lavf/qsvvpp: bypass vpp if not needed.
[libav.git] / libavcodec / hapenc.c
CommitLineData
7ca3e520
VG
1/*
2 * Vidvox Hap encoder
3 * Copyright (C) 2015 Vittorio Giovara <vittorio.giovara@gmail.com>
3ee21785 4 * Copyright (C) 2015 Tom Butterworth <bangnoise@gmail.com>
7ca3e520
VG
5 *
6 * This file is part of Libav.
7 *
8 * Libav 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 * Libav 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 Libav; 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
3ee21785
TB
46#define HAP_MAX_CHUNKS 64
47
48enum 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};
7ca3e520
VG
54
55static void compress_texture(AVCodecContext *avctx, const AVFrame *f)
56{
57 HapContext *ctx = avctx->priv_data;
58 uint8_t *out = ctx->tex_buf;
59 int i, j;
60
61 for (j = 0; j < avctx->height; j += 4) {
62 for (i = 0; i < avctx->width; i += 4) {
63 uint8_t *p = f->data[0] + i * 4 + j * f->linesize[0];
64 const int step = ctx->tex_fun(out, f->linesize[0], p);
65 out += step;
66 }
67 }
68}
69
3ee21785
TB
70/* section_length does not include the header */
71static void hap_write_section_header(PutByteContext *pbc,
72 enum HapHeaderLength header_length,
73 int section_length,
74 enum HapSectionType section_type)
75{
76 /* The first three bytes are the length of the section (not including the
77 * header) or zero if using an eight-byte header.
78 * For an eight-byte header, the length is in the last four bytes.
79 * The fourth byte stores the section type. */
80 bytestream2_put_le24(pbc, header_length == HAP_HDR_LONG ? 0 : section_length);
81 bytestream2_put_byte(pbc, section_type);
82
83 if (header_length == HAP_HDR_LONG) {
84 bytestream2_put_le32(pbc, section_length);
85 }
86}
87
88static int hap_compress_frame(AVCodecContext *avctx, uint8_t *dst)
89{
90 HapContext *ctx = avctx->priv_data;
91 int i, final_size = 0;
92
93 for (i = 0; i < ctx->chunk_count; i++) {
94 HapChunk *chunk = &ctx->chunks[i];
95 uint8_t *chunk_src, *chunk_dst;
96 int ret;
97
98 if (i == 0) {
99 chunk->compressed_offset = 0;
100 } else {
101 chunk->compressed_offset = ctx->chunks[i-1].compressed_offset
102 + ctx->chunks[i-1].compressed_size;
103 }
104 chunk->uncompressed_size = ctx->tex_size / ctx->chunk_count;
105 chunk->uncompressed_offset = i * chunk->uncompressed_size;
106 chunk->compressed_size = ctx->max_snappy;
107 chunk_src = ctx->tex_buf + chunk->uncompressed_offset;
108 chunk_dst = dst + chunk->compressed_offset;
109
110 /* Compress with snappy too, write directly on packet buffer. */
111 ret = snappy_compress(chunk_src, chunk->uncompressed_size,
112 chunk_dst, &chunk->compressed_size);
113 if (ret != SNAPPY_OK) {
114 av_log(avctx, AV_LOG_ERROR, "Snappy compress error.\n");
115 return AVERROR_BUG;
116 }
117
118 /* If there is no gain from snappy, just use the raw texture. */
119 if (chunk->compressed_size >= chunk->uncompressed_size) {
120 av_log(avctx, AV_LOG_VERBOSE,
fbe425c8 121 "Snappy buffer bigger than uncompressed (%zu >= %zu bytes).\n",
3ee21785
TB
122 chunk->compressed_size, chunk->uncompressed_size);
123 memcpy(chunk_dst, chunk_src, chunk->uncompressed_size);
124 chunk->compressor = HAP_COMP_NONE;
125 chunk->compressed_size = chunk->uncompressed_size;
126 } else {
127 chunk->compressor = HAP_COMP_SNAPPY;
128 }
129
130 final_size += chunk->compressed_size;
131 }
132
133 return final_size;
134}
135
136static int hap_decode_instructions_length(HapContext *ctx)
137{
138 /* Second-Stage Compressor Table (one byte per entry)
