// Ogg Vorbis audio decoder - v1.22 - public domain
// http://nothings.org/stb_vorbis/
//
// Original version written by Sean Barrett in 2007.
//
// Originally sponsored by RAD Game Tools. Seeking implementation
// sponsored by Phillip Bennefall, Marc Andersen, Aaron Baker,
// Elias Software, Aras Pranckevicius, and Sean Barrett.
//
// LICENSE
//
//   See end of file for license information.
//
// Limitations:
//
//   - floor 0 not supported (used in old ogg vorbis files pre-2004)
//   - lossless sample-truncation at beginning ignored
//   - cannot concatenate multiple vorbis streams
//   - sample positions are 32-bit, limiting seekable 192Khz
//       files to around 6 hours (Ogg supports 64-bit)
//
// Feature contributors:
//    Dougall Johnson (sample-exact seeking)
//
// Bugfix/warning contributors:
//    Terje Mathisen     Niklas Frykholm     Andy Hill
//    Casey Muratori     John Bolton         Gargaj
//    Laurent Gomila     Marc LeBlanc        Ronny Chevalier
//    Bernhard Wodo      Evan Balster        github:alxprd
//    Tom Beaumont       Ingo Leitgeb        Nicolas Guillemot
//    Phillip Bennefall  Rohit               Thiago Goulart
//    github:manxorist   Saga Musix          github:infatum
//    Timur Gagiev       Maxwell Koo         Peter Waller
//    github:audinowho   Dougall Johnson     David Reid
//    github:Clownacy    Pedro J. Estebanez  Remi Verschelde
//    AnthoFoxo          github:morlat       Gabriel Ravier
//
// Partial history:
//    1.22    - 2021-07-11 - various small fixes
//    1.21    - 2021-07-02 - fix bug for files with no comments
//    1.20    - 2020-07-11 - several small fixes
//    1.19    - 2020-02-05 - warnings
//    1.18    - 2020-02-02 - fix seek bugs; parse header comments; misc warnings etc.
//    1.17    - 2019-07-08 - fix CVE-2019-13217..CVE-2019-13223 (by ForAllSecure)
//    1.16    - 2019-03-04 - fix warnings
//    1.15    - 2019-02-07 - explicit failure if Ogg Skeleton data is found
//    1.14    - 2018-02-11 - delete bogus dealloca usage
//    1.13    - 2018-01-29 - fix truncation of last frame (hopefully)
//    1.12    - 2017-11-21 - limit residue begin/end to blocksize/2 to avoid large temp allocs in bad/corrupt files
//    1.11    - 2017-07-23 - fix MinGW compilation
//    1.10    - 2017-03-03 - more robust seeking; fix negative ilog(); clear error in open_memory
//    1.09    - 2016-04-04 - back out 'truncation of last frame' fix from previous version
//    1.08    - 2016-04-02 - warnings; setup memory leaks; truncation of last frame
//    1.07    - 2015-01-16 - fixes for crashes on invalid files; warning fixes; const
//    1.06    - 2015-08-31 - full, correct support for seeking API (Dougall Johnson)
//                           some crash fixes when out of memory or with corrupt files
//                           fix some inappropriately signed shifts
//    1.05    - 2015-04-19 - don't define __forceinline if it's redundant
//    1.04    - 2014-08-27 - fix missing const-correct case in API
//    1.03    - 2014-08-07 - warning fixes
//    1.02    - 2014-07-09 - declare qsort comparison as explicitly _cdecl in Windows
//    1.01    - 2014-06-18 - fix stb_vorbis_get_samples_float (interleaved was correct)
//    1.0     - 2014-05-26 - fix memory leaks; fix warnings; fix bugs in >2-channel;
//                           (API change) report sample rate for decode-full-file funcs
//
// See end of file for full version history.


//////////////////////////////////////////////////////////////////////////////
//
//  HEADER BEGINS HERE
//

#ifndef STB_VORBIS_INCLUDE_STB_VORBIS_H
#define STB_VORBIS_INCLUDE_STB_VORBIS_H

#if defined(STB_VORBIS_NO_CRT) && !defined(STB_VORBIS_NO_STDIO)
#define STB_VORBIS_NO_STDIO 1
#endif

#ifndef STB_VORBIS_NO_STDIO
#include <stdio.h>
#endif

#ifdef __cplusplus
extern "C" {
#endif

///////////   THREAD SAFETY

// Individual stb_vorbis* handles are not thread-safe; you cannot decode from
// them from multiple threads at the same time. However, you can have multiple
// stb_vorbis* handles and decode from them independently in multiple thrads.


///////////   MEMORY ALLOCATION

// normally stb_vorbis uses malloc() to allocate memory at startup,
// and alloca() to allocate temporary memory during a frame on the
// stack. (Memory consumption will depend on the amount of setup
// data in the file and how you set the compile flags for speed
// vs. size. In my test files the maximal-size usage is ~150KB.)
//
// You can modify the wrapper functions in the source (setup_malloc,
// setup_temp_malloc, temp_malloc) to change this behavior, or you
// can use a simpler allocation model: you pass in a buffer from
// which stb_vorbis will allocate _all_ its memory (including the
// temp memory). "open" may fail with a VORBIS_outofmem if you
// do not pass in enough data; there is no way to determine how
// much you do need except to succeed (at which point you can
// query get_info to find the exact amount required. yes I know
// this is lame).
//
// If you pass in a non-NULL buffer of the type below, allocation
// will occur from it as described above. Otherwise just pass NULL
// to use malloc()/alloca()

typedef struct
{
   char *alloc_buffer;
   int   alloc_buffer_length_in_bytes;
} stb_vorbis_alloc;


///////////   FUNCTIONS USEABLE WITH ALL INPUT MODES

typedef struct stb_vorbis stb_vorbis;

typedef struct
{
   unsigned int sample_rate;
   int channels;

   unsigned int setup_memory_required;
   unsigned int setup_temp_memory_required;
   unsigned int temp_memory_required;

   int max_frame_size;
} stb_vorbis_info;

typedef struct
{
   char *vendor;

   int comment_list_length;
   char **comment_list;
} stb_vorbis_comment;

// get general information about the file
extern stb_vorbis_info stb_vorbis_get_info(stb_vorbis *f);

// get ogg comments
extern stb_vorbis_comment stb_vorbis_get_comment(stb_vorbis *f);

// get the last error detected (clears it, too)
extern int stb_vorbis_get_error(stb_vorbis *f);

// close an ogg vorbis file and free all memory in use
extern void stb_vorbis_close(stb_vorbis *f);

// this function returns the offset (in samples) from the beginning of the
// file that will be returned by the next decode, if it is known, or -1
// otherwise. after a flush_pushdata() call, this may take a while before
// it becomes valid again.
// NOT WORKING YET after a seek with PULLDATA API
extern int stb_vorbis_get_sample_offset(stb_vorbis *f);

// returns the current seek point within the file, or offset from the beginning
// of the memory buffer. In pushdata mode it returns 0.
extern unsigned int stb_vorbis_get_file_offset(stb_vorbis *f);

///////////   PUSHDATA API

#ifndef STB_VORBIS_NO_PUSHDATA_API

// this API allows you to get blocks of data from any source and hand
// them to stb_vorbis. you have to buffer them; stb_vorbis will tell
// you how much it used, and you have to give it the rest next time;
// and stb_vorbis may not have enough data to work with and you will
// need to give it the same data again PLUS more. Note that the Vorbis
// specification does not bound the size of an individual frame.

extern stb_vorbis *stb_vorbis_open_pushdata(
         const unsigned char * datablock, int datablock_length_in_bytes,
         int *datablock_memory_consumed_in_bytes,
         int *error,
         const stb_vorbis_alloc *alloc_buffer);
// create a vorbis decoder by passing in the initial data block containing
//    the ogg&vorbis headers (you don't need to do parse them, just provide
//    the first N bytes of the file--you're told if it's not enough, see below)
// on success, returns an stb_vorbis *, does not set error, returns the amount of
//    data parsed/consumed on this call in *datablock_memory_consumed_in_bytes;
// on failure, returns NULL on error and sets *error, does not change *datablock_memory_consumed
// if returns NULL and *error is VORBIS_need_more_data, then the input block was
//       incomplete and you need to pass in a larger block from the start of the file

extern int stb_vorbis_decode_frame_pushdata(
         stb_vorbis *f,
         const unsigned char *datablock, int datablock_length_in_bytes,
         int *channels,             // place to write number of float * buffers
         float ***output,           // place to write float ** array of float * buffers
         int *samples               // place to write number of output samples
     );
// decode a frame of audio sample data if possible from the passed-in data block
//
// return value: number of bytes we used from datablock
//
// possible cases:
//     0 bytes used, 0 samples output (need more data)
//     N bytes used, 0 samples output (resynching the stream, keep going)
//     N bytes used, M samples output (one frame of data)
// note that after opening a file, you will ALWAYS get one N-bytes,0-sample
// frame, because Vorbis always "discards" the first frame.
//
// Note that on resynch, stb_vorbis will rarely consume all of the buffer,
// instead only datablock_length_in_bytes-3 or less. This is because it wants
// to avoid missing parts of a page header if they cross a datablock boundary,
// without writing state-machiney code to record a partial detection.
//
// The number of channels returned are stored in *channels (which can be
// NULL--it is always the same as the number of channels reported by
// get_info). *output will contain an array of float* buffers, one per
// channel. In other words, (*output)[0][0] contains the first sample from
// the first channel, and (*output)[1][0] contains the first sample from
// the second channel.
//
// *output points into stb_vorbis's internal output buffer storage; these
// buffers are owned by stb_vorbis and application code should not free
// them or modify their contents. They are transient and will be overwritten
// once you ask for more data to get decoded, so be sure to grab any data
// you need before then.

extern void stb_vorbis_flush_pushdata(stb_vorbis *f);
// inform stb_vorbis that your next datablock will not be contiguous with
// previous ones (e.g. you've seeked in the data); future attempts to decode
// frames will cause stb_vorbis to resynchronize (as noted above), and
// once it sees a valid Ogg page (typically 4-8KB, as large as 64KB), it
// will begin decoding the _next_ frame.
//
// if you want to seek using pushdata, you need to seek in your file, then
// call stb_vorbis_flush_pushdata(), then start calling decoding, then once
// decoding is returning you data, call stb_vorbis_get_sample_offset, and
// if you don't like the result, seek your file again and repeat.
#endif


//////////   PULLING INPUT API

#ifndef STB_VORBIS_NO_PULLDATA_API
// This API assumes stb_vorbis is allowed to pull data from a source--
// either a block of memory containing the _entire_ vorbis stream, or a
// FILE * that you or it create, or possibly some other reading mechanism
// if you go modify the source to replace the FILE * case with some kind
// of callback to your code. (But if you don't support seeking, you may
// just want to go ahead and use pushdata.)

#if !defined(STB_VORBIS_NO_STDIO) && !defined(STB_VORBIS_NO_INTEGER_CONVERSION)
extern int stb_vorbis_decode_filename(const char *filename, int *channels, int *sample_rate, short **output);
#endif
#if !defined(STB_VORBIS_NO_INTEGER_CONVERSION)
extern int stb_vorbis_decode_memory(const unsigned char *mem, int len, int *channels, int *sample_rate, short **output);
#endif
// decode an entire file and output the data interleaved into a malloc()ed
// buffer stored in *output. The return value is the number of samples
// decoded, or -1 if the file could not be opened or was not an ogg vorbis file.
// When you're done with it, just free() the pointer returned in *output.

extern stb_vorbis * stb_vorbis_open_memory(const unsigned char *data, int len,
                                  int *error, const stb_vorbis_alloc *alloc_buffer);
// create an ogg vorbis decoder from an ogg vorbis stream in memory (note
// this must be the entire stream!). on failure, returns NULL and sets *error

#ifndef STB_VORBIS_NO_STDIO
extern stb_vorbis * stb_vorbis_open_filename(const char *filename,
                                  int *error, const stb_vorbis_alloc *alloc_buffer);
// create an ogg vorbis decoder from a filename via fopen(). on failure,
// returns NULL and sets *error (possibly to VORBIS_file_open_failure).

extern stb_vorbis * stb_vorbis_open_file(FILE *f, int close_handle_on_close,
                                  int *error, const stb_vorbis_alloc *alloc_buffer);
// create an ogg vorbis decoder from an open FILE *, looking for a stream at
// the _current_ seek point (ftell). on failure, returns NULL and sets *error.
// note that stb_vorbis must "own" this stream; if you seek it in between
// calls to stb_vorbis, it will become confused. Moreover, if you attempt to
// perform stb_vorbis_seek_*() operations on this file, it will assume it
// owns the _entire_ rest of the file after the start point. Use the next
// function, stb_vorbis_open_file_section(), to limit it.

extern stb_vorbis * stb_vorbis_open_file_section(FILE *f, int close_handle_on_close,
                int *error, const stb_vorbis_alloc *alloc_buffer, unsigned int len);
// create an ogg vorbis decoder from an open FILE *, looking for a stream at
// the _current_ seek point (ftell); the stream will be of length 'len' bytes.
// on failure, returns NULL and sets *error. note that stb_vorbis must "own"
// this stream; if you seek it in between calls to stb_vorbis, it will become
// confused.
#endif

extern int stb_vorbis_seek_frame(stb_vorbis *f, unsigned int sample_number);
extern int stb_vorbis_seek(stb_vorbis *f, unsigned int sample_number);
// these functions seek in the Vorbis file to (approximately) 'sample_number'.
// after calling seek_frame(), the next call to get_frame_*() will include
// the specified sample. after calling stb_vorbis_seek(), the next call to
// stb_vorbis_get_samples_* will start with the specified sample. If you
// do not need to seek to EXACTLY the target sample when using get_samples_*,
// you can also use seek_frame().

extern int stb_vorbis_seek_start(stb_vorbis *f);
// this function is equivalent to stb_vorbis_seek(f,0)

extern unsigned int stb_vorbis_stream_length_in_samples(stb_vorbis *f);
extern float        stb_vorbis_stream_length_in_seconds(stb_vorbis *f);
// these functions return the total length of the vorbis stream

extern int stb_vorbis_get_frame_float(stb_vorbis *f, int *channels, float ***output);
// decode the next frame and return the number of samples. the number of
// channels returned are stored in *channels (which can be NULL--it is always
// the same as the number of channels reported by get_info). *output will
// contain an array of float* buffers, one per channel. These outputs will
// be overwritten on the next call to stb_vorbis_get_frame_*.
//
// You generally should not intermix calls to stb_vorbis_get_frame_*()
// and stb_vorbis_get_samples_*(), since the latter calls the former.

#ifndef STB_VORBIS_NO_INTEGER_CONVERSION
extern int stb_vorbis_get_frame_short_interleaved(stb_vorbis *f, int num_c, short *buffer, int num_shorts);
extern int stb_vorbis_get_frame_short            (stb_vorbis *f, int num_c, short **buffer, int num_samples);
#endif
// decode the next frame and return the number of *samples* per channel.
// Note that for interleaved data, you pass in the number of shorts (the
// size of your array), but the return value is the number of samples per
// channel, not the total number of samples.
//
// The data is coerced to the number of channels you request according to the
// channel coercion rules (see below). You must pass in the size of your
// buffer(s) so that stb_vorbis will not overwrite the end of the buffer.
// The maximum buffer size needed can be gotten from get_info(); however,
// the Vorbis I specification implies an absolute maximum of 4096 samples
// per channel.

