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 "codec_internal.h"
43 #include "encode.h"
44 #include "hap.h"
45 #include "texturedsp.h"
46
47 #define HAP_MAX_CHUNKS 64
48
49 enum HapHeaderLength {
50 /* Short header: four bytes with a 24 bit size value */
51 HAP_HDR_SHORT = 4,
52 /* Long header: eight bytes with a 32 bit size value */
53 HAP_HDR_LONG = 8,
54 };
55
compress_texture(AVCodecContext * avctx,uint8_t * out,int out_length,const AVFrame * f)56 static int compress_texture(AVCodecContext *avctx, uint8_t *out, int out_length, const AVFrame *f)
57 {
58 HapContext *ctx = avctx->priv_data;
59
60 if (ctx->tex_size > out_length)
61 return AVERROR_BUFFER_TOO_SMALL;
62
63 ctx->enc.tex_data.out = out;
64 ctx->enc.frame_data.in = f->data[0];
65 ctx->enc.stride = f->linesize[0];
66 avctx->execute2(avctx, ff_texturedsp_compress_thread, &ctx->enc, NULL, ctx->enc.slice_count);
67
68 return 0;
69 }
70
71 /* section_length does not include the header */
hap_write_section_header(PutByteContext * pbc,enum HapHeaderLength header_length,int section_length,enum HapSectionType section_type)72 static void hap_write_section_header(PutByteContext *pbc,
73 enum HapHeaderLength header_length,
74 int section_length,
75 enum HapSectionType section_type)
76 {
77 /* The first three bytes are the length of the section (not including the
78 * header) or zero if using an eight-byte header.
79 * For an eight-byte header, the length is in the last four bytes.
80 * The fourth byte stores the section type. */
81 bytestream2_put_le24(pbc, header_length == HAP_HDR_LONG ? 0 : section_length);
82 bytestream2_put_byte(pbc, section_type);
83
84 if (header_length == HAP_HDR_LONG) {
85 bytestream2_put_le32(pbc, section_length);
86 }
87 }
88
hap_compress_frame(AVCodecContext * avctx,uint8_t * dst)89 static int hap_compress_frame(AVCodecContext *avctx, uint8_t *dst)
90 {
91 HapContext *ctx = avctx->priv_data;
92 int i, final_size = 0;
93
94 for (i = 0; i < ctx->chunk_count; i++) {
95 HapChunk *chunk = &ctx->chunks[i];
96 uint8_t *chunk_src, *chunk_dst;
97 int ret;
98
99 if (i == 0) {
100 chunk->compressed_offset = 0;
101 } else {
102 chunk->compressed_offset = ctx->chunks[i-1].compressed_offset
103 + ctx->chunks[i-1].compressed_size;
104 }
105 chunk->uncompressed_size = ctx->tex_size / ctx->chunk_count;
106 chunk->uncompressed_offset = i * chunk->uncompressed_size;
107 chunk->compressed_size = ctx->max_snappy;
108 chunk_src = ctx->tex_buf + chunk->uncompressed_offset;
109 chunk_dst = dst + chunk->compressed_offset;
110
111 /* Compress with snappy too, write directly on packet buffer. */
112 ret = snappy_compress(chunk_src, chunk->uncompressed_size,
113 chunk_dst, &chunk->compressed_size);
114 if (ret != SNAPPY_OK) {
115 av_log(avctx, AV_LOG_ERROR, "Snappy compress error.\n");
116 return AVERROR_BUG;
117 }
118
119 /* If there is no gain from snappy, just use the raw texture. */
120 if (chunk->compressed_size >= chunk->uncompressed_size) {
121 av_log(avctx, AV_LOG_VERBOSE,