139 * + Chunk Size Table (four bytes per entry)
140 * + headers for both sections (short versions)
141 * = chunk_count + (4 * chunk_count) + 4 + 4 */
142 return (5 * ctx->chunk_count) + 8;
143}
144
145static int hap_header_length(HapContext *ctx)
146{
147 /* Top section header (long version) */
148 int length = HAP_HDR_LONG;
149
150 if (ctx->chunk_count > 1) {
151 /* Decode Instructions header (short) + Decode Instructions Container */
152 length += HAP_HDR_SHORT + hap_decode_instructions_length(ctx);
153 }
154
155 return length;
156}
157
158static void hap_write_frame_header(HapContext *ctx, uint8_t *dst, int frame_length)
159{
160 PutByteContext pbc;
161 int i;
162
163 bytestream2_init_writer(&pbc, dst, frame_length);
164 if (ctx->chunk_count == 1) {
165 /* Write a simple header */
166 hap_write_section_header(&pbc, HAP_HDR_LONG, frame_length - 8,
167 ctx->chunks[0].compressor | ctx->opt_tex_fmt);
168 } else {
169 /* Write a complex header with Decode Instructions Container */
170 hap_write_section_header(&pbc, HAP_HDR_LONG, frame_length - 8,
171 HAP_COMP_COMPLEX | ctx->opt_tex_fmt);
172 hap_write_section_header(&pbc, HAP_HDR_SHORT, hap_decode_instructions_length(ctx),
173 HAP_ST_DECODE_INSTRUCTIONS);
174 hap_write_section_header(&pbc, HAP_HDR_SHORT, ctx->chunk_count,
175 HAP_ST_COMPRESSOR_TABLE);
176
177 for (i = 0; i < ctx->chunk_count; i++) {
178 bytestream2_put_byte(&pbc, ctx->chunks[i].compressor >> 4);
179 }
180
181 hap_write_section_header(&pbc, HAP_HDR_SHORT, ctx->chunk_count * 4,
182 HAP_ST_SIZE_TABLE);
183
184 for (i = 0; i < ctx->chunk_count; i++) {
185 bytestream2_put_le32(&pbc, ctx->chunks[i].compressed_size);
186 }
187 }
188}
189
7ca3e520
VG
190static int hap_encode(AVCodecContext *avctx, AVPacket *pkt,
191 const AVFrame *frame, int *got_packet)
192{
193 HapContext *ctx = avctx->priv_data;
3ee21785
TB
194 int header_length = hap_header_length(ctx);
195 int final_data_size, ret;
196 int pktsize = FFMAX(ctx->tex_size, ctx->max_snappy * ctx->chunk_count) + header_length;
7ca3e520
VG
197
198 /* Allocate maximum size packet, shrink later. */
199 ret = ff_alloc_packet(pkt, pktsize);
200 if (ret < 0)
201 return ret;
202
203 /* DXTC compression. */
204 compress_texture(avctx, frame);
205
3ee21785
TB
206 /* Compress (using Snappy) the frame */
207 final_data_size = hap_compress_frame(avctx, pkt->data + header_length);
208 if (final_data_size < 0)
209 return final_data_size;
7ca3e520
VG
210
211 /* Write header at the start. */
3ee21785 212 hap_write_frame_header(ctx, pkt->data, final_data_size + header_length);
7ca3e520 213
3ee21785 214 av_shrink_packet(pkt, final_data_size + header_length);
7ca3e520
VG
215 pkt->flags |= AV_PKT_FLAG_KEY;
216 *got_packet = 1;
217 return 0;
218}
219
220static av_cold int hap_init(AVCodecContext *avctx)
221{
222 HapContext *ctx = avctx->priv_data;
223 int ratio;
3ee21785 224 int corrected_chunk_count;
7ca3e520
VG
225 int ret = av_image_check_size(avctx->width, avctx->height, 0, avctx);
226
227 if (ret < 0) {
228 av_log(avctx, AV_LOG_ERROR, "Invalid video size %dx%d.\n",
229 avctx->width, avctx->height);
230 return ret;
231 }
232
233 if (avctx->width % 4 || avctx->height % 4) {
234 av_log(avctx, AV_LOG_ERROR, "Video size %dx%d is not multiple of 4.\n",
235 avctx->width, avctx->height);
236 return AVERROR_INVALIDDATA;
237 }
238
239 ff_texturedspenc_init(&ctx->dxtc);
240
3ee21785 241 switch (ctx->opt_tex_fmt) {
7ca3e520
VG
242 case HAP_FMT_RGBDXT1:
243 ratio = 8;
