Updated ThirdParty libs

Source/ThirdParty/GLEW/glew.c

@@ -29,6 +29,8 @@
 ** THE POSSIBILITY OF SUCH DAMAGE.
 */
 
+// Modified by Lasse Oorni for Urho3D
+
 #include <glew.h>
 
 #if defined(_WIN32)
@@ -8927,6 +8929,12 @@ GLenum GLEWAPIENTRY glewContextInit (GLEW_CONTEXT_ARG_DEF_LIST)
     CONST_CAST(GLEW_VERSION_1_1)   = GLEW_VERSION_1_2   == GL_TRUE || ( major == 1 && minor >= 1 ) ? GL_TRUE : GL_FALSE;
   }
 
+  // Urho3D: GLEW does not query extensions properly on an OpenGL 3+ core context. Enable experimental mode in that case.
+  // However on OpenGL 2 we need to be strict about not using features that are not listed in extensions (for example
+  // instancing) even if the corresponding function pointers are non-null
+  if (GLEW_VERSION_3_2)
+    glewExperimental = GL_TRUE;
+
   /* query opengl extensions string */
   extStart = glGetString(GL_EXTENSIONS);
   if (extStart == 0)

Source/ThirdParty/SDL/src/video/cocoa/SDL_cocoavideo.m

@@ -18,6 +18,9 @@
      misrepresented as being the original software.
   3. This notice may not be removed or altered from any source distribution.
 */
+
+// Modified by Lasse Oorni for Urho3D
+
 #include "../../SDL_internal.h"
 
 #if SDL_VIDEO_DRIVER_COCOA
@@ -154,8 +157,10 @@ Cocoa_VideoInit(_THIS)
     Cocoa_InitKeyboard(_this);
     Cocoa_InitMouse(_this);
 
-    const char *hint = SDL_GetHint(SDL_HINT_VIDEO_MAC_FULLSCREEN_SPACES);
-    data->allow_spaces = ( (data->osversion >= 0x1070) && (!hint || (*hint != '0')) );
+    // Urho3D: disable fullscreen space due to black screen on switch
+    //const char *hint = SDL_GetHint(SDL_HINT_VIDEO_MAC_FULLSCREEN_SPACES);
+    //data->allow_spaces = ( (data->osversion >= 0x1070) && (!hint || (*hint != '0')) );
+    data->allow_spaces = SDL_FALSE;
 
     return 0;
 }

Source/ThirdParty/STB/stb_image.h

@@ -1,412 +1,517 @@
-/* stbi-1.29 - public domain JPEG/PNG reader - http://nothings.org/stb_image.c
-   when you control the images you're loading
-                                     no warranty implied; use at your own risk
-
-   QUICK NOTES:
-      Primarily of interest to game developers and other people who can
-          avoid problematic images and only need the trivial interface
-
-      JPEG baseline (no JPEG progressive)
-      PNG 8-bit only
-
-      TGA (not sure what subset, if a subset)
-      BMP non-1bpp, non-RLE
-      PSD (composited view only, no extra channels)
-
-      GIF (*comp always reports as 4-channel)
-      HDR (radiance rgbE format)
-      PIC (Softimage PIC)
-
-      - decoded from memory or through stdio FILE (define STBI_NO_STDIO to remove code)
-      - supports installable dequantizing-IDCT, YCbCr-to-RGB conversion (define STBI_SIMD)
-
-   Latest revisions:
-      1.29 (2010-08-16) various warning fixes from Aurelien Pocheville 
-      1.28 (2010-08-01) fix bug in GIF palette transparency (SpartanJ)
-      1.27 (2010-08-01) cast-to-uint8 to fix warnings (Laurent Gomila)
-                        allow trailing 0s at end of image data (Laurent Gomila)
-      1.26 (2010-07-24) fix bug in file buffering for PNG reported by SpartanJ
-      1.25 (2010-07-17) refix trans_data warning (Won Chun)
-      1.24 (2010-07-12) perf improvements reading from files
-                        minor perf improvements for jpeg
-                        deprecated type-specific functions in hope of feedback
-                        attempt to fix trans_data warning (Won Chun)
-      1.23              fixed bug in iPhone support
-      1.22 (2010-07-10) removed image *writing* support to stb_image_write.h
-                        stbi_info support from Jetro Lauha
-                        GIF support from Jean-Marc Lienher
-                        iPhone PNG-extensions from James Brown
-                        warning-fixes from Nicolas Schulz and Janez Zemva
-      1.21              fix use of 'uint8' in header (reported by jon blow)
-      1.20              added support for Softimage PIC, by Tom Seddon
-
-   See end of file for full revision history.
-
-   TODO:
-      stbi_info support for BMP,PSD,HDR,PIC
-      rewrite stbi_info and load_file variations to share file handling code
-           (current system allows individual functions to be called directly,
-           since each does all the work, but I doubt anyone uses this in practice)
-
-
- ============================    Contributors    =========================
-              
- Image formats                                Optimizations & bugfixes
-    Sean Barrett (jpeg, png, bmp)                Fabian "ryg" Giesen
-    Nicolas Schulz (hdr, psd)                                                 
-    Jonathan Dummer (tga)                     Bug fixes & warning fixes           
-    Jean-Marc Lienher (gif)                      Marc LeBlanc               
-    Tom Seddon (pic)                             Christpher Lloyd           
-    Thatcher Ulrich (psd)                        Dave Moore                 
-                                                 Won Chun                   
-                                                 the Horde3D community      
- Extensions, features                            Janez Zemva                
-    Jetro Lauha (stbi_info)                      Jonathan Blow              
-    James "moose2000" Brown (iPhone PNG)         Laurent Gomila                             
-                                                 Aruelien Pocheville
-
- If your name should be here but isn't, let Sean know.
-
-*/
-
-#ifndef STBI_INCLUDE_STB_IMAGE_H
-#define STBI_INCLUDE_STB_IMAGE_H
-
-// To get a header file for this, either cut and paste the header,
-// or create stb_image.h, #define STBI_HEADER_FILE_ONLY, and
-// then include stb_image.c from it.
-
-////   begin header file  ////////////////////////////////////////////////////
-//
-// Limitations:
-//    - no jpeg progressive support
-//    - non-HDR formats support 8-bit samples only (jpeg, png)
-//    - no delayed line count (jpeg) -- IJG doesn't support either
-//    - no 1-bit BMP
-//    - GIF always returns *comp=4
-//
-// Basic usage (see HDR discussion below):
-//    int x,y,n;
-//    unsigned char *data = stbi_load(filename, &x, &y, &n, 0);
-//    // ... process data if not NULL ... 
-//    // ... x = width, y = height, n = # 8-bit components per pixel ...
-//    // ... replace '0' with '1'..'4' to force that many components per pixel
-//    stbi_image_free(data)
-//
-// Standard parameters:
-//    int *x       -- outputs image width in pixels
-//    int *y       -- outputs image height in pixels
-//    int *comp    -- outputs # of image components in image file
-//    int req_comp -- if non-zero, # of image components requested in result
-//
-// The return value from an image loader is an 'unsigned char *' which points
-// to the pixel data. The pixel data consists of *y scanlines of *x pixels,
-// with each pixel consisting of N interleaved 8-bit components; the first
-// pixel pointed to is top-left-most in the image. There is no padding between
-// image scanlines or between pixels, regardless of format. The number of
-// components N is 'req_comp' if req_comp is non-zero, or *comp otherwise.
-// If req_comp is non-zero, *comp has the number of components that _would_
-// have been output otherwise. E.g. if you set req_comp to 4, you will always
-// get RGBA output, but you can check *comp to easily see if it's opaque.
-//
-// An output image with N components has the following components interleaved
-// in this order in each pixel:
-//
-//     N=#comp     components
-//       1           grey
-//       2           grey, alpha
-//       3           red, green, blue
-//       4           red, green, blue, alpha
-//
-// If image loading fails for any reason, the return value will be NULL,
-// and *x, *y, *comp will be unchanged. The function stbi_failure_reason()
-// can be queried for an extremely brief, end-user unfriendly explanation
-// of why the load failed. Define STBI_NO_FAILURE_STRINGS to avoid
-// compiling these strings at all, and STBI_FAILURE_USERMSG to get slightly
-// more user-friendly ones.
-//
-// Paletted PNG, BMP, GIF, and PIC images are automatically depalettized.
-//
-// ===========================================================================
-//
-// iPhone PNG support:
-//
-// By default we convert iphone-formatted PNGs back to RGB; nominally they
-// would silently load as BGR, except the existing code should have just
-// failed on such iPhone PNGs. But you can disable this conversion by
-// by calling stbi_convert_iphone_png_to_rgb(0), in which case
-// you will always just get the native iphone "format" through.
-//
-// Call stbi_set_unpremultiply_on_load(1) as well to force a divide per
-// pixel to remove any premultiplied alpha *only* if the image file explicitly
-// says there's premultiplied data (currently only happens in iPhone images,
-// and only if iPhone convert-to-rgb processing is on).
-//
-// ===========================================================================
-//
-// HDR image support   (disable by defining STBI_NO_HDR)
-//
-// stb_image now supports loading HDR images in general, and currently
-// the Radiance .HDR file format, although the support is provided
-// generically. You can still load any file through the existing interface;
-// if you attempt to load an HDR file, it will be automatically remapped to
-// LDR, assuming gamma 2.2 and an arbitrary scale factor defaulting to 1;
-// both of these constants can be reconfigured through this interface:
-//
-//     stbi_hdr_to_ldr_gamma(2.2f);
-//     stbi_hdr_to_ldr_scale(1.0f);
-//
-// (note, do not use _inverse_ constants; stbi_image will invert them
-// appropriately).
-//
-// Additionally, there is a new, parallel interface for loading files as
-// (linear) floats to preserve the full dynamic range:
-//
-//    float *data = stbi_loadf(filename, &x, &y, &n, 0);
-// 
-// If you load LDR images through this interface, those images will
-// be promoted to floating point values, run through the inverse of
-// constants corresponding to the above:
-//
-//     stbi_ldr_to_hdr_scale(1.0f);
-//     stbi_ldr_to_hdr_gamma(2.2f);
-//
-// Finally, given a filename (or an open file or memory block--see header
-// file for details) containing image data, you can query for the "most
-// appropriate" interface to use (that is, whether the image is HDR or
-// not), using:
-//
-//     stbi_is_hdr(char *filename);
-
-#ifndef STBI_NO_STDIO
-#include <stdio.h>
-#endif
-
-#define STBI_VERSION 1
-
-enum
-{
-   STBI_default = 0, // only used for req_comp
-
-   STBI_grey       = 1,
-   STBI_grey_alpha = 2,
-   STBI_rgb        = 3,
-   STBI_rgb_alpha  = 4
-};
-
-typedef unsigned char stbi_uc;
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-// PRIMARY API - works on images of any type
-
-// load image by filename, open file, or memory buffer
-extern stbi_uc *stbi_load_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp);
-
-#ifndef STBI_NO_STDIO
-extern stbi_uc *stbi_load            (char const *filename,     int *x, int *y, int *comp, int req_comp);
-extern stbi_uc *stbi_load_from_file  (FILE *f,                  int *x, int *y, int *comp, int req_comp);
-// for stbi_load_from_file, file pointer is left pointing immediately after image
-#endif
-
-#ifndef STBI_NO_HDR
-   extern float *stbi_loadf_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp);
-
-   #ifndef STBI_NO_STDIO
-   extern float *stbi_loadf            (char const *filename,   int *x, int *y, int *comp, int req_comp);
-   extern float *stbi_loadf_from_file  (FILE *f,                int *x, int *y, int *comp, int req_comp);
-   #endif
-
-   extern void   stbi_hdr_to_ldr_gamma(float gamma);
-   extern void   stbi_hdr_to_ldr_scale(float scale);
-
-   extern void   stbi_ldr_to_hdr_gamma(float gamma);
-   extern void   stbi_ldr_to_hdr_scale(float scale);
-#endif // STBI_NO_HDR
-
-// get a VERY brief reason for failure
-// NOT THREADSAFE
-extern const char *stbi_failure_reason  (void); 
-
-// free the loaded image -- this is just free()
-extern void     stbi_image_free      (void *retval_from_stbi_load);
-
-// get image dimensions & components without fully decoding
-extern int      stbi_info_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp);
-extern int      stbi_is_hdr_from_memory(stbi_uc const *buffer, int len);
-
-#ifndef STBI_NO_STDIO
-extern int      stbi_info            (char const *filename,     int *x, int *y, int *comp);
-extern int      stbi_info_from_file  (FILE *f,                  int *x, int *y, int *comp);
-
-extern int      stbi_is_hdr          (char const *filename);
-extern int      stbi_is_hdr_from_file(FILE *f);
-#endif
-
-// for image formats that explicitly notate that they have premultiplied alpha,
-// we just return the colors as stored in the file. set this flag to force
-// unpremultiplication. results are undefined if the unpremultiply overflow.
-extern void stbi_set_unpremultiply_on_load(int flag_true_if_should_unpremultiply);
-
-// indicate whether we should process iphone images back to canonical format,
-// or just pass them through "as-is"
-extern void stbi_convert_iphone_png_to_rgb(int flag_true_if_should_convert);
-
-
-// ZLIB client - used by PNG, available for other purposes
-
-extern char *stbi_zlib_decode_malloc_guesssize(const char *buffer, int len, int initial_size, int *outlen);
-extern char *stbi_zlib_decode_malloc(const char *buffer, int len, int *outlen);
-extern int   stbi_zlib_decode_buffer(char *obuffer, int olen, const char *ibuffer, int ilen);
-
-extern char *stbi_zlib_decode_noheader_malloc(const char *buffer, int len, int *outlen);
-extern int   stbi_zlib_decode_noheader_buffer(char *obuffer, int olen, const char *ibuffer, int ilen);
-
-// define new loaders
-typedef struct
-{
-   int       (*test_memory)(stbi_uc const *buffer, int len);
-   stbi_uc * (*load_from_memory)(stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp);
-   #ifndef STBI_NO_STDIO
-   int       (*test_file)(FILE *f);
-   stbi_uc * (*load_from_file)(FILE *f, int *x, int *y, int *comp, int req_comp);
-   #endif
-} stbi_loader;
-
-// register a loader by filling out the above structure (you must define ALL functions)
-// returns 1 if added or already added, 0 if not added (too many loaders)
-// NOT THREADSAFE
-extern int stbi_register_loader(stbi_loader *loader);
-
-// define faster low-level operations (typically SIMD support)
-#ifdef STBI_SIMD
-typedef void (*stbi_idct_8x8)(stbi_uc *out, int out_stride, short data[64], unsigned short *dequantize);
-// compute an integer IDCT on "input"
-//     input[x] = data[x] * dequantize[x]
-//     write results to 'out': 64 samples, each run of 8 spaced by 'out_stride'
-//                             CLAMP results to 0..255
-typedef void (*stbi_YCbCr_to_RGB_run)(stbi_uc *output, stbi_uc const  *y, stbi_uc const *cb, stbi_uc const *cr, int count, int step);
-// compute a conversion from YCbCr to RGB
-//     'count' pixels
-//     write pixels to 'output'; each pixel is 'step' bytes (either 3 or 4; if 4, write '255' as 4th), order R,G,B
-//     y: Y input channel
-//     cb: Cb input channel; scale/biased to be 0..255
-//     cr: Cr input channel; scale/biased to be 0..255
-
-extern void stbi_install_idct(stbi_idct_8x8 func);
-extern void stbi_install_YCbCr_to_RGB(stbi_YCbCr_to_RGB_run func);
-#endif // STBI_SIMD
-
-
-
-
-// TYPE-SPECIFIC ACCESS
-
-#ifdef STBI_TYPE_SPECIFIC_FUNCTIONS
-
-// is it a jpeg?
-extern int      stbi_jpeg_test_memory     (stbi_uc const *buffer, int len);
-extern stbi_uc *stbi_jpeg_load_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp);
-extern int      stbi_jpeg_info_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp);
-
-#ifndef STBI_NO_STDIO
-extern stbi_uc *stbi_jpeg_load            (char const *filename,     int *x, int *y, int *comp, int req_comp);
-extern int      stbi_jpeg_test_file       (FILE *f);
-extern stbi_uc *stbi_jpeg_load_from_file  (FILE *f,                  int *x, int *y, int *comp, int req_comp);
-
-extern int      stbi_jpeg_info            (char const *filename,     int *x, int *y, int *comp);
-extern int      stbi_jpeg_info_from_file  (FILE *f,                  int *x, int *y, int *comp);
-#endif
-
-// is it a png?
-extern int      stbi_png_test_memory      (stbi_uc const *buffer, int len);
-extern stbi_uc *stbi_png_load_from_memory (stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp);
-extern int      stbi_png_info_from_memory (stbi_uc const *buffer, int len, int *x, int *y, int *comp);
-
-#ifndef STBI_NO_STDIO
-extern stbi_uc *stbi_png_load             (char const *filename,     int *x, int *y, int *comp, int req_comp);
-extern int      stbi_png_info             (char const *filename,     int *x, int *y, int *comp);
-extern int      stbi_png_test_file        (FILE *f);
-extern stbi_uc *stbi_png_load_from_file   (FILE *f,                  int *x, int *y, int *comp, int req_comp);
-extern int      stbi_png_info_from_file   (FILE *f,                  int *x, int *y, int *comp);
-#endif
-
-// is it a bmp?
-extern int      stbi_bmp_test_memory      (stbi_uc const *buffer, int len);
-
-extern stbi_uc *stbi_bmp_load             (char const *filename,     int *x, int *y, int *comp, int req_comp);
-extern stbi_uc *stbi_bmp_load_from_memory (stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp);
-#ifndef STBI_NO_STDIO
-extern int      stbi_bmp_test_file        (FILE *f);
-extern stbi_uc *stbi_bmp_load_from_file   (FILE *f,                  int *x, int *y, int *comp, int req_comp);
-#endif
-
-// is it a tga?
-extern int      stbi_tga_test_memory      (stbi_uc const *buffer, int len);
-
-extern stbi_uc *stbi_tga_load             (char const *filename,     int *x, int *y, int *comp, int req_comp);
-extern stbi_uc *stbi_tga_load_from_memory (stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp);
-#ifndef STBI_NO_STDIO
-extern int      stbi_tga_test_file        (FILE *f);
-extern stbi_uc *stbi_tga_load_from_file   (FILE *f,                  int *x, int *y, int *comp, int req_comp);
-#endif
-
-// is it a psd?
-extern int      stbi_psd_test_memory      (stbi_uc const *buffer, int len);
-
-extern stbi_uc *stbi_psd_load             (char const *filename,     int *x, int *y, int *comp, int req_comp);
-extern stbi_uc *stbi_psd_load_from_memory (stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp);
-#ifndef STBI_NO_STDIO
-extern int      stbi_psd_test_file        (FILE *f);
-extern stbi_uc *stbi_psd_load_from_file   (FILE *f,                  int *x, int *y, int *comp, int req_comp);
-#endif
-
-// is it an hdr?
-extern int      stbi_hdr_test_memory      (stbi_uc const *buffer, int len);
-
-extern float *  stbi_hdr_load             (char const *filename,     int *x, int *y, int *comp, int req_comp);
-extern float *  stbi_hdr_load_from_memory (stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp);
-#ifndef STBI_NO_STDIO
-extern int      stbi_hdr_test_file        (FILE *f);
-extern float *  stbi_hdr_load_from_file   (FILE *f,                  int *x, int *y, int *comp, int req_comp);
-#endif
-
-// is it a pic?
-extern int      stbi_pic_test_memory      (stbi_uc const *buffer, int len);
-
-extern stbi_uc *stbi_pic_load             (char const *filename,     int *x, int *y, int *comp, int req_comp);
-extern stbi_uc *stbi_pic_load_from_memory (stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp);
-#ifndef STBI_NO_STDIO
-extern int      stbi_pic_test_file        (FILE *f);
-extern stbi_uc *stbi_pic_load_from_file   (FILE *f,                  int *x, int *y, int *comp, int req_comp);
-#endif
-
-// is it a gif?
-extern int      stbi_gif_test_memory      (stbi_uc const *buffer, int len);
-
-extern stbi_uc *stbi_gif_load             (char const *filename,     int *x, int *y, int *comp, int req_comp);
-extern stbi_uc *stbi_gif_load_from_memory (stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp);
-extern int      stbi_gif_info_from_memory (stbi_uc const *buffer, int len, int *x, int *y, int *comp);
-
-#ifndef STBI_NO_STDIO
-extern int      stbi_gif_test_file        (FILE *f);
-extern stbi_uc *stbi_gif_load_from_file   (FILE *f,                  int *x, int *y, int *comp, int req_comp);
-extern int      stbi_gif_info             (char const *filename,     int *x, int *y, int *comp);
-extern int      stbi_gif_info_from_file   (FILE *f,                  int *x, int *y, int *comp);
-#endif
-
-#endif//STBI_TYPE_SPECIFIC_FUNCTIONS
-
-
-
-
-#ifdef __cplusplus
-}
-#endif
-
-//
-//
-////   end header file   /////////////////////////////////////////////////////
-#endif // STBI_INCLUDE_STB_IMAGE_H
+/* stb_image - v2.05 - public domain image loader - http://nothings.org/stb_image.h
+                                     no warranty implied; use at your own risk
+
+   Do this:
+      #define STB_IMAGE_IMPLEMENTATION
+   before you include this file in *one* C or C++ file to create the implementation.
+
+   // i.e. it should look like this:
+   #include ...
+   #include ...
+   #include ...
+   #define STB_IMAGE_IMPLEMENTATION
+   #include "stb_image.h"
+
+   You can #define STBI_ASSERT(x) before the #include to avoid using assert.h.
+   And #define STBI_MALLOC, STBI_REALLOC, and STBI_FREE to avoid using malloc,realloc,free
+
+
+   QUICK NOTES:
+      Primarily of interest to game developers and other people who can
+          avoid problematic images and only need the trivial interface
+
+      JPEG baseline & progressive (12 bpc/arithmetic not supported, same as stock IJG lib)
+      PNG 1/2/4/8-bit-per-channel (16 bpc not supported)
+
+      TGA (not sure what subset, if a subset)
+      BMP non-1bpp, non-RLE
+      PSD (composited view only, no extra channels)
+
+      GIF (*comp always reports as 4-channel)
+      HDR (radiance rgbE format)
+      PIC (Softimage PIC)
+      PNM (PPM and PGM binary only)
+
+      - decode from memory or through FILE (define STBI_NO_STDIO to remove code)
+      - decode from arbitrary I/O callbacks
+      - SIMD acceleration on x86/x64 (SSE2) and ARM (NEON)
+
+   Full documentation under "DOCUMENTATION" below.
+
+
+   Revision 2.00 release notes:
+
+      - Progressive JPEG is now supported.
+
+      - PPM and PGM binary formats are now supported, thanks to Ken Miller.
+
+      - x86 platforms now make use of SSE2 SIMD instructions for
+        JPEG decoding, and ARM platforms can use NEON SIMD if requested.
+        This work was done by Fabian "ryg" Giesen. SSE2 is used by
+        default, but NEON must be enabled explicitly; see docs.
+
+        With other JPEG optimizations included in this version, we see
+        2x speedup on a JPEG on an x86 machine, and a 1.5x speedup
+        on a JPEG on an ARM machine, relative to previous versions of this
+        library. The same results will not obtain for all JPGs and for all
+        x86/ARM machines. (Note that progressive JPEGs are significantly
+        slower to decode than regular JPEGs.) This doesn't mean that this
+        is the fastest JPEG decoder in the land; rather, it brings it
+        closer to parity with standard libraries. If you want the fastest
+        decode, look elsewhere. (See "Philosophy" section of docs below.)
+
+        See final bullet items below for more info on SIMD.
+
+      - Added STBI_MALLOC, STBI_REALLOC, and STBI_FREE macros for replacing
+        the memory allocator. Unlike other STBI libraries, these macros don't
+        support a context parameter, so if you need to pass a context in to
+        the allocator, you'll have to store it in a global or a thread-local
+        variable.
+
+      - Split existing STBI_NO_HDR flag into two flags, STBI_NO_HDR and
+        STBI_NO_LINEAR.
+            STBI_NO_HDR:     suppress implementation of .hdr reader format
+            STBI_NO_LINEAR:  suppress high-dynamic-range light-linear float API
+
+      - You can suppress implementation of any of the decoders to reduce
+        your code footprint by #defining one or more of the following
+        symbols before creating the implementation.
+
+            STBI_NO_JPEG
+            STBI_NO_PNG
+            STBI_NO_BMP
+            STBI_NO_PSD
+            STBI_NO_TGA
+            STBI_NO_GIF
+            STBI_NO_HDR
+            STBI_NO_PIC
+            STBI_NO_PNM   (.ppm and .pgm)
+
+      - You can request *only* certain decoders and suppress all other ones
+        (this will be more forward-compatible, as addition of new decoders
+        doesn't require you to disable them explicitly):
+
+            STBI_ONLY_JPEG
+            STBI_ONLY_PNG
+            STBI_ONLY_BMP
+            STBI_ONLY_PSD
+            STBI_ONLY_TGA
+            STBI_ONLY_GIF
+            STBI_ONLY_HDR
+            STBI_ONLY_PIC
+            STBI_ONLY_PNM   (.ppm and .pgm)
+
+         Note that you can define multiples of these, and you will get all
+         of them ("only x" and "only y" is interpreted to mean "only x&y").
+
+       - If you use STBI_NO_PNG (or _ONLY_ without PNG), and you still
+         want the zlib decoder to be available, #define STBI_SUPPORT_ZLIB
+
+      - Compilation of all SIMD code can be suppressed with
+            #define STBI_NO_SIMD
+        It should not be necessary to disable SIMD unless you have issues
+        compiling (e.g. using an x86 compiler which doesn't support SSE
+        intrinsics or that doesn't support the method used to detect
+        SSE2 support at run-time), and even those can be reported as
+        bugs so I can refine the built-in compile-time checking to be
+        smarter.
+
+      - The old STBI_SIMD system which allowed installing a user-defined
+        IDCT etc. has been removed. If you need this, don't upgrade. My
+        assumption is that almost nobody was doing this, and those who
+        were will find the built-in SIMD more satisfactory anyway.
+
+      - RGB values computed for JPEG images are slightly different from
+        previous versions of stb_image. (This is due to using less
+        integer precision in SIMD.) The C code has been adjusted so
+        that the same RGB values will be computed regardless of whether
+        SIMD support is available, so your app should always produce
+        consistent results. But these results are slightly different from
+        previous versions. (Specifically, about 3% of available YCbCr values
+        will compute different RGB results from pre-1.49 versions by +-1;
+        most of the deviating values are one smaller in the G channel.)
+
+      - If you must produce consistent results with previous versions of
+        stb_image, #define STBI_JPEG_OLD and you will get the same results
+        you used to; however, you will not get the SIMD speedups for
+        the YCbCr-to-RGB conversion step (although you should still see
+        significant JPEG speedup from the other changes).
+
+        Please note that STBI_JPEG_OLD is a temporary feature; it will be
+        removed in future versions of the library. It is only intended for
+        near-term back-compatibility use.
+
+
+   Latest revision history:
+      2.05  (2015-04-19) fix bug in progressive JPEG handling, fix warning
+      2.04  (2015-04-15) try to re-enable SIMD on MinGW 64-bit
+      2.03  (2015-04-12) additional corruption checking
+                         stbi_set_flip_vertically_on_load
+                         fix NEON support; fix mingw support
+      2.02  (2015-01-19) fix incorrect assert, fix warning
+      2.01  (2015-01-17) fix various warnings
+      2.00b (2014-12-25) fix STBI_MALLOC in progressive JPEG
+      2.00  (2014-12-25) optimize JPEG, including x86 SSE2 & ARM NEON SIMD
+                         progressive JPEG
+                         PGM/PPM support
+                         STBI_MALLOC,STBI_REALLOC,STBI_FREE
+                         STBI_NO_*, STBI_ONLY_*
+                         GIF bugfix
+      1.48  (2014-12-14) fix incorrectly-named assert()
+      1.47  (2014-12-14) 1/2/4-bit PNG support (both grayscale and paletted)
+                         optimize PNG
+                         fix bug in interlaced PNG with user-specified channel count
+
+   See end of file for full revision history.
+
+
+ ============================    Contributors    =========================
+
+ Image formats                                Bug fixes & warning fixes
+    Sean Barrett (jpeg, png, bmp)                Marc LeBlanc
+    Nicolas Schulz (hdr, psd)                    Christpher Lloyd
+    Jonathan Dummer (tga)                        Dave Moore
+    Jean-Marc Lienher (gif)                      Won Chun
+    Tom Seddon (pic)                             the Horde3D community
+    Thatcher Ulrich (psd)                        Janez Zemva
+    Ken Miller (pgm, ppm)                        Jonathan Blow
+                                                 Laurent Gomila
+                                                 Aruelien Pocheville
+ Extensions, features                            Ryamond Barbiero
+    Jetro Lauha (stbi_info)                      David Woo
+    Martin "SpartanJ" Golini (stbi_info)         Martin Golini
+    James "moose2000" Brown (iPhone PNG)         Roy Eltham
+    Ben "Disch" Wenger (io callbacks)            Luke Graham
+    Omar Cornut (1/2/4-bit PNG)                  Thomas Ruf
+    Nicolas Guillemot (vertical flip)            John Bartholomew
+                                                 Ken Hamada
+ Optimizations & bugfixes                        Cort Stratton
+    Fabian "ryg" Giesen                          Blazej Dariusz Roszkowski
+    Arseny Kapoulkine                            Thibault Reuille
+                                                 Paul Du Bois
+                                                 Guillaume George
+  If your name should be here but                Jerry Jansson
+  isn't, let Sean know.                          Hayaki Saito
+                                                 Johan Duparc
+                                                 Ronny Chevalier
+                                                 Michal Cichon
+                                                 Tero Hanninen
+                                                 Sergio Gonzalez
+                                                 Cass Everitt
+                                                 Engin Manap
+                                                 Martins Mozeiko
+                                                 Joseph Thomson
+                                                 Phil Jordan
+
+License:
+   This software is in the public domain. Where that dedication is not
+   recognized, you are granted a perpetual, irrevocable license to copy
+   and modify this file however you want.
+
+*/
+
+#ifndef STBI_INCLUDE_STB_IMAGE_H
+#define STBI_INCLUDE_STB_IMAGE_H
+
+// DOCUMENTATION
+//
+// Limitations:
+//    - no 16-bit-per-channel PNG
+//    - no 12-bit-per-channel JPEG
+//    - no JPEGs with arithmetic coding
+//    - no 1-bit BMP
+//    - GIF always returns *comp=4
+//
+// Basic usage (see HDR discussion below for HDR usage):
+//    int x,y,n;
+//    unsigned char *data = stbi_load(filename, &x, &y, &n, 0);
+//    // ... process data if not NULL ...
+//    // ... x = width, y = height, n = # 8-bit components per pixel ...
+//    // ... replace '0' with '1'..'4' to force that many components per pixel
+//    // ... but 'n' will always be the number that it would have been if you said 0
+//    stbi_image_free(data)
+//
+// Standard parameters:
+//    int *x       -- outputs image width in pixels
+//    int *y       -- outputs image height in pixels
+//    int *comp    -- outputs # of image components in image file
+//    int req_comp -- if non-zero, # of image components requested in result
+//
+// The return value from an image loader is an 'unsigned char *' which points
+// to the pixel data, or NULL on an allocation failure or if the image is
+// corrupt or invalid. The pixel data consists of *y scanlines of *x pixels,
+// with each pixel consisting of N interleaved 8-bit components; the first
+// pixel pointed to is top-left-most in the image. There is no padding between
+// image scanlines or between pixels, regardless of format. The number of
+// components N is 'req_comp' if req_comp is non-zero, or *comp otherwise.
+// If req_comp is non-zero, *comp has the number of components that _would_
+// have been output otherwise. E.g. if you set req_comp to 4, you will always
+// get RGBA output, but you can check *comp to see if it's trivially opaque
+// because e.g. there were only 3 channels in the source image.
+//
+// An output image with N components has the following components interleaved
+// in this order in each pixel:
+//
+//     N=#comp     components
+//       1           grey
+//       2           grey, alpha
+//       3           red, green, blue
+//       4           red, green, blue, alpha
+//
+// If image loading fails for any reason, the return value will be NULL,
+// and *x, *y, *comp will be unchanged. The function stbi_failure_reason()
+// can be queried for an extremely brief, end-user unfriendly explanation
+// of why the load failed. Define STBI_NO_FAILURE_STRINGS to avoid
+// compiling these strings at all, and STBI_FAILURE_USERMSG to get slightly
+// more user-friendly ones.
+//
+// Paletted PNG, BMP, GIF, and PIC images are automatically depalettized.
+//
+// ===========================================================================
+//
+// Philosophy
+//
+// stb libraries are designed with the following priorities:
+//
+//    1. easy to use
+//    2. easy to maintain
+//    3. good performance
+//
+// Sometimes I let "good performance" creep up in priority over "easy to maintain",
+// and for best performance I may provide less-easy-to-use APIs that give higher
+// performance, in addition to the easy to use ones. Nevertheless, it's important
+// to keep in mind that from the standpoint of you, a client of this library,
+// all you care about is #1 and #3, and stb libraries do not emphasize #3 above all.
+//
+// Some secondary priorities arise directly from the first two, some of which
+// make more explicit reasons why performance can't be emphasized.
+//
+//    - Portable ("ease of use")
+//    - Small footprint ("easy to maintain")
+//    - No dependencies ("ease of use")
+//
+// ===========================================================================
+//
+// I/O callbacks
+//
+// I/O callbacks allow you to read from arbitrary sources, like packaged
+// files or some other source. Data read from callbacks are processed
+// through a small internal buffer (currently 128 bytes) to try to reduce
+// overhead.
+//
+// The three functions you must define are "read" (reads some bytes of data),
+// "skip" (skips some bytes of data), "eof" (reports if the stream is at the end).
+//
+// ===========================================================================
+//
+// SIMD support
+//
+// The JPEG decoder will try to automatically use SIMD kernels on x86 when
+// supported by the compiler. For ARM Neon support, you must explicitly
+// request it.
+//
+// (The old do-it-yourself SIMD API is no longer supported in the current
+// code.)
+//
+// On x86, SSE2 will automatically be used when available based on a run-time
+// test; if not, the generic C versions are used as a fall-back. On ARM targets,
+// the typical path is to have separate builds for NEON and non-NEON devices
+// (at least this is true for iOS and Android). Therefore, the NEON support is
+// toggled by a build flag: define STBI_NEON to get NEON loops.
+//
+// The output of the JPEG decoder is slightly different from versions where
+// SIMD support was introduced (that is, for versions before 1.49). The
+// difference is only +-1 in the 8-bit RGB channels, and only on a small
+// fraction of pixels. You can force the pre-1.49 behavior by defining
+// STBI_JPEG_OLD, but this will disable some of the SIMD decoding path
+// and hence cost some performance.
+//
+// If for some reason you do not want to use any of SIMD code, or if
+// you have issues compiling it, you can disable it entirely by
+// defining STBI_NO_SIMD.
+//
+// ===========================================================================
+//
+// HDR image support   (disable by defining STBI_NO_HDR)
+//
+// stb_image now supports loading HDR images in general, and currently
+// the Radiance .HDR file format, although the support is provided
+// generically. You can still load any file through the existing interface;
+// if you attempt to load an HDR file, it will be automatically remapped to
+// LDR, assuming gamma 2.2 and an arbitrary scale factor defaulting to 1;
+// both of these constants can be reconfigured through this interface:
+//
+//     stbi_hdr_to_ldr_gamma(2.2f);
+//     stbi_hdr_to_ldr_scale(1.0f);
+//
+// (note, do not use _inverse_ constants; stbi_image will invert them
+// appropriately).
+//
+// Additionally, there is a new, parallel interface for loading files as
+// (linear) floats to preserve the full dynamic range:
+//
+//    float *data = stbi_loadf(filename, &x, &y, &n, 0);
+//
+// If you load LDR images through this interface, those images will
+// be promoted to floating point values, run through the inverse of
+// constants corresponding to the above:
+//
+//     stbi_ldr_to_hdr_scale(1.0f);
+//     stbi_ldr_to_hdr_gamma(2.2f);
+//
+// Finally, given a filename (or an open file or memory block--see header
+// file for details) containing image data, you can query for the "most
+// appropriate" interface to use (that is, whether the image is HDR or
+// not), using:
+//
+//     stbi_is_hdr(char *filename);
+//
+// ===========================================================================
+//
+// iPhone PNG support:
+//
+// By default we convert iphone-formatted PNGs back to RGB, even though
+// they are internally encoded differently. You can disable this conversion
+// by by calling stbi_convert_iphone_png_to_rgb(0), in which case
+// you will always just get the native iphone "format" through (which
+// is BGR stored in RGB).
+//
+// Call stbi_set_unpremultiply_on_load(1) as well to force a divide per
+// pixel to remove any premultiplied alpha *only* if the image file explicitly
+// says there's premultiplied data (currently only happens in iPhone images,
+// and only if iPhone convert-to-rgb processing is on).
+//
+
+
+#ifndef STBI_NO_STDIO
+#include <stdio.h>
+#endif // STBI_NO_STDIO
+
+#define STBI_VERSION 1
+
+enum
+{
+   STBI_default = 0, // only used for req_comp
+
+   STBI_grey       = 1,
+   STBI_grey_alpha = 2,
+   STBI_rgb        = 3,
+   STBI_rgb_alpha  = 4
+};
+
+typedef unsigned char stbi_uc;
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#ifdef STB_IMAGE_STATIC
+#define STBIDEF static
+#else
+#define STBIDEF extern
+#endif
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// PRIMARY API - works on images of any type
+//
+
+//
+// load image by filename, open file, or memory buffer
+//
+
+typedef struct
+{
+   int      (*read)  (void *user,char *data,int size);   // fill 'data' with 'size' bytes.  return number of bytes actually read
+   void     (*skip)  (void *user,int n);                 // skip the next 'n' bytes, or 'unget' the last -n bytes if negative
+   int      (*eof)   (void *user);                       // returns nonzero if we are at end of file/data
+} stbi_io_callbacks;
+
+STBIDEF stbi_uc *stbi_load               (char              const *filename,           int *x, int *y, int *comp, int req_comp);
+STBIDEF stbi_uc *stbi_load_from_memory   (stbi_uc           const *buffer, int len   , int *x, int *y, int *comp, int req_comp);
+STBIDEF stbi_uc *stbi_load_from_callbacks(stbi_io_callbacks const *clbk  , void *user, int *x, int *y, int *comp, int req_comp);
+
+#ifndef STBI_NO_STDIO
+STBIDEF stbi_uc *stbi_load_from_file  (FILE *f,                  int *x, int *y, int *comp, int req_comp);
+// for stbi_load_from_file, file pointer is left pointing immediately after image
+#endif
+
+#ifndef STBI_NO_LINEAR
+   STBIDEF float *stbi_loadf                 (char const *filename,           int *x, int *y, int *comp, int req_comp);
+   STBIDEF float *stbi_loadf_from_memory     (stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp);
+   STBIDEF float *stbi_loadf_from_callbacks  (stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp, int req_comp);
+
+   #ifndef STBI_NO_STDIO
+   STBIDEF float *stbi_loadf_from_file  (FILE *f,                int *x, int *y, int *comp, int req_comp);
+   #endif
+#endif
+
+#ifndef STBI_NO_HDR
+   STBIDEF void   stbi_hdr_to_ldr_gamma(float gamma);
+   STBIDEF void   stbi_hdr_to_ldr_scale(float scale);
+#endif
+
+#ifndef STBI_NO_LINEAR
+   STBIDEF void   stbi_ldr_to_hdr_gamma(float gamma);
+   STBIDEF void   stbi_ldr_to_hdr_scale(float scale);
+#endif // STBI_NO_HDR
+
+// stbi_is_hdr is always defined, but always returns false if STBI_NO_HDR
+STBIDEF int    stbi_is_hdr_from_callbacks(stbi_io_callbacks const *clbk, void *user);
+STBIDEF int    stbi_is_hdr_from_memory(stbi_uc const *buffer, int len);
+#ifndef STBI_NO_STDIO
+STBIDEF int      stbi_is_hdr          (char const *filename);
+STBIDEF int      stbi_is_hdr_from_file(FILE *f);
+#endif // STBI_NO_STDIO
+
+
+// get a VERY brief reason for failure
+// NOT THREADSAFE
+STBIDEF const char *stbi_failure_reason  (void);
+
+// free the loaded image -- this is just free()
+STBIDEF void     stbi_image_free      (void *retval_from_stbi_load);
+
+// get image dimensions & components without fully decoding
+STBIDEF int      stbi_info_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp);
+STBIDEF int      stbi_info_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp);
+
+#ifndef STBI_NO_STDIO
+STBIDEF int      stbi_info            (char const *filename,     int *x, int *y, int *comp);
+STBIDEF int      stbi_info_from_file  (FILE *f,                  int *x, int *y, int *comp);
+
+#endif
+
+
+
+// for image formats that explicitly notate that they have premultiplied alpha,
+// we just return the colors as stored in the file. set this flag to force
+// unpremultiplication. results are undefined if the unpremultiply overflow.
+STBIDEF void stbi_set_unpremultiply_on_load(int flag_true_if_should_unpremultiply);
+
+// indicate whether we should process iphone images back to canonical format,
+// or just pass them through "as-is"
+STBIDEF void stbi_convert_iphone_png_to_rgb(int flag_true_if_should_convert);
+
+// flip the image vertically, so the first pixel in the output array is the bottom left
+STBIDEF void stbi_set_flip_vertically_on_load(int flag_true_if_should_flip);
+
+// ZLIB client - used by PNG, available for other purposes
+
+STBIDEF char *stbi_zlib_decode_malloc_guesssize(const char *buffer, int len, int initial_size, int *outlen);
+STBIDEF char *stbi_zlib_decode_malloc_guesssize_headerflag(const char *buffer, int len, int initial_size, int *outlen, int parse_header);
+STBIDEF char *stbi_zlib_decode_malloc(const char *buffer, int len, int *outlen);
+STBIDEF int   stbi_zlib_decode_buffer(char *obuffer, int olen, const char *ibuffer, int ilen);
+
+STBIDEF char *stbi_zlib_decode_noheader_malloc(const char *buffer, int len, int *outlen);
+STBIDEF int   stbi_zlib_decode_noheader_buffer(char *obuffer, int olen, const char *ibuffer, int ilen);
+
+
+#ifdef __cplusplus
+}
+#endif
+
+//
+//
+////   end header file   /////////////////////////////////////////////////////
+#endif // STBI_INCLUDE_STB_IMAGE_H