// Channel coercion rules:
//    Let M be the number of channels requested, and N the number of channels present,
//    and Cn be the nth channel; let stereo L be the sum of all L and center channels,
//    and stereo R be the sum of all R and center channels (channel assignment from the
//    vorbis spec).
//        M    N       output
//        1    k      sum(Ck) for all k
//        2    *      stereo L, stereo R
//        k    l      k > l, the first l channels, then 0s
//        k    l      k <= l, the first k channels
//    Note that this is not _good_ surround etc. mixing at all! It's just so
//    you get something useful.

extern int stb_vorbis_get_samples_float_interleaved(stb_vorbis *f, int channels, float *buffer, int num_floats);
extern int stb_vorbis_get_samples_float(stb_vorbis *f, int channels, float **buffer, int num_samples);
// gets num_samples samples, not necessarily on a frame boundary--this requires
// buffering so you have to supply the buffers. DOES NOT APPLY THE COERCION RULES.
// Returns the number of samples stored per channel; it may be less than requested
// at the end of the file. If there are no more samples in the file, returns 0.

#ifndef STB_VORBIS_NO_INTEGER_CONVERSION
extern int stb_vorbis_get_samples_short_interleaved(stb_vorbis *f, int channels, short *buffer, int num_shorts);
extern int stb_vorbis_get_samples_short(stb_vorbis *f, int channels, short **buffer, int num_samples);
#endif
// gets num_samples samples, not necessarily on a frame boundary--this requires
// buffering so you have to supply the buffers. Applies the coercion rules above
// to produce 'channels' channels. Returns the number of samples stored per channel;
// it may be less than requested at the end of the file. If there are no more
// samples in the file, returns 0.

#endif

////////   ERROR CODES

enum STBVorbisError
{
   VORBIS__no_error,

   VORBIS_need_more_data=1,             // not a real error

   VORBIS_invalid_api_mixing,           // can't mix API modes
   VORBIS_outofmem,                     // not enough memory
   VORBIS_feature_not_supported,        // uses floor 0
   VORBIS_too_many_channels,            // STB_VORBIS_MAX_CHANNELS is too small
   VORBIS_file_open_failure,            // fopen() failed
   VORBIS_seek_without_length,          // can't seek in unknown-length file

   VORBIS_unexpected_eof=10,            // file is truncated?
   VORBIS_seek_invalid,                 // seek past EOF

   // decoding errors (corrupt/invalid stream) -- you probably
   // don't care about the exact details of these

   // vorbis errors:
   VORBIS_invalid_setup=20,
   VORBIS_invalid_stream,

   // ogg errors:
   VORBIS_missing_capture_pattern=30,
   VORBIS_invalid_stream_structure_version,
   VORBIS_continued_packet_flag_invalid,
   VORBIS_incorrect_stream_serial_number,
   VORBIS_invalid_first_page,
   VORBIS_bad_packet_type,
   VORBIS_cant_find_last_page,
   VORBIS_seek_failed,
   VORBIS_ogg_skeleton_not_supported
};


#ifdef __cplusplus
}
#endif

#endif // STB_VORBIS_INCLUDE_STB_VORBIS_H
//
//  HEADER ENDS HERE
//
//////////////////////////////////////////////////////////////////////////////

#ifndef STB_VORBIS_HEADER_ONLY

// global configuration settings (e.g. set these in the project/makefile),
// or just set them in this file at the top (although ideally the first few
// should be visible when the header file is compiled too, although it's not
// crucial)

// STB_VORBIS_NO_PUSHDATA_API
//     does not compile the code for the various stb_vorbis_*_pushdata()
//     functions
// #define STB_VORBIS_NO_PUSHDATA_API

// STB_VORBIS_NO_PULLDATA_API
//     does not compile the code for the non-pushdata APIs
// #define STB_VORBIS_NO_PULLDATA_API

// STB_VORBIS_NO_STDIO
//     does not compile the code for the APIs that use FILE *s internally
//     or externally (implied by STB_VORBIS_NO_PULLDATA_API)
// #define STB_VORBIS_NO_STDIO

// STB_VORBIS_NO_INTEGER_CONVERSION
//     does not compile the code for converting audio sample data from
//     float to integer (implied by STB_VORBIS_NO_PULLDATA_API)
// #define STB_VORBIS_NO_INTEGER_CONVERSION

// STB_VORBIS_NO_FAST_SCALED_FLOAT
//      does not use a fast float-to-int trick to accelerate float-to-int on
//      most platforms which requires endianness be defined correctly.
//#define STB_VORBIS_NO_FAST_SCALED_FLOAT


// STB_VORBIS_MAX_CHANNELS [number]
//     globally define this to the maximum number of channels you need.
//     The spec does not put a restriction on channels except that
//     the count is stored in a byte, so 255 is the hard limit.
//     Reducing this saves about 16 bytes per value, so using 16 saves
//     (255-16)*16 or around 4KB. Plus anything other memory usage
//     I forgot to account for. Can probably go as low as 8 (7.1 audio),
//     6 (5.1 audio), or 2 (stereo only).
#ifndef STB_VORBIS_MAX_CHANNELS
#define STB_VORBIS_MAX_CHANNELS    16  // enough for anyone?
#endif

// STB_VORBIS_PUSHDATA_CRC_COUNT [number]
//     after a flush_pushdata(), stb_vorbis begins scanning for the
//     next valid page, without backtracking. when it finds something
//     that looks like a page, it streams through it and verifies its
//     CRC32. Should that validation fail, it keeps scanning. But it's
//     possible that _while_ streaming through to check the CRC32 of
//     one candidate page, it sees another candidate page. This #define
//     determines how many "overlapping" candidate pages it can search
//     at once. Note that "real" pages are typically ~4KB to ~8KB, whereas
//     garbage pages could be as big as 64KB, but probably average ~16KB.
//     So don't hose ourselves by scanning an apparent 64KB page and
//     missing a ton of real ones in the interim; so minimum of 2
#ifndef STB_VORBIS_PUSHDATA_CRC_COUNT
#define STB_VORBIS_PUSHDATA_CRC_COUNT  4
#endif

// STB_VORBIS_FAST_HUFFMAN_LENGTH [number]
//     sets the log size of the huffman-acceleration table.  Maximum
//     supported value is 24. with larger numbers, more decodings are O(1),
//     but the table size is larger so worse cache missing, so you'll have
//     to probe (and try multiple ogg vorbis files) to find the sweet spot.
#ifndef STB_VORBIS_FAST_HUFFMAN_LENGTH
#define STB_VORBIS_FAST_HUFFMAN_LENGTH   10
#endif

// STB_VORBIS_FAST_BINARY_LENGTH [number]
//     sets the log size of the binary-search acceleration table. this
//     is used in similar fashion to the fast-huffman size to set initial
//     parameters for the binary search

// STB_VORBIS_FAST_HUFFMAN_INT
//     The fast huffman tables are much more efficient if they can be
//     stored as 16-bit results instead of 32-bit results. This restricts
//     the codebooks to having only 65535 possible outcomes, though.
//     (At least, accelerated by the huffman table.)
#ifndef STB_VORBIS_FAST_HUFFMAN_INT
#define STB_VORBIS_FAST_HUFFMAN_SHORT
#endif

// STB_VORBIS_NO_HUFFMAN_BINARY_SEARCH
//     If the 'fast huffman' search doesn't succeed, then stb_vorbis falls
//     back on binary searching for the correct one. This requires storing
//     extra tables with the huffman codes in sorted order. Defining this
//     symbol trades off space for speed by forcing a linear search in the
//     non-fast case, except for "sparse" codebooks.
// #define STB_VORBIS_NO_HUFFMAN_BINARY_SEARCH

// STB_VORBIS_DIVIDES_IN_RESIDUE
//     stb_vorbis precomputes the result of the scalar residue decoding
//     that would otherwise require a divide per chunk. you can trade off
//     space for time by defining this symbol.
// #define STB_VORBIS_DIVIDES_IN_RESIDUE

// STB_VORBIS_DIVIDES_IN_CODEBOOK
//     vorbis VQ codebooks can be encoded two ways: with every case explicitly
//     stored, or with all elements being chosen from a small range of values,
//     and all values possible in all elements. By default, stb_vorbis expands
//     this latter kind out to look like the former kind for ease of decoding,
//     because otherwise an integer divide-per-vector-element is required to
//     unpack the index. If you define STB_VORBIS_DIVIDES_IN_CODEBOOK, you can
//     trade off storage for speed.
//#define STB_VORBIS_DIVIDES_IN_CODEBOOK

#ifdef STB_VORBIS_CODEBOOK_SHORTS
#error "STB_VORBIS_CODEBOOK_SHORTS is no longer supported as it produced incorrect results for some input formats"
#endif

// STB_VORBIS_DIVIDE_TABLE
//     this replaces small integer divides in the floor decode loop with
//     table lookups. made less than 1% difference, so disabled by default.

// STB_VORBIS_NO_INLINE_DECODE
//     disables the inlining of the scalar codebook fast-huffman decode.
//     might save a little codespace; useful for debugging
// #define STB_VORBIS_NO_INLINE_DECODE

// STB_VORBIS_NO_DEFER_FLOOR
//     Normally we only decode the floor without synthesizing the actual
//     full curve. We can instead synthesize the curve immediately. This
//     requires more memory and is very likely slower, so I don't think
//     you'd ever want to do it except for debugging.
// #define STB_VORBIS_NO_DEFER_FLOOR




//////////////////////////////////////////////////////////////////////////////

#ifdef STB_VORBIS_NO_PULLDATA_API
   #define STB_VORBIS_NO_INTEGER_CONVERSION
   #define STB_VORBIS_NO_STDIO
#endif

#if defined(STB_VORBIS_NO_CRT) && !defined(STB_VORBIS_NO_STDIO)
   #define STB_VORBIS_NO_STDIO 1
#endif

#ifndef STB_VORBIS_NO_INTEGER_CONVERSION
#ifndef STB_VORBIS_NO_FAST_SCALED_FLOAT

   // only need endianness for fast-float-to-int, which we don't
   // use for pushdata

   #ifndef STB_VORBIS_BIG_ENDIAN
     #define STB_VORBIS_ENDIAN  0
   #else
     #define STB_VORBIS_ENDIAN  1
   #endif

#endif
#endif


#ifndef STB_VORBIS_NO_STDIO
#include <stdio.h>
#endif

#ifndef STB_VORBIS_NO_CRT
   #include <stdlib.h>
   #include <string.h>
   #include <assert.h>
   #include <math.h>

   // find definition of alloca if it's not in stdlib.h:
   #if defined(_MSC_VER) || defined(__MINGW32__)
      #include <malloc.h>
   #endif
   #if defined(__linux__) || defined(__linux) || defined(__sun__) || defined(__EMSCRIPTEN__) || defined(__NEWLIB__)
      #include <alloca.h>
   #endif
#else // STB_VORBIS_NO_CRT
   #define NULL 0
   #define malloc(s)   0
   #define free(s)     ((void) 0)
   #define realloc(s)  0
#endif // STB_VORBIS_NO_CRT

#include <limits.h>

#ifdef __MINGW32__
   // eff you mingw:
   //     "fixed":
   //         http://sourceforge.net/p/mingw-w64/mailman/message/32882927/
   //     "no that broke the build, reverted, who cares about C":
   //         http://sourceforge.net/p/mingw-w64/mailman/message/32890381/
   #ifdef __forceinline
   #undef __forceinline
   #endif
   #define __forceinline
   #ifndef alloca
   #define alloca __builtin_alloca
   #endif
#elif !defined(_MSC_VER)
   #if __GNUC__
      #define __forceinline inline
   #else
      #define __forceinline
   #endif
#endif

#if STB_VORBIS_MAX_CHANNELS > 256
#error "Value of STB_VORBIS_MAX_CHANNELS outside of allowed range"
#endif

#if STB_VORBIS_FAST_HUFFMAN_LENGTH > 24
#error "Value of STB_VORBIS_FAST_HUFFMAN_LENGTH outside of allowed range"
#endif


#if 0
#include <crtdbg.h>
#define CHECK(f)   _CrtIsValidHeapPointer(f->channel_buffers[1])
#else
#define CHECK(f)   ((void) 0)
#endif

#define MAX_BLOCKSIZE_LOG  13   // from specification
#define MAX_BLOCKSIZE      (1 << MAX_BLOCKSIZE_LOG)


typedef unsigned char  uint8;
typedef   signed char   int8;
typedef unsigned short uint16;
typedef   signed short  int16;
typedef unsigned int   uint32;
typedef   signed int    int32;

#ifndef TRUE
#define TRUE 1
#define FALSE 0
#endif

typedef float codetype;

#ifdef _MSC_VER
#define STBV_NOTUSED(v)  (void)(v)
#else
#define STBV_NOTUSED(v)  (void)sizeof(v)
#endif

// @NOTE
//
// Some arrays below are tagged "//varies", which means it's actually
// a variable-sized piece of data, but rather than malloc I assume it's
// small enough it's better to just allocate it all together with the
// main thing
//
// Most of the variables are specified with the smallest size I could pack
// them into. It might give better performance to make them all full-sized
// integers. It should be safe to freely rearrange the structures or change
// the sizes larger--nothing relies on silently truncating etc., nor the
// order of variables.

