kolibrios-fun/programs/develop/libraries/libmpg123/frame.h

411 lines
12 KiB
C
Raw Normal View History

/*
frame: Central data structures and opmitization hooks.
copyright 2007 by the mpg123 project - free software under the terms of the LGPL 2.1
see COPYING and AUTHORS files in distribution or http://mpg123.org
initially written by Thomas Orgis
*/
#ifndef MPG123_FRAME_H
#define MPG123_FRAME_H
#include <stdio.h>
#include "config.h"
#include "mpg123.h"
#include "optimize.h"
#include "id3.h"
#include "icy.h"
#include "reader.h"
#ifdef FRAME_INDEX
#include "index.h"
#endif
#include "synths.h"
#ifdef OPT_DITHER
#include "dither.h"
int frame_dither_init(mpg123_handle *fr);
#endif
/* max = 1728 */
#define MAXFRAMESIZE 3456
struct al_table
{
short bits;
short d;
};
/* the output buffer, used to be pcm_sample, pcm_point and audiobufsize */
struct outbuffer
{
unsigned char *data; /* main data pointer, aligned */
unsigned char *p; /* read pointer */
size_t fill; /* fill from read pointer */
size_t size;
unsigned char *rdata; /* unaligned base pointer */
};
struct audioformat
{
int encoding;
int encsize; /* Size of one sample in bytes, plain int should be fine here... */
int channels;
long rate;
};
void invalidate_format(struct audioformat *af);
struct mpg123_pars_struct
{
int verbose; /* verbose level */
long flags; /* combination of above */
#ifndef NO_NTOM
long force_rate;
#endif
int down_sample;
int rva; /* (which) rva to do: 0: nothing, 1: radio/mix/track 2: album/audiophile */
long halfspeed;
long doublespeed;
long timeout;
#define NUM_CHANNELS 2
char audio_caps[NUM_CHANNELS][MPG123_RATES+1][MPG123_ENCODINGS];
/* long start_frame; */ /* frame offset to begin with */
/* long frame_number;*/ /* number of frames to decode */
#ifndef NO_ICY
long icy_interval;
#endif
double outscale;
long resync_limit;
long index_size; /* Long, because: negative values have a meaning. */
long preframes;
#ifndef NO_FEEDER
long feedpool;
long feedbuffer;
#endif
};
enum frame_state_flags
{
FRAME_ACCURATE = 0x1 /**< 0001 Positions are considered accurate. */
,FRAME_FRANKENSTEIN = 0x2 /**< 0010 This stream is concatenated. */
};
/* There is a lot to condense here... many ints can be merged as flags; though the main space is still consumed by buffers. */
struct mpg123_handle_struct
{
int fresh; /* to be moved into flags */
int new_format;
real hybrid_block[2][2][SBLIMIT*SSLIMIT];
int hybrid_blc[2];
/* the scratch vars for the decoders, sometimes real, sometimes short... sometimes int/long */
short *short_buffs[2][2];
real *real_buffs[2][2];
unsigned char *rawbuffs;
int rawbuffss;
#ifdef OPT_I486
int i486bo[2];
#endif
int bo; /* Just have it always here. */
#ifdef OPT_DITHER
int ditherindex;
float *dithernoise;
#endif
unsigned char* rawdecwin; /* the block with all decwins */
int rawdecwins; /* size of rawdecwin memory */
real *decwin; /* _the_ decode table */
#ifdef OPT_MMXORSSE
/* I am not really sure that I need both of them... used in assembler */
float *decwin_mmx;
float *decwins;
#endif
int have_eq_settings;
real equalizer[2][32];
/* for halfspeed mode */
unsigned char ssave[34];
int halfphase;
#ifndef NO_8BIT
/* a raw buffer and a pointer into the middle for signed short conversion, only allocated on demand */
unsigned char *conv16to8_buf;
unsigned char *conv16to8;
#endif
/* There's some possible memory saving for stuff that is not _really_ dynamic. */
/* layer3 */
int longLimit[9][23];
int shortLimit[9][14];
real gainpow2[256+118+4]; /* not really dynamic, just different for mmx */
/* layer2 */
real muls[27][64]; /* also used by layer 1 */
#ifndef NO_NTOM
/* decode_ntom */
unsigned long ntom_val[2];
unsigned long ntom_step;
#endif
/* special i486 fun */
#ifdef OPT_I486
int *int_buffs[2][2];
#endif
/* special altivec... */
#ifdef OPT_ALTIVEC
real *areal_buffs[4][4];
#endif
struct synth_s synths;
struct
{
#ifdef OPT_MULTI
#ifndef NO_LAYER3
#if (defined OPT_3DNOW || defined OPT_3DNOWEXT)
void (*the_dct36)(real *,real *,real *,real *,real *);
#endif
#endif
#endif