Source/ThirdParty/STB/stb_image_write.c

@@ -1,10 +1,36 @@
 #include "stb_image_write.h"
+#define STB_IMAGE_WRITE_IMPLEMENTATION
+
+#ifdef STB_IMAGE_WRITE_IMPLEMENTATION
 
 #include <stdarg.h>
 #include <stdlib.h>
 #include <stdio.h>
 #include <string.h>
+#include <math.h>
+
+#if defined(STBIW_MALLOC) && defined(STBIW_FREE) && defined(STBIW_REALLOC)
+// ok
+#elif !defined(STBIW_MALLOC) && !defined(STBIW_FREE) && !defined(STBIW_REALLOC)
+// ok
+#else
+#error "Must define all or none of STBIW_MALLOC, STBIW_FREE, and STBIW_REALLOC."
+#endif
+
+#ifndef STBIW_MALLOC
+#define STBIW_MALLOC(sz)    malloc(sz)
+#define STBIW_REALLOC(p,sz) realloc(p,sz)
+#define STBIW_FREE(p)       free(p)
+#endif
+#ifndef STBIW_MEMMOVE
+#define STBIW_MEMMOVE(a,b,sz) memmove(a,b,sz)
+#endif
+
+
+#ifndef STBIW_ASSERT
 #include <assert.h>
+#define STBIW_ASSERT(x) assert(x)
+#endif
 
 typedef unsigned int stbiw_uint32;
 typedef int stb_image_write_test[sizeof(stbiw_uint32)==4 ? 1 : -1];
@@ -23,7 +49,7 @@ static void writefv(FILE *f, const char *fmt, va_list v)
                      b[2]=(unsigned char)(x>>16); b[3]=(unsigned char)(x>>24);
                      fwrite(b,4,1,f); break; }
          default:
-            assert(0);
+            STBIW_ASSERT(0);
             return;
       }
    }
@@ -36,7 +62,7 @@ static void write3(FILE *f, unsigned char a, unsigned char b, unsigned char c)
    fwrite(arr, 3, 1, f);
 }
 
-static void write_pixels(FILE *f, int rgb_dir, int vdir, int x, int y, int comp, void *data, int write_alpha, int scanline_pad)
+static void write_pixels(FILE *f, int rgb_dir, int vdir, int x, int y, int comp, void *data, int write_alpha, int scanline_pad, int expand_mono)
 {
    unsigned char bg[3] = { 255, 0, 255}, px[3];
    stbiw_uint32 zero = 0;
@@ -45,7 +71,7 @@ static void write_pixels(FILE *f, int rgb_dir, int vdir, int x, int y, int comp,
    if (y <= 0)
       return;
 
-   if (vdir < 0) 
+   if (vdir < 0)
       j_end = -1, j = y-1;
    else
       j_end =  y, j = 0;
@@ -56,8 +82,12 @@ static void write_pixels(FILE *f, int rgb_dir, int vdir, int x, int y, int comp,
          if (write_alpha < 0)
             fwrite(&d[comp-1], 1, 1, f);
          switch (comp) {
-            case 1:
-            case 2: write3(f, d[0],d[0],d[0]);
+            case 1: fwrite(d, 1, 1, f);
+                    break;
+            case 2: if (expand_mono)
+                       write3(f, d[0],d[0],d[0]); // monochrome bmp
+                    else
+                       fwrite(d, 1, 1, f);  // monochrome TGA
                     break;
             case 4:
                if (!write_alpha) {
@@ -79,7 +109,7 @@ static void write_pixels(FILE *f, int rgb_dir, int vdir, int x, int y, int comp,
    }
 }
 
-static int outfile(char const *filename, int rgb_dir, int vdir, int x, int y, int comp, void *data, int alpha, int pad, const char *fmt, ...)
+static int outfile(char const *filename, int rgb_dir, int vdir, int x, int y, int comp, int expand_mono, void *data, int alpha, int pad, const char *fmt, ...)
 {
    FILE *f;
    if (y < 0 || x < 0) return 0;
@@ -89,7 +119,7 @@ static int outfile(char const *filename, int rgb_dir, int vdir, int x, int y, in
       va_start(v, fmt);
       writefv(f, fmt, v);
       va_end(v);
-      write_pixels(f,rgb_dir,vdir,x,y,comp,data,alpha,pad);
+      write_pixels(f,rgb_dir,vdir,x,y,comp,data,alpha,pad,expand_mono);
       fclose(f);
    }
    return f != NULL;
@@ -98,7 +128,7 @@ static int outfile(char const *filename, int rgb_dir, int vdir, int x, int y, in
 int stbi_write_bmp(char const *filename, int x, int y, int comp, const void *data)
 {
    int pad = (-x*3) & 3;
-   return outfile(filename,-1,-1,x,y,comp,(void *) data,0,pad,
+   return outfile(filename,-1,-1,x,y,comp,1,(void *) data,0,pad,
            "11 4 22 4" "4 44 22 444444",
            'B', 'M', 14+40+(x*3+pad)*y, 0,0, 14+40,  // file header
             40, x,y, 1,24, 0,0,0,0,0,0);             // bitmap header
@@ -106,48 +136,199 @@ int stbi_write_bmp(char const *filename, int x, int y, int comp, const void *dat
 
 int stbi_write_tga(char const *filename, int x, int y, int comp, const void *data)
 {
-   int has_alpha = !(comp & 1);
-   return outfile(filename, -1,-1, x, y, comp, (void *) data, has_alpha, 0,
-                  "111 221 2222 11", 0,0,2, 0,0,0, 0,0,x,y, 24+8*has_alpha, 8*has_alpha);
+   int has_alpha = (comp == 2 || comp == 4);
+   int colorbytes = has_alpha ? comp-1 : comp;
+   int format = colorbytes < 2 ? 3 : 2; // 3 color channels (RGB/RGBA) = 2, 1 color channel (Y/YA) = 3
+   return outfile(filename, -1,-1, x, y, comp, 0, (void *) data, has_alpha, 0,
+                  "111 221 2222 11", 0,0,format, 0,0,0, 0,0,x,y, (colorbytes+has_alpha)*8, has_alpha*8);
+}
+
+// *************************************************************************************************
+// Radiance RGBE HDR writer
+// by Baldur Karlsson
+#define stbiw__max(a, b)  ((a) > (b) ? (a) : (b))
+
+void stbiw__linear_to_rgbe(unsigned char *rgbe, float *linear)
+{
+   int exponent;
+   float maxcomp = stbiw__max(linear[0], stbiw__max(linear[1], linear[2]));
+
+   if (maxcomp < 1e-32) {
+      rgbe[0] = rgbe[1] = rgbe[2] = rgbe[3] = 0;
+   } else {
+      float normalize = (float) frexp(maxcomp, &exponent) * 256.0f/maxcomp;
+
+      rgbe[0] = (unsigned char)(linear[0] * normalize);
+      rgbe[1] = (unsigned char)(linear[1] * normalize);
+      rgbe[2] = (unsigned char)(linear[2] * normalize);
+      rgbe[3] = (unsigned char)(exponent + 128);
+   }
+}
+
+void stbiw__write_run_data(FILE *f, int length, unsigned char databyte)
+{
+   unsigned char lengthbyte = (unsigned char) (length+128);
+   STBIW_ASSERT(length+128 <= 255);
+   fwrite(&lengthbyte, 1, 1, f);
+   fwrite(&databyte, 1, 1, f);
+}
+
+void stbiw__write_dump_data(FILE *f, int length, unsigned char *data)
+{
+   unsigned char lengthbyte = (unsigned char )(length & 0xff);
+   STBIW_ASSERT(length <= 128); // inconsistent with spec but consistent with official code
+   fwrite(&lengthbyte, 1, 1, f);
+   fwrite(data, length, 1, f);
+}
+
+void stbiw__write_hdr_scanline(FILE *f, int width, int comp, unsigned char *scratch, const float *scanline)
+{
+   unsigned char scanlineheader[4] = { 2, 2, 0, 0 };
+   unsigned char rgbe[4];
+   float linear[3];
+   int x;
+
+   scanlineheader[2] = (width&0xff00)>>8;
+   scanlineheader[3] = (width&0x00ff);
+
+   /* skip RLE for images too small or large */
+   if (width < 8 || width >= 32768) {
+      for (x=0; x < width; x++) {
+         switch (comp) {
+            case 4: /* fallthrough */
+            case 3: linear[2] = scanline[x*comp + 2];
+                    linear[1] = scanline[x*comp + 1];
+                    linear[0] = scanline[x*comp + 0];
+                    break;
+            case 2: /* fallthrough */
+            case 1: linear[0] = linear[1] = linear[2] = scanline[x*comp + 0];
+                    break;
+         }
+         stbiw__linear_to_rgbe(rgbe, linear);
+         fwrite(rgbe, 4, 1, f);
+      }
+   } else {
+      int c,r;
+      /* encode into scratch buffer */
+      for (x=0; x < width; x++) {
+         switch(comp) {
+            case 4: /* fallthrough */
+            case 3: linear[2] = scanline[x*comp + 2];
+                    linear[1] = scanline[x*comp + 1];
+                    linear[0] = scanline[x*comp + 0];
+                    break;
+            case 2: /* fallthrough */
+            case 1: linear[0] = linear[1] = linear[2] = scanline[x*comp + 0];
+                    break;
+         }
+         stbiw__linear_to_rgbe(rgbe, linear);
+         scratch[x + width*0] = rgbe[0];
+         scratch[x + width*1] = rgbe[1];
+         scratch[x + width*2] = rgbe[2];
+         scratch[x + width*3] = rgbe[3];
+      }
+
+      fwrite(scanlineheader, 4, 1, f);
+
+      /* RLE each component separately */
+      for (c=0; c < 4; c++) {
+         unsigned char *comp = &scratch[width*c];
+
+         x = 0;
+         while (x < width) {
+            // find first run
+            r = x;
+            while (r+2 < width) {
+               if (comp[r] == comp[r+1] && comp[r] == comp[r+2])
+                  break;
+               ++r;
+            }
+            if (r+2 >= width)
+               r = width;
+            // dump up to first run
+            while (x < r) {
+               int len = r-x;
+               if (len > 128) len = 128;
+               stbiw__write_dump_data(f, len, &comp[x]);
+               x += len;
+            }
+            // if there's a run, output it
+            if (r+2 < width) { // same test as what we break out of in search loop, so only true if we break'd
+               // find next byte after run
+               while (r < width && comp[r] == comp[x])
+                  ++r;
+               // output run up to r
+               while (x < r) {
+                  int len = r-x;
+                  if (len > 127) len = 127;
+                  stbiw__write_run_data(f, len, comp[x]);
+                  x += len;
+               }
+            }
+         }
+      }
+   }
+}
+
+int stbi_write_hdr(char const *filename, int x, int y, int comp, const float *data)
+{
+   int i;
+   FILE *f;
+   if (y <= 0 || x <= 0 || data == NULL) return 0;
+   f = fopen(filename, "wb");
+   if (f) {
+      /* Each component is stored separately. Allocate scratch space for full output scanline. */
+      unsigned char *scratch = (unsigned char *) STBIW_MALLOC(x*4);
+      fprintf(f, "#?RADIANCE\n# Written by stb_image_write.h\nFORMAT=32-bit_rle_rgbe\n"      );
+      fprintf(f, "EXPOSURE=          1.0000000000000\n\n-Y %d +X %d\n"                 , y, x);
+      for(i=0; i < y; i++)
+         stbiw__write_hdr_scanline(f, x, comp, scratch, data + comp*i*x);
+      STBIW_FREE(scratch);
+      fclose(f);
+   }
+   return f != NULL;
 }
 
-// stretchy buffer; stbi__sbpush() == vector<>::push_back() -- stbi__sbcount() == vector<>::size()
-#define stbi__sbraw(a) ((int *) (a) - 2)
-#define stbi__sbm(a)   stbi__sbraw(a)[0]
-#define stbi__sbn(a)   stbi__sbraw(a)[1]
+/////////////////////////////////////////////////////////
+// PNG
+
+// stretchy buffer; stbiw__sbpush() == vector<>::push_back() -- stbiw__sbcount() == vector<>::size()
+#define stbiw__sbraw(a) ((int *) (a) - 2)
+#define stbiw__sbm(a)   stbiw__sbraw(a)[0]
+#define stbiw__sbn(a)   stbiw__sbraw(a)[1]
 
-#define stbi__sbneedgrow(a,n)  ((a)==0 || stbi__sbn(a)+n >= stbi__sbm(a))
-#define stbi__sbmaybegrow(a,n) (stbi__sbneedgrow(a,(n)) ? stbi__sbgrow(a,n) : 0)
-#define stbi__sbgrow(a,n)  stbi__sbgrowf((void **) &(a), (n), sizeof(*(a)))
+#define stbiw__sbneedgrow(a,n)  ((a)==0 || stbiw__sbn(a)+n >= stbiw__sbm(a))
+#define stbiw__sbmaybegrow(a,n) (stbiw__sbneedgrow(a,(n)) ? stbiw__sbgrow(a,n) : 0)
+#define stbiw__sbgrow(a,n)  stbiw__sbgrowf((void **) &(a), (n), sizeof(*(a)))
 
-#define stbi__sbpush(a, v)      (stbi__sbmaybegrow(a,1), (a)[stbi__sbn(a)++] = (v))
-#define stbi__sbcount(a)        ((a) ? stbi__sbn(a) : 0)
-#define stbi__sbfree(a)         ((a) ? free(stbi__sbraw(a)),0 : 0)
+#define stbiw__sbpush(a, v)      (stbiw__sbmaybegrow(a,1), (a)[stbiw__sbn(a)++] = (v))
+#define stbiw__sbcount(a)        ((a) ? stbiw__sbn(a) : 0)
+#define stbiw__sbfree(a)         ((a) ? STBIW_FREE(stbiw__sbraw(a)),0 : 0)
 
-static void *stbi__sbgrowf(void **arr, int increment, int itemsize)
+static void *stbiw__sbgrowf(void **arr, int increment, int itemsize)
 {
-   int m = *arr ? 2*stbi__sbm(*arr)+increment : increment+1;
-   void *p = realloc(*arr ? stbi__sbraw(*arr) : 0, itemsize * m + sizeof(int)*2);
-   assert(p);
+   int m = *arr ? 2*stbiw__sbm(*arr)+increment : increment+1;
+   void *p = STBIW_REALLOC(*arr ? stbiw__sbraw(*arr) : 0, itemsize * m + sizeof(int)*2);
+   STBIW_ASSERT(p);
    if (p) {
       if (!*arr) ((int *) p)[1] = 0;
       *arr = (void *) ((int *) p + 2);
-      stbi__sbm(*arr) = m;
+      stbiw__sbm(*arr) = m;
    }
    return *arr;
 }
 
-static unsigned char *stbi__zlib_flushf(unsigned char *data, unsigned int *bitbuffer, int *bitcount)
+static unsigned char *stbiw__zlib_flushf(unsigned char *data, unsigned int *bitbuffer, int *bitcount)
 {
    while (*bitcount >= 8) {
-      stbi__sbpush(data, (unsigned char) *bitbuffer);
+      stbiw__sbpush(data, (unsigned char) *bitbuffer);
       *bitbuffer >>= 8;
       *bitcount -= 8;
    }
    return data;
 }
 
