blob: 3a1bc870b13f8c267a2a3ee07ea1ed6deef29cbd
1 | /* |
2 | * Vidvox Hap encoder |
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 encoder |
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 | #include "snappy-c.h" |
34 | |
35 | #include "libavutil/frame.h" |
36 | #include "libavutil/imgutils.h" |
37 | #include "libavutil/intreadwrite.h" |
38 | #include "libavutil/opt.h" |
39 | |
40 | #include "avcodec.h" |
41 | #include "bytestream.h" |
42 | #include "hap.h" |
43 | #include "internal.h" |
44 | #include "texturedsp.h" |
45 | |
46 | #define HAP_MAX_CHUNKS 64 |
47 | |
48 | enum HapHeaderLength { |
49 | /* Short header: four bytes with a 24 bit size value */ |
50 | HAP_HDR_SHORT = 4, |
51 | /* Long header: eight bytes with a 32 bit size value */ |
52 | HAP_HDR_LONG = 8, |
53 | }; |
54 | |
55 | static int compress_texture(AVCodecContext *avctx, uint8_t *out, int out_length, const AVFrame *f) |
56 | { |
57 | HapContext *ctx = avctx->priv_data; |
58 | int i, j; |
59 | |
60 | if (ctx->tex_size > out_length) |
61 | return AVERROR_BUFFER_TOO_SMALL; |
62 | |
63 | for (j = 0; j < avctx->height; j += 4) { |
64 | for (i = 0; i < avctx->width; i += 4) { |
65 | uint8_t *p = f->data[0] + i * 4 + j * f->linesize[0]; |
66 | const int step = ctx->tex_fun(out, f->linesize[0], p); |
67 | out += step; |
68 | } |
69 | } |
70 | |
71 | return 0; |
72 | } |
73 | |
74 | /* section_length does not include the header */ |
75 | static void hap_write_section_header(PutByteContext *pbc, |
76 | enum HapHeaderLength header_length, |
77 | int section_length, |
78 | enum HapSectionType section_type) |
79 | { |
80 | /* The first three bytes are the length of the section (not including the |
81 | * header) or zero if using an eight-byte header. |
82 | * For an eight-byte header, the length is in the last four bytes. |
83 | * The fourth byte stores the section type. */ |
84 | bytestream2_put_le24(pbc, header_length == HAP_HDR_LONG ? 0 : section_length); |
85 | bytestream2_put_byte(pbc, section_type); |
86 | |
87 | if (header_length == HAP_HDR_LONG) { |
88 | bytestream2_put_le32(pbc, section_length); |
89 | } |
90 | } |
91 | |
92 | static int hap_compress_frame(AVCodecContext *avctx, uint8_t *dst) |
93 | { |
94 | HapContext *ctx = avctx->priv_data; |
95 | int i, final_size = 0; |
96 | |
97 | for (i = 0; i < ctx->chunk_count; i++) { |
98 | HapChunk *chunk = &ctx->chunks[i]; |
99 | uint8_t *chunk_src, *chunk_dst; |
100 | int ret; |
101 | |
102 | if (i == 0) { |
103 | chunk->compressed_offset = 0; |
104 | } else { |
105 | chunk->compressed_offset = ctx->chunks[i-1].compressed_offset |
106 | + ctx->chunks[i-1].compressed_size; |
107 | } |
108 | chunk->uncompressed_size = ctx->tex_size / ctx->chunk_count; |
109 | chunk->uncompressed_offset = i * chunk->uncompressed_size; |
110 | chunk->compressed_size = ctx->max_snappy; |
111 | chunk_src = ctx->tex_buf + chunk->uncompressed_offset; |
112 | chunk_dst = dst + chunk->compressed_offset; |
113 | |
114 | /* Compress with snappy too, write directly on packet buffer. */ |
115 | ret = snappy_compress(chunk_src, chunk->uncompressed_size, |
116 | chunk_dst, &chunk->compressed_size); |
117 | if (ret != SNAPPY_OK) { |
118 | av_log(avctx, AV_LOG_ERROR, "Snappy compress error.\n"); |
