pineapple/externals/ffmpeg/libavformat/idcin.c
2021-02-09 04:25:58 +01:00

378 lines
13 KiB
C
Executable file

/*
* id Quake II CIN File Demuxer
* Copyright (c) 2003 The FFmpeg project
*
* This file is part of FFmpeg.
*
* FFmpeg is free software; you can redistribute it and/or
* 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.
*
* FFmpeg is distributed in the hope that it will be useful,
* 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
* License along with FFmpeg; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
/**
* @file
* id Quake II CIN file demuxer by Mike Melanson (melanson@pcisys.net)
* For more information about the id CIN format, visit:
* http://www.csse.monash.edu.au/~timf/
*
* CIN is a somewhat quirky and ill-defined format. Here are some notes
* for anyone trying to understand the technical details of this format:
*
* The format has no definite file signature. This is problematic for a
* general-purpose media player that wants to automatically detect file
* types. However, a CIN file does start with 5 32-bit numbers that
* specify audio and video parameters. This demuxer gets around the lack
* of file signature by performing sanity checks on those parameters.
* Probabilistically, this is a reasonable solution since the number of
* valid combinations of the 5 parameters is a very small subset of the
* total 160-bit number space.
*
* Refer to the function idcin_probe() for the precise A/V parameters
* that this demuxer allows.
*
* Next, each audio and video frame has a duration of 1/14 sec. If the
* audio sample rate is a multiple of the common frequency 22050 Hz it will
* divide evenly by 14. However, if the sample rate is 11025 Hz:
* 11025 (samples/sec) / 14 (frames/sec) = 787.5 (samples/frame)
* The way the CIN stores audio in this case is by storing 787 sample
* frames in the first audio frame and 788 sample frames in the second
* audio frame. Therefore, the total number of bytes in an audio frame
* is given as:
* audio frame #0: 787 * (bytes/sample) * (# channels) bytes in frame
* audio frame #1: 788 * (bytes/sample) * (# channels) bytes in frame
* audio frame #2: 787 * (bytes/sample) * (# channels) bytes in frame
* audio frame #3: 788 * (bytes/sample) * (# channels) bytes in frame
*
* Finally, not all id CIN creation tools agree on the resolution of the
* color palette, apparently. Some creation tools specify red, green, and
* blue palette components in terms of 6-bit VGA color DAC values which
* range from 0..63. Other tools specify the RGB components as full 8-bit
* values that range from 0..255. Since there are no markers in the file to
* differentiate between the two variants, this demuxer uses the following
* heuristic:
* - load the 768 palette bytes from disk
* - assume that they will need to be shifted left by 2 bits to
* transform them from 6-bit values to 8-bit values
* - scan through all 768 palette bytes
* - if any bytes exceed 63, do not shift the bytes at all before
* transmitting them to the video decoder
*/
#include "libavutil/channel_layout.h"
#include "libavutil/imgutils.h"
#include "libavutil/intreadwrite.h"
#include "avformat.h"
#include "internal.h"
#define HUFFMAN_TABLE_SIZE (64 * 1024)
#define IDCIN_FPS 14
typedef struct IdcinDemuxContext {
int video_stream_index;
int audio_stream_index;
int audio_chunk_size1;
int audio_chunk_size2;
int block_align;
/* demux state variables */
int current_audio_chunk;
int next_chunk_is_video;
int audio_present;
int64_t first_pkt_pos;
} IdcinDemuxContext;
static int idcin_probe(const AVProbeData *p)
{
unsigned int number, sample_rate;
unsigned int w, h;
int i;
/*
* This is what you could call a "probabilistic" file check: id CIN