-static int stbi__zlib_bitrev(int code, int codebits)
+static int stbiw__zlib_bitrev(int code, int codebits)
 {
    int res=0;
    while (codebits--) {
@@ -157,7 +338,7 @@ static int stbi__zlib_bitrev(int code, int codebits)
    return res;
 }
 
-static unsigned int stbi__zlib_countm(unsigned char *a, unsigned char *b, int limit)
+static unsigned int stbiw__zlib_countm(unsigned char *a, unsigned char *b, int limit)
 {
    int i;
    for (i=0; i < limit && i < 258; ++i)
@@ -165,7 +346,7 @@ static unsigned int stbi__zlib_countm(unsigned char *a, unsigned char *b, int li
    return i;
 }
 
-static unsigned int stbi__zhash(unsigned char *data)
+static unsigned int stbiw__zhash(unsigned char *data)
 {
    stbiw_uint32 hash = data[0] + (data[1] << 8) + (data[2] << 16);
    hash ^= hash << 3;
@@ -177,19 +358,19 @@ static unsigned int stbi__zhash(unsigned char *data)
    return hash;
 }
 
-#define stbi__zlib_flush() (out = stbi__zlib_flushf(out, &bitbuf, &bitcount))
-#define stbi__zlib_add(code,codebits) \
-      (bitbuf |= (code) << bitcount, bitcount += (codebits), stbi__zlib_flush())
-#define stbi__zlib_huffa(b,c)  stbi__zlib_add(stbi__zlib_bitrev(b,c),c)
+#define stbiw__zlib_flush() (out = stbiw__zlib_flushf(out, &bitbuf, &bitcount))
+#define stbiw__zlib_add(code,codebits) \
+      (bitbuf |= (code) << bitcount, bitcount += (codebits), stbiw__zlib_flush())
+#define stbiw__zlib_huffa(b,c)  stbiw__zlib_add(stbiw__zlib_bitrev(b,c),c)
 // default huffman tables
-#define stbi__zlib_huff1(n)  stbi__zlib_huffa(0x30 + (n), 8)
-#define stbi__zlib_huff2(n)  stbi__zlib_huffa(0x190 + (n)-144, 9)
-#define stbi__zlib_huff3(n)  stbi__zlib_huffa(0 + (n)-256,7)
-#define stbi__zlib_huff4(n)  stbi__zlib_huffa(0xc0 + (n)-280,8)
-#define stbi__zlib_huff(n)  ((n) <= 143 ? stbi__zlib_huff1(n) : (n) <= 255 ? stbi__zlib_huff2(n) : (n) <= 279 ? stbi__zlib_huff3(n) : stbi__zlib_huff4(n))
-#define stbi__zlib_huffb(n) ((n) <= 143 ? stbi__zlib_huff1(n) : stbi__zlib_huff2(n))
+#define stbiw__zlib_huff1(n)  stbiw__zlib_huffa(0x30 + (n), 8)
+#define stbiw__zlib_huff2(n)  stbiw__zlib_huffa(0x190 + (n)-144, 9)
+#define stbiw__zlib_huff3(n)  stbiw__zlib_huffa(0 + (n)-256,7)
+#define stbiw__zlib_huff4(n)  stbiw__zlib_huffa(0xc0 + (n)-280,8)
+#define stbiw__zlib_huff(n)  ((n) <= 143 ? stbiw__zlib_huff1(n) : (n) <= 255 ? stbiw__zlib_huff2(n) : (n) <= 279 ? stbiw__zlib_huff3(n) : stbiw__zlib_huff4(n))
+#define stbiw__zlib_huffb(n) ((n) <= 143 ? stbiw__zlib_huff1(n) : stbiw__zlib_huff2(n))
 
-#define stbi__ZHASH   16384
+#define stbiw__ZHASH   16384
 
 unsigned char * stbi_zlib_compress(unsigned char *data, int data_len, int *out_len, int quality)
 {
@@ -200,45 +381,45 @@ unsigned char * stbi_zlib_compress(unsigned char *data, int data_len, int *out_l
    unsigned int bitbuf=0;
    int i,j, bitcount=0;
    unsigned char *out = NULL;
-   unsigned char **hash_table[stbi__ZHASH]; // 64KB on the stack!
+   unsigned char **hash_table[stbiw__ZHASH]; // 64KB on the stack!
    if (quality < 5) quality = 5;
 
-   stbi__sbpush(out, 0x78);   // DEFLATE 32K window
-   stbi__sbpush(out, 0x5e);   // FLEVEL = 1
-   stbi__zlib_add(1,1);  // BFINAL = 1
-   stbi__zlib_add(1,2);  // BTYPE = 1 -- fixed huffman
+   stbiw__sbpush(out, 0x78);   // DEFLATE 32K window
+   stbiw__sbpush(out, 0x5e);   // FLEVEL = 1
+   stbiw__zlib_add(1,1);  // BFINAL = 1
+   stbiw__zlib_add(1,2);  // BTYPE = 1 -- fixed huffman
 
-   for (i=0; i < stbi__ZHASH; ++i)
+   for (i=0; i < stbiw__ZHASH; ++i)
       hash_table[i] = NULL;
 
    i=0;
    while (i < data_len-3) {
-      // hash next 3 bytes of data to be compressed 
-      int h = stbi__zhash(data+i)&(stbi__ZHASH-1), best=3;
+      // hash next 3 bytes of data to be compressed
+      int h = stbiw__zhash(data+i)&(stbiw__ZHASH-1), best=3;
       unsigned char *bestloc = 0;
       unsigned char **hlist = hash_table[h];
-      int n = stbi__sbcount(hlist);
+      int n = stbiw__sbcount(hlist);
       for (j=0; j < n; ++j) {
          if (hlist[j]-data > i-32768) { // if entry lies within window
-            int d = stbi__zlib_countm(hlist[j], data+i, data_len-i);
+            int d = stbiw__zlib_countm(hlist[j], data+i, data_len-i);
             if (d >= best) best=d,bestloc=hlist[j];
          }
       }
       // when hash table entry is too long, delete half the entries
-      if (hash_table[h] && stbi__sbn(hash_table[h]) == 2*quality) {
-         memcpy(hash_table[h], hash_table[h]+quality, sizeof(hash_table[h][0])*quality);
-         stbi__sbn(hash_table[h]) = quality;
+      if (hash_table[h] && stbiw__sbn(hash_table[h]) == 2*quality) {
+         STBIW_MEMMOVE(hash_table[h], hash_table[h]+quality, sizeof(hash_table[h][0])*quality);
+         stbiw__sbn(hash_table[h]) = quality;
       }
-      stbi__sbpush(hash_table[h],data+i);
+      stbiw__sbpush(hash_table[h],data+i);
 
       if (bestloc) {
          // "lazy matching" - check match at *next* byte, and if it's better, do cur byte as literal
-         h = stbi__zhash(data+i+1)&(stbi__ZHASH-1);
+         h = stbiw__zhash(data+i+1)&(stbiw__ZHASH-1);
          hlist = hash_table[h];
-         n = stbi__sbcount(hlist);
+         n = stbiw__sbcount(hlist);
          for (j=0; j < n; ++j) {
             if (hlist[j]-data > i-32767) {
-               int e = stbi__zlib_countm(hlist[j], data+i+1, data_len-i-1);
+               int e = stbiw__zlib_countm(hlist[j], data+i+1, data_len-i-1);
                if (e > best) { // if next match is better, bail on current match
                   bestloc = NULL;
                   break;
@@ -248,30 +429,30 @@ unsigned char * stbi_zlib_compress(unsigned char *data, int data_len, int *out_l
       }
 
       if (bestloc) {
-         int d = data+i - bestloc; // distance back
-         assert(d <= 32767 && best <= 258);
+         int d = (int) (data+i - bestloc); // distance back
+         STBIW_ASSERT(d <= 32767 && best <= 258);
          for (j=0; best > lengthc[j+1]-1; ++j);
-         stbi__zlib_huff(j+257);
-         if (lengtheb[j]) stbi__zlib_add(best - lengthc[j], lengtheb[j]);
+         stbiw__zlib_huff(j+257);
+         if (lengtheb[j]) stbiw__zlib_add(best - lengthc[j], lengtheb[j]);
          for (j=0; d > distc[j+1]-1; ++j);
-         stbi__zlib_add(stbi__zlib_bitrev(j,5),5);
-         if (disteb[j]) stbi__zlib_add(d - distc[j], disteb[j]);
+         stbiw__zlib_add(stbiw__zlib_bitrev(j,5),5);
+         if (disteb[j]) stbiw__zlib_add(d - distc[j], disteb[j]);
          i += best;
       } else {
-         stbi__zlib_huffb(data[i]);
+         stbiw__zlib_huffb(data[i]);
          ++i;
       }
    }
    // write out final bytes
    for (;i < data_len; ++i)
-      stbi__zlib_huffb(data[i]);
-   stbi__zlib_huff(256); // end of block
+      stbiw__zlib_huffb(data[i]);
+   stbiw__zlib_huff(256); // end of block
    // pad with 0 bits to byte boundary
    while (bitcount)
-      stbi__zlib_add(0,1);
+      stbiw__zlib_add(0,1);
 
-   for (i=0; i < stbi__ZHASH; ++i)
-      (void) stbi__sbfree(hash_table[i]);
+   for (i=0; i < stbiw__ZHASH; ++i)
+      (void) stbiw__sbfree(hash_table[i]);
 
    {
       // compute adler32 on input
@@ -283,18 +464,18 @@ unsigned char * stbi_zlib_compress(unsigned char *data, int data_len, int *out_l
          j += blocklen;
          blocklen = 5552;
       }
-      stbi__sbpush(out, (unsigned char) (s2 >> 8));
-      stbi__sbpush(out, (unsigned char) s2);
-      stbi__sbpush(out, (unsigned char) (s1 >> 8));
-      stbi__sbpush(out, (unsigned char) s1);
+      stbiw__sbpush(out, (unsigned char) (s2 >> 8));
+      stbiw__sbpush(out, (unsigned char) s2);
+      stbiw__sbpush(out, (unsigned char) (s1 >> 8));
+      stbiw__sbpush(out, (unsigned char) s1);
    }
-   *out_len = stbi__sbn(out);
+   *out_len = stbiw__sbn(out);
    // make returned pointer freeable
-   memmove(stbi__sbraw(out), out, *out_len);
-   return (unsigned char *) stbi__sbraw(out);
+   STBIW_MEMMOVE(stbiw__sbraw(out), out, *out_len);
+   return (unsigned char *) stbiw__sbraw(out);
 }
 
-unsigned int stbi__crc32(unsigned char *buffer, int len)
+unsigned int stbiw__crc32(unsigned char *buffer, int len)
 {
    static unsigned int crc_table[256];
    unsigned int crc = ~0u;
@@ -308,17 +489,17 @@ unsigned int stbi__crc32(unsigned char *buffer, int len)
    return ~crc;
 }
 
-#define stbi__wpng4(o,a,b,c,d) ((o)[0]=(unsigned char)(a),(o)[1]=(unsigned char)(b),(o)[2]=(unsigned char)(c),(o)[3]=(unsigned char)(d),(o)+=4)
-#define stbi__wp32(data,v) stbi__wpng4(data, (v)>>24,(v)>>16,(v)>>8,(v));
-#define stbi__wptag(data,s) stbi__wpng4(data, s[0],s[1],s[2],s[3])
+#define stbiw__wpng4(o,a,b,c,d) ((o)[0]=(unsigned char)(a),(o)[1]=(unsigned char)(b),(o)[2]=(unsigned char)(c),(o)[3]=(unsigned char)(d),(o)+=4)
+#define stbiw__wp32(data,v) stbiw__wpng4(data, (v)>>24,(v)>>16,(v)>>8,(v));
+#define stbiw__wptag(data,s) stbiw__wpng4(data, s[0],s[1],s[2],s[3])
 
-static void stbi__wpcrc(unsigned char **data, int len)
+static void stbiw__wpcrc(unsigned char **data, int len)
 {
-   unsigned int crc = stbi__crc32(*data - len - 4, len+4);
-   stbi__wp32(*data, crc);
+   unsigned int crc = stbiw__crc32(*data - len - 4, len+4);
+   stbiw__wp32(*data, crc);
 }
 
-static unsigned char stbi__paeth(int a, int b, int c)
+static unsigned char stbiw__paeth(int a, int b, int c)
 {
    int p = a + b - c, pa = abs(p-a), pb = abs(p-b), pc = abs(p-c);
    if (pa <= pb && pa <= pc) return (unsigned char) a;
@@ -337,8 +518,8 @@ unsigned char *stbi_write_png_to_mem(unsigned char *pixels, int stride_bytes, in
    if (stride_bytes == 0)
       stride_bytes = x * n;
 
-   filt = (unsigned char *) malloc((x*n+1) * y); if (!filt) return 0;
-   line_buffer = (signed char *) malloc(x * n); if (!line_buffer) { free(filt); return 0; }
+   filt = (unsigned char *) STBIW_MALLOC((x*n+1) * y); if (!filt) return 0;
+   line_buffer = (signed char *) STBIW_MALLOC(x * n); if (!line_buffer) { STBIW_FREE(filt); return 0; }
    for (j=0; j < y; ++j) {
       static int mapping[] = { 0,1,2,3,4 };
       static int firstmap[] = { 0,1,0,5,6 };
@@ -354,7 +535,7 @@ unsigned char *stbi_write_png_to_mem(unsigned char *pixels, int stride_bytes, in
                   case 1: line_buffer[i] = z[i]; break;
                   case 2: line_buffer[i] = z[i] - z[i-stride_bytes]; break;
                   case 3: line_buffer[i] = z[i] - (z[i-stride_bytes]>>1); break;
-                  case 4: line_buffer[i] = (signed char) (z[i] - stbi__paeth(0,z[i-stride_bytes],0)); break;
+                  case 4: line_buffer[i] = (signed char) (z[i] - stbiw__paeth(0,z[i-stride_bytes],0)); break;
                   case 5: line_buffer[i] = z[i]; break;
                   case 6: line_buffer[i] = z[i]; break;
                }
@@ -364,9 +545,9 @@ unsigned char *stbi_write_png_to_mem(unsigned char *pixels, int stride_bytes, in
                   case 1: line_buffer[i] = z[i] - z[i-n]; break;
                   case 2: line_buffer[i] = z[i] - z[i-stride_bytes]; break;
                   case 3: line_buffer[i] = z[i] - ((z[i-n] + z[i-stride_bytes])>>1); break;
-                  case 4: line_buffer[i] = z[i] - stbi__paeth(z[i-n], z[i-stride_bytes], z[i-stride_bytes-n]); break;
+                  case 4: line_buffer[i] = z[i] - stbiw__paeth(z[i-n], z[i-stride_bytes], z[i-stride_bytes-n]); break;
                   case 5: line_buffer[i] = z[i] - (z[i-n]>>1); break;
-                  case 6: line_buffer[i] = z[i] - stbi__paeth(z[i-n], 0,0); break;
+                  case 6: line_buffer[i] = z[i] - stbiw__paeth(z[i-n], 0,0); break;
                }
             }
             if (p) break;
@@ -377,41 +558,43 @@ unsigned char *stbi_write_png_to_mem(unsigned char *pixels, int stride_bytes, in
       }
       // when we get here, best contains the filter type, and line_buffer contains the data
       filt[j*(x*n+1)] = (unsigned char) best;
-      memcpy(filt+j*(x*n+1)+1, line_buffer, x*n);
+      STBIW_MEMMOVE(filt+j*(x*n+1)+1, line_buffer, x*n);
    }
-   free(line_buffer);
+   STBIW_FREE(line_buffer);
    zlib = stbi_zlib_compress(filt, y*( x*n+1), &zlen, 8); // increase 8 to get smaller but use more memory
-   free(filt);
+   STBIW_FREE(filt);
    if (!zlib) return 0;
 
    // each tag requires 12 bytes of overhead
-   out = (unsigned char *) malloc(8 + 12+13 + 12+zlen + 12); 
+   out = (unsigned char *) STBIW_MALLOC(8 + 12+13 + 12+zlen + 12);
    if (!out) return 0;
    *out_len = 8 + 12+13 + 12+zlen + 12;
 
    o=out;
-   memcpy(o,sig,8); o+= 8;
-   stbi__wp32(o, 13); // header length
-   stbi__wptag(o, "IHDR");
-   stbi__wp32(o, x);
-   stbi__wp32(o, y);
+   STBIW_MEMMOVE(o,sig,8); o+= 8;
+   stbiw__wp32(o, 13); // header length
+   stbiw__wptag(o, "IHDR");
+   stbiw__wp32(o, x);
+   stbiw__wp32(o, y);
    *o++ = 8;
    *o++ = (unsigned char) ctype[n];
    *o++ = 0;
    *o++ = 0;
    *o++ = 0;
-   stbi__wpcrc(&o,13);
+   stbiw__wpcrc(&o,13);
 
-   stbi__wp32(o, zlen);
-   stbi__wptag(o, "IDAT");
-   memcpy(o, zlib, zlen); o += zlen; free(zlib);
-   stbi__wpcrc(&o, zlen);
+   stbiw__wp32(o, zlen);
+   stbiw__wptag(o, "IDAT");
+   STBIW_MEMMOVE(o, zlib, zlen);
+   o += zlen;
+   STBIW_FREE(zlib);
+   stbiw__wpcrc(&o, zlen);
 
-   stbi__wp32(o,0);
-   stbi__wptag(o, "IEND");
-   stbi__wpcrc(&o,0);
+   stbiw__wp32(o,0);
+   stbiw__wptag(o, "IEND");
+   stbiw__wpcrc(&o,0);
 
-   assert(o == out + *out_len);
+   STBIW_ASSERT(o == out + *out_len);
 
    return out;
 }
@@ -423,15 +606,28 @@ int stbi_write_png(char const *filename, int x, int y, int comp, const void *dat
    unsigned char *png = stbi_write_png_to_mem((unsigned char *) data, stride_bytes, x, y, comp, &len);
    if (!png) return 0;
    f = fopen(filename, "wb");
-   if (!f) { free(png); return 0; }
+   if (!f) { STBIW_FREE(png); return 0; }
    fwrite(png, 1, len, f);
    fclose(f);
-   free(png);
+   STBIW_FREE(png);
    return 1;
 }
+#endif // STB_IMAGE_WRITE_IMPLEMENTATION
 
 /* Revision history
-
+      0.98 (2015-04-08)
+             added STBIW_MALLOC, STBIW_ASSERT etc
+      0.97 (2015-01-18)
+             fixed HDR asserts, rewrote HDR rle logic
+      0.96 (2015-01-17)
+             add HDR output
+             fix monochrome BMP
+      0.95 (2014-08-17)
+		       add monochrome TGA output
+      0.94 (2014-05-31)
+             rename private functions to avoid conflicts with stb_image.h
+      0.93 (2014-05-27)
+             warning fixes
       0.92 (2010-08-01)
              casts to unsigned char to fix warnings
       0.91 (2010-07-17)

Source/ThirdParty/STB/stb_image_write.h

@@ -1,69 +1,97 @@
-/* stbiw-0.92 - public domain - http://nothings.org/stb/stb_image_write.h
-   writes out PNG/BMP/TGA images to C stdio - Sean Barrett 2010
-                            no warranty implied; use at your own risk
-
-
-Before including,
-
-    #define STB_IMAGE_WRITE_IMPLEMENTATION
-
-in the file that you want to have the implementation.
-
-
-ABOUT:
-
-   This header file is a library for writing images to C stdio. It could be
-   adapted to write to memory or a general streaming interface; let me know.
-
-   The PNG output is not optimal; it is 20-50% larger than the file
-   written by a decent optimizing implementation. This library is designed
-   for source code compactness and simplicitly, not optimal image file size
-   or run-time performance.
-
-USAGE:
-
-   There are three functions, one for each image file format:
-
-     int stbi_write_png(char const *filename, int w, int h, int comp, const void *data, int stride_in_bytes);
-     int stbi_write_bmp(char const *filename, int w, int h, int comp, const void *data);
-     int stbi_write_tga(char const *filename, int w, int h, int comp, const void *data);
-
-   Each function returns 0 on failure and non-0 on success.
-   
-   The functions create an image file defined by the parameters. The image
-   is a rectangle of pixels stored from left-to-right, top-to-bottom.
-   Each pixel contains 'comp' channels of data stored interleaved with 8-bits
-   per channel, in the following order: 1=Y, 2=YA, 3=RGB, 4=RGBA. (Y is
-   monochrome color.) The rectangle is 'w' pixels wide and 'h' pixels tall.
-   The *data pointer points to the first byte of the top-left-most pixel.
-   For PNG, "stride_in_bytes" is the distance in bytes from the first byte of
-   a row of pixels to the first byte of the next row of pixels.
-
-   PNG creates output files with the same number of components as the input.
-   The BMP and TGA formats expand Y to RGB in the file format. BMP does not
-   output alpha.
-   
-   PNG supports writing rectangles of data even when the bytes storing rows of
-   data are not consecutive in memory (e.g. sub-rectangles of a larger image),
-   by supplying the stride between the beginning of adjacent rows. The other
-   formats do not. (Thus you cannot write a native-format BMP through the BMP
-   writer, both because it is in BGR order and because it may have padding
-   at the end of the line.)
-*/
-
-#ifndef INCLUDE_STB_IMAGE_WRITE_H
-#define INCLUDE_STB_IMAGE_WRITE_H
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-extern int stbi_write_png(char const *filename, int w, int h, int comp, const void *data, int stride_in_bytes);
-extern int stbi_write_bmp(char const *filename, int w, int h, int comp, const void *data);
-extern int stbi_write_tga(char const *filename, int w, int h, int comp, const void *data);
-
-#ifdef __cplusplus
-}
-#endif
-
-#endif//INCLUDE_STB_IMAGE_WRITE_H
+/* stb_image_write - v0.98 - public domain - http://nothings.org/stb/stb_image_write.h
+   writes out PNG/BMP/TGA images to C stdio - Sean Barrett 2010
+                            no warranty implied; use at your own risk
+
+
+   Before #including,
+
+       #define STB_IMAGE_WRITE_IMPLEMENTATION
+
+   in the file that you want to have the implementation.
+
+   Will probably not work correctly with strict-aliasing optimizations.
+
+ABOUT:
+
+   This header file is a library for writing images to C stdio. It could be
+   adapted to write to memory or a general streaming interface; let me know.
+
+   The PNG output is not optimal; it is 20-50% larger than the file
+   written by a decent optimizing implementation. This library is designed
+   for source code compactness and simplicitly, not optimal image file size
+   or run-time performance.
+
+BUILDING:
+
+   You can #define STBIW_ASSERT(x) before the #include to avoid using assert.h.
+   You can #define STBIW_MALLOC(), STBIW_REALLOC(), and STBIW_FREE() to replace
+   malloc,realloc,free.
+   You can define STBIW_MEMMOVE() to replace memmove()
+
+USAGE:
+
+   There are four functions, one for each image file format:
+
+     int stbi_write_png(char const *filename, int w, int h, int comp, const void *data, int stride_in_bytes);
+     int stbi_write_bmp(char const *filename, int w, int h, int comp, const void *data);
+     int stbi_write_tga(char const *filename, int w, int h, int comp, const void *data);
+     int stbi_write_hdr(char const *filename, int w, int h, int comp, const void *data);
+
+   Each function returns 0 on failure and non-0 on success.
+
+   The functions create an image file defined by the parameters. The image
+   is a rectangle of pixels stored from left-to-right, top-to-bottom.
+   Each pixel contains 'comp' channels of data stored interleaved with 8-bits
+   per channel, in the following order: 1=Y, 2=YA, 3=RGB, 4=RGBA. (Y is
+   monochrome color.) The rectangle is 'w' pixels wide and 'h' pixels tall.
+   The *data pointer points to the first byte of the top-left-most pixel.
+   For PNG, "stride_in_bytes" is the distance in bytes from the first byte of
+   a row of pixels to the first byte of the next row of pixels.
+
+   PNG creates output files with the same number of components as the input.
+   The BMP format expands Y to RGB in the file format and does not
+   output alpha.
+
+   PNG supports writing rectangles of data even when the bytes storing rows of
+   data are not consecutive in memory (e.g. sub-rectangles of a larger image),
+   by supplying the stride between the beginning of adjacent rows. The other
+   formats do not. (Thus you cannot write a native-format BMP through the BMP
+   writer, both because it is in BGR order and because it may have padding
+   at the end of the line.)
+
+   HDR expects linear float data. Since the format is always 32-bit rgb(e)
+   data, alpha (if provided) is discarded, and for monochrome data it is
+   replicated across all three channels.
+
+CREDITS:
+
+   PNG/BMP/TGA
+      Sean Barrett
+   HDR
+      Baldur Karlsson
+   TGA monochrome:
+      Jean-Sebastien Guay
+   misc enhancements:
+      Tim Kelsey
+   bugfixes:
+      github:Chribba
+*/
+
+#ifndef INCLUDE_STB_IMAGE_WRITE_H
+#define INCLUDE_STB_IMAGE_WRITE_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+extern int stbi_write_png(char const *filename, int w, int h, int comp, const void  *data, int stride_in_bytes);
+extern int stbi_write_bmp(char const *filename, int w, int h, int comp, const void  *data);
+extern int stbi_write_tga(char const *filename, int w, int h, int comp, const void  *data);
+extern int stbi_write_hdr(char const *filename, int w, int h, int comp, const float *data);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif//INCLUDE_STB_IMAGE_WRITE_H
+

