hapenc.c 12.9 KB
Newer Older
1 2 3
/*
 * Vidvox Hap encoder
 * 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 10 11 12
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
13
 * FFmpeg is distributed in the hope that it will be useful,
14 15 16 17 18
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * 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 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

/**
 * @file
 * Hap encoder
 *
 * Fourcc: Hap1, Hap5, HapY
 *
 * https://github.com/Vidvox/hap/blob/master/documentation/HapVideoDRAFT.md
 */

#include <stdint.h>
#include "snappy-c.h"

#include "libavutil/frame.h"
#include "libavutil/imgutils.h"
#include "libavutil/intreadwrite.h"
#include "libavutil/opt.h"

#include "avcodec.h"
#include "bytestream.h"
#include "hap.h"
#include "internal.h"
#include "texturedsp.h"

46 47 48
#define HAP_MAX_CHUNKS 64

enum HapHeaderLength {
49 50 51 52
    /* Short header: four bytes with a 24 bit size value */
    HAP_HDR_SHORT = 4,
    /* Long header: eight bytes with a 32 bit size value */
    HAP_HDR_LONG = 8,
53
};
54

55
static int compress_texture(AVCodecContext *avctx, uint8_t *out, int out_length, const AVFrame *f)
56 57 58 59
{
    HapContext *ctx = avctx->priv_data;
    int i, j;

60 61 62
    if (ctx->tex_size > out_length)
        return AVERROR_BUFFER_TOO_SMALL;

63 64 65 66 67 68 69
    for (j = 0; j < avctx->height; j += 4) {
        for (i = 0; i < avctx->width; i += 4) {
            uint8_t *p = f->data[0] + i * 4 + j * f->linesize[0];
            const int step = ctx->tex_fun(out, f->linesize[0], p);
            out += step;
        }
    }
70 71

    return 0;
72 73
}

74 75 76 77 78 79 80 81 82 83
/* section_length does not include the header */
static void hap_write_section_header(PutByteContext *pbc,
                                     enum HapHeaderLength header_length,
                                     int section_length,
                                     enum HapSectionType section_type)
{
    /* The first three bytes are the length of the section (not including the
     * header) or zero if using an eight-byte header.
     * For an eight-byte header, the length is in the last four bytes.
     * The fourth byte stores the section type. */
84
    bytestream2_put_le24(pbc, header_length == HAP_HDR_LONG ? 0 : section_length);
85 86
    bytestream2_put_byte(pbc, section_type);

87
    if (header_length == HAP_HDR_LONG) {
88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124
        bytestream2_put_le32(pbc, section_length);
    }
}

static int hap_compress_frame(AVCodecContext *avctx, uint8_t *dst)
{
    HapContext *ctx = avctx->priv_data;
    int i, final_size = 0;

    for (i = 0; i < ctx->chunk_count; i++) {
        HapChunk *chunk = &ctx->chunks[i];
        uint8_t *chunk_src, *chunk_dst;
        int ret;

        if (i == 0) {
            chunk->compressed_offset = 0;
        } else {
            chunk->compressed_offset = ctx->chunks[i-1].compressed_offset
                                       + ctx->chunks[i-1].compressed_size;
        }
        chunk->uncompressed_size = ctx->tex_size / ctx->chunk_count;
        chunk->uncompressed_offset = i * chunk->uncompressed_size;
        chunk->compressed_size = ctx->max_snappy;
        chunk_src = ctx->tex_buf + chunk->uncompressed_offset;
        chunk_dst = dst + chunk->compressed_offset;

        /* Compress with snappy too, write directly on packet buffer. */
        ret = snappy_compress(chunk_src, chunk->uncompressed_size,
                              chunk_dst, &chunk->compressed_size);
        if (ret != SNAPPY_OK) {
            av_log(avctx, AV_LOG_ERROR, "Snappy compress error.\n");
            return AVERROR_BUG;
        }

        /* If there is no gain from snappy, just use the raw texture. */
        if (chunk->compressed_size >= chunk->uncompressed_size) {
            av_log(avctx, AV_LOG_VERBOSE,
125
                   "Snappy buffer bigger than uncompressed (%"SIZE_SPECIFIER" >= %"SIZE_SPECIFIER" bytes).\n",
126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141
                   chunk->compressed_size, chunk->uncompressed_size);
            memcpy(chunk_dst, chunk_src, chunk->uncompressed_size);
            chunk->compressor = HAP_COMP_NONE;
            chunk->compressed_size = chunk->uncompressed_size;
        } else {
            chunk->compressor = HAP_COMP_SNAPPY;
        }