122 "Snappy buffer bigger than uncompressed (%"SIZE_SPECIFIER" >= %"SIZE_SPECIFIER" bytes).\n",
123 chunk->compressed_size, chunk->uncompressed_size);
124 memcpy(chunk_dst, chunk_src, chunk->uncompressed_size);
125 chunk->compressor = HAP_COMP_NONE;
126 chunk->compressed_size = chunk->uncompressed_size;
127 } else {
128 chunk->compressor = HAP_COMP_SNAPPY;
129 }
130
131 final_size += chunk->compressed_size;
132 }
133
134 return final_size;
135 }
136
hap_decode_instructions_length(HapContext * ctx)137 static int hap_decode_instructions_length(HapContext *ctx)
138 {
139 /* Second-Stage Compressor Table (one byte per entry)
140 * + Chunk Size Table (four bytes per entry)
141 * + headers for both sections (short versions)
142 * = chunk_count + (4 * chunk_count) + 4 + 4 */
143 return (5 * ctx->chunk_count) + 8;
144 }
145
hap_header_length(HapContext * ctx)146 static int hap_header_length(HapContext *ctx)
147 {
148 /* Top section header (long version) */
149 int length = HAP_HDR_LONG;
150
151 if (ctx->chunk_count > 1) {
152 /* Decode Instructions header (short) + Decode Instructions Container */
153 length += HAP_HDR_SHORT + hap_decode_instructions_length(ctx);
154 }
155
156 return length;
157 }
158
hap_write_frame_header(HapContext * ctx,uint8_t * dst,int frame_length)159 static void hap_write_frame_header(HapContext *ctx, uint8_t *dst, int frame_length)
160 {
161 PutByteContext pbc;
162 int i;
163
164 bytestream2_init_writer(&pbc, dst, frame_length);
165 if (ctx->chunk_count == 1) {
166 /* Write a simple header */
167 hap_write_section_header(&pbc, HAP_HDR_LONG, frame_length - 8,
168 ctx->chunks[0].compressor | ctx->opt_tex_fmt);
169 } else {
170 /* Write a complex header with Decode Instructions Container */
171 hap_write_section_header(&pbc, HAP_HDR_LONG, frame_length - 8,
172 HAP_COMP_COMPLEX | ctx->opt_tex_fmt);
173 hap_write_section_header(&pbc, HAP_HDR_SHORT, hap_decode_instructions_length(ctx),
174 HAP_ST_DECODE_INSTRUCTIONS);
175 hap_write_section_header(&pbc, HAP_HDR_SHORT, ctx->chunk_count,
176 HAP_ST_COMPRESSOR_TABLE);
177
178 for (i = 0; i < ctx->chunk_count; i++) {
179 bytestream2_put_byte(&pbc, ctx->chunks[i].compressor >> 4);
180 }
181
182 hap_write_section_header(&pbc, HAP_HDR_SHORT, ctx->chunk_count * 4,
183 HAP_ST_SIZE_TABLE);
184
185 for (i = 0; i < ctx->chunk_count; i++) {
186 bytestream2_put_le32(&pbc, ctx->chunks[i].compressed_size);
187 }
188 }
189 }
190
hap_encode(AVCodecContext * avctx,AVPacket * pkt,const AVFrame * frame,int * got_packet)191 static int hap_encode(AVCodecContext *avctx, AVPacket *pkt,
192 const AVFrame *frame, int *got_packet)
193 {
194 HapContext *ctx = avctx->priv_data;
195 int header_length = hap_header_length(ctx);
196 int final_data_size, ret;
197 int pktsize = FFMAX(ctx->tex_size, ctx->max_snappy * ctx->chunk_count) + header_length;
198
199 /* Allocate maximum size packet, shrink later. */
200 ret = ff_alloc_packet(avctx, pkt, pktsize);
201 if (ret < 0)
202 return ret;