Source/ThirdParty/STB/stb_vorbis.c

@@ -1,5013 +1,5070 @@
-#include "stb_vorbis.h"
-
-// 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
-
-// STB_VORBIS_CODEBOOK_SHORTS
-//     The vorbis file format encodes VQ codebook floats as ax+b where a and
-//     b are floating point per-codebook constants, and x is a 16-bit int.
-//     Normally, stb_vorbis decodes them to floats rather than leaving them
-//     as 16-bit ints and computing ax+b while decoding. This is a speed/space
-//     tradeoff; you can save space by defining this flag.
-#ifndef STB_VORBIS_CODEBOOK_SHORTS
-#define STB_VORBIS_CODEBOOK_FLOATS
-#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>
-#if !(defined(__APPLE__) || defined(MACOSX) || defined(macintosh) || defined(Macintosh))
-#include <malloc.h>
-#endif
-#else
-#define NULL 0
-#endif
-
-#ifndef _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
-
-
-#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
-
-#ifdef STB_VORBIS_CODEBOOK_FLOATS
-typedef float codetype;
-#else
-typedef uint16 codetype;
-#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 after_previous_page_start;
-   uint32 first_decoded_sample;
-   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;
-
-  // 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;
-
-   uint32 first_audio_page_offset;
-
-   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;
-};
-
-extern int my_prof(int slot);
-//#define stb_prof my_prof
-
-#ifndef stb_prof
-#define stb_prof(x)  0
-#endif
-
-#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))
-#ifdef dealloca
-#define temp_free(f,p)                  (f->alloc.alloc_buffer ? 0 : dealloca(size))
-#else
-#define temp_free(f,p)                  0
-#endif
-#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+3) & ~3;
-   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 not a stack
-   free(p);
-}
-
-static void *setup_temp_malloc(vorb *f, int sz)
-{
-   sz = (sz+3) & ~3;
-   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, size_t sz)
-{
-   if (f->alloc.alloc_buffer) {
-      f->temp_offset += (sz+3)&~3;
-      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=i<<24, j=0; j < 8; ++j)
-         s = (s << 1) ^ (s >= (1<<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 };
-
-   // 2 compares if n < 16, 3 compares otherwise (4 if signed or n > 1<<29)
-   if (n < (1U << 14))
-        if (n < (1U <<  4))        return     0 + log2_4[n      ];
-        else if (n < (1U <<  9))      return  5 + log2_4[n >>  5];
-             else                     return 10 + log2_4[n >> 10];
-   else if (n < (1U << 24))
-             if (n < (1U << 19))      return 15 + log2_4[n >> 15];
-             else                     return 20 + log2_4[n >> 20];
-        else if (n < (1U << 29))      return 25 + log2_4[n >> 25];
-             else if (n < (1U << 31)) return 30 + log2_4[n >> 30];
-                  else                return 0; // signed n returns 0
-}
-
-#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, 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; }
-   // 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] = 1 << (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;
-      // 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) { assert(0); return FALSE; }
-      res = available[z];
-      available[z] = 0;
-      add_entry(c, bit_reverse(res), i, m++, len[i], values);
-      // propogate 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];
-         }
-      }
-   }
-}
-
-static int 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
-   assert(pow((float) r+1, dim) > entries);
-   assert((int) floor(pow((float) r, dim)) <= entries); // (int),floor() as above
-   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,y;
-} Point;
-
-int point_compare(const void *p, const void *q)
-{
-   Point *a = (Point *) p;
-   Point *b = (Point *) 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 += 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,i;
-   // 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 != ~0 || loc1 != ~0) {
-      // 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;
-      ProbedPage p;
-      len = 0;
-      for (i=0; i < f->segment_count; ++i)
-         len += f->segments[i];
-      len += 27 + f->segment_count;
-      p.page_start = f->first_audio_page_offset;
-      p.page_end = p.page_start + len;
-      p.after_previous_page_start = p.page_start;
-      p.first_decoded_sample = 0;
-      p.last_decoded_sample = loc0;
-      f->p_first = p;
-   }
-   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)
-      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 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;
-      }
-   }
-   if (f->valid_bits < 0) return 0;
-   z = f->acc & ((1 << n)-1);
-   f->acc >>= n;
-   f->valid_bits -= n;
-   return z;
-}
-
-static int32 get_bits_signed(vorb *f, int n)
-{
-   uint32 z = get_bits(f, n);
-   if (z & (1 << (n-1)))
-      z += ~((1 << n) - 1);
-   return (int32) 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 += 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);
-
-   assert(c->sorted_codewords || c->codewords);
-   // 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;
-}
-
-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);
-}
-
-#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
-
-#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
-#ifndef STB_VORBIS_CODEBOOK_FLOATS
-   #define CODEBOOK_ELEMENT(c,off)          (c->multiplicands[off] * c->delta_value + c->minimum_value)
-   #define CODEBOOK_ELEMENT_FAST(c,off)     (c->multiplicands[off] * c->delta_value)
-   #define CODEBOOK_ELEMENT_BASE(c)         (c->minimum_value)
-#else
-   #define CODEBOOK_ELEMENT(c,off)          (c->multiplicands[off])
-   #define CODEBOOK_ELEMENT_FAST(c,off)     (c->multiplicands[off])
-   #define CODEBOOK_ELEMENT_BASE(c)         (0)
-#endif
-
-static int codebook_decode_start(vorb *f, Codebook *c, int len)
-{
-   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,len);
-   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,len);
-   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;
-            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;
-               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;
-               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;
-}
-
-#ifndef STB_VORBIS_DIVIDES_IN_CODEBOOK
-static int codebook_decode_deinterleave_repeat_2(vorb *f, Codebook *c, float **outputs, 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);
-
-      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*2 + effective > len * 2) {
-         effective = len*2 - (p_inter*2 - c_inter);
-      }
-
-      {
-         z *= c->dimensions;
-         stb_prof(11);
-         if (c->sequence_p) {
-            // haven't optimized this case because I don't have any examples
-            for (i=0; i < effective; ++i) {
-               float val = CODEBOOK_ELEMENT_FAST(c,z+i) + last;
-               outputs[c_inter][p_inter] += val;
-               if (++c_inter == 2) { c_inter = 0; ++p_inter; }
-               last = val;
-            }
-         } else {
-            i=0;
-            if (c_inter == 1) {
-               float val = CODEBOOK_ELEMENT_FAST(c,z+i) + last;
-               outputs[c_inter][p_inter] += val;
-               c_inter = 0; ++p_inter;
-               ++i;
-            }
-            {
-               float *z0 = outputs[0];
-               float *z1 = outputs[1];
-               for (; i+1 < effective;) {
-                  z0[p_inter] += CODEBOOK_ELEMENT_FAST(c,z+i) + last;
-                  z1[p_inter] += CODEBOOK_ELEMENT_FAST(c,z+i+1) + last;
-                  ++p_inter;
-                  i += 2;
-               }
-            }
-            if (i < effective) {
-               float val = CODEBOOK_ELEMENT_FAST(c,z+i) + last;
-               outputs[c_inter][p_inter] += val;
-               if (++c_inter == 2) { c_inter = 0; ++p_inter; }
-            }
-         }
-      }
-
-      total_decode -= effective;
-   }
-   *c_inter_p = c_inter;
-   *p_inter_p = p_inter;
-   return TRUE;
-}
-#endif
-
-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;
-   LINE_OP(output[x], inverse_db_table[y]);
-   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]);
-   }
-}
-
-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;
-}
-
-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;
-   int n_read = r->end - 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
-
-   stb_prof(2);
-   for (i=0; i < ch; ++i)
-      if (!do_not_decode[i])
-         memset(residue_buffers[i], 0, sizeof(float) * n);
-
-   if (rtype == 2 && ch != 1) {
-      int len = ch * n;
-      for (j=0; j < ch; ++j)
-         if (!do_not_decode[j])
-            break;
-      if (j == ch)
-         goto done;
-
-      stb_prof(3);
-      for (pass=0; pass < 8; ++pass) {
-         int pcount = 0, class_set = 0;
-         if (ch == 2) {
-            stb_prof(13);
-            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
-               }
-               stb_prof(5);
-               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;
-                     stb_prof(20);  // accounts for X time
-                     #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_2(f, book, residue_buffers, &c_inter, &p_inter, n, r->part_size))
-                        goto done;
-                     #endif
-                     stb_prof(7);
-                  } else {
-                     z += r->part_size;
-                     c_inter = z & 1;
-                     p_inter = z >> 1;
-                  }
-               }
-               stb_prof(8);
-               #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE
-               ++class_set;
-               #endif
-            }
-         } else if (ch == 1) {
-            while (pcount < part_read) {
-               int z = r->begin + pcount*r->part_size;
-               int c_inter = 0, p_inter = z;
-               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;
-                     stb_prof(22);
-                     if (!codebook_decode_deinterleave_repeat(f, book, residue_buffers, ch, &c_inter, &p_inter, n, r->part_size))
-                        goto done;
-                     stb_prof(3);
-                  } else {
-                     z += r->part_size;
-                     c_inter = 0;
-                     p_inter = z;
-                  }
-               }
-               #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE
-               ++class_set;
-               #endif
-            }
-         } else {
-            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;
-                     stb_prof(22);
-                     if (!codebook_decode_deinterleave_repeat(f, book, residue_buffers, ch, &c_inter, &p_inter, n, r->part_size))
-                        goto done;
-                     stb_prof(3);
-                  } else {
-                     z += r->part_size;
-                     c_inter = z % ch;
-                     p_inter = z / ch;
-                  }
-               }
-               #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE
-               ++class_set;
-               #endif
-            }
-         }
-      }
-      goto done;
-   }
-   stb_prof(9);
-
-   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:
-   stb_prof(0);
-   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);
-}
-#else
-// 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];
-         //acc += x[j] * cos(M_PI / n * (i + 0.5) * (j + 0.5));
-      buffer[i] = acc;
-   }
-   // Urho3D: fixed freeing of stack allocated array
-   //free(x);
-}
-
-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 k_off = -8;
-   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;
-
-      k00   = z[-0] - z[-8];
-      k11   = z[-1] - z[-9];
-      z[-0] = z[-0] + z[-8];
-      z[-1] = z[-1] + z[-9];
-      z[-8] =  k00;
-      z[-9] =  k11 ;
-
-      k00    = z[ -2] - z[-10];
-      k11    = z[ -3] - z[-11];
-      z[ -2] = z[ -2] + z[-10];
-      z[ -3] = z[ -3] + z[-11];
-      z[-10] = (k00+k11) * A2;
-      z[-11] = (k11-k00) * A2;
-
-      k00    = z[-12] - z[ -4];  // reverse to avoid a unary negation
-      k11    = z[ -5] - z[-13];
-      z[ -4] = z[ -4] + z[-12];
-      z[ -5] = z[ -5] + z[-13];
-      z[-12] = k11;
-      z[-13] = k00;
-
-      k00    = z[-14] - z[ -6];  // reverse to avoid a unary negation
-      k11    = z[ -7] - z[-15];
-      z[ -6] = z[ -6] + z[-14];
-      z[ -7] = z[ -7] + z[-15];
-      z[-14] = (k00+k11) * A2;
-      z[-15] = (k00-k11) * 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 n3_4 = n - n4, 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 propogates 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_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];
-   assert(0);
-   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
-         if (finalY[j] >= 0)
-         #else
-         if (step2_flag[j])
-         #endif
-         {
-            int hy = finalY[j] * g->floor1_multiplier;
-            int hx = g->Xlist[j];
-            draw_line(target, lx,ly, hx,hy, n2);
-            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]);
-   }
-   return TRUE;
-}
-
-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];
-   int window_center;
-
-// WINDOWING
-
-   n = f->blocksize[m->blockflag];
-   window_center = n >> 1;
-
-   map = &f->mapping[m->mapping];
-
-// FLOORS
-   n2 = n >> 1;
-
-   stb_prof(1);
-   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;
-         }
-         // So we just defer everything else to later
-
-         // at this point we've decoded the floor into buffer
-      }
-   }
-   stb_prof(0);
-   // at this point we've decoded all floors
-
-   if (f->alloc.alloc_buffer)
-      assert(f->alloc.alloc_buffer_length_in_bytes == f->temp_offset);
-
-   // re-enable coupled channels if necessary
-   memcpy(really_zero_channel, zero_channel, sizeof(really_zero_channel[0]) * f->channels);
-   for (i=0; i < map->coupling_steps; ++i)
-      if (!zero_channel[map->chan[i].magnitude] || !zero_channel[map->chan[i].angle]) {
-         zero_channel[map->chan[i].magnitude] = zero_channel[map->chan[i].angle] = FALSE;
-      }
-
-// RESIDUE DECODE
-   for (i=0; i < map->submaps; ++i) {
-      float *residue_buffers[STB_VORBIS_MAX_CHANNELS];
-      int r,t;
-      uint8 do_not_decode[256];
-      int ch = 0;
-      for (j=0; j < f->channels; ++j) {
-         if (map->chan[j].mux == i) {
-            if (zero_channel[j]) {
-               do_not_decode[ch] = TRUE;
-               residue_buffers[ch] = NULL;
-            } else {
-               do_not_decode[ch] = FALSE;
-               residue_buffers[ch] = f->channel_buffers[j];
-            }
-            ++ch;
-         }
-      }
-      r = map->submap_residue[i];
-      t = f->residue_types[r];
-      decode_residue(f, residue_buffers, ch, n2, r, do_not_decode);
-   }
-
-   if (f->alloc.alloc_buffer)
-      assert(f->alloc.alloc_buffer_length_in_bytes == f->temp_offset);
-
-// INVERSE COUPLING
-   stb_prof(14);
-   for (i = map->coupling_steps-1; i >= 0; --i) {
-      int n2 = n >> 1;
-      float *m = f->channel_buffers[map->chan[i].magnitude];
-      float *a = f->channel_buffers[map->chan[i].angle    ];
-      for (j=0; j < n2; ++j) {
-         float a2,m2;
-         if (m[j] > 0)
-            if (a[j] > 0)
-               m2 = m[j], a2 = m[j] - a[j];
-            else
-               a2 = m[j], m2 = m[j] + a[j];
-         else
-            if (a[j] > 0)
-               m2 = m[j], a2 = m[j] + a[j];
-            else
-               a2 = m[j], m2 = m[j] - a[j];
-         m[j] = m2;
-         a[j] = a2;
-      }
-   }
-
-   // finish decoding the floors
-#ifndef STB_VORBIS_NO_DEFER_FLOOR
-   stb_prof(15);
-   for (i=0; i < f->channels; ++i) {
-      if (really_zero_channel[i]) {
-         memset(f->channel_buffers[i], 0, sizeof(*f->channel_buffers[i]) * n2);
-      } else {
-         do_floor(f, map, i, n, f->channel_buffers[i], f->finalY[i], NULL);
-      }
-   }
-#else
-   for (i=0; i < f->channels; ++i) {
-      if (really_zero_channel[i]) {
-         memset(f->channel_buffers[i], 0, sizeof(*f->channel_buffers[i]) * n2);
-      } else {
-         for (j=0; j < n2; ++j)
-            f->channel_buffers[i][j] *= f->floor_buffers[i][j];
-      }
-   }
-#endif
-
-// INVERSE MDCT
-   stb_prof(16);
-   for (i=0; i < f->channels; ++i)
-      inverse_mdct(f->channel_buffers[i], n, f, m->blockflag);
-   stb_prof(0);
-
-   // this shouldn't be necessary, unless we exited on an error
-   // and want to flush to get to the next packet
-   flush_packet(f);
-
-   if (f->first_decode) {
-      // assume we start so first non-discarded sample is sample 0
-      // this isn't to spec, but spec would require us to read ahead
-      // and decode the size of all current frames--could be done,
-      // but presumably it's not a commonly used feature
-      f->current_loc = -n2; // start of first frame is positioned for discard
-      // we might have to discard samples "from" the next frame too,
-      // if we're lapping a large block then a small at the start?
-      f->discard_samples_deferred = n - right_end;
-      f->current_loc_valid = TRUE;
-      f->first_decode = FALSE;
-   } else if (f->discard_samples_deferred) {
-      left_start += f->discard_samples_deferred;
-      *p_left = left_start;
-      f->discard_samples_deferred = 0;
-   } else if (f->previous_length == 0 && f->current_loc_valid) {
-      // we're recovering from a seek... that means we're going to discard
-      // the samples from this packet even though we know our position from
-      // the last page header, so we need to update the position based on
-      // the discarded samples here
-      // but wait, the code below is going to add this in itself even
-      // on a discard, so we don't need to do it here...
-   }
-
-   // check if we have ogg information about the sample # for this packet
-   if (f->last_seg_which == f->end_seg_with_known_loc) {
-      // if we have a valid current loc, and this is final:
-      if (f->current_loc_valid && (f->page_flag & PAGEFLAG_last_page)) {
-         uint32 current_end = f->known_loc_for_packet - (n-right_end);
-         // then let's infer the size of the (probably) short final frame
-         if (current_end < f->current_loc + right_end) {
-            if (current_end < f->current_loc) {
-               // negative truncation, that's impossible!
-               *len = 0;
-            } else {
-               *len = current_end - f->current_loc;
-            }
-            *len += left_start;
-            f->current_loc += *len;
-            return TRUE;
-         }
-      }
-      // otherwise, just set our sample loc
-      // guess that the ogg granule pos refers to the _middle_ of the
-      // last frame?
-      // set f->current_loc to the position of left_start
-      f->current_loc = f->known_loc_for_packet - (n2-left_start);
-      f->current_loc_valid = TRUE;
-   }
-   if (f->current_loc_valid)
-      f->current_loc += (right_start - left_start);
-
-   if (f->alloc.alloc_buffer)
-      assert(f->alloc.alloc_buffer_length_in_bytes == f->temp_offset);
-   *len = right_end;  // ignore samples after the window goes to 0
-   return TRUE;
-}
-
-static int vorbis_decode_packet(vorb *f, int *len, int *p_left, int *p_right)
-{
-   int mode, left_end, right_end;
-   if (!vorbis_decode_initial(f, p_left, &left_end, p_right, &right_end, &mode)) return 0;
-   return vorbis_decode_packet_rest(f, len, f->mode_config + mode, *p_left, left_end, *p_right, right_end, p_left);
-}
-
-static int vorbis_finish_frame(stb_vorbis *f, int len, int left, int right)
-{
-   int prev,i,j;
-   // we use right&left (the start of the right- and left-window sin()-regions)
-   // to determine how much to return, rather than inferring from the rules
-   // (same result, clearer code); 'left' indicates where our sin() window
-   // starts, therefore where the previous window's right edge starts, and
-   // therefore where to start mixing from the previous buffer. 'right'
-   // indicates where our sin() ending-window starts, therefore that's where
-   // we start saving, and where our returned-data ends.
-
-   // mixin from previous window
-   if (f->previous_length) {
-      int i,j, n = f->previous_length;
-      float *w = get_window(f, n);
-      for (i=0; i < f->channels; ++i) {
-         for (j=0; j < n; ++j)
-            f->channel_buffers[i][left+j] =
-               f->channel_buffers[i][left+j]*w[    j] +
-               f->previous_window[i][     j]*w[n-1-j];
-      }
-   }
-
-   prev = f->previous_length;
-
-   // last half of this data becomes previous window
-   f->previous_length = len - right;
-
-   // @OPTIMIZE: could avoid this copy by double-buffering the
-   // output (flipping previous_window with channel_buffers), but
-   // then previous_window would have to be 2x as large, and
-   // channel_buffers couldn't be temp mem (although they're NOT
-   // currently temp mem, they could be (unless we want to level
-   // performance by spreading out the computation))
-   for (i=0; i < f->channels; ++i)
-      for (j=0; right+j < len; ++j)
-         f->previous_window[i][j] = f->channel_buffers[i][right+j];
-
-   if (!prev)
-      // there was no previous packet, so this data isn't valid...
-      // this isn't entirely true, only the would-have-overlapped data
-      // isn't valid, but this seems to be what the spec requires
-      return 0;
-
-   // truncate a short frame
-   if (len < right) right = len;
-
-   f->samples_output += right-left;
-
-   return right - left;
-}
-
-static void vorbis_pump_first_frame(stb_vorbis *f)
-{
-   int len, right, left;
-   if (vorbis_decode_packet(f, &len, &left, &right))
-      vorbis_finish_frame(f, len, left, right);
-}
-
-#ifndef STB_VORBIS_NO_PUSHDATA_API
-static int is_whole_packet_present(stb_vorbis *f, int end_page)
-{
-   // make sure that we have the packet available before continuing...
-   // this requires a full ogg parse, but we know we can fetch from f->stream
-
-   // instead of coding this out explicitly, we could save the current read state,
-   // read the next packet with get8() until end-of-packet, check f->eof, then
-   // reset the state? but that would be slower, esp. since we'd have over 256 bytes
-   // of state to restore (primarily the page segment table)
-
-   int s = f->next_seg, first = TRUE;
-   uint8 *p = f->stream;
-
-   if (s != -1) { // if we're not starting the packet with a 'continue on next page' flag
-      for (; s < f->segment_count; ++s) {
-         p += f->segments[s];
-         if (f->segments[s] < 255)               // stop at first short segment
-            break;
-      }
-      // either this continues, or it ends it...
-      if (end_page)
-         if (s < f->segment_count-1)             return error(f, VORBIS_invalid_stream);
-      if (s == f->segment_count)
-         s = -1; // set 'crosses page' flag
-      if (p > f->stream_end)                     return error(f, VORBIS_need_more_data);
-      first = FALSE;
-   }
-   for (; s == -1;) {
-      uint8 *q; 
-      int n;
-
-      // check that we have the page header ready
-      if (p + 26 >= f->stream_end)               return error(f, VORBIS_need_more_data);
-      // validate the page
-      if (memcmp(p, ogg_page_header, 4))         return error(f, VORBIS_invalid_stream);
-      if (p[4] != 0)                             return error(f, VORBIS_invalid_stream);
-      if (first) { // the first segment must NOT have 'continued_packet', later ones MUST
-         if (f->previous_length)
-            if ((p[5] & PAGEFLAG_continued_packet))  return error(f, VORBIS_invalid_stream);
-         // if no previous length, we're resynching, so we can come in on a continued-packet,
-         // which we'll just drop
-      } else {
-         if (!(p[5] & PAGEFLAG_continued_packet)) return error(f, VORBIS_invalid_stream);
-      }
-      n = p[26]; // segment counts
-      q = p+27;  // q points to segment table
-      p = q + n; // advance past header
-      // make sure we've read the segment table
-      if (p > f->stream_end)                     return error(f, VORBIS_need_more_data);
-      for (s=0; s < n; ++s) {
-         p += q[s];
-         if (q[s] < 255)
-            break;
-      }
-      if (end_page)
-         if (s < n-1)                            return error(f, VORBIS_invalid_stream);
-      if (s == f->segment_count)
-         s = -1; // set 'crosses page' flag
-      if (p > f->stream_end)                     return error(f, VORBIS_need_more_data);
-      first = FALSE;
-   }
-   return TRUE;
-}
-#endif // !STB_VORBIS_NO_PUSHDATA_API
-
-static int start_decoder(vorb *f)
-{
-   uint8 header[6], x,y;
-   int len,i,j,k, max_submaps = 0;
-   int longest_floorlist=0;
-
-   // first page, first packet
-
-   if (!start_page(f))                              return FALSE;
-   // validate page flag
-   if (!(f->page_flag & PAGEFLAG_first_page))       return error(f, VORBIS_invalid_first_page);
-   if (f->page_flag & PAGEFLAG_last_page)           return error(f, VORBIS_invalid_first_page);
-   if (f->page_flag & PAGEFLAG_continued_packet)    return error(f, VORBIS_invalid_first_page);
-   // check for expected packet length
-   if (f->segment_count != 1)                       return error(f, VORBIS_invalid_first_page);
-   if (f->segments[0] != 30)                        return error(f, VORBIS_invalid_first_page);
-   // read packet
-   // check packet header
-   if (get8(f) != VORBIS_packet_id)                 return error(f, VORBIS_invalid_first_page);
-   if (!getn(f, header, 6))                         return error(f, VORBIS_unexpected_eof);
-   if (!vorbis_validate(header))                    return error(f, VORBIS_invalid_first_page);
-   // vorbis_version
-   if (get32(f) != 0)                               return error(f, VORBIS_invalid_first_page);
-   f->channels = get8(f); if (!f->channels)         return error(f, VORBIS_invalid_first_page);
-   if (f->channels > STB_VORBIS_MAX_CHANNELS)       return error(f, VORBIS_too_many_channels);
-   f->sample_rate = get32(f); if (!f->sample_rate)  return error(f, VORBIS_invalid_first_page);
-   get32(f); // bitrate_maximum
-   get32(f); // bitrate_nominal
-   get32(f); // bitrate_minimum
-   x = get8(f);
-   { int log0,log1;
-   log0 = x & 15;
-   log1 = x >> 4;
-   f->blocksize_0 = 1 << log0;
-   f->blocksize_1 = 1 << log1;
-   if (log0 < 6 || log0 > 13)                       return error(f, VORBIS_invalid_setup);
-   if (log1 < 6 || log1 > 13)                       return error(f, VORBIS_invalid_setup);
-   if (log0 > log1)                                 return error(f, VORBIS_invalid_setup);
-   }
-
-   // framing_flag
-   x = get8(f);
-   if (!(x & 1))                                    return error(f, VORBIS_invalid_first_page);
-
-   // second packet!
-   if (!start_page(f))                              return FALSE;
-
-   if (!start_packet(f))                            return FALSE;
-   do {
-      len = next_segment(f);
-      skip(f, len);
-      f->bytes_in_seg = 0;
-   } while (len);
-
-   // third packet!
-   if (!start_packet(f))                            return FALSE;
-
-   #ifndef STB_VORBIS_NO_PUSHDATA_API
-   if (IS_PUSH_MODE(f)) {
-      if (!is_whole_packet_present(f, TRUE)) {
-         // convert error in ogg header to write type
-         if (f->error == VORBIS_invalid_stream)
-            f->error = VORBIS_invalid_setup;
-         return FALSE;
-      }
-   }
-   #endif
-
-   crc32_init(); // always init it, to avoid multithread race conditions
-
-   if (get8_packet(f) != VORBIS_packet_setup)       return error(f, VORBIS_invalid_setup);
-   for (i=0; i < 6; ++i) header[i] = get8_packet(f);
-   if (!vorbis_validate(header))                    return error(f, VORBIS_invalid_setup);
-
-   // codebooks
-
-   f->codebook_count = get_bits(f,8) + 1;
-   f->codebooks = (Codebook *) setup_malloc(f, sizeof(*f->codebooks) * f->codebook_count);
-   if (f->codebooks == NULL)                        return error(f, VORBIS_outofmem);
-   memset(f->codebooks, 0, sizeof(*f->codebooks) * f->codebook_count);
-   for (i=0; i < f->codebook_count; ++i) {
-      uint32 *values;
-      int ordered, sorted_count;
