1
0
Fork 0
forked from fte/fteqw
fteqw/plugins/avplug/avaudio.c
Spoike 7dc01f7362 android port updated. egl now handled by native code, which means we now have proper control over everything and can default to gles2. requires android 2.0+. vulkan-on-android renderer added, but not tested.
added .ktx image support as an alternative to .dds, primarily for etc2 compression (which should work on all gles3 devices, or gl4.3), only known representations will work.
rework z-fighting workaround. now only enabled on known vanilla maps.
added splitscreen option to the singleplayer menu.

git-svn-id: https://svn.code.sf.net/p/fteqw/code/trunk@5154 fc73d0e0-1445-4013-8a0c-d673dee63da5
2017-10-13 17:50:28 +00:00

398 lines
10 KiB
C

#include "../plugin.h"
#include "../engine.h"
#include "libavcodec/avcodec.h"
#include "libavformat/avformat.h"
static cvar_t *ffmpeg_audiodecoder;
struct avaudioctx
{
//raw file
uint8_t *filedata;
size_t fileofs;
size_t filesize;
//avformat stuff
AVFormatContext *pFormatCtx;
int audioStream;
AVCodecContext *pACodecCtx;
AVFrame *pAFrame;
//decoding
int64_t lasttime;
//output audio
//we throw away data if the format changes. which is awkward, but gah.
int64_t samples_start;
int samples_channels;
int samples_speed;
int samples_width;
qbyte *samples_buffer;
size_t samples_count;
size_t samples_max;
};
static void S_AV_Purge(sfx_t *s)
{
struct avaudioctx *ctx = (struct avaudioctx*)s->decoder.buf;
s->loadstate = SLS_NOTLOADED;
// Free the audio decoder
if (ctx->pACodecCtx)
avcodec_close(ctx->pACodecCtx);
av_free(ctx->pAFrame);
// Close the video file
avformat_close_input(&ctx->pFormatCtx);
//free the decoded buffer
free(ctx->samples_buffer);
//file storage will be cleared here too
free(ctx);
memset(&s->decoder, 0, sizeof(s->decoder));
}
static sfxcache_t *S_AV_Locate(sfx_t *sfx, sfxcache_t *buf, ssamplepos_t start, int length)
{ //warning: can be called on a different thread.
struct avaudioctx *ctx = (struct avaudioctx*)sfx->decoder.buf;
AVPacket packet;
int64_t curtime;
if (!buf)
return NULL;
curtime = start + length;
// curtime = (mediatime * ctx->denum) / ctx->num;
while (1)
{
if (ctx->lasttime > curtime)
break;
// We're ahead of the previous frame. try and read the next.
if (av_read_frame(ctx->pFormatCtx, &packet) < 0)
break;
// Is this a packet from the video stream?
if(packet.stream_index==ctx->audioStream)
{
int okay;
int len;
void *odata = packet.data;
while (packet.size > 0)
{
okay = false;
len = avcodec_decode_audio4(ctx->pACodecCtx, ctx->pAFrame, &okay, &packet);
if (len < 0)
break;
packet.size -= len;
packet.data += len;
if (okay)
{
int width = 2;
int channels = ctx->pACodecCtx->channels;
unsigned int auddatasize = av_samples_get_buffer_size(NULL, ctx->pACodecCtx->channels, ctx->pAFrame->nb_samples, ctx->pACodecCtx->sample_fmt, 1);
void *auddata = ctx->pAFrame->data[0];
switch(ctx->pACodecCtx->sample_fmt)
{ //we don't support planar audio. we just treat it as mono instead.
default:
auddatasize = 0;
break;
case AV_SAMPLE_FMT_U8P:
auddatasize /= channels;
channels = 1;
case AV_SAMPLE_FMT_U8:
width = 1;
break;
case AV_SAMPLE_FMT_S16P:
auddatasize /= channels;
channels = 1;
case AV_SAMPLE_FMT_S16:
width = 2;
break;
case AV_SAMPLE_FMT_FLTP:
auddatasize /= channels;
channels = 1;
case AV_SAMPLE_FMT_FLT:
//FIXME: support float audio internally.