203
204 if (ctx->opt_compressor == HAP_COMP_NONE) {
205 /* DXTC compression directly to the packet buffer. */
206 ret = compress_texture(avctx, pkt->data + header_length, pkt->size - header_length, frame);
207 if (ret < 0)
208 return ret;
209
210 ctx->chunks[0].compressor = HAP_COMP_NONE;
211 final_data_size = ctx->tex_size;
212 } else {
213 /* DXTC compression. */
214 ret = compress_texture(avctx, ctx->tex_buf, ctx->tex_size, frame);
215 if (ret < 0)
216 return ret;
217
218 /* Compress (using Snappy) the frame */
219 final_data_size = hap_compress_frame(avctx, pkt->data + header_length);
220 if (final_data_size < 0)
221 return final_data_size;
222 }
223
224 /* Write header at the start. */
225 hap_write_frame_header(ctx, pkt->data, final_data_size + header_length);
226
227 av_shrink_packet(pkt, final_data_size + header_length);
228 *got_packet = 1;
229 return 0;
230 }
231
hap_init(AVCodecContext * avctx)232 static av_cold int hap_init(AVCodecContext *avctx)
233 {
234 HapContext *ctx = avctx->priv_data;
235 int corrected_chunk_count;
236 int ret = av_image_check_size(avctx->width, avctx->height, 0, avctx);
237
238 if (ret < 0) {
239 av_log(avctx, AV_LOG_ERROR, "Invalid video size %dx%d.\n",
240 avctx->width, avctx->height);
241 return ret;
242 }
243
244 if (avctx->width % 4 || avctx->height % 4) {
245 av_log(avctx, AV_LOG_ERROR, "Video size %dx%d is not multiple of 4.\n",
246 avctx->width, avctx->height);
247 return AVERROR_INVALIDDATA;
248 }
249
250 ff_texturedspenc_init(&ctx->dxtc);
251
252 switch (ctx->opt_tex_fmt) {
253 case HAP_FMT_RGBDXT1:
254 ctx->enc.tex_ratio = 8;
255 avctx->codec_tag = MKTAG('H', 'a', 'p', '1');
256 avctx->bits_per_coded_sample = 24;
257 ctx->enc.tex_funct = ctx->dxtc.dxt1_block;
258 break;
259 case HAP_FMT_RGBADXT5:
260 ctx->enc.tex_ratio = 16;
261 avctx->codec_tag = MKTAG('H', 'a', 'p', '5');
262 avctx->bits_per_coded_sample = 32;
263 ctx->enc.tex_funct = ctx->dxtc.dxt5_block;
264 break;
265 case HAP_FMT_YCOCGDXT5:
266 ctx->enc.tex_ratio = 16;
267 avctx->codec_tag = MKTAG('H', 'a', 'p', 'Y');
268 avctx->bits_per_coded_sample = 24;
269 ctx->enc.tex_funct = ctx->dxtc.dxt5ys_block;
270 break;
271 default:
272 av_log(avctx, AV_LOG_ERROR, "Invalid format %02X\n", ctx->opt_tex_fmt);
273 return AVERROR_INVALIDDATA;
274 }
275 ctx->enc.raw_ratio = 16;
276 ctx->enc.slice_count = av_clip(avctx->thread_count, 1, avctx->height / TEXTURE_BLOCK_H);
277
278 /* Texture compression ratio is constant, so can we computer
279 * beforehand the final size of the uncompressed buffer. */
280 ctx->tex_size = avctx->width / TEXTURE_BLOCK_W *
281 avctx->height / TEXTURE_BLOCK_H * ctx->enc.tex_ratio;
282
283 switch (ctx->opt_compressor) {
284 case HAP_COMP_NONE:
285 /* No benefit chunking uncompressed data */
286 corrected_chunk_count = 1;
287
288 ctx->max_snappy = ctx->tex_size;
289 ctx->tex_buf = NULL;
290 break;
291 case HAP_COMP_SNAPPY:
292 /* Round the chunk count to divide evenly on DXT block edges */