-      int total=0;
-      uint8 *lengths;
-      Codebook *c = f->codebooks+i;
-      x = get_bits(f, 8); if (x != 0x42)            return error(f, VORBIS_invalid_setup);
-      x = get_bits(f, 8); if (x != 0x43)            return error(f, VORBIS_invalid_setup);
-      x = get_bits(f, 8); if (x != 0x56)            return error(f, VORBIS_invalid_setup);
-      x = get_bits(f, 8);
-      c->dimensions = (get_bits(f, 8)<<8) + x;
-      x = get_bits(f, 8);
-      y = get_bits(f, 8);
-      c->entries = (get_bits(f, 8)<<16) + (y<<8) + x;
-      ordered = get_bits(f,1);
-      c->sparse = ordered ? 0 : get_bits(f,1);
-
-      if (c->sparse)
-         lengths = (uint8 *) setup_temp_malloc(f, c->entries);
-      else
-         lengths = c->codeword_lengths = (uint8 *) setup_malloc(f, c->entries);
-
-      if (!lengths) return error(f, VORBIS_outofmem);
-
-      if (ordered) {
-         int current_entry = 0;
-         int current_length = get_bits(f,5) + 1;
-         while (current_entry < c->entries) {
-            int limit = c->entries - current_entry;
-            int n = get_bits(f, ilog(limit));
-            if (current_entry + n > (int) c->entries) { return error(f, VORBIS_invalid_setup); }
-            memset(lengths + current_entry, current_length, n);
-            current_entry += n;
-            ++current_length;
-         }
-      } else {
-         for (j=0; j < c->entries; ++j) {
-            int present = c->sparse ? get_bits(f,1) : 1;
-            if (present) {
-               lengths[j] = get_bits(f, 5) + 1;
-               ++total;
-            } else {
-               lengths[j] = NO_CODE;
-            }
-         }
-      }
-
-      if (c->sparse && total >= c->entries >> 2) {
-         // convert sparse items to non-sparse!
-         if (c->entries > (int) f->setup_temp_memory_required)
-            f->setup_temp_memory_required = c->entries;
-
-         c->codeword_lengths = (uint8 *) setup_malloc(f, c->entries);
-         memcpy(c->codeword_lengths, lengths, c->entries);
-         setup_temp_free(f, lengths, c->entries); // note this is only safe if there have been no intervening temp mallocs!
-         lengths = c->codeword_lengths;
-         c->sparse = 0;
-      }
-
-      // compute the size of the sorted tables
-      if (c->sparse) {
-         sorted_count = total;
-         //assert(total != 0);
-      } else {
-         sorted_count = 0;
-         #ifndef STB_VORBIS_NO_HUFFMAN_BINARY_SEARCH
-         for (j=0; j < c->entries; ++j)
-            if (lengths[j] > STB_VORBIS_FAST_HUFFMAN_LENGTH && lengths[j] != NO_CODE)
-               ++sorted_count;
-         #endif
-      }
-
-      c->sorted_entries = sorted_count;
-      values = NULL;
-
-      if (!c->sparse) {
-         c->codewords = (uint32 *) setup_malloc(f, sizeof(c->codewords[0]) * c->entries);
-         if (!c->codewords)                  return error(f, VORBIS_outofmem);
-      } else {
-         unsigned int size;
-         if (c->sorted_entries) {
-            c->codeword_lengths = (uint8 *) setup_malloc(f, c->sorted_entries);
-            if (!c->codeword_lengths)           return error(f, VORBIS_outofmem);
-            c->codewords = (uint32 *) setup_temp_malloc(f, sizeof(*c->codewords) * c->sorted_entries);
-            if (!c->codewords)                  return error(f, VORBIS_outofmem);
-            values = (uint32 *) setup_temp_malloc(f, sizeof(*values) * c->sorted_entries);
-            if (!values)                        return error(f, VORBIS_outofmem);
-         }
-         size = c->entries + (sizeof(*c->codewords) + sizeof(*values)) * c->sorted_entries;
-         if (size > f->setup_temp_memory_required)
-            f->setup_temp_memory_required = size;
-      }
-
-      if (!compute_codewords(c, lengths, c->entries, values)) {
-         if (c->sparse) setup_temp_free(f, values, 0);
-         return error(f, VORBIS_invalid_setup);
-      }
-
-      if (c->sorted_entries) {
-         // allocate an extra slot for sentinels
-         c->sorted_codewords = (uint32 *) setup_malloc(f, sizeof(*c->sorted_codewords) * (c->sorted_entries+1));
-         // allocate an extra slot at the front so that c->sorted_values[-1] is defined
-         // so that we can catch that case without an extra if
-         c->sorted_values    = ( int   *) setup_malloc(f, sizeof(*c->sorted_values   ) * (c->sorted_entries+1));
-         if (c->sorted_values) { ++c->sorted_values; c->sorted_values[-1] = -1; }
-         compute_sorted_huffman(c, lengths, values);
-      }
-
-      if (c->sparse) {
-         setup_temp_free(f, values, sizeof(*values)*c->sorted_entries);
-         setup_temp_free(f, c->codewords, sizeof(*c->codewords)*c->sorted_entries);
-         setup_temp_free(f, lengths, c->entries);
-         c->codewords = NULL;
-      }
-
-      compute_accelerated_huffman(c);
-
-      c->lookup_type = get_bits(f, 4);
-      if (c->lookup_type > 2) return error(f, VORBIS_invalid_setup);
-      if (c->lookup_type > 0) {
-         uint16 *mults;
-         c->minimum_value = float32_unpack(get_bits(f, 32));
-         c->delta_value = float32_unpack(get_bits(f, 32));
-         c->value_bits = get_bits(f, 4)+1;
-         c->sequence_p = get_bits(f,1);
-         if (c->lookup_type == 1) {
-            c->lookup_values = lookup1_values(c->entries, c->dimensions);
-         } else {
-            c->lookup_values = c->entries * c->dimensions;
-         }
-         mults = (uint16 *) setup_temp_malloc(f, sizeof(mults[0]) * c->lookup_values);
-         if (mults == NULL) return error(f, VORBIS_outofmem);
-         for (j=0; j < (int) c->lookup_values; ++j) {
-            int q = get_bits(f, c->value_bits);
-            if (q == EOP) { setup_temp_free(f,mults,sizeof(mults[0])*c->lookup_values); return error(f, VORBIS_invalid_setup); }
-            mults[j] = q;
-         }
-
-#ifndef STB_VORBIS_DIVIDES_IN_CODEBOOK
-         if (c->lookup_type == 1) {
-            int len, sparse = c->sparse;
-            // pre-expand the lookup1-style multiplicands, to avoid a divide in the inner loop
-            if (sparse) {
-               if (c->sorted_entries == 0) goto skip;
-               c->multiplicands = (codetype *) setup_malloc(f, sizeof(c->multiplicands[0]) * c->sorted_entries * c->dimensions);
-            } else
-               c->multiplicands = (codetype *) setup_malloc(f, sizeof(c->multiplicands[0]) * c->entries        * c->dimensions);
-            if (c->multiplicands == NULL) { setup_temp_free(f,mults,sizeof(mults[0])*c->lookup_values); return error(f, VORBIS_outofmem); }
-            len = sparse ? c->sorted_entries : c->entries;
-            for (j=0; j < len; ++j) {
-               int z = sparse ? c->sorted_values[j] : j, div=1;
-               for (k=0; k < c->dimensions; ++k) {
-                  int off = (z / div) % c->lookup_values;
-                  c->multiplicands[j*c->dimensions + k] =
-                         #ifndef STB_VORBIS_CODEBOOK_FLOATS
-                            mults[off];
-                         #else
-                            mults[off]*c->delta_value + c->minimum_value;
-                            // in this case (and this case only) we could pre-expand c->sequence_p,
-                            // and throw away the decode logic for it; have to ALSO do
-                            // it in the case below, but it can only be done if
-                            //    STB_VORBIS_CODEBOOK_FLOATS
-                            //   !STB_VORBIS_DIVIDES_IN_CODEBOOK
-                         #endif
-                  div *= c->lookup_values;
-               }
-            }
-            setup_temp_free(f, mults,sizeof(mults[0])*c->lookup_values);
-            c->lookup_type = 2;
-         }
-         else
-#endif
-         {
-            c->multiplicands = (codetype *) setup_malloc(f, sizeof(c->multiplicands[0]) * c->lookup_values);
-            #ifndef STB_VORBIS_CODEBOOK_FLOATS
-            memcpy(c->multiplicands, mults, sizeof(c->multiplicands[0]) * c->lookup_values);
-            #else
-            for (j=0; j < (int) c->lookup_values; ++j)
-               c->multiplicands[j] = mults[j] * c->delta_value + c->minimum_value;
-            setup_temp_free(f, mults,sizeof(mults[0])*c->lookup_values);
-            #endif
-         }
-        skip:;
-
-         #ifdef STB_VORBIS_CODEBOOK_FLOATS
-         if (c->lookup_type == 2 && c->sequence_p) {
-            for (j=1; j < (int) c->lookup_values; ++j)
-               c->multiplicands[j] = c->multiplicands[j-1];
-            c->sequence_p = 0;
-         }
-         #endif
-      }
-   }
-
-   // time domain transfers (notused)
-
-   x = get_bits(f, 6) + 1;
-   for (i=0; i < x; ++i) {
-      uint32 z = get_bits(f, 16);
-      if (z != 0) return error(f, VORBIS_invalid_setup);
-   }
-
-   // Floors
-   f->floor_count = get_bits(f, 6)+1;
-   f->floor_config = (Floor *)  setup_malloc(f, f->floor_count * sizeof(*f->floor_config));
-   for (i=0; i < f->floor_count; ++i) {
-      f->floor_types[i] = get_bits(f, 16);
-      if (f->floor_types[i] > 1) return error(f, VORBIS_invalid_setup);
-      if (f->floor_types[i] == 0) {
-         Floor0 *g = &f->floor_config[i].floor0;
-         g->order = get_bits(f,8);
-         g->rate = get_bits(f,16);
-         g->bark_map_size = get_bits(f,16);
-         g->amplitude_bits = get_bits(f,6);
-         g->amplitude_offset = get_bits(f,8);
-         g->number_of_books = get_bits(f,4) + 1;
-         for (j=0; j < g->number_of_books; ++j)
-            g->book_list[j] = get_bits(f,8);
-         return error(f, VORBIS_feature_not_supported);
-      } else {
-         Point p[31*8+2];
-         Floor1 *g = &f->floor_config[i].floor1;
-         int max_class = -1; 
-         g->partitions = get_bits(f, 5);
-         for (j=0; j < g->partitions; ++j) {
-            g->partition_class_list[j] = get_bits(f, 4);
-            if (g->partition_class_list[j] > max_class)
-               max_class = g->partition_class_list[j];
-         }
-         for (j=0; j <= max_class; ++j) {
-            g->class_dimensions[j] = get_bits(f, 3)+1;
-            g->class_subclasses[j] = get_bits(f, 2);
-            if (g->class_subclasses[j]) {
-               g->class_masterbooks[j] = get_bits(f, 8);
-               if (g->class_masterbooks[j] >= f->codebook_count) return error(f, VORBIS_invalid_setup);
-            }
-            for (k=0; k < 1 << g->class_subclasses[j]; ++k) {
-               g->subclass_books[j][k] = get_bits(f,8)-1;
-               if (g->subclass_books[j][k] >= f->codebook_count) return error(f, VORBIS_invalid_setup);
-            }
-         }
-         g->floor1_multiplier = get_bits(f,2)+1;
-         g->rangebits = get_bits(f,4);
-         g->Xlist[0] = 0;
-         g->Xlist[1] = 1 << g->rangebits;
-         g->values = 2;
-         for (j=0; j < g->partitions; ++j) {
-            int c = g->partition_class_list[j];
-            for (k=0; k < g->class_dimensions[c]; ++k) {
-               g->Xlist[g->values] = get_bits(f, g->rangebits);
-               ++g->values;
-            }
-         }
-         // precompute the sorting
-         for (j=0; j < g->values; ++j) {
-            p[j].x = g->Xlist[j];
-            p[j].y = j;
-         }
-         qsort(p, g->values, sizeof(p[0]), point_compare);
-         for (j=0; j < g->values; ++j)
-            g->sorted_order[j] = (uint8) p[j].y;
-         // precompute the neighbors
-         for (j=2; j < g->values; ++j) {
-            int low,hi;
-            neighbors(g->Xlist, j, &low,&hi);
-            g->neighbors[j][0] = low;
-            g->neighbors[j][1] = hi;
-         }
-
-         if (g->values > longest_floorlist)
-            longest_floorlist = g->values;
-      }
-   }
-
-   // Residue
-   f->residue_count = get_bits(f, 6)+1;
-   f->residue_config = (Residue *) setup_malloc(f, f->residue_count * sizeof(*f->residue_config));
-   for (i=0; i < f->residue_count; ++i) {
-      uint8 residue_cascade[64];
-      Residue *r = f->residue_config+i;
-      f->residue_types[i] = get_bits(f, 16);
-      if (f->residue_types[i] > 2) return error(f, VORBIS_invalid_setup);
-      r->begin = get_bits(f, 24);
-      r->end = get_bits(f, 24);
-      r->part_size = get_bits(f,24)+1;
-      r->classifications = get_bits(f,6)+1;
-      r->classbook = get_bits(f,8);
-      for (j=0; j < r->classifications; ++j) {
-         uint8 high_bits=0;
-         uint8 low_bits=get_bits(f,3);
-         if (get_bits(f,1))
-            high_bits = get_bits(f,5);
-         residue_cascade[j] = high_bits*8 + low_bits;
-      }
-      r->residue_books = (short (*)[8]) setup_malloc(f, sizeof(r->residue_books[0]) * r->classifications);
-      for (j=0; j < r->classifications; ++j) {
-         for (k=0; k < 8; ++k) {
-            if (residue_cascade[j] & (1 << k)) {
-               r->residue_books[j][k] = get_bits(f, 8);
-               if (r->residue_books[j][k] >= f->codebook_count) return error(f, VORBIS_invalid_setup);
-            } else {
-               r->residue_books[j][k] = -1;
-            }
-         }
-      }
-      // precompute the classifications[] array to avoid inner-loop mod/divide
-      // call it 'classdata' since we already have r->classifications
-      r->classdata = (uint8 **) setup_malloc(f, sizeof(*r->classdata) * f->codebooks[r->classbook].entries);
-      if (!r->classdata) return error(f, VORBIS_outofmem);
-      memset(r->classdata, 0, sizeof(*r->classdata) * f->codebooks[r->classbook].entries);
-      for (j=0; j < f->codebooks[r->classbook].entries; ++j) {
-         int classwords = f->codebooks[r->classbook].dimensions;
-         int temp = j;
-         r->classdata[j] = (uint8 *) setup_malloc(f, sizeof(r->classdata[j][0]) * classwords);
-         for (k=classwords-1; k >= 0; --k) {
-            r->classdata[j][k] = temp % r->classifications;
-            temp /= r->classifications;
-         }
-      }
-   }
-
-   f->mapping_count = get_bits(f,6)+1;
-   f->mapping = (Mapping *) setup_malloc(f, f->mapping_count * sizeof(*f->mapping));
-   for (i=0; i < f->mapping_count; ++i) {
-      Mapping *m = f->mapping + i;      
-      int mapping_type = get_bits(f,16);
-      if (mapping_type != 0) return error(f, VORBIS_invalid_setup);
-      m->chan = (MappingChannel *) setup_malloc(f, f->channels * sizeof(*m->chan));
-      if (get_bits(f,1))
-         m->submaps = get_bits(f,4);
-      else
-         m->submaps = 1;
-      if (m->submaps > max_submaps)
-         max_submaps = m->submaps;
-      if (get_bits(f,1)) {
-         m->coupling_steps = get_bits(f,8)+1;
-         for (k=0; k < m->coupling_steps; ++k) {
-            m->chan[k].magnitude = get_bits(f, ilog(f->channels)-1);
-            m->chan[k].angle = get_bits(f, ilog(f->channels)-1);
-            if (m->chan[k].magnitude >= f->channels)        return error(f, VORBIS_invalid_setup);
-            if (m->chan[k].angle     >= f->channels)        return error(f, VORBIS_invalid_setup);
-            if (m->chan[k].magnitude == m->chan[k].angle)   return error(f, VORBIS_invalid_setup);
-         }
-      } else
-         m->coupling_steps = 0;
-
-      // reserved field
-      if (get_bits(f,2)) return error(f, VORBIS_invalid_setup);
-      if (m->submaps > 1) {
-         for (j=0; j < f->channels; ++j) {
-            m->chan[j].mux = get_bits(f, 4);
-            if (m->chan[j].mux >= m->submaps)                return error(f, VORBIS_invalid_setup);
-         }
-      } else
-         // @SPECIFICATION: this case is missing from the spec
-         for (j=0; j < f->channels; ++j)
-            m->chan[j].mux = 0;
-
-      for (j=0; j < m->submaps; ++j) {
-         get_bits(f,8); // discard
-         m->submap_floor[j] = get_bits(f,8);
-         m->submap_residue[j] = get_bits(f,8);
-         if (m->submap_floor[j] >= f->floor_count)      return error(f, VORBIS_invalid_setup);
-         if (m->submap_residue[j] >= f->residue_count)  return error(f, VORBIS_invalid_setup);
-      }
-   }
-
-   // Modes
-   f->mode_count = get_bits(f, 6)+1;
-   for (i=0; i < f->mode_count; ++i) {
-      Mode *m = f->mode_config+i;
-      m->blockflag = get_bits(f,1);
-      m->windowtype = get_bits(f,16);
-      m->transformtype = get_bits(f,16);
-      m->mapping = get_bits(f,8);
-      if (m->windowtype != 0)                 return error(f, VORBIS_invalid_setup);
-      if (m->transformtype != 0)              return error(f, VORBIS_invalid_setup);
-      if (m->mapping >= f->mapping_count)     return error(f, VORBIS_invalid_setup);
-   }
-
-   flush_packet(f);
-
-   f->previous_length = 0;
-
-   for (i=0; i < f->channels; ++i) {
-      f->channel_buffers[i] = (float *) setup_malloc(f, sizeof(float) * f->blocksize_1);
-      f->previous_window[i] = (float *) setup_malloc(f, sizeof(float) * f->blocksize_1/2);
-      f->finalY[i]          = (int16 *) setup_malloc(f, sizeof(int16) * longest_floorlist);
-      #ifdef STB_VORBIS_NO_DEFER_FLOOR
-      f->floor_buffers[i]   = (float *) setup_malloc(f, sizeof(float) * f->blocksize_1/2);
-      #endif
-   }
-
-   if (!init_blocksize(f, 0, f->blocksize_0)) return FALSE;
-   if (!init_blocksize(f, 1, f->blocksize_1)) return FALSE;
-   f->blocksize[0] = f->blocksize_0;
-   f->blocksize[1] = f->blocksize_1;
-
-#ifdef STB_VORBIS_DIVIDE_TABLE
-   if (integer_divide_table[1][1]==0)
-      for (i=0; i < DIVTAB_NUMER; ++i)
-         for (j=1; j < DIVTAB_DENOM; ++j)
-            integer_divide_table[i][j] = i / j;
-#endif
-
-   // compute how much temporary memory is needed
-
-   // 1.
-   {
-      uint32 imdct_mem = (f->blocksize_1 * sizeof(float) >> 1);
-      uint32 classify_mem;
-      int i,max_part_read=0;
-      for (i=0; i < f->residue_count; ++i) {
-         Residue *r = f->residue_config + i;
-         int n_read = r->end - r->begin;
-         int part_read = n_read / r->part_size;
-         if (part_read > max_part_read)
-            max_part_read = part_read;
-      }
-      #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE
-      classify_mem = f->channels * (sizeof(void*) + max_part_read * sizeof(uint8 *));
-      #else
-      classify_mem = f->channels * (sizeof(void*) + max_part_read * sizeof(int *));
-      #endif
-
-      f->temp_memory_required = classify_mem;
-      if (imdct_mem > f->temp_memory_required)
-         f->temp_memory_required = imdct_mem;
-   }
-
-   f->first_decode = TRUE;
-
-   if (f->alloc.alloc_buffer) {
-      assert(f->temp_offset == f->alloc.alloc_buffer_length_in_bytes);
-      // check if there's enough temp memory so we don't error later
-      if (f->setup_offset + sizeof(*f) + f->temp_memory_required > (unsigned) f->temp_offset)
-         return error(f, VORBIS_outofmem);
-   }
-
-   f->first_audio_page_offset = stb_vorbis_get_file_offset(f);
-
-   return TRUE;
-}
-
-static void vorbis_deinit(stb_vorbis *p)
-{
-   int i,j;
-   for (i=0; i < p->residue_count; ++i) {
-      Residue *r = p->residue_config+i;
-      if (r->classdata) {
-         for (j=0; j < p->codebooks[r->classbook].entries; ++j)
-            setup_free(p, r->classdata[j]);
-         setup_free(p, r->classdata);
-      }
-      setup_free(p, r->residue_books);
-   }
-
-   if (p->codebooks) {
-      for (i=0; i < p->codebook_count; ++i) {
-         Codebook *c = p->codebooks + i;
-         setup_free(p, c->codeword_lengths);
-         setup_free(p, c->multiplicands);
-         setup_free(p, c->codewords);
-         setup_free(p, c->sorted_codewords);
-         // c->sorted_values[-1] is the first entry in the array
-         setup_free(p, c->sorted_values ? c->sorted_values-1 : NULL);
-      }
-      setup_free(p, p->codebooks);
-   }
-   setup_free(p, p->floor_config);
-   setup_free(p, p->residue_config);
-   for (i=0; i < p->mapping_count; ++i)
-      setup_free(p, p->mapping[i].chan);
-   setup_free(p, p->mapping);
-   for (i=0; i < p->channels; ++i) {
-      setup_free(p, p->channel_buffers[i]);
-      setup_free(p, p->previous_window[i]);
-      #ifdef STB_VORBIS_NO_DEFER_FLOOR
-      setup_free(p, p->floor_buffers[i]);
-      #endif
-      setup_free(p, p->finalY[i]);
-   }
-   for (i=0; i < 2; ++i) {
-      setup_free(p, p->A[i]);
-      setup_free(p, p->B[i]);
-      setup_free(p, p->C[i]);
-      setup_free(p, p->window[i]);
-      // Urho3D: fixed memory leak
-      setup_free(p, p->bit_reverse[i]);
-   }
-   #ifndef STB_VORBIS_NO_STDIO
-   if (p->close_on_free) fclose(p->f);
-   #endif
-}
-
-void stb_vorbis_close(stb_vorbis *p)
-{
-   if (p == NULL) return;
-   vorbis_deinit(p);
-   setup_free(p,p);
-}
-
-static void vorbis_init(stb_vorbis *p, stb_vorbis_alloc *z)
-{
-   memset(p, 0, sizeof(*p)); // NULL out all malloc'd pointers to start
-   if (z) {
-      p->alloc = *z;
-      p->alloc.alloc_buffer_length_in_bytes = (p->alloc.alloc_buffer_length_in_bytes+3) & ~3;
-      p->temp_offset = p->alloc.alloc_buffer_length_in_bytes;
-   }
-   p->eof = 0;
-   p->error = VORBIS__no_error;
-   p->stream = NULL;
-   p->codebooks = NULL;
-   p->page_crc_tests = -1;
-   #ifndef STB_VORBIS_NO_STDIO
-   p->close_on_free = FALSE;
-   p->f = NULL;
-   #endif
-}
-
-int stb_vorbis_get_sample_offset(stb_vorbis *f)
-{
-   if (f->current_loc_valid)
-      return f->current_loc;
-   else
-      return -1;
-}
-
-stb_vorbis_info stb_vorbis_get_info(stb_vorbis *f)
-{
-   stb_vorbis_info d;
-   d.channels = f->channels;
-   d.sample_rate = f->sample_rate;
-   d.setup_memory_required = f->setup_memory_required;
-   d.setup_temp_memory_required = f->setup_temp_memory_required;
-   d.temp_memory_required = f->temp_memory_required;
-   d.max_frame_size = f->blocksize_1 >> 1;
-   return d;
-}
-
-int stb_vorbis_get_error(stb_vorbis *f)
-{
-   int e = f->error;
-   f->error = VORBIS__no_error;
-   return e;
-}
-
-static stb_vorbis * vorbis_alloc(stb_vorbis *f)
-{
-   stb_vorbis *p = (stb_vorbis *) setup_malloc(f, sizeof(*p));
-   return p;
-}
-
-#ifndef STB_VORBIS_NO_PUSHDATA_API
-
-void stb_vorbis_flush_pushdata(stb_vorbis *f)
-{
-   f->previous_length = 0;
-   f->page_crc_tests  = 0;
-   f->discard_samples_deferred = 0;
-   f->current_loc_valid = FALSE;
-   f->first_decode = FALSE;
-   f->samples_output = 0;
-   f->channel_buffer_start = 0;
-   f->channel_buffer_end = 0;
-}
-
-static int vorbis_search_for_page_pushdata(vorb *f, uint8 *data, int data_len)
-{
-   int i,n;
-   for (i=0; i < f->page_crc_tests; ++i)
-      f->scan[i].bytes_done = 0;
-
-   // if we have room for more scans, search for them first, because
-   // they may cause us to stop early if their header is incomplete
-   if (f->page_crc_tests < STB_VORBIS_PUSHDATA_CRC_COUNT) {
-      if (data_len < 4) return 0;
-      data_len -= 3; // need to look for 4-byte sequence, so don't miss
-                     // one that straddles a boundary
-      for (i=0; i < data_len; ++i) {
-         if (data[i] == 0x4f) {
-            if (0==memcmp(data+i, ogg_page_header, 4)) {
-               int j,len;
-               uint32 crc;
-               // make sure we have the whole page header
-               if (i+26 >= data_len || i+27+data[i+26] >= data_len) {
-                  // only read up to this page start, so hopefully we'll
-                  // have the whole page header start next time
-                  data_len = i;
-                  break;
-               }
-               // ok, we have it all; compute the length of the page
-               len = 27 + data[i+26];
-               for (j=0; j < data[i+26]; ++j)
-                  len += data[i+27+j];
-               // scan everything up to the embedded crc (which we must 0)
-               crc = 0;
-               for (j=0; j < 22; ++j)
-                  crc = crc32_update(crc, data[i+j]);
-               // now process 4 0-bytes
-               for (   ; j < 26; ++j)
-                  crc = crc32_update(crc, 0);
-               // len is the total number of bytes we need to scan
-               n = f->page_crc_tests++;
-               f->scan[n].bytes_left = len-j;
-               f->scan[n].crc_so_far = crc;
-               f->scan[n].goal_crc = data[i+22] + (data[i+23] << 8) + (data[i+24]<<16) + (data[i+25]<<24);
-               // if the last frame on a page is continued to the next, then
-               // we can't recover the sample_loc immediately
-               if (data[i+27+data[i+26]-1] == 255)
-                  f->scan[n].sample_loc = ~0;
-               else
-                  f->scan[n].sample_loc = data[i+6] + (data[i+7] << 8) + (data[i+ 8]<<16) + (data[i+ 9]<<24);
-               f->scan[n].bytes_done = i+j;
-               if (f->page_crc_tests == STB_VORBIS_PUSHDATA_CRC_COUNT)
-                  break;
-               // keep going if we still have room for more
-            }
-         }
-      }
-   }
-
-   for (i=0; i < f->page_crc_tests;) {
-      uint32 crc;
-      int j;
-      int n = f->scan[i].bytes_done;
-      int m = f->scan[i].bytes_left;
-      if (m > data_len - n) m = data_len - n;
-      // m is the bytes to scan in the current chunk
-      crc = f->scan[i].crc_so_far;
-      for (j=0; j < m; ++j)
-         crc = crc32_update(crc, data[n+j]);
-      f->scan[i].bytes_left -= m;
-      f->scan[i].crc_so_far = crc;
-      if (f->scan[i].bytes_left == 0) {
-         // does it match?
-         if (f->scan[i].crc_so_far == f->scan[i].goal_crc) {
-            // Houston, we have page
-            data_len = n+m; // consumption amount is wherever that scan ended
-            f->page_crc_tests = -1; // drop out of page scan mode
-            f->previous_length = 0; // decode-but-don't-output one frame
-            f->next_seg = -1;       // start a new page
-            f->current_loc = f->scan[i].sample_loc; // set the current sample location
-                                    // to the amount we'd have decoded had we decoded this page
-            f->current_loc_valid = f->current_loc != ~0;
-            return data_len;
-         }
-         // delete entry
-         f->scan[i] = f->scan[--f->page_crc_tests];
-      } else {
-         ++i;
-      }
-   }
-
-   return data_len;
-}
-
-// return value: number of bytes we used
-int stb_vorbis_decode_frame_pushdata(
-         stb_vorbis *f,                 // the file we're decoding
-         uint8 *data, int data_len,     // the memory available for decoding
-         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
-     )
-{
-   int i;
-   int len,right,left;
-
-   if (!IS_PUSH_MODE(f)) return error(f, VORBIS_invalid_api_mixing);
-
-   if (f->page_crc_tests >= 0) {
-      *samples = 0;
-      return vorbis_search_for_page_pushdata(f, data, data_len);
-   }
-
-   f->stream     = data;
-   f->stream_end = data + data_len;
-   f->error      = VORBIS__no_error;
-
-   // check that we have the entire packet in memory
-   if (!is_whole_packet_present(f, FALSE)) {
-      *samples = 0;
-      return 0;
-   }
-
-   if (!vorbis_decode_packet(f, &len, &left, &right)) {
-      // save the actual error we encountered
-      enum STBVorbisError error = f->error;
-      if (error == VORBIS_bad_packet_type) {
-         // flush and resynch
-         f->error = VORBIS__no_error;
-         while (get8_packet(f) != EOP)
-            if (f->eof) break;
-         *samples = 0;
-         return f->stream - data;
-      }
-      if (error == VORBIS_continued_packet_flag_invalid) {
-         if (f->previous_length == 0) {
-            // we may be resynching, in which case it's ok to hit one
-            // of these; just discard the packet
-            f->error = VORBIS__no_error;
-            while (get8_packet(f) != EOP)
-               if (f->eof) break;
-            *samples = 0;
-            return f->stream - data;
-         }
-      }
-      // if we get an error while parsing, what to do?
-      // well, it DEFINITELY won't work to continue from where we are!
-      stb_vorbis_flush_pushdata(f);
-      // restore the error that actually made us bail
-      f->error = error;
-      *samples = 0;
-      return 1;
-   }
-
-   // success!
-   len = vorbis_finish_frame(f, len, left, right);
-   for (i=0; i < f->channels; ++i)
-      f->outputs[i] = f->channel_buffers[i] + left;
-
-   if (channels) *channels = f->channels;
-   *samples = len;
-   *output = f->outputs;
-   return f->stream - data;
-}
-
-stb_vorbis *stb_vorbis_open_pushdata(
-         unsigned char *data, int data_len, // the memory available for decoding
-         int *data_used,              // only defined if result is not NULL
-         int *error, stb_vorbis_alloc *alloc)