enum optdec type;
enum optcla class;
} cpu_opts;
int verbose; /* 0: nothing, 1: just print chosen decoder, 2: be verbose */
const struct al_table *alloc;
/* The runtime-chosen decoding, based on input and output format. */
func_synth synth;
func_synth_stereo synth_stereo;
func_synth_mono synth_mono;
/* Yes, this function is runtime-switched, too. */
void (*make_decode_tables)(mpg123_handle *fr); /* That is the volume control. */
int stereo; /* I _think_ 1 for mono and 2 for stereo */
int jsbound;
#define SINGLE_STEREO -1
#define SINGLE_LEFT 0
#define SINGLE_RIGHT 1
#define SINGLE_MIX 3
int single;
int II_sblimit;
int down_sample_sblimit;
int lsf; /* 0: MPEG 1.0; 1: MPEG 2.0/2.5 -- both used as bool and array index! */
/* Many flags in disguise as integers... wasting bytes. */
int mpeg25;
int down_sample;
int header_change;
int lay;
int (*do_layer)(mpg123_handle *);
int error_protection;
int bitrate_index;
int sampling_frequency;
int padding;
int extension;
int mode;
int mode_ext;
int copyright;
int original;
int emphasis;
int framesize; /* computed framesize */
int freesize; /* free format frame size */
enum mpg123_vbr vbr; /* 1 if variable bitrate was detected */
off_t num; /* frame offset ... */
off_t input_offset; /* byte offset of this frame in input stream */
off_t playnum; /* playback offset... includes repetitions, reset at seeks */
off_t audio_start; /* The byte offset in the file where audio data begins. */
int state_flags;
char silent_resync; /* Do not complain for the next n resyncs. */
unsigned char* xing_toc; /* The seek TOC from Xing header. */
int freeformat;
long freeformat_framesize;
/* bitstream info; bsi */
int bitindex;
unsigned char *wordpointer;
/* temporary storage for getbits stuff */
unsigned long ultmp;
unsigned char uctmp;
/* rva data, used in common.c, set in id3.c */
double maxoutburst; /* The maximum amplitude in current sample represenation. */
double lastscale;
struct
{
int level[2];
float gain[2];
float peak[2];
} rva;
/* input data */
off_t track_frames;
off_t track_samples;
double mean_framesize;
off_t mean_frames;
int fsizeold;
int ssize;
unsigned int bitreservoir;
unsigned char bsspace[2][MAXFRAMESIZE+512]; /* MAXFRAMESIZE */
unsigned char *bsbuf;
unsigned char *bsbufold;
int bsnum;
/* That is the header matching the last read frame body. */
unsigned long oldhead;
/* That is the header that is supposedly the first of the stream. */
unsigned long firsthead;
int abr_rate;
#ifdef FRAME_INDEX
struct frame_index index;
#endif
/* output data */
struct outbuffer buffer;
struct audioformat af;
int own_buffer;
size_t outblock; /* number of bytes that this frame produces (upper bound) */
int to_decode; /* this frame holds data to be decoded */
int to_ignore; /* the same, somehow */
off_t firstframe; /* start decoding from here */
off_t lastframe; /* last frame to decode (for gapless or num_frames limit) */
off_t ignoreframe; /* frames to decode but discard before firstframe */
#ifdef GAPLESS
off_t gapless_frames; /* frame count for the gapless part */
off_t firstoff; /* number of samples to ignore from firstframe */
off_t lastoff; /* number of samples to use from lastframe */
off_t begin_s; /* overall begin offset in samples */
off_t begin_os;
off_t end_s; /* overall end offset in samples */
off_t end_os;
off_t fullend_os; /* gapless_frames translated to output samples */
#endif
unsigned int crc; /* Well, I need a safe 16bit type, actually. But wider doesn't hurt. */
struct reader *rd; /* pointer to the reading functions */
struct reader_data rdat; /* reader data and state info */
struct mpg123_pars_struct p;
int err;
int decoder_change;
int delayed_change;
long clip;
/* the meta crap */
int metaflags;
unsigned char id3buf[128];
#ifndef NO_ID3V2
mpg123_id3v2 id3v2;
#endif
#ifndef NO_ICY
struct icy_meta icy;
#endif
/*
More variables needed for decoders, layerX.c.
This time it is not about static variables but about the need for alignment which cannot be guaranteed on the stack by certain compilers (Sun Studio).