293 corrected_chunk_count = av_clip(ctx->opt_chunk_count, 1, HAP_MAX_CHUNKS);
294 while ((ctx->tex_size / ctx->enc.tex_ratio) % corrected_chunk_count != 0) {
295 corrected_chunk_count--;
296 }
297
298 ctx->max_snappy = snappy_max_compressed_length(ctx->tex_size / corrected_chunk_count);
299 ctx->tex_buf = av_malloc(ctx->tex_size);
300 if (!ctx->tex_buf) {
301 return AVERROR(ENOMEM);
302 }
303 break;
304 default:
305 av_log(avctx, AV_LOG_ERROR, "Invalid compresor %02X\n", ctx->opt_compressor);
306 return AVERROR_INVALIDDATA;
307 }
308 if (corrected_chunk_count != ctx->opt_chunk_count) {
309 av_log(avctx, AV_LOG_INFO, "%d chunks requested but %d used.\n",
310 ctx->opt_chunk_count, corrected_chunk_count);
311 }
312 ret = ff_hap_set_chunk_count(ctx, corrected_chunk_count, 1);
313 if (ret != 0)
314 return ret;
315
316 return 0;
317 }
318
hap_close(AVCodecContext * avctx)319 static av_cold int hap_close(AVCodecContext *avctx)
320 {
321 HapContext *ctx = avctx->priv_data;
322
323 ff_hap_free_context(ctx);
324
325 return 0;
326 }
327
328 #define OFFSET(x) offsetof(HapContext, x)
329 #define FLAGS AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
330 static const AVOption options[] = {
331 { "format", NULL, OFFSET(opt_tex_fmt), AV_OPT_TYPE_INT, { .i64 = HAP_FMT_RGBDXT1 }, HAP_FMT_RGBDXT1, HAP_FMT_YCOCGDXT5, FLAGS, "format" },
332 { "hap", "Hap 1 (DXT1 textures)", 0, AV_OPT_TYPE_CONST, { .i64 = HAP_FMT_RGBDXT1 }, 0, 0, FLAGS, "format" },
333 { "hap_alpha", "Hap Alpha (DXT5 textures)", 0, AV_OPT_TYPE_CONST, { .i64 = HAP_FMT_RGBADXT5 }, 0, 0, FLAGS, "format" },
334 { "hap_q", "Hap Q (DXT5-YCoCg textures)", 0, AV_OPT_TYPE_CONST, { .i64 = HAP_FMT_YCOCGDXT5 }, 0, 0, FLAGS, "format" },
335 { "chunks", "chunk count", OFFSET(opt_chunk_count), AV_OPT_TYPE_INT, {.i64 = 1 }, 1, HAP_MAX_CHUNKS, FLAGS, },
336 { "compressor", "second-stage compressor", OFFSET(opt_compressor), AV_OPT_TYPE_INT, { .i64 = HAP_COMP_SNAPPY }, HAP_COMP_NONE, HAP_COMP_SNAPPY, FLAGS, "compressor" },
337 { "none", "None", 0, AV_OPT_TYPE_CONST, { .i64 = HAP_COMP_NONE }, 0, 0, FLAGS, "compressor" },
338 { "snappy", "Snappy", 0, AV_OPT_TYPE_CONST, { .i64 = HAP_COMP_SNAPPY }, 0, 0, FLAGS, "compressor" },
339 { NULL },
340 };
341
342 static const AVClass hapenc_class = {
343 .class_name = "Hap encoder",
344 .item_name = av_default_item_name,
345 .option = options,
346 .version = LIBAVUTIL_VERSION_INT,
347 };
348
349 const FFCodec ff_hap_encoder = {
350 .p.name = "hap",
351 .p.long_name = NULL_IF_CONFIG_SMALL("Vidvox Hap"),
352 .p.type = AVMEDIA_TYPE_VIDEO,
353 .p.id = AV_CODEC_ID_HAP,
354 .priv_data_size = sizeof(HapContext),
355 .p.priv_class = &hapenc_class,
356 .p.capabilities = AV_CODEC_CAP_SLICE_THREADS,
357 .init = hap_init,
358 FF_CODEC_ENCODE_CB(hap_encode),
359 .close = hap_close,
360 .p.pix_fmts = (const enum AVPixelFormat[]) {
361 AV_PIX_FMT_RGBA, AV_PIX_FMT_NONE,
362 },
363 .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE |
364 FF_CODEC_CAP_INIT_CLEANUP,
365 };
366