119 | return AVERROR_BUG; |
120 | } |
121 | |
122 | /* If there is no gain from snappy, just use the raw texture. */ |
123 | if (chunk->compressed_size >= chunk->uncompressed_size) { |
124 | av_log(avctx, AV_LOG_VERBOSE, |
125 | "Snappy buffer bigger than uncompressed (%"SIZE_SPECIFIER" >= %"SIZE_SPECIFIER" bytes).\n", |
126 | chunk->compressed_size, chunk->uncompressed_size); |
127 | memcpy(chunk_dst, chunk_src, chunk->uncompressed_size); |
128 | chunk->compressor = HAP_COMP_NONE; |
129 | chunk->compressed_size = chunk->uncompressed_size; |
130 | } else { |
131 | chunk->compressor = HAP_COMP_SNAPPY; |
132 | } |
133 | |
134 | final_size += chunk->compressed_size; |
135 | } |
136 | |
137 | return final_size; |
138 | } |
139 | |
140 | static int hap_decode_instructions_length(HapContext *ctx) |
141 | { |
142 | /* Second-Stage Compressor Table (one byte per entry) |
143 | * + Chunk Size Table (four bytes per entry) |
144 | * + headers for both sections (short versions) |
145 | * = chunk_count + (4 * chunk_count) + 4 + 4 */ |
146 | return (5 * ctx->chunk_count) + 8; |
147 | } |
148 | |
149 | static int hap_header_length(HapContext *ctx) |
150 | { |
151 | /* Top section header (long version) */ |
152 | int length = HAP_HDR_LONG; |
153 | |
154 | if (ctx->chunk_count > 1) { |
155 | /* Decode Instructions header (short) + Decode Instructions Container */ |
156 | length += HAP_HDR_SHORT + hap_decode_instructions_length(ctx); |
157 | } |
158 | |
159 | return length; |
160 | } |
161 | |
162 | static void hap_write_frame_header(HapContext *ctx, uint8_t *dst, int frame_length) |
163 | { |
164 | PutByteContext pbc; |
165 | int i; |
166 | |
167 | bytestream2_init_writer(&pbc, dst, frame_length); |
168 | if (ctx->chunk_count == 1) { |
169 | /* Write a simple header */ |
170 | hap_write_section_header(&pbc, HAP_HDR_LONG, frame_length - 8, |
171 | ctx->chunks[0].compressor | ctx->opt_tex_fmt); |
172 | } else { |
173 | /* Write a complex header with Decode Instructions Container */ |
174 | hap_write_section_header(&pbc, HAP_HDR_LONG, frame_length - 8, |
175 | HAP_COMP_COMPLEX | ctx->opt_tex_fmt); |
176 | hap_write_section_header(&pbc, HAP_HDR_SHORT, hap_decode_instructions_length(ctx), |
177 | HAP_ST_DECODE_INSTRUCTIONS); |
178 | hap_write_section_header(&pbc, HAP_HDR_SHORT, ctx->chunk_count, |
179 | HAP_ST_COMPRESSOR_TABLE); |
180 | |
181 | for (i = 0; i < ctx->chunk_count; i++) { |
182 | bytestream2_put_byte(&pbc, ctx->chunks[i].compressor >> 4); |
183 | } |
184 | |
185 | hap_write_section_header(&pbc, HAP_HDR_SHORT, ctx->chunk_count * 4, |
186 | HAP_ST_SIZE_TABLE); |
187 | |
188 | for (i = 0; i < ctx->chunk_count; i++) { |
189 | bytestream2_put_le32(&pbc, ctx->chunks[i].compressed_size); |
190 | } |
191 | } |
192 | } |
193 | |
194 | static int hap_encode(AVCodecContext *avctx, AVPacket *pkt, |
195 | const AVFrame *frame, int *got_packet) |
196 | { |
197 | HapContext *ctx = avctx->priv_data; |
198 | int header_length = hap_header_length(ctx); |
199 | int final_data_size, ret; |
200 | int pktsize = FFMAX(ctx->tex_size, ctx->max_snappy * ctx->chunk_count) + header_length; |
201 | |
202 | /* Allocate maximum size packet, shrink later. */ |