244 avctx->codec_tag = MKTAG('H', 'a', 'p', '1');
9f5d6f46 245 avctx->bits_per_coded_sample = 24;
7ca3e520
VG
246 ctx->tex_fun = ctx->dxtc.dxt1_block;
247 break;
248 case HAP_FMT_RGBADXT5:
249 ratio = 4;
250 avctx->codec_tag = MKTAG('H', 'a', 'p', '5');
9f5d6f46 251 avctx->bits_per_coded_sample = 32;
7ca3e520
VG
252 ctx->tex_fun = ctx->dxtc.dxt5_block;
253 break;
254 case HAP_FMT_YCOCGDXT5:
255 ratio = 4;
256 avctx->codec_tag = MKTAG('H', 'a', 'p', 'Y');
9f5d6f46 257 avctx->bits_per_coded_sample = 24;
7ca3e520
VG
258 ctx->tex_fun = ctx->dxtc.dxt5ys_block;
259 break;
260 default:
3ee21785 261 av_log(avctx, AV_LOG_ERROR, "Invalid format %02X\n", ctx->opt_tex_fmt);
7ca3e520
VG
262 return AVERROR_INVALIDDATA;
263 }
264
265 /* Texture compression ratio is constant, so can we computer
266 * beforehand the final size of the uncompressed buffer. */
267 ctx->tex_size = FFALIGN(avctx->width, TEXTURE_BLOCK_W) *
268 FFALIGN(avctx->height, TEXTURE_BLOCK_H) * 4 / ratio;
3ee21785
TB
269
270 /* Round the chunk count to divide evenly on DXT block edges */
271 corrected_chunk_count = av_clip(ctx->opt_chunk_count, 1, HAP_MAX_CHUNKS);
272 while ((ctx->tex_size / (64 / ratio)) % corrected_chunk_count != 0) {
273 corrected_chunk_count--;
274 }
275 if (corrected_chunk_count != ctx->opt_chunk_count) {
276 av_log(avctx, AV_LOG_INFO, "%d chunks requested but %d used.\n",
277 ctx->opt_chunk_count, corrected_chunk_count);
278 }
279 ret = ff_hap_set_chunk_count(ctx, corrected_chunk_count, 1);
280 if (ret != 0)
281 return ret;
282
283 ctx->max_snappy = snappy_max_compressed_length(ctx->tex_size / corrected_chunk_count);
7ca3e520
VG
284
285 ctx->tex_buf = av_malloc(ctx->tex_size);
286 if (!ctx->tex_buf)
287 return AVERROR(ENOMEM);
288
289 return 0;
290}
291
292static av_cold int hap_close(AVCodecContext *avctx)
293{
294 HapContext *ctx = avctx->priv_data;
295
3ee21785 296 ff_hap_free_context(ctx);
7ca3e520
VG
297
298 return 0;
299}
300
301#define OFFSET(x) offsetof(HapContext, x)
302#define FLAGS AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
303static const AVOption options[] = {
3ee21785 304 { "format", NULL, OFFSET(opt_tex_fmt), AV_OPT_TYPE_INT, { .i64 = HAP_FMT_RGBDXT1 }, HAP_FMT_RGBDXT1, HAP_FMT_YCOCGDXT5, FLAGS, "format" },
7ca3e520
VG
305 { "hap", "Hap 1 (DXT1 textures)", 0, AV_OPT_TYPE_CONST, { .i64 = HAP_FMT_RGBDXT1 }, 0, 0, FLAGS, "format" },
306 { "hap_alpha", "Hap Alpha (DXT5 textures)", 0, AV_OPT_TYPE_CONST, { .i64 = HAP_FMT_RGBADXT5 }, 0, 0, FLAGS, "format" },
307 { "hap_q", "Hap Q (DXT5-YCoCg textures)", 0, AV_OPT_TYPE_CONST, { .i64 = HAP_FMT_YCOCGDXT5 }, 0, 0, FLAGS, "format" },
3ee21785 308 { "chunks", "chunk count", OFFSET(opt_chunk_count), AV_OPT_TYPE_INT, {.i64 = 1 }, 1, HAP_MAX_CHUNKS, FLAGS, },
7ca3e520
VG
309 { NULL },
310};
311
312static const AVClass hapenc_class = {
313 .class_name = "Hap encoder",
314 .item_name = av_default_item_name,
315 .option = options,
316 .version = LIBAVUTIL_VERSION_INT,
317};
318
319AVCodec ff_hap_encoder = {
320 .name = "hap",
321 .long_name = NULL_IF_CONFIG_SMALL("Vidvox Hap encoder"),
322 .type = AVMEDIA_TYPE_VIDEO,
323 .id = AV_CODEC_ID_HAP,
324 .priv_data_size = sizeof(HapContext),
325 .priv_class = &hapenc_class,
326 .init = hap_init,
327 .encode2 = hap_encode,
328 .close = hap_close,
329 .pix_fmts = (const enum AVPixelFormat[]) {
330 AV_PIX_FMT_RGBA, AV_PIX_FMT_NONE,
331 },
332 .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE |
333 FF_CODEC_CAP_INIT_CLEANUP,
334};