#define FAST_HUFFMAN_TABLE_SIZE   (1 << STB_VORBIS_FAST_HUFFMAN_LENGTH)
#define FAST_HUFFMAN_TABLE_MASK   (FAST_HUFFMAN_TABLE_SIZE - 1)

typedef struct
{
   int dimensions, entries;
   uint8 *codeword_lengths;
   float  minimum_value;
   float  delta_value;
   uint8  value_bits;
   uint8  lookup_type;
   uint8  sequence_p;
   uint8  sparse;
   uint32 lookup_values;
   codetype *multiplicands;
   uint32 *codewords;
   #ifdef STB_VORBIS_FAST_HUFFMAN_SHORT
    int16  fast_huffman[FAST_HUFFMAN_TABLE_SIZE];
   #else
    int32  fast_huffman[FAST_HUFFMAN_TABLE_SIZE];
   #endif
   uint32 *sorted_codewords;
   int    *sorted_values;
   int     sorted_entries;
} Codebook;

typedef struct
{
   uint8 order;
   uint16 rate;
   uint16 bark_map_size;
   uint8 amplitude_bits;
   uint8 amplitude_offset;
   uint8 number_of_books;
   uint8 book_list[16]; // varies
} Floor0;

typedef struct
{
   uint8 partitions;
   uint8 partition_class_list[32]; // varies
   uint8 class_dimensions[16]; // varies
   uint8 class_subclasses[16]; // varies
   uint8 class_masterbooks[16]; // varies
   int16 subclass_books[16][8]; // varies
   uint16 Xlist[31*8+2]; // varies
   uint8 sorted_order[31*8+2];
   uint8 neighbors[31*8+2][2];
   uint8 floor1_multiplier;
   uint8 rangebits;
   int values;
} Floor1;

typedef union
{
   Floor0 floor0;
   Floor1 floor1;
} Floor;

typedef struct
{
   uint32 begin, end;
   uint32 part_size;
   uint8 classifications;
   uint8 classbook;
   uint8 **classdata;
   int16 (*residue_books)[8];
} Residue;

typedef struct
{
   uint8 magnitude;
   uint8 angle;
   uint8 mux;
} MappingChannel;

typedef struct
{
   uint16 coupling_steps;
   MappingChannel *chan;
   uint8  submaps;
   uint8  submap_floor[15]; // varies
   uint8  submap_residue[15]; // varies
} Mapping;

typedef struct
{
   uint8 blockflag;
   uint8 mapping;
   uint16 windowtype;
   uint16 transformtype;
} Mode;

typedef struct
{
   uint32  goal_crc;    // expected crc if match
   int     bytes_left;  // bytes left in packet
   uint32  crc_so_far;  // running crc
   int     bytes_done;  // bytes processed in _current_ chunk
   uint32  sample_loc;  // granule pos encoded in page
} CRCscan;

typedef struct
{
   uint32 page_start, page_end;
   uint32 last_decoded_sample;
} ProbedPage;

struct stb_vorbis
{
  // user-accessible info
   unsigned int sample_rate;
   int channels;

   unsigned int setup_memory_required;
   unsigned int temp_memory_required;
   unsigned int setup_temp_memory_required;

   char *vendor;
   int comment_list_length;
   char **comment_list;

  // input config
#ifndef STB_VORBIS_NO_STDIO
   FILE *f;
   uint32 f_start;
   int close_on_free;
#endif

   uint8 *stream;
   uint8 *stream_start;
   uint8 *stream_end;

   uint32 stream_len;

   uint8  push_mode;

   // the page to seek to when seeking to start, may be zero
   uint32 first_audio_page_offset;

   // p_first is the page on which the first audio packet ends
   // (but not necessarily the page on which it starts)
   ProbedPage p_first, p_last;

  // memory management
   stb_vorbis_alloc alloc;
   int setup_offset;
   int temp_offset;

  // run-time results
   int eof;
   enum STBVorbisError error;

  // user-useful data

  // header info
   int blocksize[2];
   int blocksize_0, blocksize_1;
   int codebook_count;
   Codebook *codebooks;
   int floor_count;
   uint16 floor_types[64]; // varies
   Floor *floor_config;
   int residue_count;
   uint16 residue_types[64]; // varies
   Residue *residue_config;
   int mapping_count;
   Mapping *mapping;
   int mode_count;
   Mode mode_config[64];  // varies

   uint32 total_samples;

  // decode buffer
   float *channel_buffers[STB_VORBIS_MAX_CHANNELS];
   float *outputs        [STB_VORBIS_MAX_CHANNELS];

   float *previous_window[STB_VORBIS_MAX_CHANNELS];
   int previous_length;

   #ifndef STB_VORBIS_NO_DEFER_FLOOR
   int16 *finalY[STB_VORBIS_MAX_CHANNELS];
   #else
   float *floor_buffers[STB_VORBIS_MAX_CHANNELS];
   #endif

   uint32 current_loc; // sample location of next frame to decode
   int    current_loc_valid;

  // per-blocksize precomputed data

   // twiddle factors
   float *A[2],*B[2],*C[2];
   float *window[2];
   uint16 *bit_reverse[2];

  // current page/packet/segment streaming info
   uint32 serial; // stream serial number for verification
   int last_page;
   int segment_count;
   uint8 segments[255];
   uint8 page_flag;
   uint8 bytes_in_seg;
   uint8 first_decode;
   int next_seg;
   int last_seg;  // flag that we're on the last segment
   int last_seg_which; // what was the segment number of the last seg?
   uint32 acc;
   int valid_bits;
   int packet_bytes;
   int end_seg_with_known_loc;
   uint32 known_loc_for_packet;
   int discard_samples_deferred;
   uint32 samples_output;

  // push mode scanning
   int page_crc_tests; // only in push_mode: number of tests active; -1 if not searching
#ifndef STB_VORBIS_NO_PUSHDATA_API
   CRCscan scan[STB_VORBIS_PUSHDATA_CRC_COUNT];
#endif

  // sample-access
   int channel_buffer_start;
   int channel_buffer_end;
};

#if defined(STB_VORBIS_NO_PUSHDATA_API)
   #define IS_PUSH_MODE(f)   FALSE
#elif defined(STB_VORBIS_NO_PULLDATA_API)
   #define IS_PUSH_MODE(f)   TRUE
#else
   #define IS_PUSH_MODE(f)   ((f)->push_mode)
#endif

typedef struct stb_vorbis vorb;

static int error(vorb *f, enum STBVorbisError e)
{
   f->error = e;
   if (!f->eof && e != VORBIS_need_more_data) {
      f->error=e; // breakpoint for debugging
   }
   return 0;
}


// these functions are used for allocating temporary memory
// while decoding. if you can afford the stack space, use
// alloca(); otherwise, provide a temp buffer and it will
// allocate out of those.

#define array_size_required(count,size)  (count*(sizeof(void *)+(size)))

#define temp_alloc(f,size)              (f->alloc.alloc_buffer ? setup_temp_malloc(f,size) : alloca(size))
#define temp_free(f,p)                  (void)0
#define temp_alloc_save(f)              ((f)->temp_offset)
#define temp_alloc_restore(f,p)         ((f)->temp_offset = (p))

#define temp_block_array(f,count,size)  make_block_array(temp_alloc(f,array_size_required(count,size)), count, size)

// given a sufficiently large block of memory, make an array of pointers to subblocks of it
static void *make_block_array(void *mem, int count, int size)
{
   int i;
   void ** p = (void **) mem;
   char *q = (char *) (p + count);
   for (i=0; i < count; ++i) {
      p[i] = q;
      q += size;
   }
   return p;
}

static void *setup_malloc(vorb *f, int sz)
{
   sz = (sz+7) & ~7; // round up to nearest 8 for alignment of future allocs.
   f->setup_memory_required += sz;
   if (f->alloc.alloc_buffer) {
      void *p = (char *) f->alloc.alloc_buffer + f->setup_offset;
      if (f->setup_offset + sz > f->temp_offset) return NULL;
      f->setup_offset += sz;
      return p;
   }
   return sz ? malloc(sz) : NULL;
}

static void setup_free(vorb *f, void *p)
{
   if (f->alloc.alloc_buffer) return; // do nothing; setup mem is a stack
   free(p);
}

static void *setup_temp_malloc(vorb *f, int sz)
{
   sz = (sz+7) & ~7; // round up to nearest 8 for alignment of future allocs.
   if (f->alloc.alloc_buffer) {
      if (f->temp_offset - sz < f->setup_offset) return NULL;
      f->temp_offset -= sz;
      return (char *) f->alloc.alloc_buffer + f->temp_offset;
   }
   return malloc(sz);
}

static void setup_temp_free(vorb *f, void *p, int sz)
{
   if (f->alloc.alloc_buffer) {
      f->temp_offset += (sz+7)&~7;
      return;
   }
   free(p);
}

#define CRC32_POLY    0x04c11db7   // from spec

static uint32 crc_table[256];
static void crc32_init(void)
{
   int i,j;
   uint32 s;
   for(i=0; i < 256; i++) {
      for (s=(uint32) i << 24, j=0; j < 8; ++j)
         s = (s << 1) ^ (s >= (1U<<31) ? CRC32_POLY : 0);
      crc_table[i] = s;
   }
}

static __forceinline uint32 crc32_update(uint32 crc, uint8 byte)
{
   return (crc << 8) ^ crc_table[byte ^ (crc >> 24)];
}


// used in setup, and for huffman that doesn't go fast path
static unsigned int bit_reverse(unsigned int n)
{
  n = ((n & 0xAAAAAAAA) >>  1) | ((n & 0x55555555) << 1);
  n = ((n & 0xCCCCCCCC) >>  2) | ((n & 0x33333333) << 2);
  n = ((n & 0xF0F0F0F0) >>  4) | ((n & 0x0F0F0F0F) << 4);
  n = ((n & 0xFF00FF00) >>  8) | ((n & 0x00FF00FF) << 8);
  return (n >> 16) | (n << 16);
}

static float square(float x)
{
   return x*x;
}

// this is a weird definition of log2() for which log2(1) = 1, log2(2) = 2, log2(4) = 3
// as required by the specification. fast(?) implementation from stb.h
// @OPTIMIZE: called multiple times per-packet with "constants"; move to setup
static int ilog(int32 n)
{
   static signed char log2_4[16] = { 0,1,2,2,3,3,3,3,4,4,4,4,4,4,4,4 };

   if (n < 0) return 0; // signed n returns 0

   // 2 compares if n < 16, 3 compares otherwise (4 if signed or n > 1<<29)
   if (n < (1 << 14))
        if (n < (1 <<  4))            return  0 + log2_4[n      ];
        else if (n < (1 <<  9))       return  5 + log2_4[n >>  5];
             else                     return 10 + log2_4[n >> 10];
   else if (n < (1 << 24))
             if (n < (1 << 19))       return 15 + log2_4[n >> 15];
             else                     return 20 + log2_4[n >> 20];
        else if (n < (1 << 29))       return 25 + log2_4[n >> 25];
             else                     return 30 + log2_4[n >> 30];
}

#ifndef M_PI
  #define M_PI  3.14159265358979323846264f  // from CRC
#endif

// code length assigned to a value with no huffman encoding
#define NO_CODE   255

/////////////////////// LEAF SETUP FUNCTIONS //////////////////////////
//
// these functions are only called at setup, and only a few times
// per file

static float float32_unpack(uint32 x)
{
   // from the specification
   uint32 mantissa = x & 0x1fffff;
   uint32 sign = x & 0x80000000;
   uint32 exp = (x & 0x7fe00000) >> 21;
   double res = sign ? -(double)mantissa : (double)mantissa;
   return (float) ldexp((float)res, (int)exp-788);
}


// zlib & jpeg huffman tables assume that the output symbols
// can either be arbitrarily arranged, or have monotonically
// increasing frequencies--they rely on the lengths being sorted;
// this makes for a very simple generation algorithm.
// vorbis allows a huffman table with non-sorted lengths. This
// requires a more sophisticated construction, since symbols in
// order do not map to huffman codes "in order".
static void add_entry(Codebook *c, uint32 huff_code, int symbol, int count, int len, uint32 *values)
{
   if (!c->sparse) {
      c->codewords      [symbol] = huff_code;
   } else {
      c->codewords       [count] = huff_code;
      c->codeword_lengths[count] = len;
      values             [count] = symbol;
   }
}

static int compute_codewords(Codebook *c, uint8 *len, int n, uint32 *values)
{
   int i,k,m=0;
   uint32 available[32];

   memset(available, 0, sizeof(available));
   // find the first entry
   for (k=0; k < n; ++k) if (len[k] < NO_CODE) break;
   if (k == n) { assert(c->sorted_entries == 0); return TRUE; }
   assert(len[k] < 32); // no error return required, code reading lens checks this
   // add to the list
   add_entry(c, 0, k, m++, len[k], values);
   // add all available leaves
   for (i=1; i <= len[k]; ++i)
      available[i] = 1U << (32-i);
   // note that the above code treats the first case specially,
   // but it's really the same as the following code, so they
   // could probably be combined (except the initial code is 0,
   // and I use 0 in available[] to mean 'empty')
   for (i=k+1; i < n; ++i) {
      uint32 res;
      int z = len[i], y;
      if (z == NO_CODE) continue;
      assert(z < 32); // no error return required, code reading lens checks this
      // find lowest available leaf (should always be earliest,
      // which is what the specification calls for)
      // note that this property, and the fact we can never have
      // more than one free leaf at a given level, isn't totally
      // trivial to prove, but it seems true and the assert never
      // fires, so!
      while (z > 0 && !available[z]) --z;
      if (z == 0) { return FALSE; }
      res = available[z];
      available[z] = 0;
      add_entry(c, bit_reverse(res), i, m++, len[i], values);
      // propagate availability up the tree
      if (z != len[i]) {
         for (y=len[i]; y > z; --y) {
            assert(available[y] == 0);
            available[y] = res + (1 << (32-y));
         }
      }
   }
   return TRUE;
}

// accelerated huffman table allows fast O(1) match of all symbols
// of length <= STB_VORBIS_FAST_HUFFMAN_LENGTH
static void compute_accelerated_huffman(Codebook *c)
{
   int i, len;
   for (i=0; i < FAST_HUFFMAN_TABLE_SIZE; ++i)
      c->fast_huffman[i] = -1;

   len = c->sparse ? c->sorted_entries : c->entries;
   #ifdef STB_VORBIS_FAST_HUFFMAN_SHORT
   if (len > 32767) len = 32767; // largest possible value we can encode!
   #endif
   for (i=0; i < len; ++i) {
      if (c->codeword_lengths[i] <= STB_VORBIS_FAST_HUFFMAN_LENGTH) {
         uint32 z = c->sparse ? bit_reverse(c->sorted_codewords[i]) : c->codewords[i];
         // set table entries for all bit combinations in the higher bits
         while (z < FAST_HUFFMAN_TABLE_SIZE) {
             c->fast_huffman[z] = i;
             z += 1 << c->codeword_lengths[i];
         }
      }
   }
}

#ifdef _MSC_VER
#define STBV_CDECL __cdecl
#else
#define STBV_CDECL
#endif

static int STBV_CDECL uint32_compare(const void *p, const void *q)
{
   uint32 x = * (uint32 *) p;
   uint32 y = * (uint32 *) q;
   return x < y ? -1 : x > y;
}

static int include_in_sort(Codebook *c, uint8 len)
{
   if (c->sparse) { assert(len != NO_CODE); return TRUE; }
   if (len == NO_CODE) return FALSE;
   if (len > STB_VORBIS_FAST_HUFFMAN_LENGTH) return TRUE;
   return FALSE;
}

// if the fast table above doesn't work, we want to binary
// search them... need to reverse the bits
static void compute_sorted_huffman(Codebook *c, uint8 *lengths, uint32 *values)
{
   int i, len;
   // build a list of all the entries
   // OPTIMIZATION: don't include the short ones, since they'll be caught by FAST_HUFFMAN.
   // this is kind of a frivolous optimization--I don't see any performance improvement,
   // but it's like 4 extra lines of code, so.
   if (!c->sparse) {
      int k = 0;
      for (i=0; i < c->entries; ++i)
         if (include_in_sort(c, lengths[i]))
            c->sorted_codewords[k++] = bit_reverse(c->codewords[i]);
      assert(k == c->sorted_entries);
   } else {
      for (i=0; i < c->sorted_entries; ++i)
         c->sorted_codewords[i] = bit_reverse(c->codewords[i]);
   }

   qsort(c->sorted_codewords, c->sorted_entries, sizeof(c->sorted_codewords[0]), uint32_compare);
   c->sorted_codewords[c->sorted_entries] = 0xffffffff;

   len = c->sparse ? c->sorted_entries : c->entries;
   // now we need to indicate how they correspond; we could either
   //   #1: sort a different data structure that says who they correspond to
   //   #2: for each sorted entry, search the original list to find who corresponds
   //   #3: for each original entry, find the sorted entry
   // #1 requires extra storage, #2 is slow, #3 can use binary search!
   for (i=0; i < len; ++i) {
      int huff_len = c->sparse ? lengths[values[i]] : lengths[i];
      if (include_in_sort(c,huff_len)) {
         uint32 code = bit_reverse(c->codewords[i]);
         int x=0, n=c->sorted_entries;
         while (n > 1) {
            // invariant: sc[x] <= code < sc[x+n]
            int m = x + (n >> 1);
            if (c->sorted_codewords[m] <= code) {
               x = m;
               n -= (n>>1);
            } else {
               n >>= 1;
            }
         }
         assert(c->sorted_codewords[x] == code);
         if (c->sparse) {
            c->sorted_values[x] = values[i];
            c->codeword_lengths[x] = huff_len;
         } else {
            c->sorted_values[x] = i;
         }
      }
   }
}

