mirror of
https://git.rip/DMCA_FUCKER/re3.git
synced 2024-11-14 02:49:16 +00:00
160 lines
5.9 KiB
C
160 lines
5.9 KiB
C
/*
|
|
libmpg123: MPEG Audio Decoder library
|
|
|
|
separate header just for audio format definitions not tied to
|
|
library code
|
|
|
|
copyright 1995-2020 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
|
|
*/
|
|
|
|
#ifndef MPG123_ENC_H
|
|
#define MPG123_ENC_H
|
|
|
|
/** \file fmt123.h Audio format definitions. */
|
|
|
|
/** \defgroup mpg123_enc mpg123 PCM sample encodings
|
|
* These are definitions for audio formats used by libmpg123 and
|
|
* libout123.
|
|
*
|
|
* @{
|
|
*/
|
|
|
|
/** An enum over all sample types possibly known to mpg123.
|
|
* The values are designed as bit flags to allow bitmasking for encoding
|
|
* families.
|
|
* This is also why the enum is not used as type for actual encoding variables,
|
|
* plain integers (at least 16 bit, 15 bit being used) cover the possible
|
|
* combinations of these flags.
|
|
*
|
|
* Note that (your build of) libmpg123 does not necessarily support all these.
|
|
* Usually, you can expect the 8bit encodings and signed 16 bit.
|
|
* Also 32bit float will be usual beginning with mpg123-1.7.0 .
|
|
* What you should bear in mind is that (SSE, etc) optimized routines may be
|
|
* absent for some formats. We do have SSE for 16, 32 bit and float, though.
|
|
* 24 bit integer is done via postprocessing of 32 bit output -- just cutting
|
|
* the last byte, no rounding, even. If you want better, do it yourself.
|
|
*
|
|
* All formats are in native byte order. If you need different endinaness, you
|
|
* can simply postprocess the output buffers (libmpg123 wouldn't do anything
|
|
* else). The macro MPG123_SAMPLESIZE() can be helpful there.
|
|
*/
|
|
enum mpg123_enc_enum
|
|
{
|
|
/* 0000 0000 0000 1111 Some 8 bit integer encoding. */
|
|
MPG123_ENC_8 = 0x00f
|
|
/* 0000 0000 0100 0000 Some 16 bit integer encoding. */
|
|
, MPG123_ENC_16 = 0x040
|
|
/* 0100 0000 0000 0000 Some 24 bit integer encoding. */
|
|
, MPG123_ENC_24 = 0x4000
|
|
/* 0000 0001 0000 0000 Some 32 bit integer encoding. */
|
|
, MPG123_ENC_32 = 0x100
|
|
/* 0000 0000 1000 0000 Some signed integer encoding. */
|
|
, MPG123_ENC_SIGNED = 0x080
|
|
/* 0000 1110 0000 0000 Some float encoding. */
|
|
, MPG123_ENC_FLOAT = 0xe00
|
|
/* 0000 0000 1101 0000 signed 16 bit */
|
|
, MPG123_ENC_SIGNED_16 = (MPG123_ENC_16|MPG123_ENC_SIGNED|0x10)
|
|
/* 0000 0000 0110 0000 unsigned 16 bit */
|
|
, MPG123_ENC_UNSIGNED_16 = (MPG123_ENC_16|0x20)
|
|
/* 0000 0000 0000 0001 unsigned 8 bit */
|
|
, MPG123_ENC_UNSIGNED_8 = 0x01
|
|
/* 0000 0000 1000 0010 signed 8 bit */
|
|
, MPG123_ENC_SIGNED_8 = (MPG123_ENC_SIGNED|0x02)
|
|
/* 0000 0000 0000 0100 ulaw 8 bit */
|
|
, MPG123_ENC_ULAW_8 = 0x04
|
|
/* 0000 0000 0000 1000 alaw 8 bit */
|
|
, MPG123_ENC_ALAW_8 = 0x08
|
|
/* 0001 0001 1000 0000 signed 32 bit */
|
|
, MPG123_ENC_SIGNED_32 = MPG123_ENC_32|MPG123_ENC_SIGNED|0x1000
|
|
/* 0010 0001 0000 0000 unsigned 32 bit */
|
|
, MPG123_ENC_UNSIGNED_32 = MPG123_ENC_32|0x2000
|
|
/* 0101 0000 1000 0000 signed 24 bit */
|
|
, MPG123_ENC_SIGNED_24 = MPG123_ENC_24|MPG123_ENC_SIGNED|0x1000
|
|
/* 0110 0000 0000 0000 unsigned 24 bit */
|
|
, MPG123_ENC_UNSIGNED_24 = MPG123_ENC_24|0x2000
|
|
/* 0000 0010 0000 0000 32bit float */
|
|
, MPG123_ENC_FLOAT_32 = 0x200
|
|
/* 0000 0100 0000 0000 64bit float */
|
|
, MPG123_ENC_FLOAT_64 = 0x400
|
|
/* Any possibly known encoding from the list above. */
|
|
, MPG123_ENC_ANY = ( MPG123_ENC_SIGNED_16 | MPG123_ENC_UNSIGNED_16
|
|
| MPG123_ENC_UNSIGNED_8 | MPG123_ENC_SIGNED_8
|
|
| MPG123_ENC_ULAW_8 | MPG123_ENC_ALAW_8
|
|
| MPG123_ENC_SIGNED_32 | MPG123_ENC_UNSIGNED_32
|
|
| MPG123_ENC_SIGNED_24 | MPG123_ENC_UNSIGNED_24
|
|
| MPG123_ENC_FLOAT_32 | MPG123_ENC_FLOAT_64 )
|
|
};
|
|
|
|
/** Get size of one PCM sample with given encoding.