{
float *in = (void*)auddata;
signed short *out = (void*)auddata;
int v;
unsigned int i;
for (i = 0; i < auddatasize/sizeof(*in); i++)
{
v = (short)(in[i]*32767);
if (v < -32767)
v = -32767;
else if (v > 32767)
v = 32767;
out[i] = v;
}
auddatasize/=2;
width = 2;
}
case AV_SAMPLE_FMT_DBLP:
auddatasize /= channels;
channels = 1;
case AV_SAMPLE_FMT_DBL:
{
double *in = (double*)auddata;
signed short *out = (void*)auddata;
int v;
unsigned int i;
for (i = 0; i < auddatasize/sizeof(*in); i++)
{
v = (short)(in[i]*32767);
if (v < -32767)
v = -32767;
else if (v > 32767)
v = 32767;
out[i] = v;
}
auddatasize/=4;
width = 2;
}
break;
}
if (ctx->samples_channels != channels || ctx->samples_speed != ctx->pACodecCtx->sample_rate || ctx->samples_width != width)
{ //something changed, update
ctx->samples_channels = channels;
ctx->samples_speed = ctx->pACodecCtx->sample_rate;
ctx->samples_width = width;
//and discard any decoded audio. this might loose some.
ctx->samples_start += ctx->samples_count;
ctx->samples_count = 0;
}
if (ctx->samples_max < (ctx->samples_count*ctx->samples_width*ctx->samples_channels)+auddatasize)
{
ctx->samples_max = (ctx->samples_count*ctx->samples_width*ctx->samples_channels)+auddatasize;
ctx->samples_max *= 2; //slop
ctx->samples_buffer = realloc(ctx->samples_buffer, ctx->samples_max);
}
if (width == 1)
{ //FTE uses signed 8bit audio. ffmpeg uses unsigned 8bit audio. *sigh*.
char *out = (char*)(ctx->samples_buffer + ctx->samples_count*(ctx->samples_width*ctx->samples_channels));
unsigned char *in = auddata;
int i;
for (i = 0; i < auddatasize; i++)
out[i] = in[i]-128;
}
else
memcpy(ctx->samples_buffer + ctx->samples_count*(ctx->samples_width*ctx->samples_channels), auddata, auddatasize);
ctx->samples_count += auddatasize/(ctx->samples_width*ctx->samples_channels);
}
}
packet.data = odata;
}
// Free the packet that was allocated by av_read_frame
av_packet_unref(&packet);
}
buf->length = ctx->samples_count;
buf->speed = ctx->samples_speed;
buf->width = ctx->samples_width;
buf->numchannels = ctx->samples_channels;
buf->soundoffset = ctx->samples_start;
buf->data = ctx->samples_buffer;
//if we couldn't return any new data, then we're at an eof, return NULL to signal that.
if (start == buf->soundoffset + buf->length && length > 0)
return NULL;
return buf;
}
static float S_AV_Query(struct sfx_s *sfx, struct sfxcache_s *buf, char *title, size_t titlesize)
{
struct avaudioctx *ctx = (struct avaudioctx*)sfx->decoder.buf;
if (!ctx)
return -1;
if (buf)
{
buf->data = NULL;
buf->soundoffset = 0;
buf->length = 0;
buf->numchannels = ctx->samples_channels;
buf->speed = ctx->samples_speed;
buf->width = ctx->samples_width;
}
return ctx->pFormatCtx->duration / (float)AV_TIME_BASE;
}
static int AVIO_Mem_Read(void *opaque, uint8_t *buf, int buf_size)
{
struct avaudioctx *ctx = opaque;
if (ctx->fileofs > ctx->filesize)
buf_size = 0;
if (buf_size > ctx->filesize-ctx->fileofs)
buf_size = ctx->filesize-ctx->fileofs;
if (buf_size > 0)
{
memcpy(buf, ctx->filedata + ctx->fileofs, buf_size);
ctx->fileofs += buf_size;
return buf_size;
}
return 0;
}
static int64_t AVIO_Mem_Seek(void *opaque, int64_t offset, int whence)
{
struct avaudioctx *ctx = opaque;
whence &= ~AVSEEK_FORCE;
switch(whence)
{
default:
return -1;
case SEEK_SET:
ctx->fileofs = offset;
break;
case SEEK_CUR:
ctx->fileofs += offset;
break;
case SEEK_END:
ctx->fileofs = ctx->filesize + offset;
break;
case AVSEEK_SIZE:
return ctx->filesize;
}
if (ctx->fileofs < 0)
ctx->fileofs = 0;
return ctx->fileofs;
}
/*const char *COM_GetFileExtension (const char *in)
{
const char *dot;
for (dot = in + strlen(in); dot >= in && *dot != '.'; dot--)
;
if (dot < in)
return "";
in = dot+1;
return in;
}*/
static qboolean QDECL S_LoadAVSound (sfx_t *s, qbyte *data, size_t datalen, int sndspeed)
{
struct avaudioctx *ctx;
int i;
AVCodec *pCodec;
const int iBufSize = 4 * 1024;
if (!ffmpeg_audiodecoder)
ffmpeg_audiodecoder = pCvar_GetNVFDG("ffmpeg_audiodecoder_wip", "0", 0, "Enables the use of ffmpeg's decoder for pure audio files.", "ffmpeg");
if (!ffmpeg_audiodecoder->value /* && *ffmpeg_audiodecoder.string */)
return false;
if (!data || !datalen)
return false;
if (datalen >= 4 && !strncmp(data, "RIFF", 4))
return false; //ignore it if it looks like a wav file. that means we don't need to figure out how to calculate loopstart.