// only run while parsing the header (3 times)
static int vorbis_validate(uint8 *data)
{
   static uint8 vorbis[6] = { 'v', 'o', 'r', 'b', 'i', 's' };
   return memcmp(data, vorbis, 6) == 0;
}

// called from setup only, once per code book
// (formula implied by specification)
static int lookup1_values(int entries, int dim)
{
   int r = (int) floor(exp((float) log((float) entries) / dim));
   if ((int) floor(pow((float) r+1, dim)) <= entries)   // (int) cast for MinGW warning;
      ++r;                                              // floor() to avoid _ftol() when non-CRT
   if (pow((float) r+1, dim) <= entries)
      return -1;
   if ((int) floor(pow((float) r, dim)) > entries)
      return -1;
   return r;
}

// called twice per file
static void compute_twiddle_factors(int n, float *A, float *B, float *C)
{
   int n4 = n >> 2, n8 = n >> 3;
   int k,k2;

   for (k=k2=0; k < n4; ++k,k2+=2) {
      A[k2  ] = (float)  cos(4*k*M_PI/n);
      A[k2+1] = (float) -sin(4*k*M_PI/n);
      B[k2  ] = (float)  cos((k2+1)*M_PI/n/2) * 0.5f;
      B[k2+1] = (float)  sin((k2+1)*M_PI/n/2) * 0.5f;
   }
   for (k=k2=0; k < n8; ++k,k2+=2) {
      C[k2  ] = (float)  cos(2*(k2+1)*M_PI/n);
      C[k2+1] = (float) -sin(2*(k2+1)*M_PI/n);
   }
}

static void compute_window(int n, float *window)
{
   int n2 = n >> 1, i;
   for (i=0; i < n2; ++i)
      window[i] = (float) sin(0.5 * M_PI * square((float) sin((i - 0 + 0.5) / n2 * 0.5 * M_PI)));
}

static void compute_bitreverse(int n, uint16 *rev)
{
   int ld = ilog(n) - 1; // ilog is off-by-one from normal definitions
   int i, n8 = n >> 3;
   for (i=0; i < n8; ++i)
      rev[i] = (bit_reverse(i) >> (32-ld+3)) << 2;
}

static int init_blocksize(vorb *f, int b, int n)
{
   int n2 = n >> 1, n4 = n >> 2, n8 = n >> 3;
   f->A[b] = (float *) setup_malloc(f, sizeof(float) * n2);
   f->B[b] = (float *) setup_malloc(f, sizeof(float) * n2);
   f->C[b] = (float *) setup_malloc(f, sizeof(float) * n4);
   if (!f->A[b] || !f->B[b] || !f->C[b]) return error(f, VORBIS_outofmem);
   compute_twiddle_factors(n, f->A[b], f->B[b], f->C[b]);
   f->window[b] = (float *) setup_malloc(f, sizeof(float) * n2);
   if (!f->window[b]) return error(f, VORBIS_outofmem);
   compute_window(n, f->window[b]);
   f->bit_reverse[b] = (uint16 *) setup_malloc(f, sizeof(uint16) * n8);
   if (!f->bit_reverse[b]) return error(f, VORBIS_outofmem);
   compute_bitreverse(n, f->bit_reverse[b]);
   return TRUE;
}

static void neighbors(uint16 *x, int n, int *plow, int *phigh)
{
   int low = -1;
   int high = 65536;
   int i;
   for (i=0; i < n; ++i) {
      if (x[i] > low  && x[i] < x[n]) { *plow  = i; low = x[i]; }
      if (x[i] < high && x[i] > x[n]) { *phigh = i; high = x[i]; }
   }
}

// this has been repurposed so y is now the original index instead of y
typedef struct
{
   uint16 x,id;
} stbv__floor_ordering;

static int STBV_CDECL point_compare(const void *p, const void *q)
{
   stbv__floor_ordering *a = (stbv__floor_ordering *) p;
   stbv__floor_ordering *b = (stbv__floor_ordering *) q;
   return a->x < b->x ? -1 : a->x > b->x;
}

//
/////////////////////// END LEAF SETUP FUNCTIONS //////////////////////////


#if defined(STB_VORBIS_NO_STDIO)
   #define USE_MEMORY(z)    TRUE
#else
   #define USE_MEMORY(z)    ((z)->stream)
#endif

static uint8 get8(vorb *z)
{
   if (USE_MEMORY(z)) {
      if (z->stream >= z->stream_end) { z->eof = TRUE; return 0; }
      return *z->stream++;
   }

   #ifndef STB_VORBIS_NO_STDIO
   {
   int c = fgetc(z->f);
   if (c == EOF) { z->eof = TRUE; return 0; }
   return c;
   }
   #endif
}

static uint32 get32(vorb *f)
{
   uint32 x;
   x = get8(f);
   x += get8(f) << 8;
   x += get8(f) << 16;
   x += (uint32) get8(f) << 24;
   return x;
}

static int getn(vorb *z, uint8 *data, int n)
{
   if (USE_MEMORY(z)) {
      if (z->stream+n > z->stream_end) { z->eof = 1; return 0; }
      memcpy(data, z->stream, n);
      z->stream += n;
      return 1;
   }

   #ifndef STB_VORBIS_NO_STDIO
   if (fread(data, n, 1, z->f) == 1)
      return 1;
   else {
      z->eof = 1;
      return 0;
   }
   #endif
}

static void skip(vorb *z, int n)
{
   if (USE_MEMORY(z)) {
      z->stream += n;
      if (z->stream >= z->stream_end) z->eof = 1;
      return;
   }
   #ifndef STB_VORBIS_NO_STDIO
   {
      long x = ftell(z->f);
      fseek(z->f, x+n, SEEK_SET);
   }
   #endif
}

static int set_file_offset(stb_vorbis *f, unsigned int loc)
{
   #ifndef STB_VORBIS_NO_PUSHDATA_API
   if (f->push_mode) return 0;
   #endif
   f->eof = 0;
   if (USE_MEMORY(f)) {
      if (f->stream_start + loc >= f->stream_end || f->stream_start + loc < f->stream_start) {
         f->stream = f->stream_end;
         f->eof = 1;
         return 0;
      } else {
         f->stream = f->stream_start + loc;
         return 1;
      }
   }
   #ifndef STB_VORBIS_NO_STDIO
   if (loc + f->f_start < loc || loc >= 0x80000000) {
      loc = 0x7fffffff;
      f->eof = 1;
   } else {
      loc += f->f_start;
   }
   if (!fseek(f->f, loc, SEEK_SET))
      return 1;
   f->eof = 1;
   fseek(f->f, f->f_start, SEEK_END);
   return 0;
   #endif
}


static uint8 ogg_page_header[4] = { 0x4f, 0x67, 0x67, 0x53 };

static int capture_pattern(vorb *f)
{
   if (0x4f != get8(f)) return FALSE;
   if (0x67 != get8(f)) return FALSE;
   if (0x67 != get8(f)) return FALSE;
   if (0x53 != get8(f)) return FALSE;
   return TRUE;
}

#define PAGEFLAG_continued_packet   1
#define PAGEFLAG_first_page         2
#define PAGEFLAG_last_page          4

static int start_page_no_capturepattern(vorb *f)
{
   uint32 loc0,loc1,n;
   if (f->first_decode && !IS_PUSH_MODE(f)) {
      f->p_first.page_start = stb_vorbis_get_file_offset(f) - 4;
   }
   // stream structure version
   if (0 != get8(f)) return error(f, VORBIS_invalid_stream_structure_version);
   // header flag
   f->page_flag = get8(f);
   // absolute granule position
   loc0 = get32(f);
   loc1 = get32(f);
   // @TODO: validate loc0,loc1 as valid positions?
   // stream serial number -- vorbis doesn't interleave, so discard
   get32(f);
   //if (f->serial != get32(f)) return error(f, VORBIS_incorrect_stream_serial_number);
   // page sequence number
   n = get32(f);
   f->last_page = n;
   // CRC32
   get32(f);
   // page_segments
   f->segment_count = get8(f);
   if (!getn(f, f->segments, f->segment_count))
      return error(f, VORBIS_unexpected_eof);
   // assume we _don't_ know any the sample position of any segments
   f->end_seg_with_known_loc = -2;
   if (loc0 != ~0U || loc1 != ~0U) {
      int i;
      // determine which packet is the last one that will complete
      for (i=f->segment_count-1; i >= 0; --i)
         if (f->segments[i] < 255)
            break;
      // 'i' is now the index of the _last_ segment of a packet that ends
      if (i >= 0) {
         f->end_seg_with_known_loc = i;
         f->known_loc_for_packet   = loc0;
      }
   }
   if (f->first_decode) {
      int i,len;
      len = 0;
      for (i=0; i < f->segment_count; ++i)
         len += f->segments[i];
      len += 27 + f->segment_count;
      f->p_first.page_end = f->p_first.page_start + len;
      f->p_first.last_decoded_sample = loc0;
   }
   f->next_seg = 0;
   return TRUE;
}

static int start_page(vorb *f)
{
   if (!capture_pattern(f)) return error(f, VORBIS_missing_capture_pattern);
   return start_page_no_capturepattern(f);
}

static int start_packet(vorb *f)
{
   while (f->next_seg == -1) {
      if (!start_page(f)) return FALSE;
      if (f->page_flag & PAGEFLAG_continued_packet)
         return error(f, VORBIS_continued_packet_flag_invalid);
   }
   f->last_seg = FALSE;
   f->valid_bits = 0;
   f->packet_bytes = 0;
   f->bytes_in_seg = 0;
   // f->next_seg is now valid
   return TRUE;
}

static int maybe_start_packet(vorb *f)
{
   if (f->next_seg == -1) {
      int x = get8(f);
      if (f->eof) return FALSE; // EOF at page boundary is not an error!
      if (0x4f != x      ) return error(f, VORBIS_missing_capture_pattern);
      if (0x67 != get8(f)) return error(f, VORBIS_missing_capture_pattern);
      if (0x67 != get8(f)) return error(f, VORBIS_missing_capture_pattern);
      if (0x53 != get8(f)) return error(f, VORBIS_missing_capture_pattern);
      if (!start_page_no_capturepattern(f)) return FALSE;
      if (f->page_flag & PAGEFLAG_continued_packet) {
         // set up enough state that we can read this packet if we want,
         // e.g. during recovery
         f->last_seg = FALSE;
         f->bytes_in_seg = 0;
         return error(f, VORBIS_continued_packet_flag_invalid);
      }
   }
   return start_packet(f);
}

static int next_segment(vorb *f)
{
   int len;
   if (f->last_seg) return 0;
   if (f->next_seg == -1) {
      f->last_seg_which = f->segment_count-1; // in case start_page fails
      if (!start_page(f)) { f->last_seg = 1; return 0; }
      if (!(f->page_flag & PAGEFLAG_continued_packet)) return error(f, VORBIS_continued_packet_flag_invalid);
   }
   len = f->segments[f->next_seg++];
   if (len < 255) {
      f->last_seg = TRUE;
      f->last_seg_which = f->next_seg-1;
   }
   if (f->next_seg >= f->segment_count)
      f->next_seg = -1;
   assert(f->bytes_in_seg == 0);
   f->bytes_in_seg = len;
   return len;
}

#define EOP    (-1)
#define INVALID_BITS  (-1)

static int get8_packet_raw(vorb *f)
{
   if (!f->bytes_in_seg) {  // CLANG!
      if (f->last_seg) return EOP;
      else if (!next_segment(f)) return EOP;
   }
   assert(f->bytes_in_seg > 0);
   --f->bytes_in_seg;
   ++f->packet_bytes;
   return get8(f);
}

static int get8_packet(vorb *f)
{
   int x = get8_packet_raw(f);
   f->valid_bits = 0;
   return x;
}

static int get32_packet(vorb *f)
{
   uint32 x;
   x = get8_packet(f);
   x += get8_packet(f) << 8;
   x += get8_packet(f) << 16;
   x += (uint32) get8_packet(f) << 24;
   return x;
}

static void flush_packet(vorb *f)
{
   while (get8_packet_raw(f) != EOP);
}

// @OPTIMIZE: this is the secondary bit decoder, so it's probably not as important
// as the huffman decoder?
static uint32 get_bits(vorb *f, int n)
{
   uint32 z;

   if (f->valid_bits < 0) return 0;
   if (f->valid_bits < n) {
      if (n > 24) {
         // the accumulator technique below would not work correctly in this case
         z = get_bits(f, 24);
         z += get_bits(f, n-24) << 24;
         return z;
      }
      if (f->valid_bits == 0) f->acc = 0;
      while (f->valid_bits < n) {
         int z = get8_packet_raw(f);
         if (z == EOP) {
            f->valid_bits = INVALID_BITS;
            return 0;
         }
         f->acc += z << f->valid_bits;
         f->valid_bits += 8;
      }
   }

   assert(f->valid_bits >= n);
   z = f->acc & ((1 << n)-1);
   f->acc >>= n;
   f->valid_bits -= n;
   return z;
}

// @OPTIMIZE: primary accumulator for huffman
// expand the buffer to as many bits as possible without reading off end of packet
// it might be nice to allow f->valid_bits and f->acc to be stored in registers,
// e.g. cache them locally and decode locally
static __forceinline void prep_huffman(vorb *f)
{
   if (f->valid_bits <= 24) {
      if (f->valid_bits == 0) f->acc = 0;
      do {
         int z;
         if (f->last_seg && !f->bytes_in_seg) return;
         z = get8_packet_raw(f);
         if (z == EOP) return;
         f->acc += (unsigned) z << f->valid_bits;
         f->valid_bits += 8;
      } while (f->valid_bits <= 24);
   }
}

enum
{
   VORBIS_packet_id = 1,
   VORBIS_packet_comment = 3,
   VORBIS_packet_setup = 5
};

static int codebook_decode_scalar_raw(vorb *f, Codebook *c)
{
   int i;
   prep_huffman(f);

   if (c->codewords == NULL && c->sorted_codewords == NULL)
      return -1;