|
|
* This is included both in libmpg123 and libout123. Both offer
|
|
* an API function to provide the macro results from library
|
|
* compile-time, not that of you application. This most likely
|
|
* does not matter as I do not expect any fresh PCM sample
|
|
* encoding to appear. But who knows? Perhaps the encoding type
|
|
* will be abused for funny things in future, not even plain PCM.
|
|
* And, by the way: Thomas really likes the ?: operator.
|
|
* \param enc the encoding (mpg123_enc_enum value)
|
|
* \return size of one sample in bytes
|
|
*/
|
|
#define MPG123_SAMPLESIZE(enc) ( \
|
|
(enc) < 1 \
|
|
? 0 \
|
|
: ( (enc) & MPG123_ENC_8 \
|
|
? 1 \
|
|
: ( (enc) & MPG123_ENC_16 \
|
|
? 2 \
|
|
: ( (enc) & MPG123_ENC_24 \
|
|
? 3 \
|
|
: ( ( (enc) & MPG123_ENC_32 \
|
|
|| (enc) == MPG123_ENC_FLOAT_32 ) \
|
|
? 4 \
|
|
: ( (enc) == MPG123_ENC_FLOAT_64 \
|
|
? 8 \
|
|
: 0 \
|
|
) ) ) ) ) )
|
|
|
|
/** Representation of zero in differing encodings.
|
|
* This exists to define proper silence in various encodings without
|
|
* having to link to libsyn123 to do actual conversions at runtime.
|
|
* You have to handle big/little endian order yourself, though.
|
|
* This takes the shortcut that any signed encoding has a zero with
|
|
* all-zero bits. Unsigned linear encodings just have the highest bit set
|
|
* (2^(n-1) for n bits), while the nonlinear 8-bit ones are special.
|
|
* \param enc the encoding (mpg123_enc_enum value)
|
|
* \param siz bytes per sample (return value of MPG123_SAMPLESIZE(enc))
|
|
* \param off byte (octet) offset counted from LSB
|
|
* \return unsigned byte value for the designated octet
|
|
*/
|
|
#define MPG123_ZEROSAMPLE(enc, siz, off) ( \
|
|
(enc) == MPG123_ENC_ULAW_8 \
|
|
? (off == 0 ? 0xff : 0x00) \
|
|
: ( (enc) == MPG123_ENC_ALAW_8 \
|
|
? (off == 0 ? 0xd5 : 0x00) \
|
|
: ( (((enc) & (MPG123_ENC_SIGNED|MPG123_ENC_FLOAT)) || (siz) != ((off)+1)) \
|
|
? 0x00 \
|
|
: 0x80 \
|
|
) ) )
|
|
|
|
/** Structure defining an audio format.
|
|
* Providing the members as individual function arguments to define a certain
|
|
* output format is easy enough. This struct makes is more comfortable to deal
|
|
* with a list of formats.
|
|
* Negative values for the members might be used to communicate use of default
|
|
* values.
|
|
*/
|
|
struct mpg123_fmt
|
|
{
|
|
long rate; /**< sampling rate in Hz */
|
|
int channels; /**< channel count */
|
|
/** encoding code, can be single value or bitwise or of members of
|
|
* mpg123_enc_enum */
|
|
int encoding;
|
|
};
|
|
|
|
/* @} */
|
|
|
|
#endif
|
|
|