* files don't have a definite file signature. In lieu of such a marker,
* perform sanity checks on the 5 32-bit header fields:
* width, height: greater than 0, less than or equal to 1024
* audio sample rate: greater than or equal to 8000, less than or
* equal to 48000, or 0 for no audio
* audio sample width (bytes/sample): 0 for no audio, or 1 or 2
* audio channels: 0 for no audio, or 1 or 2
*/
/* check we have enough data to do all checks, otherwise the
0-padding may cause a wrong recognition */
if (p->buf_size < 20 + HUFFMAN_TABLE_SIZE + 12)
return 0;
/* check the video width */
w = AV_RL32(&p->buf[0]);
if ((w == 0) || (w > 1024))
return 0;
/* check the video height */
h = AV_RL32(&p->buf[4]);
if ((h == 0) || (h > 1024))
return 0;
/* check the audio sample rate */
sample_rate = AV_RL32(&p->buf[8]);
if (sample_rate && (sample_rate < 8000 || sample_rate > 48000))
return 0;
/* check the audio bytes/sample */
number = AV_RL32(&p->buf[12]);
if (number > 2 || sample_rate && !number)
return 0;
/* check the audio channels */
number = AV_RL32(&p->buf[16]);
if (number > 2 || sample_rate && !number)
return 0;
i = 20 + HUFFMAN_TABLE_SIZE;
if (AV_RL32(&p->buf[i]) == 1)
i += 768;
if (i+12 > p->buf_size || AV_RL32(&p->buf[i+8]) != w*h)
return 1;
/* return half certainty since this check is a bit sketchy */
return AVPROBE_SCORE_EXTENSION;
}
static int idcin_read_header(AVFormatContext *s)
{
AVIOContext *pb = s->pb;
IdcinDemuxContext *idcin = s->priv_data;
AVStream *st;
unsigned int width, height;
unsigned int sample_rate, bytes_per_sample, channels;
int ret;
/* get the 5 header parameters */
width = avio_rl32(pb);
height = avio_rl32(pb);
sample_rate = avio_rl32(pb);
bytes_per_sample = avio_rl32(pb);
channels = avio_rl32(pb);
if (s->pb->eof_reached) {
av_log(s, AV_LOG_ERROR, "incomplete header\n");
return s->pb->error ? s->pb->error : AVERROR_EOF;
}
if (av_image_check_size(width, height, 0, s) < 0)
return AVERROR_INVALIDDATA;
if (sample_rate > 0) {
if (sample_rate < 14 || sample_rate > INT_MAX) {
av_log(s, AV_LOG_ERROR, "invalid sample rate: %u\n", sample_rate);
return AVERROR_INVALIDDATA;
}
if (bytes_per_sample < 1 || bytes_per_sample > 2) {
av_log(s, AV_LOG_ERROR, "invalid bytes per sample: %u\n",
bytes_per_sample);
return AVERROR_INVALIDDATA;
}
if (channels < 1 || channels > 2) {
av_log(s, AV_LOG_ERROR, "invalid channels: %u\n", channels);
return AVERROR_INVALIDDATA;
}
idcin->audio_present = 1;
} else {
/* if sample rate is 0, assume no audio */
idcin->audio_present = 0;
}
st = avformat_new_stream(s, NULL);
if (!st)
return AVERROR(ENOMEM);
avpriv_set_pts_info(st, 33, 1, IDCIN_FPS);
st->start_time = 0;
idcin->video_stream_index = st->index;
st->codecpar->codec_type = AVMEDIA_TYPE_VIDEO;
st->codecpar->codec_id = AV_CODEC_ID_IDCIN;
st->codecpar->codec_tag = 0; /* no fourcc */
st->codecpar->width = width;
st->codecpar->height = height;
/* load up the Huffman tables into extradata */
if ((ret = ff_get_extradata(s, st->codecpar, pb, HUFFMAN_TABLE_SIZE)) < 0)
return ret;
if (idcin->audio_present) {
idcin->audio_present = 1;
st = avformat_new_stream(s, NULL);
if (!st)
return AVERROR(ENOMEM);
avpriv_set_pts_info(st, 63, 1, sample_rate);
st->start_time = 0;
idcin->audio_stream_index = st->index;
st->codecpar->codec_type = AVMEDIA_TYPE_AUDIO;
st->codecpar->codec_tag = 1;
st->codecpar->channels = channels;
st->codecpar->channel_layout = channels > 1 ? AV_CH_LAYOUT_STEREO :
AV_CH_LAYOUT_MONO;
st->codecpar->sample_rate = sample_rate;
st->codecpar->bits_per_coded_sample = bytes_per_sample * 8;
st->codecpar->bit_rate = sample_rate * bytes_per_sample * 8 * channels;
st->codecpar->block_align = idcin->block_align = bytes_per_sample * channels;