   // cases to use binary search: sorted_codewords && !c->codewords
   //                             sorted_codewords && c->entries > 8
   if (c->entries > 8 ? c->sorted_codewords!=NULL : !c->codewords) {
      // binary search
      uint32 code = bit_reverse(f->acc);
      int x=0, n=c->sorted_entries, len;

      while (n > 1) {
         // invariant: sc[x] <= code < sc[x+n]
         int m = x + (n >> 1);
         if (c->sorted_codewords[m] <= code) {
            x = m;
            n -= (n>>1);
         } else {
            n >>= 1;
         }
      }
      // x is now the sorted index
      if (!c->sparse) x = c->sorted_values[x];
      // x is now sorted index if sparse, or symbol otherwise
      len = c->codeword_lengths[x];
      if (f->valid_bits >= len) {
         f->acc >>= len;
         f->valid_bits -= len;
         return x;
      }

      f->valid_bits = 0;
      return -1;
   }

   // if small, linear search
   assert(!c->sparse);
   for (i=0; i < c->entries; ++i) {
      if (c->codeword_lengths[i] == NO_CODE) continue;
      if (c->codewords[i] == (f->acc & ((1 << c->codeword_lengths[i])-1))) {
         if (f->valid_bits >= c->codeword_lengths[i]) {
            f->acc >>= c->codeword_lengths[i];
            f->valid_bits -= c->codeword_lengths[i];
            return i;
         }
         f->valid_bits = 0;
         return -1;
      }
   }

   error(f, VORBIS_invalid_stream);
   f->valid_bits = 0;
   return -1;
}

#ifndef STB_VORBIS_NO_INLINE_DECODE

#define DECODE_RAW(var, f,c)                                  \
   if (f->valid_bits < STB_VORBIS_FAST_HUFFMAN_LENGTH)        \
      prep_huffman(f);                                        \
   var = f->acc & FAST_HUFFMAN_TABLE_MASK;                    \
   var = c->fast_huffman[var];                                \
   if (var >= 0) {                                            \
      int n = c->codeword_lengths[var];                       \
      f->acc >>= n;                                           \
      f->valid_bits -= n;                                     \
      if (f->valid_bits < 0) { f->valid_bits = 0; var = -1; } \
   } else {                                                   \
      var = codebook_decode_scalar_raw(f,c);                  \
   }

#else

static int codebook_decode_scalar(vorb *f, Codebook *c)
{
   int i;
   if (f->valid_bits < STB_VORBIS_FAST_HUFFMAN_LENGTH)
      prep_huffman(f);
   // fast huffman table lookup
   i = f->acc & FAST_HUFFMAN_TABLE_MASK;
   i = c->fast_huffman[i];
   if (i >= 0) {
      f->acc >>= c->codeword_lengths[i];
      f->valid_bits -= c->codeword_lengths[i];
      if (f->valid_bits < 0) { f->valid_bits = 0; return -1; }
      return i;
   }
   return codebook_decode_scalar_raw(f,c);
}

#define DECODE_RAW(var,f,c)    var = codebook_decode_scalar(f,c);

#endif

#define DECODE(var,f,c)                                       \
   DECODE_RAW(var,f,c)                                        \
   if (c->sparse) var = c->sorted_values[var];

#ifndef STB_VORBIS_DIVIDES_IN_CODEBOOK
  #define DECODE_VQ(var,f,c)   DECODE_RAW(var,f,c)
#else
  #define DECODE_VQ(var,f,c)   DECODE(var,f,c)
#endif






// CODEBOOK_ELEMENT_FAST is an optimization for the CODEBOOK_FLOATS case
// where we avoid one addition
#define CODEBOOK_ELEMENT(c,off)          (c->multiplicands[off])
#define CODEBOOK_ELEMENT_FAST(c,off)     (c->multiplicands[off])
#define CODEBOOK_ELEMENT_BASE(c)         (0)

static int codebook_decode_start(vorb *f, Codebook *c)
{
   int z = -1;

   // type 0 is only legal in a scalar context
   if (c->lookup_type == 0)
      error(f, VORBIS_invalid_stream);
   else {
      DECODE_VQ(z,f,c);
      if (c->sparse) assert(z < c->sorted_entries);
      if (z < 0) {  // check for EOP
         if (!f->bytes_in_seg)
            if (f->last_seg)
               return z;
         error(f, VORBIS_invalid_stream);
      }
   }
   return z;
}

static int codebook_decode(vorb *f, Codebook *c, float *output, int len)
{
   int i,z = codebook_decode_start(f,c);
   if (z < 0) return FALSE;
   if (len > c->dimensions) len = c->dimensions;

#ifdef STB_VORBIS_DIVIDES_IN_CODEBOOK
   if (c->lookup_type == 1) {
      float last = CODEBOOK_ELEMENT_BASE(c);
      int div = 1;
      for (i=0; i < len; ++i) {
         int off = (z / div) % c->lookup_values;
         float val = CODEBOOK_ELEMENT_FAST(c,off) + last;
         output[i] += val;
         if (c->sequence_p) last = val + c->minimum_value;
         div *= c->lookup_values;
      }
      return TRUE;
   }
#endif

   z *= c->dimensions;
   if (c->sequence_p) {
      float last = CODEBOOK_ELEMENT_BASE(c);
      for (i=0; i < len; ++i) {
         float val = CODEBOOK_ELEMENT_FAST(c,z+i) + last;
         output[i] += val;
         last = val + c->minimum_value;
      }
   } else {
      float last = CODEBOOK_ELEMENT_BASE(c);
      for (i=0; i < len; ++i) {
         output[i] += CODEBOOK_ELEMENT_FAST(c,z+i) + last;
      }
   }

   return TRUE;
}

static int codebook_decode_step(vorb *f, Codebook *c, float *output, int len, int step)
{
   int i,z = codebook_decode_start(f,c);
   float last = CODEBOOK_ELEMENT_BASE(c);
   if (z < 0) return FALSE;
   if (len > c->dimensions) len = c->dimensions;

#ifdef STB_VORBIS_DIVIDES_IN_CODEBOOK
   if (c->lookup_type == 1) {
      int div = 1;
      for (i=0; i < len; ++i) {
         int off = (z / div) % c->lookup_values;
         float val = CODEBOOK_ELEMENT_FAST(c,off) + last;
         output[i*step] += val;
         if (c->sequence_p) last = val;
         div *= c->lookup_values;
      }
      return TRUE;
   }
#endif

   z *= c->dimensions;
   for (i=0; i < len; ++i) {
      float val = CODEBOOK_ELEMENT_FAST(c,z+i) + last;
      output[i*step] += val;
      if (c->sequence_p) last = val;
   }

   return TRUE;
}

static int codebook_decode_deinterleave_repeat(vorb *f, Codebook *c, float **outputs, int ch, int *c_inter_p, int *p_inter_p, int len, int total_decode)
{
   int c_inter = *c_inter_p;
   int p_inter = *p_inter_p;
   int i,z, effective = c->dimensions;

   // type 0 is only legal in a scalar context
   if (c->lookup_type == 0)   return error(f, VORBIS_invalid_stream);

   while (total_decode > 0) {
      float last = CODEBOOK_ELEMENT_BASE(c);
      DECODE_VQ(z,f,c);
      #ifndef STB_VORBIS_DIVIDES_IN_CODEBOOK
      assert(!c->sparse || z < c->sorted_entries);
      #endif
      if (z < 0) {
         if (!f->bytes_in_seg)
            if (f->last_seg) return FALSE;
         return error(f, VORBIS_invalid_stream);
      }

      // if this will take us off the end of the buffers, stop short!
      // we check by computing the length of the virtual interleaved
      // buffer (len*ch), our current offset within it (p_inter*ch)+(c_inter),
      // and the length we'll be using (effective)
      if (c_inter + p_inter*ch + effective > len * ch) {
         effective = len*ch - (p_inter*ch - c_inter);
      }

   #ifdef STB_VORBIS_DIVIDES_IN_CODEBOOK
      if (c->lookup_type == 1) {
         int div = 1;
         for (i=0; i < effective; ++i) {
            int off = (z / div) % c->lookup_values;
            float val = CODEBOOK_ELEMENT_FAST(c,off) + last;
            if (outputs[c_inter])
               outputs[c_inter][p_inter] += val;
            if (++c_inter == ch) { c_inter = 0; ++p_inter; }
            if (c->sequence_p) last = val;
            div *= c->lookup_values;
         }
      } else
   #endif
      {
         z *= c->dimensions;
         if (c->sequence_p) {
            for (i=0; i < effective; ++i) {
               float val = CODEBOOK_ELEMENT_FAST(c,z+i) + last;
               if (outputs[c_inter])
                  outputs[c_inter][p_inter] += val;
               if (++c_inter == ch) { c_inter = 0; ++p_inter; }
               last = val;
            }
         } else {
            for (i=0; i < effective; ++i) {
               float val = CODEBOOK_ELEMENT_FAST(c,z+i) + last;
               if (outputs[c_inter])
                  outputs[c_inter][p_inter] += val;
               if (++c_inter == ch) { c_inter = 0; ++p_inter; }
            }
         }
      }

      total_decode -= effective;
   }
   *c_inter_p = c_inter;
   *p_inter_p = p_inter;
   return TRUE;
}

static int predict_point(int x, int x0, int x1, int y0, int y1)
{
   int dy = y1 - y0;
   int adx = x1 - x0;
   // @OPTIMIZE: force int division to round in the right direction... is this necessary on x86?
   int err = abs(dy) * (x - x0);
   int off = err / adx;
   return dy < 0 ? y0 - off : y0 + off;
}

// the following table is block-copied from the specification
static float inverse_db_table[256] =
{
  1.0649863e-07f, 1.1341951e-07f, 1.2079015e-07f, 1.2863978e-07f,
  1.3699951e-07f, 1.4590251e-07f, 1.5538408e-07f, 1.6548181e-07f,
  1.7623575e-07f, 1.8768855e-07f, 1.9988561e-07f, 2.1287530e-07f,
  2.2670913e-07f, 2.4144197e-07f, 2.5713223e-07f, 2.7384213e-07f,
  2.9163793e-07f, 3.1059021e-07f, 3.3077411e-07f, 3.5226968e-07f,
  3.7516214e-07f, 3.9954229e-07f, 4.2550680e-07f, 4.5315863e-07f,
  4.8260743e-07f, 5.1396998e-07f, 5.4737065e-07f, 5.8294187e-07f,
  6.2082472e-07f, 6.6116941e-07f, 7.0413592e-07f, 7.4989464e-07f,
  7.9862701e-07f, 8.5052630e-07f, 9.0579828e-07f, 9.6466216e-07f,
  1.0273513e-06f, 1.0941144e-06f, 1.1652161e-06f, 1.2409384e-06f,
  1.3215816e-06f, 1.4074654e-06f, 1.4989305e-06f, 1.5963394e-06f,
  1.7000785e-06f, 1.8105592e-06f, 1.9282195e-06f, 2.0535261e-06f,
  2.1869758e-06f, 2.3290978e-06f, 2.4804557e-06f, 2.6416497e-06f,
  2.8133190e-06f, 2.9961443e-06f, 3.1908506e-06f, 3.3982101e-06f,
  3.6190449e-06f, 3.8542308e-06f, 4.1047004e-06f, 4.3714470e-06f,
  4.6555282e-06f, 4.9580707e-06f, 5.2802740e-06f, 5.6234160e-06f,
  5.9888572e-06f, 6.3780469e-06f, 6.7925283e-06f, 7.2339451e-06f,
  7.7040476e-06f, 8.2047000e-06f, 8.7378876e-06f, 9.3057248e-06f,
  9.9104632e-06f, 1.0554501e-05f, 1.1240392e-05f, 1.1970856e-05f,
  1.2748789e-05f, 1.3577278e-05f, 1.4459606e-05f, 1.5399272e-05f,
  1.6400004e-05f, 1.7465768e-05f, 1.8600792e-05f, 1.9809576e-05f,
  2.1096914e-05f, 2.2467911e-05f, 2.3928002e-05f, 2.5482978e-05f,
  2.7139006e-05f, 2.8902651e-05f, 3.0780908e-05f, 3.2781225e-05f,
  3.4911534e-05f, 3.7180282e-05f, 3.9596466e-05f, 4.2169667e-05f,
  4.4910090e-05f, 4.7828601e-05f, 5.0936773e-05f, 5.4246931e-05f,
  5.7772202e-05f, 6.1526565e-05f, 6.5524908e-05f, 6.9783085e-05f,
  7.4317983e-05f, 7.9147585e-05f, 8.4291040e-05f, 8.9768747e-05f,
  9.5602426e-05f, 0.00010181521f, 0.00010843174f, 0.00011547824f,
  0.00012298267f, 0.00013097477f, 0.00013948625f, 0.00014855085f,
  0.00015820453f, 0.00016848555f, 0.00017943469f, 0.00019109536f,
  0.00020351382f, 0.00021673929f, 0.00023082423f, 0.00024582449f,
  0.00026179955f, 0.00027881276f, 0.00029693158f, 0.00031622787f,
  0.00033677814f, 0.00035866388f, 0.00038197188f, 0.00040679456f,
  0.00043323036f, 0.00046138411f, 0.00049136745f, 0.00052329927f,
  0.00055730621f, 0.00059352311f, 0.00063209358f, 0.00067317058f,
  0.00071691700f, 0.00076350630f, 0.00081312324f, 0.00086596457f,
  0.00092223983f, 0.00098217216f, 0.0010459992f,  0.0011139742f,
  0.0011863665f,  0.0012634633f,  0.0013455702f,  0.0014330129f,
  0.0015261382f,  0.0016253153f,  0.0017309374f,  0.0018434235f,
  0.0019632195f,  0.0020908006f,  0.0022266726f,  0.0023713743f,
  0.0025254795f,  0.0026895994f,  0.0028643847f,  0.0030505286f,
  0.0032487691f,  0.0034598925f,  0.0036847358f,  0.0039241906f,
  0.0041792066f,  0.0044507950f,  0.0047400328f,  0.0050480668f,
  0.0053761186f,  0.0057254891f,  0.0060975636f,  0.0064938176f,
  0.0069158225f,  0.0073652516f,  0.0078438871f,  0.0083536271f,
  0.0088964928f,  0.009474637f,   0.010090352f,   0.010746080f,
  0.011444421f,   0.012188144f,   0.012980198f,   0.013823725f,
  0.014722068f,   0.015678791f,   0.016697687f,   0.017782797f,
  0.018938423f,   0.020169149f,   0.021479854f,   0.022875735f,
  0.024362330f,   0.025945531f,   0.027631618f,   0.029427276f,
  0.031339626f,   0.033376252f,   0.035545228f,   0.037855157f,
  0.040315199f,   0.042935108f,   0.045725273f,   0.048696758f,
  0.051861348f,   0.055231591f,   0.058820850f,   0.062643361f,
  0.066714279f,   0.071049749f,   0.075666962f,   0.080584227f,
  0.085821044f,   0.091398179f,   0.097337747f,   0.10366330f,
  0.11039993f,    0.11757434f,    0.12521498f,    0.13335215f,
  0.14201813f,    0.15124727f,    0.16107617f,    0.17154380f,
  0.18269168f,    0.19456402f,    0.20720788f,    0.22067342f,
  0.23501402f,    0.25028656f,    0.26655159f,    0.28387361f,
  0.30232132f,    0.32196786f,    0.34289114f,    0.36517414f,
  0.38890521f,    0.41417847f,    0.44109412f,    0.46975890f,
  0.50028648f,    0.53279791f,    0.56742212f,    0.60429640f,
  0.64356699f,    0.68538959f,    0.72993007f,    0.77736504f,
  0.82788260f,    0.88168307f,    0.9389798f,     1.0f
};


// @OPTIMIZE: if you want to replace this bresenham line-drawing routine,
// note that you must produce bit-identical output to decode correctly;
// this specific sequence of operations is specified in the spec (it's
// drawing integer-quantized frequency-space lines that the encoder
// expects to be exactly the same)
//     ... also, isn't the whole point of Bresenham's algorithm to NOT
// have to divide in the setup? sigh.
#ifndef STB_VORBIS_NO_DEFER_FLOOR
#define LINE_OP(a,b)   a *= b
#else
#define LINE_OP(a,b)   a = b
#endif