203 | ret = ff_alloc_packet2(avctx, pkt, pktsize, header_length); |
204 | if (ret < 0) |
205 | return ret; |
206 | |
207 | if (ctx->opt_compressor == HAP_COMP_NONE) { |
208 | /* DXTC compression directly to the packet buffer. */ |
209 | ret = compress_texture(avctx, pkt->data + header_length, pkt->size - header_length, frame); |
210 | if (ret < 0) |
211 | return ret; |
212 | |
213 | ctx->chunks[0].compressor = HAP_COMP_NONE; |
214 | final_data_size = ctx->tex_size; |
215 | } else { |
216 | /* DXTC compression. */ |
217 | ret = compress_texture(avctx, ctx->tex_buf, ctx->tex_size, frame); |
218 | if (ret < 0) |
219 | return ret; |
220 | |
221 | /* Compress (using Snappy) the frame */ |
222 | final_data_size = hap_compress_frame(avctx, pkt->data + header_length); |
223 | if (final_data_size < 0) |
224 | return final_data_size; |
225 | } |
226 | |
227 | /* Write header at the start. */ |
228 | hap_write_frame_header(ctx, pkt->data, final_data_size + header_length); |
229 | |
230 | av_shrink_packet(pkt, final_data_size + header_length); |
231 | pkt->flags |= AV_PKT_FLAG_KEY; |
232 | *got_packet = 1; |
233 | return 0; |
234 | } |
235 | |
236 | static av_cold int hap_init(AVCodecContext *avctx) |
237 | { |
238 | HapContext *ctx = avctx->priv_data; |
239 | int ratio; |
240 | int corrected_chunk_count; |
241 | int ret = av_image_check_size(avctx->width, avctx->height, 0, avctx); |
242 | |
243 | if (ret < 0) { |
244 | av_log(avctx, AV_LOG_ERROR, "Invalid video size %dx%d.\n", |
245 | avctx->width, avctx->height); |
246 | return ret; |
247 | } |
248 | |
249 | if (avctx->width % 4 || avctx->height % 4) { |
250 | av_log(avctx, AV_LOG_ERROR, "Video size %dx%d is not multiple of 4.\n", |
251 | avctx->width, avctx->height); |
252 | return AVERROR_INVALIDDATA; |
253 | } |
254 | |
255 | ff_texturedspenc_init(&ctx->dxtc); |
256 | |
257 | switch (ctx->opt_tex_fmt) { |
258 | case HAP_FMT_RGBDXT1: |
259 | ratio = 8; |
260 | avctx->codec_tag = MKTAG('H', 'a', 'p', '1'); |
261 | avctx->bits_per_coded_sample = 24; |
262 | ctx->tex_fun = ctx->dxtc.dxt1_block; |
263 | break; |
264 | case HAP_FMT_RGBADXT5: |
265 | ratio = 4; |
266 | avctx->codec_tag = MKTAG('H', 'a', 'p', '5'); |
267 | avctx->bits_per_coded_sample = 32; |
268 | ctx->tex_fun = ctx->dxtc.dxt5_block; |
269 | break; |
270 | case HAP_FMT_YCOCGDXT5: |
271 | ratio = 4; |
272 | avctx->codec_tag = MKTAG('H', 'a', 'p', 'Y'); |
273 | avctx->bits_per_coded_sample = 24; |
274 | ctx->tex_fun = ctx->dxtc.dxt5ys_block; |
275 | break; |
276 | default: |
277 | av_log(avctx, AV_LOG_ERROR, "Invalid format %02X\n", ctx->opt_tex_fmt); |
278 | return AVERROR_INVALIDDATA; |
279 | } |
280 | |
281 | /* Texture compression ratio is constant, so can we computer |
282 | * beforehand the final size of the uncompressed buffer. */ |
283 | ctx->tex_size = FFALIGN(avctx->width, TEXTURE_BLOCK_W) * |
284 | FFALIGN(avctx->height, TEXTURE_BLOCK_H) * 4 / ratio; |
285 | |
286 | switch (ctx->opt_compressor) { |
287 | case HAP_COMP_NONE: |
288 | /* No benefit chunking uncompressed data */ |
289 | corrected_chunk_count = 1; |
290 | |
291 | ctx->max_snappy = ctx->tex_size; |
292 | ctx->tex_buf = NULL; |
293 | break; |
294 | case HAP_COMP_SNAPPY: |