-{
-   stb_vorbis *f, p;
-   vorbis_init(&p, alloc);
-   p.stream     = data;
-   p.stream_end = data + data_len;
-   p.push_mode  = TRUE;
-   if (!start_decoder(&p)) {
-      if (p.eof)
-         *error = VORBIS_need_more_data;
-      else
-         *error = p.error;
-      return NULL;
-   }
-   f = vorbis_alloc(&p);
-   if (f) {
-      *f = p;
-      *data_used = f->stream - data;
-      *error = 0;
-      return f;
-   } else {
-      vorbis_deinit(&p);
-      return NULL;
-   }
-}
-#endif // STB_VORBIS_NO_PUSHDATA_API
-
-unsigned int stb_vorbis_get_file_offset(stb_vorbis *f)
-{
-   #ifndef STB_VORBIS_NO_PUSHDATA_API
-   if (f->push_mode) return 0;
-   #endif
-   if (USE_MEMORY(f)) return f->stream - f->stream_start;
-   #ifndef STB_VORBIS_NO_STDIO
-   return ftell(f->f) - f->f_start;
-   #endif
-}
-
-#ifndef STB_VORBIS_NO_PULLDATA_API
-//
-// DATA-PULLING API
-//
-
-static uint32 vorbis_find_page(stb_vorbis *f, uint32 *end, uint32 *last)
-{
-   for(;;) {
-      int n;
-      if (f->eof) return 0;
-      n = get8(f);
-      if (n == 0x4f) { // page header
-         unsigned int retry_loc = stb_vorbis_get_file_offset(f);
-         int i;
-         // check if we're off the end of a file_section stream
-         if (retry_loc - 25 > f->stream_len)
-            return 0;
-         // check the rest of the header
-         for (i=1; i < 4; ++i)
-            if (get8(f) != ogg_page_header[i])
-               break;
-         if (f->eof) return 0;
-         if (i == 4) {
-            uint8 header[27];
-            uint32 i, crc, goal, len;
-            for (i=0; i < 4; ++i)
-               header[i] = ogg_page_header[i];
-            for (; i < 27; ++i)
-               header[i] = get8(f);
-            if (f->eof) return 0;
-            if (header[4] != 0) goto invalid;
-            goal = header[22] + (header[23] << 8) + (header[24]<<16) + (header[25]<<24);
-            for (i=22; i < 26; ++i)
-               header[i] = 0;
-            crc = 0;
-            for (i=0; i < 27; ++i)
-               crc = crc32_update(crc, header[i]);
-            len = 0;
-            for (i=0; i < header[26]; ++i) {
-               int s = get8(f);
-               crc = crc32_update(crc, s);
-               len += s;
-            }
-            if (len && f->eof) return 0;
-            for (i=0; i < len; ++i)
-               crc = crc32_update(crc, get8(f));
-            // finished parsing probable page
-            if (crc == goal) {
-               // we could now check that it's either got the last
-               // page flag set, OR it's followed by the capture
-               // pattern, but I guess TECHNICALLY you could have
-               // a file with garbage between each ogg page and recover
-               // from it automatically? So even though that paranoia
-               // might decrease the chance of an invalid decode by
-               // another 2^32, not worth it since it would hose those
-               // invalid-but-useful files?
-               if (end)
-                  *end = stb_vorbis_get_file_offset(f);
-               if (last)
-                  if (header[5] & 0x04)
-                     *last = 1;
-                  else
-                     *last = 0;
-               set_file_offset(f, retry_loc-1);
-               return 1;
-            }
-         }
-        invalid:
-         // not a valid page, so rewind and look for next one
-         set_file_offset(f, retry_loc);
-      }
-   }
-}
-
-// seek is implemented with 'interpolation search'--this is like
-// binary search, but we use the data values to estimate the likely
-// location of the data item (plus a bit of a bias so when the
-// estimation is wrong we don't waste overly much time)
-
-#define SAMPLE_unknown  0xffffffff
-
-
-// ogg vorbis, in its insane infinite wisdom, only provides
-// information about the sample at the END of the page.
-// therefore we COULD have the data we need in the current
-// page, and not know it. we could just use the end location
-// as our only knowledge for bounds, seek back, and eventually
-// the binary search finds it. or we can try to be smart and
-// not waste time trying to locate more pages. we try to be
-// smart, since this data is already in memory anyway, so
-// doing needless I/O would be crazy!
-static int vorbis_analyze_page(stb_vorbis *f, ProbedPage *z)
-{
-   uint8 header[27], lacing[255];
-   uint8 packet_type[255];
-   int num_packet, packet_start, previous =0;
-   int i,len;
-   uint32 samples;
-
-   // record where the page starts
-   z->page_start = stb_vorbis_get_file_offset(f);
-
-   // parse the header
-   getn(f, header, 27);
-   assert(header[0] == 'O' && header[1] == 'g' && header[2] == 'g' && header[3] == 'S');
-   getn(f, lacing, header[26]);
-
-   // determine the length of the payload
-   len = 0;
-   for (i=0; i < header[26]; ++i)
-      len += lacing[i];
-
-   // this implies where the page ends
-   z->page_end = z->page_start + 27 + header[26] + len;
-
-   // read the last-decoded sample out of the data
-   z->last_decoded_sample = header[6] + (header[7] << 8) + (header[8] << 16) + (header[9] << 16);
-
-   if (header[5] & 4) {
-      // if this is the last page, it's not possible to work
-      // backwards to figure out the first sample! whoops! fuck.
-      z->first_decoded_sample = SAMPLE_unknown;
-      set_file_offset(f, z->page_start);
-      return 1;
-   }
-
-   // scan through the frames to determine the sample-count of each one...
-   // our goal is the sample # of the first fully-decoded sample on the
-   // page, which is the first decoded sample of the 2nd page
-
-   num_packet=0;
-
-   packet_start = ((header[5] & 1) == 0);
-
-   for (i=0; i < header[26]; ++i) {
-      if (packet_start) {
-         uint8 n,b,m;
-         if (lacing[i] == 0) goto bail; // trying to read from zero-length packet
-         n = get8(f);
-         // if bottom bit is non-zero, we've got corruption
-         if (n & 1) goto bail;
-         n >>= 1;
-         b = ilog(f->mode_count-1);
-         m = n >> b;
-         n &= (1 << b)-1;
-         if (n >= f->mode_count) goto bail;
-         if (num_packet == 0 && f->mode_config[n].blockflag)
-            previous = (m & 1);
-         packet_type[num_packet++] = f->mode_config[n].blockflag;
-         skip(f, lacing[i]-1);
-      } else
-         skip(f, lacing[i]);
-      packet_start = (lacing[i] < 255);
-   }
-
-   // now that we know the sizes of all the pages, we can start determining
-   // how much sample data there is.
-
-   samples = 0;
-
-   // for the last packet, we step by its whole length, because the definition
-   // is that we encoded the end sample loc of the 'last packet completed',
-   // where 'completed' refers to packets being split, and we are left to guess
-   // what 'end sample loc' means. we assume it means ignoring the fact that
-   // the last half of the data is useless without windowing against the next
-   // packet... (so it's not REALLY complete in that sense)
-   if (num_packet > 1)
-      samples += f->blocksize[packet_type[num_packet-1]];
-
-   for (i=num_packet-2; i >= 1; --i) {
-      // now, for this packet, how many samples do we have that
-      // do not overlap the following packet?
-      if (packet_type[i] == 1)
-         if (packet_type[i+1] == 1)
-            samples += f->blocksize_1 >> 1;
-         else
-            samples += ((f->blocksize_1 - f->blocksize_0) >> 2) + (f->blocksize_0 >> 1);
-      else
-         samples += f->blocksize_0 >> 1;
-   }
-   // now, at this point, we've rewound to the very beginning of the
-   // _second_ packet. if we entirely discard the first packet after
-   // a seek, this will be exactly the right sample number. HOWEVER!
-   // we can't as easily compute this number for the LAST page. The
-   // only way to get the sample offset of the LAST page is to use
-   // the end loc from the previous page. But what that returns us
-   // is _exactly_ the place where we get our first non-overlapped
-   // sample. (I think. Stupid spec for being ambiguous.) So for
-   // consistency it's better to do that here, too. However, that
-   // will then require us to NOT discard all of the first frame we
-   // decode, in some cases, which means an even weirder frame size
-   // and extra code. what a fucking pain.
-   
-   // we're going to discard the first packet if we
-   // start the seek here, so we don't care about it. (we could actually
-   // do better; if the first packet is long, and the previous packet
-   // is short, there's actually data in the first half of the first
-   // packet that doesn't need discarding... but not worth paying the
-   // effort of tracking that of that here and in the seeking logic)
-   // except crap, if we infer it from the _previous_ packet's end
-   // location, we DO need to use that definition... and we HAVE to
-   // infer the start loc of the LAST packet from the previous packet's
-   // end location. fuck you, ogg vorbis.
-
-   z->first_decoded_sample = z->last_decoded_sample - samples;
-
-   // restore file state to where we were
-   set_file_offset(f, z->page_start);
-   return 1;
-
-   // restore file state to where we were
-  bail:
-   set_file_offset(f, z->page_start);
-   return 0;
-}
-
-static int vorbis_seek_frame_from_page(stb_vorbis *f, uint32 page_start, uint32 first_sample, uint32 target_sample, int fine)
-{
-   int left_start, left_end, right_start, right_end, mode,i;
-   int frame=0;
-   uint32 frame_start;
-   int frames_to_skip, data_to_skip;
-
-   // first_sample is the sample # of the first sample that doesn't
-   // overlap the previous page... note that this requires us to
-   // _partially_ discard the first packet! bleh.
-   set_file_offset(f, page_start);
-
-   f->next_seg = -1;  // force page resync
-
-   frame_start = first_sample;
-   // frame start is where the previous packet's last decoded sample
-   // was, which corresponds to left_end... EXCEPT if the previous
-   // packet was long and this packet is short? Probably a bug here.
-
-
-   // now, we can start decoding frames... we'll only FAKE decode them,
-   // until we find the frame that contains our sample; then we'll rewind,
-   // and try again
-   for (;;) {
-      int start;
-
-      if (!vorbis_decode_initial(f, &left_start, &left_end, &right_start, &right_end, &mode))
-         return error(f, VORBIS_seek_failed);
-
-      if (frame == 0)
-         start = left_end;
-      else
-         start = left_start;
-
-      // the window starts at left_start; the last valid sample we generate
-      // before the next frame's window start is right_start-1
-      if (target_sample < frame_start + right_start-start)
-         break;
-
-      flush_packet(f);
-      if (f->eof)
-         return error(f, VORBIS_seek_failed);
-
-      frame_start += right_start - start;
-
-      ++frame;
-   }
-
-   // ok, at this point, the sample we want is contained in frame #'frame'
-
-   // to decode frame #'frame' normally, we have to decode the
-   // previous frame first... but if it's the FIRST frame of the page
-   // we can't. if it's the first frame, it means it falls in the part
-   // of the first frame that doesn't overlap either of the other frames.
-   // so, if we have to handle that case for the first frame, we might
-   // as well handle it for all of them, so:
-   if (target_sample > frame_start + (left_end - left_start)) {
-      // so what we want to do is go ahead and just immediately decode
-      // this frame, but then make it so the next get_frame_float() uses
-      // this already-decoded data? or do we want to go ahead and rewind,
-      // and leave a flag saying to skip the first N data? let's do that
-      frames_to_skip = frame;  // if this is frame #1, skip 1 frame (#0)
-      data_to_skip = left_end - left_start;
-   } else {
-      // otherwise, we want to skip frames 0, 1, 2, ... frame-2
-      // (which means frame-2+1 total frames) then decode frame-1,
-      // then leave frame pending
-      frames_to_skip = frame - 1;
-      assert(frames_to_skip >= 0);
-      data_to_skip = -1;      
-   }
-
-   set_file_offset(f, page_start);
-   f->next_seg = - 1; // force page resync
-
-   for (i=0; i < frames_to_skip; ++i) {
-      maybe_start_packet(f);
-      flush_packet(f);
-   }
-
-   if (data_to_skip >= 0) {
-      int i,j,n = f->blocksize_0 >> 1;
-      f->discard_samples_deferred = data_to_skip;
-      for (i=0; i < f->channels; ++i)
-         for (j=0; j < n; ++j)
-            f->previous_window[i][j] = 0;
-      f->previous_length = n;
-      frame_start += data_to_skip;
-   } else {
-      f->previous_length = 0;
-      vorbis_pump_first_frame(f);
-   }
-
-   // at this point, the NEXT decoded frame will generate the desired sample
-   if (fine) {
-      // so if we're doing sample accurate streaming, we want to go ahead and decode it!
-      if (target_sample != frame_start) {
-         int n;
-         stb_vorbis_get_frame_float(f, &n, NULL);
-         assert(target_sample > frame_start);
-         assert(f->channel_buffer_start + (int) (target_sample-frame_start) < f->channel_buffer_end);
-         f->channel_buffer_start += (target_sample - frame_start);
-      }
-   }
-
-   return 0;
-}
-
-static int vorbis_seek_base(stb_vorbis *f, unsigned int sample_number, int fine)
-{
-   ProbedPage p[2],q;
-   if (IS_PUSH_MODE(f)) return error(f, VORBIS_invalid_api_mixing);
-
-   // do we know the location of the last page?
-   if (f->p_last.page_start == 0) {
-      uint32 z = stb_vorbis_stream_length_in_samples(f);
-      if (z == 0) return error(f, VORBIS_cant_find_last_page);
-   }
-
-   p[0] = f->p_first;
-   p[1] = f->p_last;
-
-   if (sample_number >= f->p_last.last_decoded_sample)
-      sample_number = f->p_last.last_decoded_sample-1;
-
-   if (sample_number < f->p_first.last_decoded_sample) {
-      vorbis_seek_frame_from_page(f, p[0].page_start, 0, sample_number, fine);
-      return 0;
-   } else {
-      int attempts=0;
-      while (p[0].page_end < p[1].page_start) {
-         uint32 probe;
-         uint32 start_offset, end_offset;
-         uint32 start_sample, end_sample;
-
-         // copy these into local variables so we can tweak them
-         // if any are unknown
-         start_offset = p[0].page_end;
-         end_offset   = p[1].after_previous_page_start; // an address known to seek to page p[1]
-         start_sample = p[0].last_decoded_sample;
-         end_sample   = p[1].last_decoded_sample;
-
-         // currently there is no such tweaking logic needed/possible?
-         if (start_sample == SAMPLE_unknown || end_sample == SAMPLE_unknown)
-            return error(f, VORBIS_seek_failed);
-
-         // now we want to lerp between these for the target samples...
-      
-         // step 1: we need to bias towards the page start...
-         if (start_offset + 4000 < end_offset)
-            end_offset -= 4000;
-
-         // now compute an interpolated search loc
-         probe = start_offset + (int) floor((float) (end_offset - start_offset) / (end_sample - start_sample) * (sample_number - start_sample));
-
-         // next we need to bias towards binary search...
-         // code is a little wonky to allow for full 32-bit unsigned values
-         if (attempts >= 4) {
-            uint32 probe2 = start_offset + ((end_offset - start_offset) >> 1);
-            if (attempts >= 8)
-               probe = probe2;
-            else if (probe < probe2)
-               probe = probe + ((probe2 - probe) >> 1);
-            else
-               probe = probe2 + ((probe - probe2) >> 1);
-         }
-         ++attempts;
-
-         set_file_offset(f, probe);
-         if (!vorbis_find_page(f, NULL, NULL))   return error(f, VORBIS_seek_failed);
-         if (!vorbis_analyze_page(f, &q))        return error(f, VORBIS_seek_failed);
-         q.after_previous_page_start = probe;
-
-         // it's possible we've just found the last page again
-         if (q.page_start == p[1].page_start) {
-            p[1] = q;
-            continue;
-         }
-
-         if (sample_number < q.last_decoded_sample)
-            p[1] = q;
-         else
-            p[0] = q;
-      }
-
-      if (p[0].last_decoded_sample <= sample_number && sample_number < p[1].last_decoded_sample) {
-         vorbis_seek_frame_from_page(f, p[1].page_start, p[0].last_decoded_sample, sample_number, fine);
-         return 0;
-      }
-      return error(f, VORBIS_seek_failed);
-   }
-}
-
-int stb_vorbis_seek_frame(stb_vorbis *f, unsigned int sample_number)
-{
-   return vorbis_seek_base(f, sample_number, FALSE);
-}
-
-int stb_vorbis_seek(stb_vorbis *f, unsigned int sample_number)
-{
-   return vorbis_seek_base(f, sample_number, TRUE);
-}
-
-void stb_vorbis_seek_start(stb_vorbis *f)
-{
-   if (IS_PUSH_MODE(f)) { error(f, VORBIS_invalid_api_mixing); return; }
-   set_file_offset(f, f->first_audio_page_offset);
-   f->previous_length = 0;
-   f->first_decode = TRUE;
-   f->next_seg = -1;
-   vorbis_pump_first_frame(f);
-}
-
-unsigned int stb_vorbis_stream_length_in_samples(stb_vorbis *f)
-{
-   unsigned int restore_offset, previous_safe;
-   unsigned int end, last_page_loc;
-
-   if (IS_PUSH_MODE(f)) return error(f, VORBIS_invalid_api_mixing);
-   if (!f->total_samples) {
-      int last;
-      uint32 lo,hi;
-      char header[6];
-
-      // first, store the current decode position so we can restore it
-      restore_offset = stb_vorbis_get_file_offset(f);
-
-      // now we want to seek back 64K from the end (the last page must
-      // be at most a little less than 64K, but let's allow a little slop)
-      if (f->stream_len >= 65536 && f->stream_len-65536 >= f->first_audio_page_offset)
-         previous_safe = f->stream_len - 65536;
-      else
-         previous_safe = f->first_audio_page_offset;
-
-      set_file_offset(f, previous_safe);
-      // previous_safe is now our candidate 'earliest known place that seeking
-      // to will lead to the final page'
-
-      if (!vorbis_find_page(f, &end, (int unsigned *)&last)) {
-         // if we can't find a page, we're hosed!
-         f->error = VORBIS_cant_find_last_page;
-         f->total_samples = 0xffffffff;
-         goto done;
-      }
-
-      // check if there are more pages
-      last_page_loc = stb_vorbis_get_file_offset(f);
-
-      // stop when the last_page flag is set, not when we reach eof;
-      // this allows us to stop short of a 'file_section' end without
-      // explicitly checking the length of the section
-      while (!last) {
-         set_file_offset(f, end);
-         if (!vorbis_find_page(f, &end, (int unsigned *)&last)) {
-            // the last page we found didn't have the 'last page' flag
-            // set. whoops!
-            break;
-         }
-         previous_safe = last_page_loc+1;
-         last_page_loc = stb_vorbis_get_file_offset(f);
-      }
-
-      set_file_offset(f, last_page_loc);
-
-      // parse the header
-      getn(f, (unsigned char *)header, 6);
-      // extract the absolute granule position
-      lo = get32(f);
-      hi = get32(f);
-      if (lo == 0xffffffff && hi == 0xffffffff) {
-         f->error = VORBIS_cant_find_last_page;
-         f->total_samples = SAMPLE_unknown;
-         goto done;
-      }
-      if (hi)
-         lo = 0xfffffffe; // saturate
-      f->total_samples = lo;
-
-      f->p_last.page_start = last_page_loc;
-      f->p_last.page_end   = end;
-      f->p_last.last_decoded_sample = lo;
-      f->p_last.first_decoded_sample = SAMPLE_unknown;
-      f->p_last.after_previous_page_start = previous_safe;
-
-     done:
-      set_file_offset(f, restore_offset);
-   }
-   return f->total_samples == SAMPLE_unknown ? 0 : f->total_samples;
-}
-
-float stb_vorbis_stream_length_in_seconds(stb_vorbis *f)
-{
-   return stb_vorbis_stream_length_in_samples(f) / (float) f->sample_rate;
-}
-
-
-
-int stb_vorbis_get_frame_float(stb_vorbis *f, int *channels, float ***output)
-{
-   int len, right,left,i;
-   if (IS_PUSH_MODE(f)) return error(f, VORBIS_invalid_api_mixing);
-
-   if (!vorbis_decode_packet(f, &len, &left, &right)) {
-      f->channel_buffer_start = f->channel_buffer_end = 0;
-      return 0;
-   }
-
-   len = vorbis_finish_frame(f, len, left, right);
-   for (i=0; i < f->channels; ++i)
-      f->outputs[i] = f->channel_buffers[i] + left;
-
-   f->channel_buffer_start = left;
-   f->channel_buffer_end   = left+len;
-
-   if (channels) *channels = f->channels;
-   if (output)   *output = f->outputs;
-   return len;
-}
-
-#ifndef STB_VORBIS_NO_STDIO
-
-stb_vorbis * stb_vorbis_open_file_section(FILE *file, int close_on_free, int *error, stb_vorbis_alloc *alloc, unsigned int length)
-{
-   stb_vorbis *f, p;
-   vorbis_init(&p, alloc);
-   p.f = file;
-   p.f_start = ftell(file);
-   p.stream_len   = length;
-   p.close_on_free = close_on_free;
-   if (start_decoder(&p)) {
-      f = vorbis_alloc(&p);
-      if (f) {
-         *f = p;
-         vorbis_pump_first_frame(f);
-         return f;
-      }
-   }
-   if (error) *error = p.error;
-   vorbis_deinit(&p);
-   return NULL;
-}
-
-stb_vorbis * stb_vorbis_open_file(FILE *file, int close_on_free, int *error, stb_vorbis_alloc *alloc)
-{
-   unsigned int len, start;
-   start = ftell(file);
-   fseek(file, 0, SEEK_END);
-   len = ftell(file) - start;
-   fseek(file, start, SEEK_SET);
-   return stb_vorbis_open_file_section(file, close_on_free, error, alloc, len);
-}
-
-stb_vorbis * stb_vorbis_open_filename(char *filename, int *error, stb_vorbis_alloc *alloc)
-{
-   FILE *f = fopen(filename, "rb");
-   if (f) 
-      return stb_vorbis_open_file(f, TRUE, error, alloc);
-   if (error) *error = VORBIS_file_open_failure;
-   return NULL;
-}
-#endif // STB_VORBIS_NO_STDIO
-
-stb_vorbis * stb_vorbis_open_memory(unsigned char *data, int len, int *error, stb_vorbis_alloc *alloc)
-{
-   stb_vorbis *f, p;
-   if (data == NULL) return NULL;
-   vorbis_init(&p, alloc);
-   p.stream = data;
-   p.stream_end = data + len;
-   p.stream_start = p.stream;
-   p.stream_len = len;
-   p.push_mode = FALSE;
-   if (start_decoder(&p)) {
-      f = vorbis_alloc(&p);
-      if (f) {
-         *f = p;
-         vorbis_pump_first_frame(f);
-         return f;
-      }
-   }
-   if (error) *error = p.error;
-   vorbis_deinit(&p);
-   return NULL;
-}
-
-#ifndef STB_VORBIS_NO_INTEGER_CONVERSION
-#define PLAYBACK_MONO     1
-#define PLAYBACK_LEFT     2
-#define PLAYBACK_RIGHT    4
-
-#define L  (PLAYBACK_LEFT  | PLAYBACK_MONO)
-#define C  (PLAYBACK_LEFT  | PLAYBACK_RIGHT | PLAYBACK_MONO)
-#define R  (PLAYBACK_RIGHT | PLAYBACK_MONO)
-
-static int8 channel_position[7][6] =
-{
-   { 0 },
-   { C },
-   { L, R },
-   { L, C, R },
-   { L, R, L, R },
-   { L, C, R, L, R },
-   { L, C, R, L, R, C },
-};
-
-
-#ifndef STB_VORBIS_NO_FAST_SCALED_FLOAT
-   typedef union {
-      float f;
-      int i;
-   } float_conv;
-   typedef char stb_vorbis_float_size_test[sizeof(float)==4 && sizeof(int) == 4];
-   #define FASTDEF(x) float_conv x
-   // add (1<<23) to convert to int, then divide by 2^SHIFT, then add 0.5/2^SHIFT to round
-   #define MAGIC(SHIFT) (1.5f * (1 << (23-SHIFT)) + 0.5f/(1 << SHIFT))
-   #define ADDEND(SHIFT) (((150-SHIFT) << 23) + (1 << 22))
-   #define FAST_SCALED_FLOAT_TO_INT(temp,x,s) (temp.f = (x) + MAGIC(s), temp.i - ADDEND(s))
-   #define check_endianness()  
-#else
-   #define FAST_SCALED_FLOAT_TO_INT(temp,x,s) ((int) ((x) * (1 << (s))))
-   #define check_endianness()
-   #define FASTDEF(x)
-#endif
-
-static void copy_samples(short *dest, float *src, int len)
-{
-   int i;
-   check_endianness();
-   for (i=0; i < len; ++i) {
-      FASTDEF(temp);
-      int v = FAST_SCALED_FLOAT_TO_INT(temp, src[i],15);
-      if ((unsigned int) (v + 32768) > 65535)
-         v = v < 0 ? -32768 : 32767;
-      dest[i] = v;
-   }
-}
-
-static void compute_samples(int mask, short *output, int num_c, float **data, int d_offset, int len)
-{
-   #define BUFFER_SIZE  32
-   float buffer[BUFFER_SIZE];
-   int i,j,o,n = BUFFER_SIZE;
-   check_endianness();
-   for (o = 0; o < len; o += BUFFER_SIZE) {
-      memset(buffer, 0, sizeof(buffer));
-      if (o + n > len) n = len - o;
-      for (j=0; j < num_c; ++j) {
-         if (channel_position[num_c][j] & mask) {
-            for (i=0; i < n; ++i)
-               buffer[i] += data[j][d_offset+o+i];
-         }
-      }
-      for (i=0; i < n; ++i) {
-         FASTDEF(temp);
-         int v = FAST_SCALED_FLOAT_TO_INT(temp,buffer[i],15);
-         if ((unsigned int) (v + 32768) > 65535)
-            v = v < 0 ? -32768 : 32767;
-         output[o+i] = v;
-      }
-   }
-}
-
-static int channel_selector[3][2] = { {0}, {PLAYBACK_MONO}, {PLAYBACK_LEFT, PLAYBACK_RIGHT} };
-static void compute_stereo_samples(short *output, int num_c, float **data, int d_offset, int len)
-{
-   #define BUFFER_SIZE  32
-   float buffer[BUFFER_SIZE];
-   int i,j,o,n = BUFFER_SIZE >> 1;
-   // o is the offset in the source data
-   check_endianness();
-   for (o = 0; o < len; o += BUFFER_SIZE >> 1) {
-      // o2 is the offset in the output data
-      int o2 = o << 1;
-      memset(buffer, 0, sizeof(buffer));
-      if (o + n > len) n = len - o;
-      for (j=0; j < num_c; ++j) {
-         int m = channel_position[num_c][j] & (PLAYBACK_LEFT | PLAYBACK_RIGHT);
-         if (m == (PLAYBACK_LEFT | PLAYBACK_RIGHT)) {
-            for (i=0; i < n; ++i) {
-               buffer[i*2+0] += data[j][d_offset+o+i];
-               buffer[i*2+1] += data[j][d_offset+o+i];
-            }
-         } else if (m == PLAYBACK_LEFT) {
-            for (i=0; i < n; ++i) {
-               buffer[i*2+0] += data[j][d_offset+o+i];
-            }
-         } else if (m == PLAYBACK_RIGHT) {
-            for (i=0; i < n; ++i) {
-               buffer[i*2+1] += data[j][d_offset+o+i];
-            }
-         }
-      }
-      for (i=0; i < (n<<1); ++i) {
-         FASTDEF(temp);
-         int v = FAST_SCALED_FLOAT_TO_INT(temp,buffer[i],15);
-         if ((unsigned int) (v + 32768) > 65535)
-            v = v < 0 ? -32768 : 32767;
-         output[o2+i] = v;
-      }
-   }
-}
-
-static void convert_samples_short(int buf_c, short **buffer, int b_offset, int data_c, float **data, int d_offset, int samples)
-{
-   int i;
-   if (buf_c != data_c && buf_c <= 2 && data_c <= 6) {
-      static int channel_selector[3][2] = { {0}, {PLAYBACK_MONO}, {PLAYBACK_LEFT, PLAYBACK_RIGHT} };
-      for (i=0; i < buf_c; ++i)
-         compute_samples(channel_selector[buf_c][i], buffer[i]+b_offset, data_c, data, d_offset, samples);
-   } else {
-      int limit = buf_c < data_c ? buf_c : data_c;
-      for (i=0; i < limit; ++i)
-         copy_samples(buffer[i]+b_offset, data[i], samples);
-      for (   ; i < buf_c; ++i)
-         memset(buffer[i]+b_offset, 0, sizeof(short) * samples);
-   }
-}
-
-int stb_vorbis_get_frame_short(stb_vorbis *f, int num_c, short **buffer, int num_samples)
-{
-   float **output;
-   int len = stb_vorbis_get_frame_float(f, NULL, &output);
-   if (len > num_samples) len = num_samples;
-   if (len)
-      convert_samples_short(num_c, buffer, 0, f->channels, output, 0, len);
-   return len;
-}
-
-static void convert_channels_short_interleaved(int buf_c, short *buffer, int data_c, float **data, int d_offset, int len)
-{
-   int i;
-   check_endianness();
-   if (buf_c != data_c && buf_c <= 2 && data_c <= 6) {
-      assert(buf_c == 2);
-      for (i=0; i < buf_c; ++i)
-         compute_stereo_samples(buffer, data_c, data, d_offset, len);
-   } else {
-      int limit = buf_c < data_c ? buf_c : data_c;
-      int j;
-      for (j=0; j < len; ++j) {
-         for (i=0; i < limit; ++i) {
-            FASTDEF(temp);
-            float f = data[i][d_offset+j];
-            int v = FAST_SCALED_FLOAT_TO_INT(temp, f,15);//data[i][d_offset+j],15);
-            if ((unsigned int) (v + 32768) > 65535)
-               v = v < 0 ? -32768 : 32767;
-            *buffer++ = v;
-         }
-         for (   ; i < buf_c; ++i)
-            *buffer++ = 0;
-      }
-   }
-}
-
-int stb_vorbis_get_frame_short_interleaved(stb_vorbis *f, int num_c, short *buffer, int num_shorts)
-{
-   float **output;
-   int len;
-   if (num_c == 1) return stb_vorbis_get_frame_short(f,num_c,&buffer, num_shorts);
-   len = stb_vorbis_get_frame_float(f, NULL, &output);
-   if (len) {
-      if (len*num_c > num_shorts) len = num_shorts / num_c;
-      convert_channels_short_interleaved(num_c, buffer, f->channels, output, 0, len);