#ifdef STB_VORBIS_DIVIDE_TABLE
#define DIVTAB_NUMER   32
#define DIVTAB_DENOM   64
int8 integer_divide_table[DIVTAB_NUMER][DIVTAB_DENOM]; // 2KB
#endif

static __forceinline void draw_line(float *output, int x0, int y0, int x1, int y1, int n)
{
   int dy = y1 - y0;
   int adx = x1 - x0;
   int ady = abs(dy);
   int base;
   int x=x0,y=y0;
   int err = 0;
   int sy;

#ifdef STB_VORBIS_DIVIDE_TABLE
   if (adx < DIVTAB_DENOM && ady < DIVTAB_NUMER) {
      if (dy < 0) {
         base = -integer_divide_table[ady][adx];
         sy = base-1;
      } else {
         base =  integer_divide_table[ady][adx];
         sy = base+1;
      }
   } else {
      base = dy / adx;
      if (dy < 0)
         sy = base - 1;
      else
         sy = base+1;
   }
#else
   base = dy / adx;
   if (dy < 0)
      sy = base - 1;
   else
      sy = base+1;
#endif
   ady -= abs(base) * adx;
   if (x1 > n) x1 = n;
   if (x < x1) {
      LINE_OP(output[x], inverse_db_table[y&255]);
      for (++x; x < x1; ++x) {
         err += ady;
         if (err >= adx) {
            err -= adx;
            y += sy;
         } else
            y += base;
         LINE_OP(output[x], inverse_db_table[y&255]);
      }
   }
}

static int residue_decode(vorb *f, Codebook *book, float *target, int offset, int n, int rtype)
{
   int k;
   if (rtype == 0) {
      int step = n / book->dimensions;
      for (k=0; k < step; ++k)
         if (!codebook_decode_step(f, book, target+offset+k, n-offset-k, step))
            return FALSE;
   } else {
      for (k=0; k < n; ) {
         if (!codebook_decode(f, book, target+offset, n-k))
            return FALSE;
         k += book->dimensions;
         offset += book->dimensions;
      }
   }
   return TRUE;
}

// n is 1/2 of the blocksize --
// specification: "Correct per-vector decode length is [n]/2"
static void decode_residue(vorb *f, float *residue_buffers[], int ch, int n, int rn, uint8 *do_not_decode)
{
   int i,j,pass;
   Residue *r = f->residue_config + rn;
   int rtype = f->residue_types[rn];
   int c = r->classbook;
   int classwords = f->codebooks[c].dimensions;
   unsigned int actual_size = rtype == 2 ? n*2 : n;
   unsigned int limit_r_begin = (r->begin < actual_size ? r->begin : actual_size);
   unsigned int limit_r_end   = (r->end   < actual_size ? r->end   : actual_size);
   int n_read = limit_r_end - limit_r_begin;
   int part_read = n_read / r->part_size;
   int temp_alloc_point = temp_alloc_save(f);
   #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE
   uint8 ***part_classdata = (uint8 ***) temp_block_array(f,f->channels, part_read * sizeof(**part_classdata));
   #else
   int **classifications = (int **) temp_block_array(f,f->channels, part_read * sizeof(**classifications));
   #endif

   CHECK(f);

   for (i=0; i < ch; ++i)
      if (!do_not_decode[i])
         memset(residue_buffers[i], 0, sizeof(float) * n);

   if (rtype == 2 && ch != 1) {
      for (j=0; j < ch; ++j)
         if (!do_not_decode[j])
            break;
      if (j == ch)
         goto done;

      for (pass=0; pass < 8; ++pass) {
         int pcount = 0, class_set = 0;
         if (ch == 2) {
            while (pcount < part_read) {
               int z = r->begin + pcount*r->part_size;
               int c_inter = (z & 1), p_inter = z>>1;
               if (pass == 0) {
                  Codebook *c = f->codebooks+r->classbook;
                  int q;
                  DECODE(q,f,c);
                  if (q == EOP) goto done;
                  #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE
                  part_classdata[0][class_set] = r->classdata[q];
                  #else
                  for (i=classwords-1; i >= 0; --i) {
                     classifications[0][i+pcount] = q % r->classifications;
                     q /= r->classifications;
                  }
                  #endif
               }
               for (i=0; i < classwords && pcount < part_read; ++i, ++pcount) {
                  int z = r->begin + pcount*r->part_size;
                  #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE
                  int c = part_classdata[0][class_set][i];
                  #else
                  int c = classifications[0][pcount];
                  #endif
                  int b = r->residue_books[c][pass];
                  if (b >= 0) {
                     Codebook *book = f->codebooks + b;
                     #ifdef STB_VORBIS_DIVIDES_IN_CODEBOOK
                     if (!codebook_decode_deinterleave_repeat(f, book, residue_buffers, ch, &c_inter, &p_inter, n, r->part_size))
                        goto done;
                     #else
                     // saves 1%
                     if (!codebook_decode_deinterleave_repeat(f, book, residue_buffers, ch, &c_inter, &p_inter, n, r->part_size))
                        goto done;
                     #endif
                  } else {
                     z += r->part_size;
                     c_inter = z & 1;
                     p_inter = z >> 1;
                  }
               }
               #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE
               ++class_set;
               #endif
            }
         } else if (ch > 2) {
            while (pcount < part_read) {
               int z = r->begin + pcount*r->part_size;
               int c_inter = z % ch, p_inter = z/ch;
               if (pass == 0) {
                  Codebook *c = f->codebooks+r->classbook;
                  int q;
                  DECODE(q,f,c);
                  if (q == EOP) goto done;
                  #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE
                  part_classdata[0][class_set] = r->classdata[q];
                  #else
                  for (i=classwords-1; i >= 0; --i) {
                     classifications[0][i+pcount] = q % r->classifications;
                     q /= r->classifications;
                  }
                  #endif
               }
               for (i=0; i < classwords && pcount < part_read; ++i, ++pcount) {
                  int z = r->begin + pcount*r->part_size;
                  #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE
                  int c = part_classdata[0][class_set][i];
                  #else
                  int c = classifications[0][pcount];
                  #endif
                  int b = r->residue_books[c][pass];
                  if (b >= 0) {
                     Codebook *book = f->codebooks + b;
                     if (!codebook_decode_deinterleave_repeat(f, book, residue_buffers, ch, &c_inter, &p_inter, n, r->part_size))
                        goto done;
                  } else {
                     z += r->part_size;
                     c_inter = z % ch;
                     p_inter = z / ch;
                  }
               }
               #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE
               ++class_set;
               #endif
            }
         }
      }
      goto done;
   }
   CHECK(f);

   for (pass=0; pass < 8; ++pass) {
      int pcount = 0, class_set=0;
      while (pcount < part_read) {
         if (pass == 0) {
            for (j=0; j < ch; ++j) {
               if (!do_not_decode[j]) {
                  Codebook *c = f->codebooks+r->classbook;
                  int temp;
                  DECODE(temp,f,c);
                  if (temp == EOP) goto done;
                  #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE
                  part_classdata[j][class_set] = r->classdata[temp];
                  #else
                  for (i=classwords-1; i >= 0; --i) {
                     classifications[j][i+pcount] = temp % r->classifications;
                     temp /= r->classifications;
                  }
                  #endif
               }
            }
         }
         for (i=0; i < classwords && pcount < part_read; ++i, ++pcount) {
            for (j=0; j < ch; ++j) {
               if (!do_not_decode[j]) {
                  #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE
                  int c = part_classdata[j][class_set][i];
                  #else
                  int c = classifications[j][pcount];
                  #endif
                  int b = r->residue_books[c][pass];
                  if (b >= 0) {
                     float *target = residue_buffers[j];
                     int offset = r->begin + pcount * r->part_size;
                     int n = r->part_size;
                     Codebook *book = f->codebooks + b;
                     if (!residue_decode(f, book, target, offset, n, rtype))
                        goto done;
                  }
               }
            }
         }
         #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE
         ++class_set;
         #endif
      }
   }
  done:
   CHECK(f);
   #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE
   temp_free(f,part_classdata);
   #else
   temp_free(f,classifications);
   #endif
   temp_alloc_restore(f,temp_alloc_point);
}


#if 0
// slow way for debugging
void inverse_mdct_slow(float *buffer, int n)
{
   int i,j;
   int n2 = n >> 1;
   float *x = (float *) malloc(sizeof(*x) * n2);
   memcpy(x, buffer, sizeof(*x) * n2);
   for (i=0; i < n; ++i) {
      float acc = 0;
      for (j=0; j < n2; ++j)
         // formula from paper:
         //acc += n/4.0f * x[j] * (float) cos(M_PI / 2 / n * (2 * i + 1 + n/2.0)*(2*j+1));
         // formula from wikipedia
         //acc += 2.0f / n2 * x[j] * (float) cos(M_PI/n2 * (i + 0.5 + n2/2)*(j + 0.5));
         // these are equivalent, except the formula from the paper inverts the multiplier!
         // however, what actually works is NO MULTIPLIER!?!
         //acc += 64 * 2.0f / n2 * x[j] * (float) cos(M_PI/n2 * (i + 0.5 + n2/2)*(j + 0.5));
         acc += x[j] * (float) cos(M_PI / 2 / n * (2 * i + 1 + n/2.0)*(2*j+1));
      buffer[i] = acc;
   }
   free(x);
}
#elif 0
// same as above, but just barely able to run in real time on modern machines
void inverse_mdct_slow(float *buffer, int n, vorb *f, int blocktype)
{
   float mcos[16384];
   int i,j;
   int n2 = n >> 1, nmask = (n << 2) -1;
   float *x = (float *) malloc(sizeof(*x) * n2);
   memcpy(x, buffer, sizeof(*x) * n2);
   for (i=0; i < 4*n; ++i)
      mcos[i] = (float) cos(M_PI / 2 * i / n);

   for (i=0; i < n; ++i) {
      float acc = 0;
      for (j=0; j < n2; ++j)
         acc += x[j] * mcos[(2 * i + 1 + n2)*(2*j+1) & nmask];
      buffer[i] = acc;
   }
   free(x);
}
#elif 0
// transform to use a slow dct-iv; this is STILL basically trivial,
// but only requires half as many ops
void dct_iv_slow(float *buffer, int n)
{
   float mcos[16384];
   float x[2048];
   int i,j;
   int n2 = n >> 1, nmask = (n << 3) - 1;
   memcpy(x, buffer, sizeof(*x) * n);
   for (i=0; i < 8*n; ++i)
      mcos[i] = (float) cos(M_PI / 4 * i / n);
   for (i=0; i < n; ++i) {
      float acc = 0;
      for (j=0; j < n; ++j)
         acc += x[j] * mcos[((2 * i + 1)*(2*j+1)) & nmask];
      buffer[i] = acc;
   }
}

void inverse_mdct_slow(float *buffer, int n, vorb *f, int blocktype)
{
   int i, n4 = n >> 2, n2 = n >> 1, n3_4 = n - n4;
   float temp[4096];

   memcpy(temp, buffer, n2 * sizeof(float));
   dct_iv_slow(temp, n2);  // returns -c'-d, a-b'

   for (i=0; i < n4  ; ++i) buffer[i] = temp[i+n4];            // a-b'
   for (   ; i < n3_4; ++i) buffer[i] = -temp[n3_4 - i - 1];   // b-a', c+d'
   for (   ; i < n   ; ++i) buffer[i] = -temp[i - n3_4];       // c'+d
}
#endif

#ifndef LIBVORBIS_MDCT
#define LIBVORBIS_MDCT 0
#endif

#if LIBVORBIS_MDCT
// directly call the vorbis MDCT using an interface documented
// by Jeff Roberts... useful for performance comparison
typedef struct
{
  int n;
  int log2n;

  float *trig;
  int   *bitrev;

  float scale;
} mdct_lookup;

extern void mdct_init(mdct_lookup *lookup, int n);
extern void mdct_clear(mdct_lookup *l);
extern void mdct_backward(mdct_lookup *init, float *in, float *out);

mdct_lookup M1,M2;

void inverse_mdct(float *buffer, int n, vorb *f, int blocktype)
{
   mdct_lookup *M;
   if (M1.n == n) M = &M1;
   else if (M2.n == n) M = &M2;
   else if (M1.n == 0) { mdct_init(&M1, n); M = &M1; }
   else {
      if (M2.n) __asm int 3;
      mdct_init(&M2, n);
      M = &M2;
   }

   mdct_backward(M, buffer, buffer);
}
#endif


// the following were split out into separate functions while optimizing;
// they could be pushed back up but eh. __forceinline showed no change;
// they're probably already being inlined.
static void imdct_step3_iter0_loop(int n, float *e, int i_off, int k_off, float *A)
{
   float *ee0 = e + i_off;
   float *ee2 = ee0 + k_off;
   int i;

   assert((n & 3) == 0);
   for (i=(n>>2); i > 0; --i) {
      float k00_20, k01_21;
      k00_20  = ee0[ 0] - ee2[ 0];
      k01_21  = ee0[-1] - ee2[-1];
      ee0[ 0] += ee2[ 0];//ee0[ 0] = ee0[ 0] + ee2[ 0];
      ee0[-1] += ee2[-1];//ee0[-1] = ee0[-1] + ee2[-1];
      ee2[ 0] = k00_20 * A[0] - k01_21 * A[1];
      ee2[-1] = k01_21 * A[0] + k00_20 * A[1];
      A += 8;

      k00_20  = ee0[-2] - ee2[-2];
      k01_21  = ee0[-3] - ee2[-3];
      ee0[-2] += ee2[-2];//ee0[-2] = ee0[-2] + ee2[-2];
      ee0[-3] += ee2[-3];//ee0[-3] = ee0[-3] + ee2[-3];
      ee2[-2] = k00_20 * A[0] - k01_21 * A[1];
      ee2[-3] = k01_21 * A[0] + k00_20 * A[1];
      A += 8;

      k00_20  = ee0[-4] - ee2[-4];
      k01_21  = ee0[-5] - ee2[-5];
      ee0[-4] += ee2[-4];//ee0[-4] = ee0[-4] + ee2[-4];
      ee0[-5] += ee2[-5];//ee0[-5] = ee0[-5] + ee2[-5];
      ee2[-4] = k00_20 * A[0] - k01_21 * A[1];
      ee2[-5] = k01_21 * A[0] + k00_20 * A[1];
      A += 8;

      k00_20  = ee0[-6] - ee2[-6];
      k01_21  = ee0[-7] - ee2[-7];
      ee0[-6] += ee2[-6];//ee0[-6] = ee0[-6] + ee2[-6];
      ee0[-7] += ee2[-7];//ee0[-7] = ee0[-7] + ee2[-7];
      ee2[-6] = k00_20 * A[0] - k01_21 * A[1];
      ee2[-7] = k01_21 * A[0] + k00_20 * A[1];
      A += 8;
      ee0 -= 8;
      ee2 -= 8;
   }
}

static void imdct_step3_inner_r_loop(int lim, float *e, int d0, int k_off, float *A, int k1)
{
   int i;
   float k00_20, k01_21;

   float *e0 = e + d0;
   float *e2 = e0 + k_off;

   for (i=lim >> 2; i > 0; --i) {
      k00_20 = e0[-0] - e2[-0];
      k01_21 = e0[-1] - e2[-1];
      e0[-0] += e2[-0];//e0[-0] = e0[-0] + e2[-0];
      e0[-1] += e2[-1];//e0[-1] = e0[-1] + e2[-1];
      e2[-0] = (k00_20)*A[0] - (k01_21) * A[1];
      e2[-1] = (k01_21)*A[0] + (k00_20) * A[1];

