reallocp(&ctx->tex_buf, ctx->tex_size); |
|
337 |
if (ret < 0) |
|
338 |
return ret; |
|
339 |
||
340 |
avctx->execute2(avctx, decompress_chunks_thread, NULL, |
|
341 |
ctx->chunk_results, ctx->chunk_count); |
|
342 |
||
343 |
for (i = 0; i < ctx->chunk_count; i++) { |
|
344 |
if (ctx->chunk_results[i] < 0) |
|
345 |
return ctx->chunk_results[i]; |
|
346 |
}
|
|
347 |
||
348 |
ctx->tex_data = ctx->tex_buf; |
|
349 |
tex_size = ctx->tex_size; |
|
350 |
}
|
|
351 |
||
352 |
if (tex_size < (avctx->coded_width / TEXTURE_BLOCK_W) |
|
353 |
*(avctx->coded_height / TEXTURE_BLOCK_H) |
|
354 |
*ctx->tex_rat) { |
|
355 |
av_log(avctx, AV_LOG_ERROR, "Insufficient data\n"); |
|
356 |
return AVERROR_INVALIDDATA; |
|
357 |
}
|
|
358 |
||
359 |
/* Use the decompress function on the texture, one block per thread */
|
|
360 |
avctx->execute2(avctx, decompress_texture_thread, tframe.f, NULL, ctx->slice_count); |
|
361 |
||
362 |
/* Frame is ready to be output */
|
|
363 |
tframe.f->pict_type = AV_PICTURE_TYPE_I; |
|
364 |
tframe.f->key_frame = 1; |
|
365 |
*got_frame = 1; |
|
366 |
||
367 |
return avpkt->size; |
|
368 |
}
|
|
369 |
||
370 |
static av_cold int hap_init(AVCodecContext *avctx) |
|
371 |
{
|
|
372 |
HapContext *ctx = avctx->priv_data; |
|
373 |
const char *texture_name; |
|
374 |
int ret = av_image_check_size(avctx->width, avctx->height, 0, avctx); |
|
375 |
||
376 |
if (ret < 0) { |
|
377 |
av_log(avctx, AV_LOG_ERROR, "Invalid video size %dx%d.\n", |
|
378 |
avctx->width, avctx->height); |
|
379 |
return ret; |
|
380 |
}
|
|
381 |
||
382 |
/* Since codec is based on 4x4 blocks, size is aligned to 4 */
|
|
383 |
avctx->coded_width = FFALIGN(avctx->width, TEXTURE_BLOCK_W); |
|
384 |
avctx->coded_height = FFALIGN(avctx->height, TEXTURE_BLOCK_H); |
|
385 |
||
386 |
/* Technically only one mode has alpha, but 32 bits are easier to handle */
|
|
387 |
avctx->pix_fmt = AV_PIX_FMT_RGBA; |
|
388 |
||
389 |
ff_texturedsp_init(&ctx->dxtc); |
|
390 |
||
391 |
switch (avctx->codec_tag) { |
|
392 |
case MKTAG('H','a','p','1'): |
|
393 |
texture_name = "DXT1"; |
|
394 |
ctx->tex_rat = 8; |
|
395 |
ctx->tex_fun = ctx->dxtc.dxt1_block; |
|
396 |
break; |
|
397 |
case MKTAG('H','a','p','5'): |
|
398 |
texture_name = "DXT5"; |
|
399 |
ctx->tex_rat = 16; |
|
400 |
ctx->tex_fun = ctx->dxtc.dxt5_block; |
|
401 |
break; |
|
402 |
case MKTAG('H','a','p','Y'): |
|
403 |
texture_name = "DXT5-YCoCg-scaled"; |
|
404 |
ctx->tex_rat = 16; |
|
405 |
ctx->tex_fun = ctx->dxtc.dxt5ys_block; |
|
406 |
break; |
|
407 |
default: |
|
408 |
return AVERROR_DECODER_NOT_FOUND; |
|
409 |
}
|
|
410 |
||
411 |
av_log(avctx, AV_LOG_DEBUG, "%s texture\n", texture_name); |
|
412 |
||
413 |
ctx->slice_count = av_clip(avctx->thread_count, 1, |
|
414 |
avctx->coded_height / TEXTURE_BLOCK_H); |
|
415 |
||
416 |
return 0; |
|
417 |
}
|
|
418 |
||
419 |
static av_cold int hap_close(AVCodecContext *avctx) |
|
420 |
{
|
|
421 |
HapContext *ctx = avctx->priv_data; |
|
422 |
||
423 |
ff_hap_free_context(ctx); |
|
424 |
||
425 |
return 0; |
|
426 |
}
|
|
427 |
||
428 |
AVCodec ff_hap_decoder = { |
|
429 |
.name = "hap", |
|
430 |
.long_name = NULL_IF_CONFIG_SMALL("Vidvox Hap decoder"), |
|
431 |
.type = AVMEDIA_TYPE_VIDEO, |
|
432 |
.id = AV_CODEC_ID_HAP, |
|
433 |
.init = hap_init, |
|
434 |
.decode = hap_decode, |
|
435 |
.close = hap_close, |
|
436 |
.priv_data_size = sizeof(HapContext), |
|
437 |
.capabilities = AV_CODEC_CAP_FRAME_THREADS | AV_CODEC_CAP_SLICE_THREADS | |
|
438 |
AV_CODEC_CAP_DR1, |
|
439 |
.caps_internal = FF_CODEC_CAP_INIT_THREADSAFE | |
|
440 |
FF_CODEC_CAP_INIT_CLEANUP, |
|
441 |
};
|