        final_size += chunk->compressed_size;
    }

    return final_size;
}

static int hap_decode_instructions_length(HapContext *ctx)
{
142 143 144 145
    /*    Second-Stage Compressor Table (one byte per entry)
     *  + Chunk Size Table (four bytes per entry)
     *  + headers for both sections (short versions)
     *  = chunk_count + (4 * chunk_count) + 4 + 4 */
146 147 148 149 150 151
    return (5 * ctx->chunk_count) + 8;
}

static int hap_header_length(HapContext *ctx)
{
    /* Top section header (long version) */
152
    int length = HAP_HDR_LONG;
153 154 155

    if (ctx->chunk_count > 1) {
        /* Decode Instructions header (short) + Decode Instructions Container */
156
        length += HAP_HDR_SHORT + hap_decode_instructions_length(ctx);
157 158 159 160 161 162 163 164 165 166 167 168 169
    }

    return length;
}

static void hap_write_frame_header(HapContext *ctx, uint8_t *dst, int frame_length)
{
    PutByteContext pbc;
    int i;

    bytestream2_init_writer(&pbc, dst, frame_length);
    if (ctx->chunk_count == 1) {
        /* Write a simple header */
170
        hap_write_section_header(&pbc, HAP_HDR_LONG, frame_length - 8,
171 172 173
                                 ctx->chunks[0].compressor | ctx->opt_tex_fmt);
    } else {
        /* 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,
179 180 181 182 183 184
                                 HAP_ST_COMPRESSOR_TABLE);

        for (i = 0; i < ctx->chunk_count; i++) {
            bytestream2_put_byte(&pbc, ctx->chunks[i].compressor >> 4);
        }

185
        hap_write_section_header(&pbc, HAP_HDR_SHORT, ctx->chunk_count * 4,
186 187 188 189 190 191 192 193
                                 HAP_ST_SIZE_TABLE);

        for (i = 0; i < ctx->chunk_count; i++) {
            bytestream2_put_le32(&pbc, ctx->chunks[i].compressed_size);
        }
    }
}

194 195 196 197
static int hap_encode(AVCodecContext *avctx, AVPacket *pkt,
                      const AVFrame *frame, int *got_packet)
{
    HapContext *ctx = avctx->priv_data;
198 199 200
    int header_length = hap_header_length(ctx);
    int final_data_size, ret;
    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);
204 205 206
    if (ret < 0)
        return ret;

207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225
    if (ctx->opt_compressor == HAP_COMP_NONE) {
        /* DXTC compression directly to the packet buffer. */
        ret = compress_texture(avctx, pkt->data + header_length, pkt->size - header_length, frame);
        if (ret < 0)
            return ret;

        ctx->chunks[0].compressor = HAP_COMP_NONE;
        final_data_size = ctx->tex_size;
    } else {
        /* DXTC compression. */
        ret = compress_texture(avctx, ctx->tex_buf, ctx->tex_size, frame);
        if (ret < 0)
            return ret;

        /* Compress (using Snappy) the frame */
        final_data_size = hap_compress_frame(avctx, pkt->data + header_length);
        if (final_data_size < 0)
            return final_data_size;
    }
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);
231 232 233 234 235 236 237 238 239
    pkt->flags |= AV_PKT_FLAG_KEY;
    *got_packet = 1;
    return 0;
}

static av_cold int hap_init(AVCodecContext *avctx)
{
    HapContext *ctx = avctx->priv_data;
    int ratio;
240
    int corrected_chunk_count;
241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256
    int ret = av_image_check_size(avctx->width, avctx->height, 0, avctx);

    if (ret < 0) {
        av_log(avctx, AV_LOG_ERROR, "Invalid video size %dx%d.\n",
               avctx->width, avctx->height);
        return ret;
    }

    if (avctx->width % 4 || avctx->height % 4) {
        av_log(avctx, AV_LOG_ERROR, "Video size %dx%d is not multiple of 4.\n",
               avctx->width, avctx->height);
        return AVERROR_INVALIDDATA;
    }

    ff_texturedspenc_init(&ctx->dxtc);

257
    switch (ctx->opt_tex_fmt) {
258 259 260
    case HAP_FMT_RGBDXT1:
        ratio = 8;
        avctx->codec_tag = MKTAG('H', 'a', 'p', '1');
261
        avctx->bits_per_coded_sample = 24;
262 263 264 265 266
        ctx->tex_fun = ctx->dxtc.dxt1_block;
        break;
    case HAP_FMT_RGBADXT5:
        ratio = 4;
        avctx->codec_tag = MKTAG('H', 'a', 'p', '5');
267
        avctx->bits_per_coded_sample = 32;
268 269 270 271 272
        ctx->tex_fun = ctx->dxtc.dxt5_block;
        break;
    case HAP_FMT_YCOCGDXT5:
        ratio = 4;
        avctx->codec_tag = MKTAG('H', 'a', 'p', 'Y');
273
        avctx->bits_per_coded_sample = 24;
274 275 276
        ctx->tex_fun = ctx->dxtc.dxt5ys_block;
        break;
    default:
277
        av_log(avctx, AV_LOG_ERROR, "Invalid format %02X\n", ctx->opt_tex_fmt);
278 279 280 281 282 283 284
        return AVERROR_INVALIDDATA;
    }