      A += k1;

      k00_20 = e0[-2] - e2[-2];
      k01_21 = e0[-3] - e2[-3];
      e0[-2] += e2[-2];//e0[-2] = e0[-2] + e2[-2];
      e0[-3] += e2[-3];//e0[-3] = e0[-3] + e2[-3];
      e2[-2] = (k00_20)*A[0] - (k01_21) * A[1];
      e2[-3] = (k01_21)*A[0] + (k00_20) * A[1];

      A += k1;

      k00_20 = e0[-4] - e2[-4];
      k01_21 = e0[-5] - e2[-5];
      e0[-4] += e2[-4];//e0[-4] = e0[-4] + e2[-4];
      e0[-5] += e2[-5];//e0[-5] = e0[-5] + e2[-5];
      e2[-4] = (k00_20)*A[0] - (k01_21) * A[1];
      e2[-5] = (k01_21)*A[0] + (k00_20) * A[1];

      A += k1;

      k00_20 = e0[-6] - e2[-6];
      k01_21 = e0[-7] - e2[-7];
      e0[-6] += e2[-6];//e0[-6] = e0[-6] + e2[-6];
      e0[-7] += e2[-7];//e0[-7] = e0[-7] + e2[-7];
      e2[-6] = (k00_20)*A[0] - (k01_21) * A[1];
      e2[-7] = (k01_21)*A[0] + (k00_20) * A[1];

      e0 -= 8;
      e2 -= 8;

      A += k1;
   }
}

static void imdct_step3_inner_s_loop(int n, float *e, int i_off, int k_off, float *A, int a_off, int k0)
{
   int i;
   float A0 = A[0];
   float A1 = A[0+1];
   float A2 = A[0+a_off];
   float A3 = A[0+a_off+1];
   float A4 = A[0+a_off*2+0];
   float A5 = A[0+a_off*2+1];
   float A6 = A[0+a_off*3+0];
   float A7 = A[0+a_off*3+1];

   float k00,k11;

   float *ee0 = e  +i_off;
   float *ee2 = ee0+k_off;

   for (i=n; i > 0; --i) {
      k00     = ee0[ 0] - ee2[ 0];
      k11     = ee0[-1] - ee2[-1];
      ee0[ 0] =  ee0[ 0] + ee2[ 0];
      ee0[-1] =  ee0[-1] + ee2[-1];
      ee2[ 0] = (k00) * A0 - (k11) * A1;
      ee2[-1] = (k11) * A0 + (k00) * A1;

      k00     = ee0[-2] - ee2[-2];
      k11     = ee0[-3] - ee2[-3];
      ee0[-2] =  ee0[-2] + ee2[-2];
      ee0[-3] =  ee0[-3] + ee2[-3];
      ee2[-2] = (k00) * A2 - (k11) * A3;
      ee2[-3] = (k11) * A2 + (k00) * A3;

      k00     = ee0[-4] - ee2[-4];
      k11     = ee0[-5] - ee2[-5];
      ee0[-4] =  ee0[-4] + ee2[-4];
      ee0[-5] =  ee0[-5] + ee2[-5];
      ee2[-4] = (k00) * A4 - (k11) * A5;
      ee2[-5] = (k11) * A4 + (k00) * A5;

      k00     = ee0[-6] - ee2[-6];
      k11     = ee0[-7] - ee2[-7];
      ee0[-6] =  ee0[-6] + ee2[-6];
      ee0[-7] =  ee0[-7] + ee2[-7];
      ee2[-6] = (k00) * A6 - (k11) * A7;
      ee2[-7] = (k11) * A6 + (k00) * A7;

      ee0 -= k0;
      ee2 -= k0;
   }
}

static __forceinline void iter_54(float *z)
{
   float k00,k11,k22,k33;
   float y0,y1,y2,y3;

   k00  = z[ 0] - z[-4];
   y0   = z[ 0] + z[-4];
   y2   = z[-2] + z[-6];
   k22  = z[-2] - z[-6];

   z[-0] = y0 + y2;      // z0 + z4 + z2 + z6
   z[-2] = y0 - y2;      // z0 + z4 - z2 - z6

   // done with y0,y2

   k33  = z[-3] - z[-7];

   z[-4] = k00 + k33;    // z0 - z4 + z3 - z7
   z[-6] = k00 - k33;    // z0 - z4 - z3 + z7

   // done with k33

   k11  = z[-1] - z[-5];
   y1   = z[-1] + z[-5];
   y3   = z[-3] + z[-7];

   z[-1] = y1 + y3;      // z1 + z5 + z3 + z7
   z[-3] = y1 - y3;      // z1 + z5 - z3 - z7
   z[-5] = k11 - k22;    // z1 - z5 + z2 - z6
   z[-7] = k11 + k22;    // z1 - z5 - z2 + z6
}

static void imdct_step3_inner_s_loop_ld654(int n, float *e, int i_off, float *A, int base_n)
{
   int a_off = base_n >> 3;
   float A2 = A[0+a_off];
   float *z = e + i_off;
   float *base = z - 16 * n;

   while (z > base) {
      float k00,k11;
      float l00,l11;

      k00    = z[-0] - z[ -8];
      k11    = z[-1] - z[ -9];
      l00    = z[-2] - z[-10];
      l11    = z[-3] - z[-11];
      z[ -0] = z[-0] + z[ -8];
      z[ -1] = z[-1] + z[ -9];
      z[ -2] = z[-2] + z[-10];
      z[ -3] = z[-3] + z[-11];
      z[ -8] = k00;
      z[ -9] = k11;
      z[-10] = (l00+l11) * A2;
      z[-11] = (l11-l00) * A2;

      k00    = z[ -4] - z[-12];
      k11    = z[ -5] - z[-13];
      l00    = z[ -6] - z[-14];
      l11    = z[ -7] - z[-15];
      z[ -4] = z[ -4] + z[-12];
      z[ -5] = z[ -5] + z[-13];
      z[ -6] = z[ -6] + z[-14];
      z[ -7] = z[ -7] + z[-15];
      z[-12] = k11;
      z[-13] = -k00;
      z[-14] = (l11-l00) * A2;
      z[-15] = (l00+l11) * -A2;

      iter_54(z);
      iter_54(z-8);
      z -= 16;
   }
}

static void inverse_mdct(float *buffer, int n, vorb *f, int blocktype)
{
   int n2 = n >> 1, n4 = n >> 2, n8 = n >> 3, l;
   int ld;
   // @OPTIMIZE: reduce register pressure by using fewer variables?
   int save_point = temp_alloc_save(f);
   float *buf2 = (float *) temp_alloc(f, n2 * sizeof(*buf2));
   float *u=NULL,*v=NULL;
   // twiddle factors
   float *A = f->A[blocktype];

   // IMDCT algorithm from "The use of multirate filter banks for coding of high quality digital audio"
   // See notes about bugs in that paper in less-optimal implementation 'inverse_mdct_old' after this function.

   // kernel from paper


   // merged:
   //   copy and reflect spectral data
   //   step 0

   // note that it turns out that the items added together during
   // this step are, in fact, being added to themselves (as reflected
   // by step 0). inexplicable inefficiency! this became obvious
   // once I combined the passes.

   // so there's a missing 'times 2' here (for adding X to itself).
   // this propagates through linearly to the end, where the numbers
   // are 1/2 too small, and need to be compensated for.

   {
      float *d,*e, *AA, *e_stop;
      d = &buf2[n2-2];
      AA = A;
      e = &buffer[0];
      e_stop = &buffer[n2];
      while (e != e_stop) {
         d[1] = (e[0] * AA[0] - e[2]*AA[1]);
         d[0] = (e[0] * AA[1] + e[2]*AA[0]);
         d -= 2;
         AA += 2;
         e += 4;
      }

      e = &buffer[n2-3];
      while (d >= buf2) {
         d[1] = (-e[2] * AA[0] - -e[0]*AA[1]);
         d[0] = (-e[2] * AA[1] + -e[0]*AA[0]);
         d -= 2;
         AA += 2;
         e -= 4;
      }
   }

   // now we use symbolic names for these, so that we can
   // possibly swap their meaning as we change which operations
   // are in place

   u = buffer;
   v = buf2;

   // step 2    (paper output is w, now u)
   // this could be in place, but the data ends up in the wrong
   // place... _somebody_'s got to swap it, so this is nominated
   {
      float *AA = &A[n2-8];
      float *d0,*d1, *e0, *e1;

      e0 = &v[n4];
      e1 = &v[0];

      d0 = &u[n4];
      d1 = &u[0];

      while (AA >= A) {
         float v40_20, v41_21;

         v41_21 = e0[1] - e1[1];
         v40_20 = e0[0] - e1[0];
         d0[1]  = e0[1] + e1[1];
         d0[0]  = e0[0] + e1[0];
         d1[1]  = v41_21*AA[4] - v40_20*AA[5];
         d1[0]  = v40_20*AA[4] + v41_21*AA[5];

         v41_21 = e0[3] - e1[3];
         v40_20 = e0[2] - e1[2];
         d0[3]  = e0[3] + e1[3];
         d0[2]  = e0[2] + e1[2];
         d1[3]  = v41_21*AA[0] - v40_20*AA[1];
         d1[2]  = v40_20*AA[0] + v41_21*AA[1];

         AA -= 8;

         d0 += 4;
         d1 += 4;
         e0 += 4;
         e1 += 4;
      }
   }

   // step 3
   ld = ilog(n) - 1; // ilog is off-by-one from normal definitions

   // optimized step 3:

   // the original step3 loop can be nested r inside s or s inside r;
   // it's written originally as s inside r, but this is dumb when r
   // iterates many times, and s few. So I have two copies of it and
   // switch between them halfway.

   // this is iteration 0 of step 3
   imdct_step3_iter0_loop(n >> 4, u, n2-1-n4*0, -(n >> 3), A);
   imdct_step3_iter0_loop(n >> 4, u, n2-1-n4*1, -(n >> 3), A);

   // this is iteration 1 of step 3
   imdct_step3_inner_r_loop(n >> 5, u, n2-1 - n8*0, -(n >> 4), A, 16);
   imdct_step3_inner_r_loop(n >> 5, u, n2-1 - n8*1, -(n >> 4), A, 16);
   imdct_step3_inner_r_loop(n >> 5, u, n2-1 - n8*2, -(n >> 4), A, 16);
   imdct_step3_inner_r_loop(n >> 5, u, n2-1 - n8*3, -(n >> 4), A, 16);

   l=2;
   for (; l < (ld-3)>>1; ++l) {
      int k0 = n >> (l+2), k0_2 = k0>>1;
      int lim = 1 << (l+1);
      int i;
      for (i=0; i < lim; ++i)
         imdct_step3_inner_r_loop(n >> (l+4), u, n2-1 - k0*i, -k0_2, A, 1 << (l+3));
   }

   for (; l < ld-6; ++l) {
      int k0 = n >> (l+2), k1 = 1 << (l+3), k0_2 = k0>>1;
      int rlim = n >> (l+6), r;
      int lim = 1 << (l+1);
      int i_off;
      float *A0 = A;
      i_off = n2-1;
      for (r=rlim; r > 0; --r) {
         imdct_step3_inner_s_loop(lim, u, i_off, -k0_2, A0, k1, k0);
         A0 += k1*4;
         i_off -= 8;
      }
   }

   // iterations with count:
   //   ld-6,-5,-4 all interleaved together
   //       the big win comes from getting rid of needless flops
   //         due to the constants on pass 5 & 4 being all 1 and 0;
   //       combining them to be simultaneous to improve cache made little difference
   imdct_step3_inner_s_loop_ld654(n >> 5, u, n2-1, A, n);

   // output is u

   // step 4, 5, and 6
   // cannot be in-place because of step 5
   {
      uint16 *bitrev = f->bit_reverse[blocktype];
      // weirdly, I'd have thought reading sequentially and writing
      // erratically would have been better than vice-versa, but in
      // fact that's not what my testing showed. (That is, with
      // j = bitreverse(i), do you read i and write j, or read j and write i.)

      float *d0 = &v[n4-4];
      float *d1 = &v[n2-4];
      while (d0 >= v) {
         int k4;

         k4 = bitrev[0];
         d1[3] = u[k4+0];
         d1[2] = u[k4+1];
         d0[3] = u[k4+2];
         d0[2] = u[k4+3];

         k4 = bitrev[1];
         d1[1] = u[k4+0];
         d1[0] = u[k4+1];
         d0[1] = u[k4+2];
         d0[0] = u[k4+3];

         d0 -= 4;
         d1 -= 4;
         bitrev += 2;
      }
   }
   // (paper output is u, now v)


   // data must be in buf2
   assert(v == buf2);

   // step 7   (paper output is v, now v)
   // this is now in place
   {
      float *C = f->C[blocktype];
      float *d, *e;

      d = v;
      e = v + n2 - 4;

      while (d < e) {
         float a02,a11,b0,b1,b2,b3;

         a02 = d[0] - e[2];
         a11 = d[1] + e[3];

         b0 = C[1]*a02 + C[0]*a11;
         b1 = C[1]*a11 - C[0]*a02;

         b2 = d[0] + e[ 2];
         b3 = d[1] - e[ 3];

         d[0] = b2 + b0;
         d[1] = b3 + b1;
         e[2] = b2 - b0;
         e[3] = b1 - b3;

         a02 = d[2] - e[0];
         a11 = d[3] + e[1];

         b0 = C[3]*a02 + C[2]*a11;
         b1 = C[3]*a11 - C[2]*a02;

         b2 = d[2] + e[ 0];
         b3 = d[3] - e[ 1];

         d[2] = b2 + b0;
         d[3] = b3 + b1;
         e[0] = b2 - b0;
         e[1] = b1 - b3;

         C += 4;
         d += 4;
         e -= 4;
      }
   }

   // data must be in buf2


   // step 8+decode   (paper output is X, now buffer)
   // this generates pairs of data a la 8 and pushes them directly through
   // the decode kernel (pushing rather than pulling) to avoid having
   // to make another pass later

   // this cannot POSSIBLY be in place, so we refer to the buffers directly

   {
      float *d0,*d1,*d2,*d3;

      float *B = f->B[blocktype] + n2 - 8;
      float *e = buf2 + n2 - 8;
      d0 = &buffer[0];
      d1 = &buffer[n2-4];
      d2 = &buffer[n2];
      d3 = &buffer[n-4];
      while (e >= v) {
         float p0,p1,p2,p3;

         p3 =  e[6]*B[7] - e[7]*B[6];
         p2 = -e[6]*B[6] - e[7]*B[7];

         d0[0] =   p3;
         d1[3] = - p3;
         d2[0] =   p2;
         d3[3] =   p2;

         p1 =  e[4]*B[5] - e[5]*B[4];
         p0 = -e[4]*B[4] - e[5]*B[5];

         d0[1] =   p1;
         d1[2] = - p1;
         d2[1] =   p0;
         d3[2] =   p0;

         p3 =  e[2]*B[3] - e[3]*B[2];
         p2 = -e[2]*B[2] - e[3]*B[3];

         d0[2] =   p3;
         d1[1] = - p3;
         d2[2] =   p2;
         d3[1] =   p2;

         p1 =  e[0]*B[1] - e[1]*B[0];
         p0 = -e[0]*B[0] - e[1]*B[1];

         d0[3] =   p1;
         d1[0] = - p1;
         d2[3] =   p0;
         d3[0] =   p0;

         B -= 8;
         e -= 8;
         d0 += 4;
         d2 += 4;
         d1 -= 4;
         d3 -= 4;
      }
   }

   temp_free(f,buf2);
   temp_alloc_restore(f,save_point);
}

#if 0
// this is the original version of the above code, if you want to optimize it from scratch
void inverse_mdct_naive(float *buffer, int n)
{
   float s;
   float A[1 << 12], B[1 << 12], C[1 << 11];
   int i,k,k2,k4, n2 = n >> 1, n4 = n >> 2, n8 = n >> 3, l;
   int n3_4 = n - n4, ld;
   // how can they claim this only uses N words?!
   // oh, because they're only used sparsely, whoops
   float u[1 << 13], X[1 << 13], v[1 << 13], w[1 << 13];
   // set up twiddle factors

   for (k=k2=0; k < n4; ++k,k2+=2) {
      A[k2  ] = (float)  cos(4*k*M_PI/n);
      A[k2+1] = (float) -sin(4*k*M_PI/n);
      B[k2  ] = (float)  cos((k2+1)*M_PI/n/2);
      B[k2+1] = (float)  sin((k2+1)*M_PI/n/2);
   }
   for (k=k2=0; k < n8; ++k,k2+=2) {
      C[k2  ] = (float)  cos(2*(k2+1)*M_PI/n);
      C[k2+1] = (float) -sin(2*(k2+1)*M_PI/n);
   }

   // IMDCT algorithm from "The use of multirate filter banks for coding of high quality digital audio"
   // Note there are bugs in that pseudocode, presumably due to them attempting
   // to rename the arrays nicely rather than representing the way their actual
   // implementation bounces buffers back and forth. As a result, even in the
   // "some formulars corrected" version, a direct implementation fails. These
   // are noted below as "paper bug".