We do not require the compiler to align stuff for our hand-written assembly. We only hope that it's able to align stuff for SSE and similar ops it generates itself.
*/
/*
Those layer-specific structs could actually share memory, as they are not in use simultaneously. One might allocate on decoder switch, too.
They all reside in one lump of memory (after each other), allocated to layerscratch.
*/
real *layerscratch;
#ifndef NO_LAYER1
struct
{
real (*fraction)[SBLIMIT]; /* ALIGNED(16) real fraction[2][SBLIMIT]; */
} layer1;
#endif
#ifndef NO_LAYER2
struct
{
real (*fraction)[4][SBLIMIT]; /* ALIGNED(16) real fraction[2][4][SBLIMIT] */
} layer2;
#endif
#ifndef NO_LAYER3
/* These are significant chunks of memory already... */
struct
{
real (*hybrid_in)[SBLIMIT][SSLIMIT]; /* ALIGNED(16) real hybridIn[2][SBLIMIT][SSLIMIT]; */
real (*hybrid_out)[SSLIMIT][SBLIMIT]; /* ALIGNED(16) real hybridOut[2][SSLIMIT][SBLIMIT]; */
} layer3;
#endif
/* A place for storing additional data for the large file wrapper.
This is cruft! */
void *wrapperdata;
/* A callback used to properly destruct the wrapper data. */
void (*wrapperclean)(void*);
};
/* generic init, does not include dynamic buffers */
void frame_init(mpg123_handle *fr);
void frame_init_par(mpg123_handle *fr, mpg123_pars *mp);
/* output buffer and format */
int frame_outbuffer(mpg123_handle *fr);
int frame_output_format(mpg123_handle *fr);
int frame_buffers(mpg123_handle *fr); /* various decoder buffers, needed once */
int frame_reset(mpg123_handle* fr); /* reset for next track */
int frame_buffers_reset(mpg123_handle *fr);
void frame_exit(mpg123_handle *fr); /* end, free all buffers */
/* Index functions... */
/* Well... print it... */
int mpg123_print_index(mpg123_handle *fr, FILE* out);
/* Find a seek position in index. */
off_t frame_index_find(mpg123_handle *fr, off_t want_frame, off_t* get_frame);
/* Apply index_size setting. */
int frame_index_setup(mpg123_handle *fr);
void do_volume(mpg123_handle *fr, double factor);
void do_rva(mpg123_handle *fr);
/* samples per frame ...
Layer I
Layer II
Layer III
MPEG-1
384
1152
1152
MPEG-2 LSF
384
1152
576
MPEG 2.5
384
1152
576
*/
#define spf(fr) ((fr)->lay == 1 ? 384 : ((fr)->lay==2 ? 1152 : ((fr)->lsf || (fr)->mpeg25 ? 576 : 1152)))
#ifdef GAPLESS
/* well, I take that one for granted... at least layer3 */
#define GAPLESS_DELAY 529
void frame_gapless_init(mpg123_handle *fr, off_t framecount, off_t bskip, off_t eskip);
void frame_gapless_realinit(mpg123_handle *fr);
void frame_gapless_update(mpg123_handle *mh, off_t total_samples);
/*void frame_gapless_position(mpg123_handle* fr);
void frame_gapless_bytify(mpg123_handle *fr);
void frame_gapless_ignore(mpg123_handle *fr, off_t frames);*/
/* void frame_gapless_buffercheck(mpg123_handle *fr); */
#endif
/* Number of samples the decoding of the current frame should yield. */
off_t frame_expect_outsamples(mpg123_handle *fr);
/* Skip this frame... do some fake action to get away without actually decoding it. */
void frame_skip(mpg123_handle *fr);
/*
Seeking core functions:
- convert input sample offset to output sample offset
- convert frame offset to output sample offset
- get leading frame offset for output sample offset
The offsets are "unadjusted"/internal; resampling is being taken care of.
*/
off_t frame_ins2outs(mpg123_handle *fr, off_t ins);
off_t frame_outs(mpg123_handle *fr, off_t num);
/* This one just computes the expected sample count for _this_ frame. */
off_t frame_expect_outsampels(mpg123_handle *fr);
off_t frame_offset(mpg123_handle *fr, off_t outs);
void frame_set_frameseek(mpg123_handle *fr, off_t fe);
void frame_set_seek(mpg123_handle *fr, off_t sp);
off_t frame_tell_seek(mpg123_handle *fr);
/* Take a copy of the Xing VBR TOC for fuzzy seeking. */
int frame_fill_toc(mpg123_handle *fr, unsigned char* in);
#endif