    /* Texture compression ratio is constant, so can we computer
     * beforehand the final size of the uncompressed buffer. */
    ctx->tex_size   = FFALIGN(avctx->width,  TEXTURE_BLOCK_W) *
                      FFALIGN(avctx->height, TEXTURE_BLOCK_H) * 4 / ratio;
285

286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309
    switch (ctx->opt_compressor) {
    case HAP_COMP_NONE:
        /* No benefit chunking uncompressed data */
        corrected_chunk_count = 1;

        ctx->max_snappy = ctx->tex_size;
        ctx->tex_buf = NULL;
        break;
    case HAP_COMP_SNAPPY:
        /* Round the chunk count to divide evenly on DXT block edges */
        corrected_chunk_count = av_clip(ctx->opt_chunk_count, 1, HAP_MAX_CHUNKS);
        while ((ctx->tex_size / (64 / ratio)) % corrected_chunk_count != 0) {
            corrected_chunk_count--;
        }

        ctx->max_snappy = snappy_max_compressed_length(ctx->tex_size / corrected_chunk_count);
        ctx->tex_buf = av_malloc(ctx->tex_size);
        if (!ctx->tex_buf) {
            return AVERROR(ENOMEM);
        }
        break;
    default:
        av_log(avctx, AV_LOG_ERROR, "Invalid compresor %02X\n", ctx->opt_compressor);
        return AVERROR_INVALIDDATA;
310 311 312 313 314 315 316 317 318
    }
    if (corrected_chunk_count != ctx->opt_chunk_count) {
        av_log(avctx, AV_LOG_INFO, "%d chunks requested but %d used.\n",
                                    ctx->opt_chunk_count, corrected_chunk_count);
    }
    ret = ff_hap_set_chunk_count(ctx, corrected_chunk_count, 1);
    if (ret != 0)
        return ret;

319 320 321 322 323 324 325
    return 0;
}

static av_cold int hap_close(AVCodecContext *avctx)
{
    HapContext *ctx = avctx->priv_data;

326
    ff_hap_free_context(ctx);
327 328 329 330 331 332 333

    return 0;
}

#define OFFSET(x) offsetof(HapContext, x)
#define FLAGS     AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
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" },
335 336 337
        { "hap",       "Hap 1 (DXT1 textures)", 0, AV_OPT_TYPE_CONST, { .i64 = HAP_FMT_RGBDXT1   }, 0, 0, FLAGS, "format" },
        { "hap_alpha", "Hap Alpha (DXT5 textures)", 0, AV_OPT_TYPE_CONST, { .i64 = HAP_FMT_RGBADXT5  }, 0, 0, FLAGS, "format" },
        { "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, },
339 340 341
    { "compressor", "second-stage compressor", OFFSET(opt_compressor), AV_OPT_TYPE_INT, { .i64 = HAP_COMP_SNAPPY }, HAP_COMP_NONE, HAP_COMP_SNAPPY, FLAGS, "compressor" },
        { "none",       "None", 0, AV_OPT_TYPE_CONST, { .i64 = HAP_COMP_NONE }, 0, 0, FLAGS, "compressor" },
        { "snappy",     "Snappy", 0, AV_OPT_TYPE_CONST, { .i64 = HAP_COMP_SNAPPY }, 0, 0, FLAGS, "compressor" },
342 343 344 345 346 347 348 349 350 351 352 353
    { NULL },
};

static const AVClass hapenc_class = {
    .class_name = "Hap encoder",
    .item_name  = av_default_item_name,
    .option     = options,
    .version    = LIBAVUTIL_VERSION_INT,
};

AVCodec ff_hap_encoder = {
    .name           = "hap",
354
    .long_name      = NULL_IF_CONFIG_SMALL("Vidvox Hap"),
355 356 357 358 359 360 361 362 363 364 365 366 367
    .type           = AVMEDIA_TYPE_VIDEO,
    .id             = AV_CODEC_ID_HAP,
    .priv_data_size = sizeof(HapContext),
    .priv_class     = &hapenc_class,
    .init           = hap_init,
    .encode2        = hap_encode,
    .close          = hap_close,
    .pix_fmts       = (const enum AVPixelFormat[]) {
        AV_PIX_FMT_RGBA, AV_PIX_FMT_NONE,
    },
    .caps_internal  = FF_CODEC_CAP_INIT_THREADSAFE |
                      FF_CODEC_CAP_INIT_CLEANUP,
};