   // copy and reflect spectral data
   for (k=0; k < n2; ++k) u[k] = buffer[k];
   for (   ; k < n ; ++k) u[k] = -buffer[n - k - 1];
   // kernel from paper
   // step 1
   for (k=k2=k4=0; k < n4; k+=1, k2+=2, k4+=4) {
      v[n-k4-1] = (u[k4] - u[n-k4-1]) * A[k2]   - (u[k4+2] - u[n-k4-3])*A[k2+1];
      v[n-k4-3] = (u[k4] - u[n-k4-1]) * A[k2+1] + (u[k4+2] - u[n-k4-3])*A[k2];
   }
   // step 2
   for (k=k4=0; k < n8; k+=1, k4+=4) {
      w[n2+3+k4] = v[n2+3+k4] + v[k4+3];
      w[n2+1+k4] = v[n2+1+k4] + v[k4+1];
      w[k4+3]    = (v[n2+3+k4] - v[k4+3])*A[n2-4-k4] - (v[n2+1+k4]-v[k4+1])*A[n2-3-k4];
      w[k4+1]    = (v[n2+1+k4] - v[k4+1])*A[n2-4-k4] + (v[n2+3+k4]-v[k4+3])*A[n2-3-k4];
   }
   // step 3
   ld = ilog(n) - 1; // ilog is off-by-one from normal definitions
   for (l=0; l < ld-3; ++l) {
      int k0 = n >> (l+2), k1 = 1 << (l+3);
      int rlim = n >> (l+4), r4, r;
      int s2lim = 1 << (l+2), s2;
      for (r=r4=0; r < rlim; r4+=4,++r) {
         for (s2=0; s2 < s2lim; s2+=2) {
            u[n-1-k0*s2-r4] = w[n-1-k0*s2-r4] + w[n-1-k0*(s2+1)-r4];
            u[n-3-k0*s2-r4] = w[n-3-k0*s2-r4] + w[n-3-k0*(s2+1)-r4];
            u[n-1-k0*(s2+1)-r4] = (w[n-1-k0*s2-r4] - w[n-1-k0*(s2+1)-r4]) * A[r*k1]
                                - (w[n-3-k0*s2-r4] - w[n-3-k0*(s2+1)-r4]) * A[r*k1+1];
            u[n-3-k0*(s2+1)-r4] = (w[n-3-k0*s2-r4] - w[n-3-k0*(s2+1)-r4]) * A[r*k1]
                                + (w[n-1-k0*s2-r4] - w[n-1-k0*(s2+1)-r4]) * A[r*k1+1];
         }
      }
      if (l+1 < ld-3) {
         // paper bug: ping-ponging of u&w here is omitted
         memcpy(w, u, sizeof(u));
      }
   }

   // step 4
   for (i=0; i < n8; ++i) {
      int j = bit_reverse(i) >> (32-ld+3);
      assert(j < n8);
      if (i == j) {
         // paper bug: original code probably swapped in place; if copying,
         //            need to directly copy in this case
         int i8 = i << 3;
         v[i8+1] = u[i8+1];
         v[i8+3] = u[i8+3];
         v[i8+5] = u[i8+5];
         v[i8+7] = u[i8+7];
      } else if (i < j) {
         int i8 = i << 3, j8 = j << 3;
         v[j8+1] = u[i8+1], v[i8+1] = u[j8 + 1];
         v[j8+3] = u[i8+3], v[i8+3] = u[j8 + 3];
         v[j8+5] = u[i8+5], v[i8+5] = u[j8 + 5];
         v[j8+7] = u[i8+7], v[i8+7] = u[j8 + 7];
      }
   }
   // step 5
   for (k=0; k < n2; ++k) {
      w[k] = v[k*2+1];
   }
   // step 6
   for (k=k2=k4=0; k < n8; ++k, k2 += 2, k4 += 4) {
      u[n-1-k2] = w[k4];
      u[n-2-k2] = w[k4+1];
      u[n3_4 - 1 - k2] = w[k4+2];
      u[n3_4 - 2 - k2] = w[k4+3];
   }
   // step 7
   for (k=k2=0; k < n8; ++k, k2 += 2) {
      v[n2 + k2 ] = ( u[n2 + k2] + u[n-2-k2] + C[k2+1]*(u[n2+k2]-u[n-2-k2]) + C[k2]*(u[n2+k2+1]+u[n-2-k2+1]))/2;
      v[n-2 - k2] = ( u[n2 + k2] + u[n-2-k2] - C[k2+1]*(u[n2+k2]-u[n-2-k2]) - C[k2]*(u[n2+k2+1]+u[n-2-k2+1]))/2;
      v[n2+1+ k2] = ( u[n2+1+k2] - u[n-1-k2] + C[k2+1]*(u[n2+1+k2]+u[n-1-k2]) - C[k2]*(u[n2+k2]-u[n-2-k2]))/2;
      v[n-1 - k2] = (-u[n2+1+k2] + u[n-1-k2] + C[k2+1]*(u[n2+1+k2]+u[n-1-k2]) - C[k2]*(u[n2+k2]-u[n-2-k2]))/2;
   }
   // step 8
   for (k=k2=0; k < n4; ++k,k2 += 2) {
      X[k]      = v[k2+n2]*B[k2  ] + v[k2+1+n2]*B[k2+1];
      X[n2-1-k] = v[k2+n2]*B[k2+1] - v[k2+1+n2]*B[k2  ];
   }

   // decode kernel to output
   // determined the following value experimentally
   // (by first figuring out what made inverse_mdct_slow work); then matching that here
   // (probably vorbis encoder premultiplies by n or n/2, to save it on the decoder?)
   s = 0.5; // theoretically would be n4

   // [[[ note! the s value of 0.5 is compensated for by the B[] in the current code,
   //     so it needs to use the "old" B values to behave correctly, or else
   //     set s to 1.0 ]]]
   for (i=0; i < n4  ; ++i) buffer[i] = s * X[i+n4];
   for (   ; i < n3_4; ++i) buffer[i] = -s * X[n3_4 - i - 1];
   for (   ; i < n   ; ++i) buffer[i] = -s * X[i - n3_4];
}
#endif

static float *get_window(vorb *f, int len)
{
   len <<= 1;
   if (len == f->blocksize_0) return f->window[0];
   if (len == f->blocksize_1) return f->window[1];
   return NULL;
}

#ifndef STB_VORBIS_NO_DEFER_FLOOR
typedef int16 YTYPE;
#else
typedef int YTYPE;
#endif
static int do_floor(vorb *f, Mapping *map, int i, int n, float *target, YTYPE *finalY, uint8 *step2_flag)
{
   int n2 = n >> 1;
   int s = map->chan[i].mux, floor;
   floor = map->submap_floor[s];
   if (f->floor_types[floor] == 0) {
      return error(f, VORBIS_invalid_stream);
   } else {
      Floor1 *g = &f->floor_config[floor].floor1;
      int j,q;
      int lx = 0, ly = finalY[0] * g->floor1_multiplier;
      for (q=1; q < g->values; ++q) {
         j = g->sorted_order[q];
         #ifndef STB_VORBIS_NO_DEFER_FLOOR
         STBV_NOTUSED(step2_flag);
         if (finalY[j] >= 0)
         #else
         if (step2_flag[j])
         #endif
         {
            int hy = finalY[j] * g->floor1_multiplier;
            int hx = g->Xlist[j];
            if (lx != hx)
               draw_line(target, lx,ly, hx,hy, n2);
            CHECK(f);
            lx = hx, ly = hy;
         }
      }
      if (lx < n2) {
         // optimization of: draw_line(target, lx,ly, n,ly, n2);
         for (j=lx; j < n2; ++j)
            LINE_OP(target[j], inverse_db_table[ly]);
         CHECK(f);
      }
   }
   return TRUE;
}

// The meaning of "left" and "right"
//
// For a given frame:
//     we compute samples from 0..n
//     window_center is n/2
//     we'll window and mix the samples from left_start to left_end with data from the previous frame
//     all of the samples from left_end to right_start can be output without mixing; however,
//        this interval is 0-length except when transitioning between short and long frames
//     all of the samples from right_start to right_end need to be mixed with the next frame,
//        which we don't have, so those get saved in a buffer
//     frame N's right_end-right_start, the number of samples to mix with the next frame,
//        has to be the same as frame N+1's left_end-left_start (which they are by
//        construction)

static int vorbis_decode_initial(vorb *f, int *p_left_start, int *p_left_end, int *p_right_start, int *p_right_end, int *mode)
{
   Mode *m;
   int i, n, prev, next, window_center;
   f->channel_buffer_start = f->channel_buffer_end = 0;

  retry:
   if (f->eof) return FALSE;
   if (!maybe_start_packet(f))
      return FALSE;
   // check packet type
   if (get_bits(f,1) != 0) {
      if (IS_PUSH_MODE(f))
         return error(f,VORBIS_bad_packet_type);
      while (EOP != get8_packet(f));
      goto retry;
   }

   if (f->alloc.alloc_buffer)
      assert(f->alloc.alloc_buffer_length_in_bytes == f->temp_offset);

   i = get_bits(f, ilog(f->mode_count-1));
   if (i == EOP) return FALSE;
   if (i >= f->mode_count) return FALSE;
   *mode = i;
   m = f->mode_config + i;
   if (m->blockflag) {
      n = f->blocksize_1;
      prev = get_bits(f,1);
      next = get_bits(f,1);
   } else {
      prev = next = 0;
      n = f->blocksize_0;
   }

// WINDOWING

   window_center = n >> 1;
   if (m->blockflag && !prev) {
      *p_left_start = (n - f->blocksize_0) >> 2;
      *p_left_end   = (n + f->blocksize_0) >> 2;
   } else {
      *p_left_start = 0;
      *p_left_end   = window_center;
   }
   if (m->blockflag && !next) {
      *p_right_start = (n*3 - f->blocksize_0) >> 2;
      *p_right_end   = (n*3 + f->blocksize_0) >> 2;
   } else {
      *p_right_start = window_center;
      *p_right_end   = n;
   }

   return TRUE;
}

static int vorbis_decode_packet_rest(vorb *f, int *len, Mode *m, int left_start, int left_end, int right_start, int right_end, int *p_left)
{
   Mapping *map;
   int i,j,k,n,n2;
   int zero_channel[256];
   int really_zero_channel[256];

// WINDOWING

   STBV_NOTUSED(left_end);
   n = f->blocksize[m->blockflag];
   map = &f->mapping[m->mapping];

// FLOORS
   n2 = n >> 1;

   CHECK(f);

   for (i=0; i < f->channels; ++i) {
      int s = map->chan[i].mux, floor;
      zero_channel[i] = FALSE;
      floor = map->submap_floor[s];
      if (f->floor_types[floor] == 0) {
         return error(f, VORBIS_invalid_stream);
      } else {
         Floor1 *g = &f->floor_config[floor].floor1;
         if (get_bits(f, 1)) {
            short *finalY;
            uint8 step2_flag[256];
            static int range_list[4] = { 256, 128, 86, 64 };
            int range = range_list[g->floor1_multiplier-1];
            int offset = 2;
            finalY = f->finalY[i];
            finalY[0] = get_bits(f, ilog(range)-1);
            finalY[1] = get_bits(f, ilog(range)-1);
            for (j=0; j < g->partitions; ++j) {
               int pclass = g->partition_class_list[j];
               int cdim = g->class_dimensions[pclass];
               int cbits = g->class_subclasses[pclass];
               int csub = (1 << cbits)-1;
               int cval = 0;
               if (cbits) {
                  Codebook *c = f->codebooks + g->class_masterbooks[pclass];
                  DECODE(cval,f,c);
               }
               for (k=0; k < cdim; ++k) {
                  int book = g->subclass_books[pclass][cval & csub];
                  cval = cval >> cbits;
                  if (book >= 0) {
                     int temp;
                     Codebook *c = f->codebooks + book;
                     DECODE(temp,f,c);
                     finalY[offset++] = temp;
                  } else
                     finalY[offset++] = 0;
               }
            }
            if (f->valid_bits == INVALID_BITS) goto error; // behavior according to spec
            step2_flag[0] = step2_flag[1] = 1;
            for (j=2; j < g->values; ++j) {
               int low, high, pred, highroom, lowroom, room, val;
               low = g->neighbors[j][0];
               high = g->neighbors[j][1];
               //neighbors(g->Xlist, j, &low, &high);
               pred = predict_point(g->Xlist[j], g->Xlist[low], g->Xlist[high], finalY[low], finalY[high]);
               val = finalY[j];
               highroom = range - pred;
               lowroom = pred;
               if (highroom < lowroom)
                  room = highroom * 2;
               else
                  room = lowroom * 2;
               if (val) {
                  step2_flag[low] = step2_flag[high] = 1;
                  step2_flag[j] = 1;
                  if (val >= room)
                     if (highroom > lowroom)
                        finalY[j] = val - lowroom + pred;
                     else
                        finalY[j] = pred - val + highroom - 1;
                  else
                     if (val & 1)
                        finalY[j] = pred - ((val+1)>>1);
                     else
                        finalY[j] = pred + (val>>1);
               } else {
                  step2_flag[j] = 0;
                  finalY[j] = pred;
               }
            }

#ifdef STB_VORBIS_NO_DEFER_FLOOR
            do_floor(f, map, i, n, f->floor_buffers[i], finalY, step2_flag);
#else
            // defer final floor computation until _after_ residue
            for (j=0; j < g->values; ++j) {
               if (!step2_flag[j])
                  finalY[j] = -1;
            }
#endif
         } else {
           error:
            zero_channel[i] = TRUE;