-   }
-   return len;
-}
-
-int stb_vorbis_get_samples_short_interleaved(stb_vorbis *f, int channels, short *buffer, int num_shorts)
-{
-   float **outputs;
-   int len = num_shorts / channels;
-   int n=0;
-   int z = f->channels;
-   if (z > channels) z = channels;
-   while (n < len) {
-      int k = f->channel_buffer_end - f->channel_buffer_start;
-      if (n+k >= len) k = len - n;
-      if (k)
-         convert_channels_short_interleaved(channels, buffer, f->channels, f->channel_buffers, f->channel_buffer_start, k);
-      buffer += k*channels;
-      n += k;
-      f->channel_buffer_start += k;
-      if (n == len) break;
-      if (!stb_vorbis_get_frame_float(f, NULL, &outputs)) break;
-   }
-   return n;
-}
-
-int stb_vorbis_get_samples_short(stb_vorbis *f, int channels, short **buffer, int len)
-{
-   float **outputs;
-   int n=0;
-   int z = f->channels;
-   if (z > channels) z = channels;
-   while (n < len) {
-      int k = f->channel_buffer_end - f->channel_buffer_start;
-      if (n+k >= len) k = len - n;
-      if (k)
-         convert_samples_short(channels, buffer, n, f->channels, f->channel_buffers, f->channel_buffer_start, k);
-      n += k;
-      f->channel_buffer_start += k;
-      if (n == len) break;
-      if (!stb_vorbis_get_frame_float(f, NULL, &outputs)) break;
-   }
-   return n;
-}
-
-#ifndef STB_VORBIS_NO_STDIO
-int stb_vorbis_decode_filename(char *filename, int *channels, short **output)
-{
-   int data_len, offset, total, limit, error;
-   short *data;
-   stb_vorbis *v = stb_vorbis_open_filename(filename, &error, NULL);
-   if (v == NULL) return -1;
-   limit = v->channels * 4096;
-   *channels = v->channels;
-   offset = data_len = 0;
-   total = limit;
-   data = (short *) malloc(total * sizeof(*data));
-   if (data == NULL) {
-      stb_vorbis_close(v);
-      return -2;
-   }
-   for (;;) {
-      int n = stb_vorbis_get_frame_short_interleaved(v, v->channels, data+offset, total-offset);
-      if (n == 0) break;
-      data_len += n;
-      offset += n * v->channels;
-      if (offset + limit > total) {
-         short *data2;
-         total *= 2;
-         data2 = (short *) realloc(data, total * sizeof(*data));
-         if (data2 == NULL) {
-            free(data);
-            stb_vorbis_close(v);
-            return -2;
-         }
-         data = data2;
-      }
-   }
-   *output = data;
-   return data_len;
-}
-#endif // NO_STDIO
-
-int stb_vorbis_decode_memory(uint8 *mem, int len, int *channels, short **output)
-{
-   int data_len, offset, total, limit, error;
-   short *data;
-   stb_vorbis *v = stb_vorbis_open_memory(mem, len, &error, NULL);
-   if (v == NULL) return -1;
-   limit = v->channels * 4096;
-   *channels = v->channels;
-   offset = data_len = 0;
-   total = limit;
-   data = (short *) malloc(total * sizeof(*data));
-   if (data == NULL) {
-      stb_vorbis_close(v);
-      return -2;
-   }
-   for (;;) {
-      int n = stb_vorbis_get_frame_short_interleaved(v, v->channels, data+offset, total-offset);
-      if (n == 0) break;
-      data_len += n;
-      offset += n * v->channels;
-      if (offset + limit > total) {
-         short *data2;
-         total *= 2;
-         data2 = (short *) realloc(data, total * sizeof(*data));
-         if (data2 == NULL) {
-            free(data);
-            stb_vorbis_close(v);
-            return -2;
-         }
-         data = data2;
-      }
-   }
-   *output = data;
-   return data_len;
-}
-#endif
-
-int stb_vorbis_get_samples_float_interleaved(stb_vorbis *f, int channels, float *buffer, int num_floats)
-{
-   float **outputs;
-   int len = num_floats / channels;
-   int n=0;
-   int z = f->channels;
-   if (z > channels) z = channels;
-   while (n < len) {
-      int i,j;
-      int k = f->channel_buffer_end - f->channel_buffer_start;
-      if (n+k >= len) k = len - n;
-      for (j=0; j < k; ++j) {
-         for (i=0; i < z; ++i)
-            *buffer++ = f->channel_buffers[i][f->channel_buffer_start+j];
-         for (   ; i < channels; ++i)
-            *buffer++ = 0;
-      }
-      n += k;
-      f->channel_buffer_start += k;
-      if (n == len) break;
-      if (!stb_vorbis_get_frame_float(f, NULL, &outputs)) break;
-   }
-   return n;
-}
-
-int stb_vorbis_get_samples_float(stb_vorbis *f, int channels, float **buffer, int num_samples)
-{
-   float **outputs;
-   int n=0;
-   int z = f->channels;
-   if (z > channels) z = channels;
-   while (n < num_samples) {
-      int i;
-      int k = f->channel_buffer_end - f->channel_buffer_start;
-      if (n+k >= num_samples) k = num_samples - n;
-      if (k) {
-         for (i=0; i < z; ++i)
-            memcpy(buffer[i]+n, f->channel_buffers+f->channel_buffer_start, sizeof(float)*k);
-         for (   ; i < channels; ++i)
-            memset(buffer[i]+n, 0, sizeof(float) * k);
-      }
-      n += k;
-      f->channel_buffer_start += k;
-      if (n == num_samples) break;
-      if (!stb_vorbis_get_frame_float(f, NULL, &outputs)) break;
-   }
-   return n;
-}
-#endif // STB_VORBIS_NO_PULLDATA_API
+#include "stb_vorbis.h"
+
+#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
+
+// STB_VORBIS_CODEBOOK_SHORTS
+//     The vorbis file format encodes VQ codebook floats as ax+b where a and
+//     b are floating point per-codebook constants, and x is a 16-bit int.
+//     Normally, stb_vorbis decodes them to floats rather than leaving them
+//     as 16-bit ints and computing ax+b while decoding. This is a speed/space
+//     tradeoff; you can save space by defining this flag.
+#ifndef STB_VORBIS_CODEBOOK_SHORTS
+#define STB_VORBIS_CODEBOOK_FLOATS
+#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>
+#if !(defined(__APPLE__) || defined(MACOSX) || defined(macintosh) || defined(Macintosh))
+#include <malloc.h>
+#endif
+#else
+#define NULL 0
+#endif
+
+#ifndef _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
+
+
+#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
+
+#ifdef STB_VORBIS_CODEBOOK_FLOATS
+typedef float codetype;
+#else
+typedef uint16 codetype;
+#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 after_previous_page_start;
+   uint32 first_decoded_sample;
+   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;
+
+  // 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;
+
+   uint32 first_audio_page_offset;
+
+   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;
+};
+
+extern int my_prof(int slot);
+//#define stb_prof my_prof
+
+#ifndef stb_prof
+#define stb_prof(x)  ((void) 0)
+#endif
+
+#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))
+#ifdef dealloca
+#define temp_free(f,p)                  (f->alloc.alloc_buffer ? 0 : dealloca(size))
+#else
+#define temp_free(f,p)                  0
+#endif
+#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+3) & ~3;
+   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 not a stack
+   free(p);
+}
+
+static void *setup_temp_malloc(vorb *f, int sz)
+{
+   sz = (sz+3) & ~3;
+   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+3)&~3;
+      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=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 };
+
+   // 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 if (n < (1 << 31)) return 30 + log2_4[n >> 30];
+                  else                return 0; // signed n returns 0
+}
+
+#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, 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; }
+   // 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] = 1 << (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;
+      // 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) { assert(0); return FALSE; }
+      res = available[z];
+      available[z] = 0;
+      add_entry(c, bit_reverse(res), i, m++, len[i], values);
+      // propogate 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
+   assert(pow((float) r+1, dim) > entries);
+   assert((int) floor(pow((float) r, dim)) <= entries); // (int),floor() as above
+   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,y;
+} Point;
+
+static int STBV_CDECL point_compare(const void *p, const void *q)
+{
+   Point *a = (Point *) p;
+   Point *b = (Point *) 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 += 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;
+   // 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;
+      ProbedPage p;
+      len = 0;
+      for (i=0; i < f->segment_count; ++i)
+         len += f->segments[i];
+      len += 27 + f->segment_count;
+      p.page_start = f->first_audio_page_offset;
+      p.page_end = p.page_start + len;
+      p.after_previous_page_start = p.page_start;
+      p.first_decoded_sample = 0;
+      p.last_decoded_sample = loc0;
+      f->p_first = p;
+   }
+   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 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;
+      }
+   }
+   if (f->valid_bits < 0) return 0;
+   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 += 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);
+
+   assert(c->sorted_codewords || c->codewords);
+   // 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
+#ifndef STB_VORBIS_CODEBOOK_FLOATS
+   #define CODEBOOK_ELEMENT(c,off)          (c->multiplicands[off] * c->delta_value + c->minimum_value)
+   #define CODEBOOK_ELEMENT_FAST(c,off)     (c->multiplicands[off] * c->delta_value)
+   #define CODEBOOK_ELEMENT_BASE(c)         (c->minimum_value)
+#else
+   #define CODEBOOK_ELEMENT(c,off)          (c->multiplicands[off])
+   #define CODEBOOK_ELEMENT_FAST(c,off)     (c->multiplicands[off])
+   #define CODEBOOK_ELEMENT_BASE(c)         (0)
+#endif
+
+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;
+}
+
+#ifndef STB_VORBIS_DIVIDES_IN_CODEBOOK
+static int codebook_decode_deinterleave_repeat_2(vorb *f, Codebook *c, float **outputs, 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);
+
+      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*2 + effective > len * 2) {
+         effective = len*2 - (p_inter*2 - c_inter);
+      }
+
+      {
+         z *= c->dimensions;
+         stb_prof(11);
+         if (c->sequence_p) {
+            // haven't optimized this case because I don't have any examples
+            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 == 2) { c_inter = 0; ++p_inter; }
+               last = val;
+            }
+         } else {
+            i=0;
+            if (c_inter == 1) {
+               float val = CODEBOOK_ELEMENT_FAST(c,z+i) + last;
+               if (outputs[c_inter])
+                  outputs[c_inter][p_inter] += val;
+               c_inter = 0; ++p_inter;
+               ++i;
+            }
+            {
+               float *z0 = outputs[0];
+               float *z1 = outputs[1];
+               for (; i+1 < effective;) {
+                  float v0 = CODEBOOK_ELEMENT_FAST(c,z+i) + last;
+                  float v1 = CODEBOOK_ELEMENT_FAST(c,z+i+1) + last;
+                  if (z0)
+                     z0[p_inter] += v0;
+                  if (z1)
+                     z1[p_inter] += v1;
+                  ++p_inter;
+                  i += 2;
+               }
+            }
+            if (i < effective) {
+               float val = CODEBOOK_ELEMENT_FAST(c,z+i) + last;
+               if (outputs[c_inter])
+                  outputs[c_inter][p_inter] += val;
+               if (++c_inter == 2) { c_inter = 0; ++p_inter; }
+            }
+         }
+      }
+
+      total_decode -= effective;
+   }
+   *c_inter_p = c_inter;
+   *p_inter_p = p_inter;
+   return TRUE;
+}
+#endif
+
+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;
+   LINE_OP(output[x], inverse_db_table[y]);
+   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]);
+   }
+}
+
+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;
+}
+
+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;
+   int n_read = r->end - 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
+
+   stb_prof(2);
+   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;
+
+      stb_prof(3);
+      for (pass=0; pass < 8; ++pass) {
+         int pcount = 0, class_set = 0;
+         if (ch == 2) {
+            stb_prof(13);
+            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
+               }
+               stb_prof(5);
+               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;
+                     stb_prof(20);  // accounts for X time
+                     #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_2(f, book, residue_buffers, &c_inter, &p_inter, n, r->part_size))
+                        goto done;
+                     #endif
+                     stb_prof(7);
+                  } else {
+                     z += r->part_size;
+                     c_inter = z & 1;
+                     p_inter = z >> 1;
+                  }
+               }
+               stb_prof(8);
+               #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE
+               ++class_set;
+               #endif
+            }
+         } else if (ch == 1) {
+            while (pcount < part_read) {
+               int z = r->begin + pcount*r->part_size;
+               int c_inter = 0, p_inter = z;
+               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;
+                     stb_prof(22);
+                     if (!codebook_decode_deinterleave_repeat(f, book, residue_buffers, ch, &c_inter, &p_inter, n, r->part_size))
+                        goto done;
+                     stb_prof(3);
+                  } else {
+                     z += r->part_size;
+                     c_inter = 0;
+                     p_inter = z;
+                  }
+               }
+               #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE
+               ++class_set;
+               #endif
+            }
+         } else {
+            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;
+                     stb_prof(22);
+                     if (!codebook_decode_deinterleave_repeat(f, book, residue_buffers, ch, &c_inter, &p_inter, n, r->part_size))
+                        goto done;
+                     stb_prof(3);
+                  } else {
+                     z += r->part_size;
+                     c_inter = z % ch;
+                     p_inter = z / ch;
+                  }
+               }
+               #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE
+               ++class_set;
+               #endif
+            }
+         }
+      }
+      goto done;
+   }
+   stb_prof(9);
+
+   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:
+   stb_prof(0);
+   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;
+
+      k00   = z[-0] - z[-8];
+      k11   = z[-1] - z[-9];
+      z[-0] = z[-0] + z[-8];
+      z[-1] = z[-1] + z[-9];
+      z[-8] =  k00;
+      z[-9] =  k11 ;
+
+      k00    = z[ -2] - z[-10];
+      k11    = z[ -3] - z[-11];
+      z[ -2] = z[ -2] + z[-10];
+      z[ -3] = z[ -3] + z[-11];
+      z[-10] = (k00+k11) * A2;
+      z[-11] = (k11-k00) * A2;
+
+      k00    = z[-12] - z[ -4];  // reverse to avoid a unary negation
+      k11    = z[ -5] - z[-13];
+      z[ -4] = z[ -4] + z[-12];
+      z[ -5] = z[ -5] + z[-13];
+      z[-12] = k11;
+      z[-13] = k00;
+
+      k00    = z[-14] - z[ -6];  // reverse to avoid a unary negation
+      k11    = z[ -7] - z[-15];
+      z[ -6] = z[ -6] + z[-14];
+      z[ -7] = z[ -7] + z[-15];
+      z[-14] = (k00+k11) * A2;
+      z[-15] = (k00-k11) * 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 propogates 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_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];
+   assert(0);
+   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
+         if (finalY[j] >= 0)
+         #else
+         if (step2_flag[j])
+         #endif
+         {
+            int hy = finalY[j] * g->floor1_multiplier;
+            int hx = g->Xlist[j];
+            draw_line(target, lx,ly, hx,hy, n2);
+            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]);
+   }
+   return TRUE;
+}
+
+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
+
+   n = f->blocksize[m->blockflag];
+   map = &f->mapping[m->mapping];
+
+// FLOORS
+   n2 = n >> 1;
+
+   stb_prof(1);
+   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;
+         }
+         // So we just defer everything else to later
+
+         // at this point we've decoded the floor into buffer
+      }
+   }
+   stb_prof(0);
+   // at this point we've decoded all floors
+
+   if (f->alloc.alloc_buffer)
+      assert(f->alloc.alloc_buffer_length_in_bytes == f->temp_offset);
+
+   // re-enable coupled channels if necessary
+   memcpy(really_zero_channel, zero_channel, sizeof(really_zero_channel[0]) * f->channels);
+   for (i=0; i < map->coupling_steps; ++i)
+      if (!zero_channel[map->chan[i].magnitude] || !zero_channel[map->chan[i].angle]) {
+         zero_channel[map->chan[i].magnitude] = zero_channel[map->chan[i].angle] = FALSE;
+      }
+
+// RESIDUE DECODE
+   for (i=0; i < map->submaps; ++i) {
+      float *residue_buffers[STB_VORBIS_MAX_CHANNELS];
+      int r;
+      uint8 do_not_decode[256];
+      int ch = 0;
+      for (j=0; j < f->channels; ++j) {
+         if (map->chan[j].mux == i) {
+            if (zero_channel[j]) {
+               do_not_decode[ch] = TRUE;
+               residue_buffers[ch] = NULL;
+            } else {
+               do_not_decode[ch] = FALSE;
+               residue_buffers[ch] = f->channel_buffers[j];
+            }
+            ++ch;
+         }
+      }
+      r = map->submap_residue[i];
+      decode_residue(f, residue_buffers, ch, n2, r, do_not_decode);
+   }
+
+   if (f->alloc.alloc_buffer)
+      assert(f->alloc.alloc_buffer_length_in_bytes == f->temp_offset);
+
+// INVERSE COUPLING
+   stb_prof(14);
+   for (i = map->coupling_steps-1; i >= 0; --i) {
+      int n2 = n >> 1;
+      float *m = f->channel_buffers[map->chan[i].magnitude];
+      float *a = f->channel_buffers[map->chan[i].angle    ];
+      for (j=0; j < n2; ++j) {
+         float a2,m2;
+         if (m[j] > 0)
+            if (a[j] > 0)
+               m2 = m[j], a2 = m[j] - a[j];
+            else
+               a2 = m[j], m2 = m[j] + a[j];
+         else
+            if (a[j] > 0)
+               m2 = m[j], a2 = m[j] + a[j];
+            else
+               a2 = m[j], m2 = m[j] - a[j];
+         m[j] = m2;
+         a[j] = a2;
+      }
+   }
+
+   // finish decoding the floors
+#ifndef STB_VORBIS_NO_DEFER_FLOOR
+   stb_prof(15);
+   for (i=0; i < f->channels; ++i) {
+      if (really_zero_channel[i]) {
+         memset(f->channel_buffers[i], 0, sizeof(*f->channel_buffers[i]) * n2);
+      } else {
+         do_floor(f, map, i, n, f->channel_buffers[i], f->finalY[i], NULL);
+      }
+   }
+#else
+   for (i=0; i < f->channels; ++i) {
+      if (really_zero_channel[i]) {
+         memset(f->channel_buffers[i], 0, sizeof(*f->channel_buffers[i]) * n2);
+      } else {
+         for (j=0; j < n2; ++j)
+            f->channel_buffers[i][j] *= f->floor_buffers[i][j];
+      }
+   }
+#endif
+
+// INVERSE MDCT
+   stb_prof(16);
+   for (i=0; i < f->channels; ++i)
+      inverse_mdct(f->channel_buffers[i], n, f, m->blockflag);
+   stb_prof(0);
+
+   // this shouldn't be necessary, unless we exited on an error
+   // and want to flush to get to the next packet
+   flush_packet(f);
+
+   if (f->first_decode) {
+      // assume we start so first non-discarded sample is sample 0
+      // this isn't to spec, but spec would require us to read ahead
+      // and decode the size of all current frames--could be done,
+      // but presumably it's not a commonly used feature
+      f->current_loc = -n2; // start of first frame is positioned for discard
+      // we might have to discard samples "from" the next frame too,
+      // if we're lapping a large block then a small at the start?
+      f->discard_samples_deferred = n - right_end;
+      f->current_loc_valid = TRUE;
+      f->first_decode = FALSE;
+   } else if (f->discard_samples_deferred) {
+      left_start += f->discard_samples_deferred;
+      *p_left = left_start;
+      f->discard_samples_deferred = 0;
+   } else if (f->previous_length == 0 && f->current_loc_valid) {
+      // we're recovering from a seek... that means we're going to discard
+      // the samples from this packet even though we know our position from
+      // the last page header, so we need to update the position based on
+      // the discarded samples here
+      // but wait, the code below is going to add this in itself even
+      // on a discard, so we don't need to do it here...
+   }
+
+   // check if we have ogg information about the sample # for this packet
+   if (f->last_seg_which == f->end_seg_with_known_loc) {
+      // if we have a valid current loc, and this is final:
+      if (f->current_loc_valid && (f->page_flag & PAGEFLAG_last_page)) {
+         uint32 current_end = f->known_loc_for_packet - (n-right_end);
+         // then let's infer the size of the (probably) short final frame
+         if (current_end < f->current_loc + right_end) {
+            if (current_end < f->current_loc) {
+               // negative truncation, that's impossible!
+               *len = 0;
+            } else {
+               *len = current_end - f->current_loc;
+            }
+            *len += left_start;
+            f->current_loc += *len;
+            return TRUE;
+         }
+      }
+      // otherwise, just set our sample loc
+      // guess that the ogg granule pos refers to the _middle_ of the
+      // last frame?
+      // set f->current_loc to the position of left_start
+      f->current_loc = f->known_loc_for_packet - (n2-left_start);
+      f->current_loc_valid = TRUE;
+   }
+   if (f->current_loc_valid)
+      f->current_loc += (right_start - left_start);
+
+   if (f->alloc.alloc_buffer)
+      assert(f->alloc.alloc_buffer_length_in_bytes == f->temp_offset);
+   *len = right_end;  // ignore samples after the window goes to 0
+   return TRUE;
+}
+
+static int vorbis_decode_packet(vorb *f, int *len, int *p_left, int *p_right)
+{
+   int mode, left_end, right_end;
+   if (!vorbis_decode_initial(f, p_left, &left_end, p_right, &right_end, &mode)) return 0;
+   return vorbis_decode_packet_rest(f, len, f->mode_config + mode, *p_left, left_end, *p_right, right_end, p_left);
+}
+
+static int vorbis_finish_frame(stb_vorbis *f, int len, int left, int right)
+{
+   int prev,i,j;
+   // we use right&left (the start of the right- and left-window sin()-regions)
+   // to determine how much to return, rather than inferring from the rules
+   // (same result, clearer code); 'left' indicates where our sin() window
+   // starts, therefore where the previous window's right edge starts, and
+   // therefore where to start mixing from the previous buffer. 'right'
+   // indicates where our sin() ending-window starts, therefore that's where
+   // we start saving, and where our returned-data ends.
+
+   // mixin from previous window
+   if (f->previous_length) {
+      int i,j, n = f->previous_length;
+      float *w = get_window(f, n);
+      for (i=0; i < f->channels; ++i) {
+         for (j=0; j < n; ++j)
+            f->channel_buffers[i][left+j] =
+               f->channel_buffers[i][left+j]*w[    j] +
+               f->previous_window[i][     j]*w[n-1-j];
+      }
+   }
+
+   prev = f->previous_length;
+
+   // last half of this data becomes previous window
+   f->previous_length = len - right;
+
+   // @OPTIMIZE: could avoid this copy by double-buffering the
+   // output (flipping previous_window with channel_buffers), but
+   // then previous_window would have to be 2x as large, and
+   // channel_buffers couldn't be temp mem (although they're NOT
+   // currently temp mem, they could be (unless we want to level
+   // performance by spreading out the computation))
+   for (i=0; i < f->channels; ++i)
+      for (j=0; right+j < len; ++j)
+         f->previous_window[i][j] = f->channel_buffers[i][right+j];
+
+   if (!prev)
+      // there was no previous packet, so this data isn't valid...
+      // this isn't entirely true, only the would-have-overlapped data
+      // isn't valid, but this seems to be what the spec requires
+      return 0;
+
+   // truncate a short frame
+   if (len < right) right = len;
+
+   f->samples_output += right-left;
+
+   return right - left;
+}
+
+static void vorbis_pump_first_frame(stb_vorbis *f)
+{
+   int len, right, left;
+   if (vorbis_decode_packet(f, &len, &left, &right))
+      vorbis_finish_frame(f, len, left, right);
+}
+
+#ifndef STB_VORBIS_NO_PUSHDATA_API
+static int is_whole_packet_present(stb_vorbis *f, int end_page)
+{
+   // make sure that we have the packet available before continuing...
+   // this requires a full ogg parse, but we know we can fetch from f->stream
+
+   // instead of coding this out explicitly, we could save the current read state,
+   // read the next packet with get8() until end-of-packet, check f->eof, then
+   // reset the state? but that would be slower, esp. since we'd have over 256 bytes
+   // of state to restore (primarily the page segment table)
+
+   int s = f->next_seg, first = TRUE;
+   uint8 *p = f->stream;
+
+   if (s != -1) { // if we're not starting the packet with a 'continue on next page' flag
+      for (; s < f->segment_count; ++s) {
+         p += f->segments[s];
+         if (f->segments[s] < 255)               // stop at first short segment
+            break;
+      }
+      // either this continues, or it ends it...
+      if (end_page)
+         if (s < f->segment_count-1)             return error(f, VORBIS_invalid_stream);
+      if (s == f->segment_count)
+         s = -1; // set 'crosses page' flag
+      if (p > f->stream_end)                     return error(f, VORBIS_need_more_data);
+      first = FALSE;
+   }
+   for (; s == -1;) {
+      uint8 *q; 
+      int n;
+
+      // check that we have the page header ready
+      if (p + 26 >= f->stream_end)               return error(f, VORBIS_need_more_data);
+      // validate the page
+      if (memcmp(p, ogg_page_header, 4))         return error(f, VORBIS_invalid_stream);
+      if (p[4] != 0)                             return error(f, VORBIS_invalid_stream);
+      if (first) { // the first segment must NOT have 'continued_packet', later ones MUST
+         if (f->previous_length)
+            if ((p[5] & PAGEFLAG_continued_packet))  return error(f, VORBIS_invalid_stream);
+         // if no previous length, we're resynching, so we can come in on a continued-packet,
+         // which we'll just drop
+      } else {
+         if (!(p[5] & PAGEFLAG_continued_packet)) return error(f, VORBIS_invalid_stream);
+      }
+      n = p[26]; // segment counts
+      q = p+27;  // q points to segment table
+      p = q + n; // advance past header
+      // make sure we've read the segment table
+      if (p > f->stream_end)                     return error(f, VORBIS_need_more_data);
+      for (s=0; s < n; ++s) {
+         p += q[s];
+         if (q[s] < 255)
+            break;
+      }
+      if (end_page)
+         if (s < n-1)                            return error(f, VORBIS_invalid_stream);
+      if (s == n)
+         s = -1; // set 'crosses page' flag
+      if (p > f->stream_end)                     return error(f, VORBIS_need_more_data);
+      first = FALSE;
+   }
+   return TRUE;
+}
+#endif // !STB_VORBIS_NO_PUSHDATA_API
+
+static int start_decoder(vorb *f)
+{
+   uint8 header[6], x,y;
+   int len,i,j,k, max_submaps = 0;
+   int longest_floorlist=0;
+
+   // first page, first packet
+
+   if (!start_page(f))                              return FALSE;
+   // validate page flag
+   if (!(f->page_flag & PAGEFLAG_first_page))       return error(f, VORBIS_invalid_first_page);
+   if (f->page_flag & PAGEFLAG_last_page)           return error(f, VORBIS_invalid_first_page);
+   if (f->page_flag & PAGEFLAG_continued_packet)    return error(f, VORBIS_invalid_first_page);
+   // check for expected packet length
+   if (f->segment_count != 1)                       return error(f, VORBIS_invalid_first_page);
+   if (f->segments[0] != 30)                        return error(f, VORBIS_invalid_first_page);
+   // read packet
+   // check packet header
+   if (get8(f) != VORBIS_packet_id)                 return error(f, VORBIS_invalid_first_page);
+   if (!getn(f, header, 6))                         return error(f, VORBIS_unexpected_eof);