295 | /* Round the chunk count to divide evenly on DXT block edges */ |
296 | corrected_chunk_count = av_clip(ctx->opt_chunk_count, 1, HAP_MAX_CHUNKS); |
297 | while ((ctx->tex_size / (64 / ratio)) % corrected_chunk_count != 0) { |
298 | corrected_chunk_count--; |
299 | } |
300 | |
301 | ctx->max_snappy = snappy_max_compressed_length(ctx->tex_size / corrected_chunk_count); |
302 | ctx->tex_buf = av_malloc(ctx->tex_size); |
303 | if (!ctx->tex_buf) { |
304 | return AVERROR(ENOMEM); |
305 | } |
306 | break; |
307 | default: |
308 | av_log(avctx, AV_LOG_ERROR, "Invalid compresor %02X\n", ctx->opt_compressor); |
309 | return AVERROR_INVALIDDATA; |
310 | } |
311 | if (corrected_chunk_count != ctx->opt_chunk_count) { |
312 | av_log(avctx, AV_LOG_INFO, "%d chunks requested but %d used.\n", |
313 | ctx->opt_chunk_count, corrected_chunk_count); |
314 | } |
315 | ret = ff_hap_set_chunk_count(ctx, corrected_chunk_count, 1); |
316 | if (ret != 0) |
317 | return ret; |
318 | |
319 | return 0; |
320 | } |
321 | |
322 | static av_cold int hap_close(AVCodecContext *avctx) |
323 | { |
324 | HapContext *ctx = avctx->priv_data; |
325 | |
326 | ff_hap_free_context(ctx); |
327 | |
328 | return 0; |
329 | } |
330 | |
331 | #define OFFSET(x) offsetof(HapContext, x) |
332 | #define FLAGS AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM |
333 | static const AVOption options[] = { |
334 | { "format", NULL, OFFSET(opt_tex_fmt), AV_OPT_TYPE_INT, { .i64 = HAP_FMT_RGBDXT1 }, HAP_FMT_RGBDXT1, HAP_FMT_YCOCGDXT5, FLAGS, "format" }, |
335 | { "hap", "Hap 1 (DXT1 textures)", 0, AV_OPT_TYPE_CONST, { .i64 = HAP_FMT_RGBDXT1 }, 0, 0, FLAGS, "format" }, |
336 | { "hap_alpha", "Hap Alpha (DXT5 textures)", 0, AV_OPT_TYPE_CONST, { .i64 = HAP_FMT_RGBADXT5 }, 0, 0, FLAGS, "format" }, |
337 | { "hap_q", "Hap Q (DXT5-YCoCg textures)", 0, AV_OPT_TYPE_CONST, { .i64 = HAP_FMT_YCOCGDXT5 }, 0, 0, FLAGS, "format" }, |
338 | { "chunks", "chunk count", OFFSET(opt_chunk_count), AV_OPT_TYPE_INT, {.i64 = 1 }, 1, HAP_MAX_CHUNKS, FLAGS, }, |
339 | { "compressor", "second-stage compressor", OFFSET(opt_compressor), AV_OPT_TYPE_INT, { .i64 = HAP_COMP_SNAPPY }, HAP_COMP_NONE, HAP_COMP_SNAPPY, FLAGS, "compressor" }, |
340 | { "none", "None", 0, AV_OPT_TYPE_CONST, { .i64 = HAP_COMP_NONE }, 0, 0, FLAGS, "compressor" }, |
341 | { "snappy", "Snappy", 0, AV_OPT_TYPE_CONST, { .i64 = HAP_COMP_SNAPPY }, 0, 0, FLAGS, "compressor" }, |
342 | { NULL }, |
343 | }; |
344 | |
345 | static const AVClass hapenc_class = { |
346 | .class_name = "Hap encoder", |
347 | .item_name = av_default_item_name, |
348 | .option = options, |
349 | .version = LIBAVUTIL_VERSION_INT, |
350 | }; |
351 | |
352 | AVCodec ff_hap_encoder = { |
353 | .name = "hap", |
354 | .long_name = NULL_IF_CONFIG_SMALL("Vidvox Hap"), |
355 | .type = AVMEDIA_TYPE_VIDEO, |
356 | .id = AV_CODEC_ID_HAP, |
357 | .priv_data_size = sizeof(HapContext), |
358 | .priv_class = &hapenc_class, |
359 | .init = hap_init, |
360 | .encode2 = hap_encode, |
361 | .close = hap_close, |
362 | .pix_fmts = (const enum AVPixelFormat[]) { |
363 | AV_PIX_FMT_RGBA, AV_PIX_FMT_NONE, |
364 | }, |
365 | .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE | |
366 | FF_CODEC_CAP_INIT_CLEANUP, |
367 | }; |
368 |