// if (strcasecmp(COM_GetFileExtension(s->name), "wav")) //don't do .wav - I've no idea how to read the loopstart tag with ffmpeg.
// return false;
s->decoder.buf = ctx = malloc(sizeof(*ctx) + datalen);
if (!ctx)
return false; //o.O
memset(ctx, 0, sizeof(*ctx));
// Create internal io buffer for FFmpeg
ctx->filedata = data; //defer that copy
ctx->filesize = datalen; //defer that copy
ctx->pFormatCtx = avformat_alloc_context();
ctx->pFormatCtx->pb = avio_alloc_context(av_malloc(iBufSize), iBufSize, 0, ctx, AVIO_Mem_Read, 0, AVIO_Mem_Seek);
// Open file
if(avformat_open_input(&ctx->pFormatCtx, s->name, NULL, NULL)==0)
{
// Retrieve stream information
if(avformat_find_stream_info(ctx->pFormatCtx, NULL)>=0)
{
ctx->audioStream=-1;
for(i=0; i<ctx->pFormatCtx->nb_streams; i++)
if(ctx->pFormatCtx->streams[i]->codec->codec_type==AVMEDIA_TYPE_AUDIO)
{
ctx->audioStream=i;
break;
}
if(ctx->audioStream!=-1)
{
ctx->pACodecCtx=ctx->pFormatCtx->streams[ctx->audioStream]->codec;
pCodec=avcodec_find_decoder(ctx->pACodecCtx->codec_id);
ctx->pAFrame=av_frame_alloc();
if(pCodec!=NULL && ctx->pAFrame && avcodec_open2(ctx->pACodecCtx, pCodec, NULL) >= 0)
{ //success
}
else
ctx->audioStream = -1;
}
}
if (ctx->audioStream != -1)
{
//sucky copy
ctx->filedata = (uint8_t*)(ctx+1);
memcpy(ctx->filedata, data, datalen);
s->decoder.ended = S_AV_Purge;
s->decoder.purge = S_AV_Purge;
s->decoder.decodedata = S_AV_Locate;
s->decoder.querydata = S_AV_Query;
return true;
}
}
S_AV_Purge(s);
return false;
}
static qboolean AVAudio_Init(void)
{
if (!pPlug_ExportNative("S_LoadSound", S_LoadAVSound))
{
Con_Printf("avplug: Engine doesn't support audio decoder plugins\n");
return false;
}
return true;
}
//generic module stuff. this has to go somewhere.
static void AVLogCallback(void *avcl, int level, const char *fmt, va_list vl)
{ //needs to be reenterant
#ifdef _DEBUG
char string[1024];
Q_vsnprintf (string, sizeof(string), fmt, vl);
pCon_Print(string);
#endif
}
//get the encoder/decoders to register themselves with the engine, then make sure avformat/avcodec have registered all they have to give.
qboolean AVEnc_Init(void);
qboolean AVDec_Init(void);
qintptr_t Plug_Init(qintptr_t *args)
{
qboolean okay = false;
okay |= AVAudio_Init();
okay |= AVDec_Init();
okay |= AVEnc_Init();
if (okay)
{
av_register_all();
avcodec_register_all();
av_log_set_level(AV_LOG_WARNING);
av_log_set_callback(AVLogCallback);
}
return okay;
}