if (bytes_per_sample == 1)
st->codecpar->codec_id = AV_CODEC_ID_PCM_U8;
else
st->codecpar->codec_id = AV_CODEC_ID_PCM_S16LE;
if (sample_rate % 14 != 0) {
idcin->audio_chunk_size1 = (sample_rate / 14) *
bytes_per_sample * channels;
idcin->audio_chunk_size2 = (sample_rate / 14 + 1) *
bytes_per_sample * channels;
} else {
idcin->audio_chunk_size1 = idcin->audio_chunk_size2 =
(sample_rate / 14) * bytes_per_sample * channels;
}
idcin->current_audio_chunk = 0;
}
idcin->next_chunk_is_video = 1;
idcin->first_pkt_pos = avio_tell(s->pb);
return 0;
}
static int idcin_read_packet(AVFormatContext *s,
AVPacket *pkt)
{
int ret;
unsigned int command;
unsigned int chunk_size;
IdcinDemuxContext *idcin = s->priv_data;
AVIOContext *pb = s->pb;
int i;
int palette_scale;
unsigned char r, g, b;
unsigned char palette_buffer[768];
uint32_t palette[256];
if (avio_feof(s->pb))
return s->pb->error ? s->pb->error : AVERROR_EOF;
if (idcin->next_chunk_is_video) {
command = avio_rl32(pb);
if (command == 2) {
return AVERROR(EIO);
} else if (command == 1) {
/* trigger a palette change */
ret = avio_read(pb, palette_buffer, 768);
if (ret < 0) {
return ret;
} else if (ret != 768) {
av_log(s, AV_LOG_ERROR, "incomplete packet\n");
return AVERROR(EIO);
}
/* scale the palette as necessary */
palette_scale = 2;
for (i = 0; i < 768; i++)
if (palette_buffer[i] > 63) {
palette_scale = 0;
break;
}
for (i = 0; i < 256; i++) {
r = palette_buffer[i * 3 ] << palette_scale;
g = palette_buffer[i * 3 + 1] << palette_scale;
b = palette_buffer[i * 3 + 2] << palette_scale;
palette[i] = (0xFFU << 24) | (r << 16) | (g << 8) | (b);
if (palette_scale == 2)
palette[i] |= palette[i] >> 6 & 0x30303;
}
}
if (s->pb->eof_reached) {
av_log(s, AV_LOG_ERROR, "incomplete packet\n");
return s->pb->error ? s->pb->error : AVERROR_EOF;
}
chunk_size = avio_rl32(pb);
if (chunk_size < 4 || chunk_size > INT_MAX - 4) {
av_log(s, AV_LOG_ERROR, "invalid chunk size: %u\n", chunk_size);
return AVERROR_INVALIDDATA;
}
/* skip the number of decoded bytes (always equal to width * height) */
avio_skip(pb, 4);
chunk_size -= 4;
ret= av_get_packet(pb, pkt, chunk_size);
if (ret < 0)
return ret;
else if (ret != chunk_size) {
av_log(s, AV_LOG_ERROR, "incomplete packet\n");
return AVERROR(EIO);
}
if (command == 1) {
uint8_t *pal;
pal = av_packet_new_side_data(pkt, AV_PKT_DATA_PALETTE,
AVPALETTE_SIZE);
if (!pal) {
return AVERROR(ENOMEM);
}
memcpy(pal, palette, AVPALETTE_SIZE);
pkt->flags |= AV_PKT_FLAG_KEY;
}
pkt->stream_index = idcin->video_stream_index;
pkt->duration = 1;
} else {
/* send out the audio chunk */
if (idcin->current_audio_chunk)
chunk_size = idcin->audio_chunk_size2;
else
chunk_size = idcin->audio_chunk_size1;
ret= av_get_packet(pb, pkt, chunk_size);
if (ret < 0)
return ret;
pkt->stream_index = idcin->audio_stream_index;
pkt->duration = chunk_size / idcin->block_align;
idcin->current_audio_chunk ^= 1;
}
if (idcin->audio_present)
idcin->next_chunk_is_video ^= 1;
return 0;
}
static int idcin_read_seek(AVFormatContext *s, int stream_index,
int64_t timestamp, int flags)
{
IdcinDemuxContext *idcin = s->priv_data;
if (idcin->first_pkt_pos > 0) {
int64_t ret = avio_seek(s->pb, idcin->first_pkt_pos, SEEK_SET);
if (ret < 0)
return ret;
ff_update_cur_dts(s, s->streams[idcin->video_stream_index], 0);
idcin->next_chunk_is_video = 1;
idcin->current_audio_chunk = 0;
return 0;
}
return -1;
}
AVInputFormat ff_idcin_demuxer = {
.name = "idcin",
.long_name = NULL_IF_CONFIG_SMALL("id Cinematic"),
.priv_data_size = sizeof(IdcinDemuxContext),
.read_probe = idcin_probe,
.read_header = idcin_read_header,
.read_packet = idcin_read_packet,
.read_seek = idcin_read_seek,
.flags = AVFMT_NO_BYTE_SEEK,
};