+   if (!vorbis_validate(header))                    return error(f, VORBIS_invalid_first_page);
+   // vorbis_version
+   if (get32(f) != 0)                               return error(f, VORBIS_invalid_first_page);
+   f->channels = get8(f); if (!f->channels)         return error(f, VORBIS_invalid_first_page);
+   if (f->channels > STB_VORBIS_MAX_CHANNELS)       return error(f, VORBIS_too_many_channels);
+   f->sample_rate = get32(f); if (!f->sample_rate)  return error(f, VORBIS_invalid_first_page);
+   get32(f); // bitrate_maximum
+   get32(f); // bitrate_nominal
+   get32(f); // bitrate_minimum
+   x = get8(f);
+   { int log0,log1;
+   log0 = x & 15;
+   log1 = x >> 4;
+   f->blocksize_0 = 1 << log0;
+   f->blocksize_1 = 1 << log1;
+   if (log0 < 6 || log0 > 13)                       return error(f, VORBIS_invalid_setup);
+   if (log1 < 6 || log1 > 13)                       return error(f, VORBIS_invalid_setup);
+   if (log0 > log1)                                 return error(f, VORBIS_invalid_setup);
+   }
+
+   // framing_flag
+   x = get8(f);
+   if (!(x & 1))                                    return error(f, VORBIS_invalid_first_page);
+
+   // second packet!
+   if (!start_page(f))                              return FALSE;
+
+   if (!start_packet(f))                            return FALSE;
+   do {
+      len = next_segment(f);
+      skip(f, len);
+      f->bytes_in_seg = 0;
+   } while (len);
+
+   // third packet!
+   if (!start_packet(f))                            return FALSE;
+
+   #ifndef STB_VORBIS_NO_PUSHDATA_API
+   if (IS_PUSH_MODE(f)) {
+      if (!is_whole_packet_present(f, TRUE)) {
+         // convert error in ogg header to write type
+         if (f->error == VORBIS_invalid_stream)
+            f->error = VORBIS_invalid_setup;
+         return FALSE;
+      }
+   }
+   #endif
+
+   crc32_init(); // always init it, to avoid multithread race conditions
+
+   if (get8_packet(f) != VORBIS_packet_setup)       return error(f, VORBIS_invalid_setup);
+   for (i=0; i < 6; ++i) header[i] = get8_packet(f);
+   if (!vorbis_validate(header))                    return error(f, VORBIS_invalid_setup);
+
+   // codebooks
+
+   f->codebook_count = get_bits(f,8) + 1;
+   f->codebooks = (Codebook *) setup_malloc(f, sizeof(*f->codebooks) * f->codebook_count);
+   if (f->codebooks == NULL)                        return error(f, VORBIS_outofmem);
+   memset(f->codebooks, 0, sizeof(*f->codebooks) * f->codebook_count);
+   for (i=0; i < f->codebook_count; ++i) {
+      uint32 *values;
+      int ordered, sorted_count;
+      int total=0;
+      uint8 *lengths;
+      Codebook *c = f->codebooks+i;
+      x = get_bits(f, 8); if (x != 0x42)            return error(f, VORBIS_invalid_setup);
+      x = get_bits(f, 8); if (x != 0x43)            return error(f, VORBIS_invalid_setup);
+      x = get_bits(f, 8); if (x != 0x56)            return error(f, VORBIS_invalid_setup);
+      x = get_bits(f, 8);
+      c->dimensions = (get_bits(f, 8)<<8) + x;
+      x = get_bits(f, 8);
+      y = get_bits(f, 8);
+      c->entries = (get_bits(f, 8)<<16) + (y<<8) + x;
+      ordered = get_bits(f,1);
+      c->sparse = ordered ? 0 : get_bits(f,1);
+
+      if (c->sparse)
+         lengths = (uint8 *) setup_temp_malloc(f, c->entries);
+      else
+         lengths = c->codeword_lengths = (uint8 *) setup_malloc(f, c->entries);
+
+      if (!lengths) return error(f, VORBIS_outofmem);
+
+      if (ordered) {
+         int current_entry = 0;
+         int current_length = get_bits(f,5) + 1;
+         while (current_entry < c->entries) {
+            int limit = c->entries - current_entry;
+            int n = get_bits(f, ilog(limit));
+            if (current_entry + n > (int) c->entries) { return error(f, VORBIS_invalid_setup); }
+            memset(lengths + current_entry, current_length, n);
+            current_entry += n;
+            ++current_length;
+         }
+      } else {
+         for (j=0; j < c->entries; ++j) {
+            int present = c->sparse ? get_bits(f,1) : 1;
+            if (present) {
+               lengths[j] = get_bits(f, 5) + 1;
+               ++total;
+            } else {
+               lengths[j] = NO_CODE;
+            }
+         }
+      }
+
+      if (c->sparse && total >= c->entries >> 2) {
+         // convert sparse items to non-sparse!
+         if (c->entries > (int) f->setup_temp_memory_required)
+            f->setup_temp_memory_required = c->entries;
+
+         c->codeword_lengths = (uint8 *) setup_malloc(f, c->entries);
+         memcpy(c->codeword_lengths, lengths, c->entries);
+         setup_temp_free(f, lengths, c->entries); // note this is only safe if there have been no intervening temp mallocs!
+         lengths = c->codeword_lengths;
+         c->sparse = 0;
+      }
+
+      // compute the size of the sorted tables
+      if (c->sparse) {
+         sorted_count = total;
+      } else {
+         sorted_count = 0;
+         #ifndef STB_VORBIS_NO_HUFFMAN_BINARY_SEARCH
+         for (j=0; j < c->entries; ++j)
+            if (lengths[j] > STB_VORBIS_FAST_HUFFMAN_LENGTH && lengths[j] != NO_CODE)
+               ++sorted_count;
+         #endif
+      }
+
+      c->sorted_entries = sorted_count;
+      values = NULL;
+
+      if (!c->sparse) {
+         c->codewords = (uint32 *) setup_malloc(f, sizeof(c->codewords[0]) * c->entries);
+         if (!c->codewords)                  return error(f, VORBIS_outofmem);
+      } else {
+         unsigned int size;
+         if (c->sorted_entries) {
+            c->codeword_lengths = (uint8 *) setup_malloc(f, c->sorted_entries);
+            if (!c->codeword_lengths)           return error(f, VORBIS_outofmem);
+            c->codewords = (uint32 *) setup_temp_malloc(f, sizeof(*c->codewords) * c->sorted_entries);
+            if (!c->codewords)                  return error(f, VORBIS_outofmem);
+            values = (uint32 *) setup_temp_malloc(f, sizeof(*values) * c->sorted_entries);
+            if (!values)                        return error(f, VORBIS_outofmem);
+         }
+         size = c->entries + (sizeof(*c->codewords) + sizeof(*values)) * c->sorted_entries;
+         if (size > f->setup_temp_memory_required)
+            f->setup_temp_memory_required = size;
+      }
+
+      if (!compute_codewords(c, lengths, c->entries, values)) {
+         if (c->sparse) setup_temp_free(f, values, 0);
+         return error(f, VORBIS_invalid_setup);
+      }
+
+      if (c->sorted_entries) {
+         // allocate an extra slot for sentinels
+         c->sorted_codewords = (uint32 *) setup_malloc(f, sizeof(*c->sorted_codewords) * (c->sorted_entries+1));
+         // allocate an extra slot at the front so that c->sorted_values[-1] is defined
+         // so that we can catch that case without an extra if
+         c->sorted_values    = ( int   *) setup_malloc(f, sizeof(*c->sorted_values   ) * (c->sorted_entries+1));
+         if (c->sorted_values) { ++c->sorted_values; c->sorted_values[-1] = -1; }
+         compute_sorted_huffman(c, lengths, values);
+      }
+
+      if (c->sparse) {
+         setup_temp_free(f, values, sizeof(*values)*c->sorted_entries);
+         setup_temp_free(f, c->codewords, sizeof(*c->codewords)*c->sorted_entries);
+         setup_temp_free(f, lengths, c->entries);
+         c->codewords = NULL;
+      }
+
+      compute_accelerated_huffman(c);
+
+      c->lookup_type = get_bits(f, 4);
+      if (c->lookup_type > 2) return error(f, VORBIS_invalid_setup);
+      if (c->lookup_type > 0) {
+         uint16 *mults;
+         c->minimum_value = float32_unpack(get_bits(f, 32));
+         c->delta_value = float32_unpack(get_bits(f, 32));
+         c->value_bits = get_bits(f, 4)+1;
+         c->sequence_p = get_bits(f,1);
+         if (c->lookup_type == 1) {
+            c->lookup_values = lookup1_values(c->entries, c->dimensions);
+         } else {
+            c->lookup_values = c->entries * c->dimensions;
+         }
+         mults = (uint16 *) setup_temp_malloc(f, sizeof(mults[0]) * c->lookup_values);
+         if (mults == NULL) return error(f, VORBIS_outofmem);
+         for (j=0; j < (int) c->lookup_values; ++j) {
+            int q = get_bits(f, c->value_bits);
+            if (q == EOP) { setup_temp_free(f,mults,sizeof(mults[0])*c->lookup_values); return error(f, VORBIS_invalid_setup); }
+            mults[j] = q;
+         }
+
+#ifndef STB_VORBIS_DIVIDES_IN_CODEBOOK
+         if (c->lookup_type == 1) {
+            int len, sparse = c->sparse;
+            // pre-expand the lookup1-style multiplicands, to avoid a divide in the inner loop
+            if (sparse) {
+               if (c->sorted_entries == 0) goto skip;
+               c->multiplicands = (codetype *) setup_malloc(f, sizeof(c->multiplicands[0]) * c->sorted_entries * c->dimensions);
+            } else
+               c->multiplicands = (codetype *) setup_malloc(f, sizeof(c->multiplicands[0]) * c->entries        * c->dimensions);
+            if (c->multiplicands == NULL) { setup_temp_free(f,mults,sizeof(mults[0])*c->lookup_values); return error(f, VORBIS_outofmem); }
+            len = sparse ? c->sorted_entries : c->entries;
+            for (j=0; j < len; ++j) {
+               int z = sparse ? c->sorted_values[j] : j, div=1;
+               for (k=0; k < c->dimensions; ++k) {
+                  int off = (z / div) % c->lookup_values;
+                  c->multiplicands[j*c->dimensions + k] =
+                         #ifndef STB_VORBIS_CODEBOOK_FLOATS
+                            mults[off];
+                         #else
+                            mults[off]*c->delta_value + c->minimum_value;
+                            // in this case (and this case only) we could pre-expand c->sequence_p,
+                            // and throw away the decode logic for it; have to ALSO do
+                            // it in the case below, but it can only be done if
+                            //    STB_VORBIS_CODEBOOK_FLOATS
+                            //   !STB_VORBIS_DIVIDES_IN_CODEBOOK
+                         #endif
+                  div *= c->lookup_values;
+               }
+            }
+            setup_temp_free(f, mults,sizeof(mults[0])*c->lookup_values);
+            c->lookup_type = 2;
+         }
+         else
+#endif
+         {
+            c->multiplicands = (codetype *) setup_malloc(f, sizeof(c->multiplicands[0]) * c->lookup_values);
+            #ifndef STB_VORBIS_CODEBOOK_FLOATS
+            memcpy(c->multiplicands, mults, sizeof(c->multiplicands[0]) * c->lookup_values);
+            #else
+            for (j=0; j < (int) c->lookup_values; ++j)
+               c->multiplicands[j] = mults[j] * c->delta_value + c->minimum_value;
+            #endif
+            setup_temp_free(f, mults,sizeof(mults[0])*c->lookup_values);
+         }
+#ifndef STB_VORBIS_DIVIDES_IN_CODEBOOK
+        skip:;
+#endif
+
+         #ifdef STB_VORBIS_CODEBOOK_FLOATS
+         if (c->lookup_type == 2 && c->sequence_p) {
+            for (j=1; j < (int) c->lookup_values; ++j)
+               c->multiplicands[j] = c->multiplicands[j-1];
+            c->sequence_p = 0;
+         }
+         #endif
+      }
+   }
+
+   // time domain transfers (notused)
+
+   x = get_bits(f, 6) + 1;
+   for (i=0; i < x; ++i) {
+      uint32 z = get_bits(f, 16);
+      if (z != 0) return error(f, VORBIS_invalid_setup);
+   }
+
+   // Floors
+   f->floor_count = get_bits(f, 6)+1;
+   f->floor_config = (Floor *)  setup_malloc(f, f->floor_count * sizeof(*f->floor_config));
+   for (i=0; i < f->floor_count; ++i) {
+      f->floor_types[i] = get_bits(f, 16);
+      if (f->floor_types[i] > 1) return error(f, VORBIS_invalid_setup);
+      if (f->floor_types[i] == 0) {
+         Floor0 *g = &f->floor_config[i].floor0;
+         g->order = get_bits(f,8);
+         g->rate = get_bits(f,16);
+         g->bark_map_size = get_bits(f,16);
+         g->amplitude_bits = get_bits(f,6);
+         g->amplitude_offset = get_bits(f,8);
+         g->number_of_books = get_bits(f,4) + 1;
+         for (j=0; j < g->number_of_books; ++j)
+            g->book_list[j] = get_bits(f,8);
+         return error(f, VORBIS_feature_not_supported);
+      } else {
+         Point p[31*8+2];
+         Floor1 *g = &f->floor_config[i].floor1;
+         int max_class = -1; 
+         g->partitions = get_bits(f, 5);
+         for (j=0; j < g->partitions; ++j) {
+            g->partition_class_list[j] = get_bits(f, 4);
+            if (g->partition_class_list[j] > max_class)
+               max_class = g->partition_class_list[j];
+         }
+         for (j=0; j <= max_class; ++j) {
+            g->class_dimensions[j] = get_bits(f, 3)+1;
+            g->class_subclasses[j] = get_bits(f, 2);
+            if (g->class_subclasses[j]) {
+               g->class_masterbooks[j] = get_bits(f, 8);
+               if (g->class_masterbooks[j] >= f->codebook_count) return error(f, VORBIS_invalid_setup);
+            }
+            for (k=0; k < 1 << g->class_subclasses[j]; ++k) {
+               g->subclass_books[j][k] = get_bits(f,8)-1;
+               if (g->subclass_books[j][k] >= f->codebook_count) return error(f, VORBIS_invalid_setup);
+            }
+         }
+         g->floor1_multiplier = get_bits(f,2)+1;
+         g->rangebits = get_bits(f,4);
+         g->Xlist[0] = 0;
+         g->Xlist[1] = 1 << g->rangebits;
+         g->values = 2;
+         for (j=0; j < g->partitions; ++j) {
+            int c = g->partition_class_list[j];
+            for (k=0; k < g->class_dimensions[c]; ++k) {
+               g->Xlist[g->values] = get_bits(f, g->rangebits);
+               ++g->values;
+            }
+         }
+         // precompute the sorting
+         for (j=0; j < g->values; ++j) {
+            p[j].x = g->Xlist[j];
+            p[j].y = j;
+         }
+         qsort(p, g->values, sizeof(p[0]), point_compare);
+         for (j=0; j < g->values; ++j)
+            g->sorted_order[j] = (uint8) p[j].y;
+         // precompute the neighbors
+         for (j=2; j < g->values; ++j) {
+            int low,hi;
+            neighbors(g->Xlist, j, &low,&hi);
+            g->neighbors[j][0] = low;
+            g->neighbors[j][1] = hi;
+         }
+
+         if (g->values > longest_floorlist)
+            longest_floorlist = g->values;
+      }
+   }
+
+   // Residue
+   f->residue_count = get_bits(f, 6)+1;
+   f->residue_config = (Residue *) setup_malloc(f, f->residue_count * sizeof(*f->residue_config));
+   for (i=0; i < f->residue_count; ++i) {
+      uint8 residue_cascade[64];
+      Residue *r = f->residue_config+i;
+      f->residue_types[i] = get_bits(f, 16);
+      if (f->residue_types[i] > 2) return error(f, VORBIS_invalid_setup);
+      r->begin = get_bits(f, 24);
+      r->end = get_bits(f, 24);
+      r->part_size = get_bits(f,24)+1;
+      r->classifications = get_bits(f,6)+1;
+      r->classbook = get_bits(f,8);
+      for (j=0; j < r->classifications; ++j) {
+         uint8 high_bits=0;
+         uint8 low_bits=get_bits(f,3);
+         if (get_bits(f,1))
+            high_bits = get_bits(f,5);
+         residue_cascade[j] = high_bits*8 + low_bits;
+      }
+      r->residue_books = (short (*)[8]) setup_malloc(f, sizeof(r->residue_books[0]) * r->classifications);
+      for (j=0; j < r->classifications; ++j) {
+         for (k=0; k < 8; ++k) {
+            if (residue_cascade[j] & (1 << k)) {
+               r->residue_books[j][k] = get_bits(f, 8);
+               if (r->residue_books[j][k] >= f->codebook_count) return error(f, VORBIS_invalid_setup);
+            } else {
+               r->residue_books[j][k] = -1;
+            }
+         }
+      }
+      // precompute the classifications[] array to avoid inner-loop mod/divide
+      // call it 'classdata' since we already have r->classifications
+      r->classdata = (uint8 **) setup_malloc(f, sizeof(*r->classdata) * f->codebooks[r->classbook].entries);
+      if (!r->classdata) return error(f, VORBIS_outofmem);
+      memset(r->classdata, 0, sizeof(*r->classdata) * f->codebooks[r->classbook].entries);
+      for (j=0; j < f->codebooks[r->classbook].entries; ++j) {
+         int classwords = f->codebooks[r->classbook].dimensions;
+         int temp = j;
+         r->classdata[j] = (uint8 *) setup_malloc(f, sizeof(r->classdata[j][0]) * classwords);
+         for (k=classwords-1; k >= 0; --k) {
+            r->classdata[j][k] = temp % r->classifications;
+            temp /= r->classifications;
+         }
+      }
+   }
+
+   f->mapping_count = get_bits(f,6)+1;
+   f->mapping = (Mapping *) setup_malloc(f, f->mapping_count * sizeof(*f->mapping));
+   for (i=0; i < f->mapping_count; ++i) {
+      Mapping *m = f->mapping + i;      
+      int mapping_type = get_bits(f,16);
+      if (mapping_type != 0) return error(f, VORBIS_invalid_setup);
+      m->chan = (MappingChannel *) setup_malloc(f, f->channels * sizeof(*m->chan));
+      if (get_bits(f,1))
+         m->submaps = get_bits(f,4)+1;
+      else
+         m->submaps = 1;
+      if (m->submaps > max_submaps)
+         max_submaps = m->submaps;
+      if (get_bits(f,1)) {
+         m->coupling_steps = get_bits(f,8)+1;
+         for (k=0; k < m->coupling_steps; ++k) {
+            m->chan[k].magnitude = get_bits(f, ilog(f->channels-1));
+            m->chan[k].angle = get_bits(f, ilog(f->channels-1));
+            if (m->chan[k].magnitude >= f->channels)        return error(f, VORBIS_invalid_setup);
+            if (m->chan[k].angle     >= f->channels)        return error(f, VORBIS_invalid_setup);
+            if (m->chan[k].magnitude == m->chan[k].angle)   return error(f, VORBIS_invalid_setup);
+         }
+      } else
+         m->coupling_steps = 0;
+
+      // reserved field
+      if (get_bits(f,2)) return error(f, VORBIS_invalid_setup);
+      if (m->submaps > 1) {
+         for (j=0; j < f->channels; ++j) {
+            m->chan[j].mux = get_bits(f, 4);
+            if (m->chan[j].mux >= m->submaps)                return error(f, VORBIS_invalid_setup);
+         }
+      } else
+         // @SPECIFICATION: this case is missing from the spec
+         for (j=0; j < f->channels; ++j)
+            m->chan[j].mux = 0;
+
+      for (j=0; j < m->submaps; ++j) {
+         get_bits(f,8); // discard
+         m->submap_floor[j] = get_bits(f,8);
+         m->submap_residue[j] = get_bits(f,8);
+         if (m->submap_floor[j] >= f->floor_count)      return error(f, VORBIS_invalid_setup);
+         if (m->submap_residue[j] >= f->residue_count)  return error(f, VORBIS_invalid_setup);
+      }
+   }
+
+   // Modes
+   f->mode_count = get_bits(f, 6)+1;
+   for (i=0; i < f->mode_count; ++i) {
+      Mode *m = f->mode_config+i;
+      m->blockflag = get_bits(f,1);
+      m->windowtype = get_bits(f,16);
+      m->transformtype = get_bits(f,16);
+      m->mapping = get_bits(f,8);
+      if (m->windowtype != 0)                 return error(f, VORBIS_invalid_setup);
+      if (m->transformtype != 0)              return error(f, VORBIS_invalid_setup);
+      if (m->mapping >= f->mapping_count)     return error(f, VORBIS_invalid_setup);
+   }
+
+   flush_packet(f);
+
+   f->previous_length = 0;
+
+   for (i=0; i < f->channels; ++i) {
+      f->channel_buffers[i] = (float *) setup_malloc(f, sizeof(float) * f->blocksize_1);
+      f->previous_window[i] = (float *) setup_malloc(f, sizeof(float) * f->blocksize_1/2);
+      f->finalY[i]          = (int16 *) setup_malloc(f, sizeof(int16) * longest_floorlist);
+      #ifdef STB_VORBIS_NO_DEFER_FLOOR
+      f->floor_buffers[i]   = (float *) setup_malloc(f, sizeof(float) * f->blocksize_1/2);
+      #endif
+   }
+
+   if (!init_blocksize(f, 0, f->blocksize_0)) return FALSE;
+   if (!init_blocksize(f, 1, f->blocksize_1)) return FALSE;
+   f->blocksize[0] = f->blocksize_0;
+   f->blocksize[1] = f->blocksize_1;
+
+#ifdef STB_VORBIS_DIVIDE_TABLE
+   if (integer_divide_table[1][1]==0)
+      for (i=0; i < DIVTAB_NUMER; ++i)
+         for (j=1; j < DIVTAB_DENOM; ++j)
+            integer_divide_table[i][j] = i / j;
+#endif
+
+   // compute how much temporary memory is needed
+
+   // 1.
+   {
+      uint32 imdct_mem = (f->blocksize_1 * sizeof(float) >> 1);
+      uint32 classify_mem;
+      int i,max_part_read=0;
+      for (i=0; i < f->residue_count; ++i) {
+         Residue *r = f->residue_config + i;
+         int n_read = r->end - r->begin;
+         int part_read = n_read / r->part_size;
+         if (part_read > max_part_read)
+            max_part_read = part_read;
+      }
+      #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE
+      classify_mem = f->channels * (sizeof(void*) + max_part_read * sizeof(uint8 *));
+      #else
+      classify_mem = f->channels * (sizeof(void*) + max_part_read * sizeof(int *));
+      #endif
+
+      f->temp_memory_required = classify_mem;
+      if (imdct_mem > f->temp_memory_required)
+         f->temp_memory_required = imdct_mem;
+   }
+
+   f->first_decode = TRUE;
+
+   if (f->alloc.alloc_buffer) {
+      assert(f->temp_offset == f->alloc.alloc_buffer_length_in_bytes);
+      // check if there's enough temp memory so we don't error later
+      if (f->setup_offset + sizeof(*f) + f->temp_memory_required > (unsigned) f->temp_offset)
+         return error(f, VORBIS_outofmem);
+   }
+
+   f->first_audio_page_offset = stb_vorbis_get_file_offset(f);
+
+   return TRUE;
+}
+
+static void vorbis_deinit(stb_vorbis *p)
+{
+   int i,j;
+   for (i=0; i < p->residue_count; ++i) {
+      Residue *r = p->residue_config+i;
+      if (r->classdata) {
+         for (j=0; j < p->codebooks[r->classbook].entries; ++j)
+            setup_free(p, r->classdata[j]);
+         setup_free(p, r->classdata);
+      }
+      setup_free(p, r->residue_books);
+   }
+
+   if (p->codebooks) {
+      for (i=0; i < p->codebook_count; ++i) {
+         Codebook *c = p->codebooks + i;
+         setup_free(p, c->codeword_lengths);
+         setup_free(p, c->multiplicands);
+         setup_free(p, c->codewords);
+         setup_free(p, c->sorted_codewords);
+         // c->sorted_values[-1] is the first entry in the array
+         setup_free(p, c->sorted_values ? c->sorted_values-1 : NULL);
+      }
+      setup_free(p, p->codebooks);
+   }
+   setup_free(p, p->floor_config);
+   setup_free(p, p->residue_config);
+   for (i=0; i < p->mapping_count; ++i)
+      setup_free(p, p->mapping[i].chan);
+   setup_free(p, p->mapping);
+   for (i=0; i < p->channels; ++i) {
+      setup_free(p, p->channel_buffers[i]);
+      setup_free(p, p->previous_window[i]);
+      #ifdef STB_VORBIS_NO_DEFER_FLOOR
+      setup_free(p, p->floor_buffers[i]);
+      #endif
+      setup_free(p, p->finalY[i]);
+   }
+   for (i=0; i < 2; ++i) {
+      setup_free(p, p->A[i]);
+      setup_free(p, p->B[i]);
+      setup_free(p, p->C[i]);
+      setup_free(p, p->window[i]);
+      setup_free(p, p->bit_reverse[i]);
+   }
+   #ifndef STB_VORBIS_NO_STDIO
+   if (p->close_on_free) fclose(p->f);
+   #endif
+}
+
+void stb_vorbis_close(stb_vorbis *p)
+{
+   if (p == NULL) return;
+   vorbis_deinit(p);
+   setup_free(p,p);
+}
+
+static void vorbis_init(stb_vorbis *p, stb_vorbis_alloc *z)
+{
+   memset(p, 0, sizeof(*p)); // NULL out all malloc'd pointers to start
+   if (z) {
+      p->alloc = *z;
+      p->alloc.alloc_buffer_length_in_bytes = (p->alloc.alloc_buffer_length_in_bytes+3) & ~3;
+      p->temp_offset = p->alloc.alloc_buffer_length_in_bytes;
+   }
+   p->eof = 0;
+   p->error = VORBIS__no_error;
+   p->stream = NULL;
+   p->codebooks = NULL;
+   p->page_crc_tests = -1;
+   #ifndef STB_VORBIS_NO_STDIO
+   p->close_on_free = FALSE;
+   p->f = NULL;
+   #endif
+}
+
+int stb_vorbis_get_sample_offset(stb_vorbis *f)
+{
+   if (f->current_loc_valid)
+      return f->current_loc;
+   else
+      return -1;
+}
+
+stb_vorbis_info stb_vorbis_get_info(stb_vorbis *f)
+{
+   stb_vorbis_info d;
+   d.channels = f->channels;
+   d.sample_rate = f->sample_rate;
+   d.setup_memory_required = f->setup_memory_required;
+   d.setup_temp_memory_required = f->setup_temp_memory_required;
+   d.temp_memory_required = f->temp_memory_required;
+   d.max_frame_size = f->blocksize_1 >> 1;
+   return d;
+}
+
+int stb_vorbis_get_error(stb_vorbis *f)
+{
+   int e = f->error;
+   f->error = VORBIS__no_error;
+   return e;
+}
+
+static stb_vorbis * vorbis_alloc(stb_vorbis *f)
+{
+   stb_vorbis *p = (stb_vorbis *) setup_malloc(f, sizeof(*p));
+   return p;
+}
+
+#ifndef STB_VORBIS_NO_PUSHDATA_API
+
+void stb_vorbis_flush_pushdata(stb_vorbis *f)
+{
+   f->previous_length = 0;
+   f->page_crc_tests  = 0;
+   f->discard_samples_deferred = 0;
+   f->current_loc_valid = FALSE;
+   f->first_decode = FALSE;
+   f->samples_output = 0;
+   f->channel_buffer_start = 0;
+   f->channel_buffer_end = 0;
+}
+
+static int vorbis_search_for_page_pushdata(vorb *f, uint8 *data, int data_len)
+{
+   int i,n;
+   for (i=0; i < f->page_crc_tests; ++i)
+      f->scan[i].bytes_done = 0;
+
+   // if we have room for more scans, search for them first, because
+   // they may cause us to stop early if their header is incomplete
+   if (f->page_crc_tests < STB_VORBIS_PUSHDATA_CRC_COUNT) {
+      if (data_len < 4) return 0;
+      data_len -= 3; // need to look for 4-byte sequence, so don't miss
+                     // one that straddles a boundary
+      for (i=0; i < data_len; ++i) {
+         if (data[i] == 0x4f) {
+            if (0==memcmp(data+i, ogg_page_header, 4)) {
+               int j,len;
+               uint32 crc;
+               // make sure we have the whole page header
+               if (i+26 >= data_len || i+27+data[i+26] >= data_len) {
+                  // only read up to this page start, so hopefully we'll
+                  // have the whole page header start next time
+                  data_len = i;
+                  break;
+               }
+               // ok, we have it all; compute the length of the page
+               len = 27 + data[i+26];
+               for (j=0; j < data[i+26]; ++j)
+                  len += data[i+27+j];
+               // scan everything up to the embedded crc (which we must 0)
+               crc = 0;
+               for (j=0; j < 22; ++j)
+                  crc = crc32_update(crc, data[i+j]);
+               // now process 4 0-bytes
+               for (   ; j < 26; ++j)
+                  crc = crc32_update(crc, 0);
+               // len is the total number of bytes we need to scan
+               n = f->page_crc_tests++;
+               f->scan[n].bytes_left = len-j;
+               f->scan[n].crc_so_far = crc;
+               f->scan[n].goal_crc = data[i+22] + (data[i+23] << 8) + (data[i+24]<<16) + (data[i+25]<<24);
+               // if the last frame on a page is continued to the next, then
+               // we can't recover the sample_loc immediately
+               if (data[i+27+data[i+26]-1] == 255)
+                  f->scan[n].sample_loc = ~0;
+               else
+                  f->scan[n].sample_loc = data[i+6] + (data[i+7] << 8) + (data[i+ 8]<<16) + (data[i+ 9]<<24);
+               f->scan[n].bytes_done = i+j;
+               if (f->page_crc_tests == STB_VORBIS_PUSHDATA_CRC_COUNT)
+                  break;
+               // keep going if we still have room for more
+            }
+         }
+      }
+   }
+
+   for (i=0; i < f->page_crc_tests;) {
+      uint32 crc;
+      int j;
+      int n = f->scan[i].bytes_done;
+      int m = f->scan[i].bytes_left;
+      if (m > data_len - n) m = data_len - n;
+      // m is the bytes to scan in the current chunk
+      crc = f->scan[i].crc_so_far;
+      for (j=0; j < m; ++j)
+         crc = crc32_update(crc, data[n+j]);
+      f->scan[i].bytes_left -= m;
+      f->scan[i].crc_so_far = crc;
+      if (f->scan[i].bytes_left == 0) {
+         // does it match?
+         if (f->scan[i].crc_so_far == f->scan[i].goal_crc) {
+            // Houston, we have page
+            data_len = n+m; // consumption amount is wherever that scan ended
+            f->page_crc_tests = -1; // drop out of page scan mode
+            f->previous_length = 0; // decode-but-don't-output one frame
+            f->next_seg = -1;       // start a new page
+            f->current_loc = f->scan[i].sample_loc; // set the current sample location
+                                    // to the amount we'd have decoded had we decoded this page
+            f->current_loc_valid = f->current_loc != ~0U;
+            return data_len;
+         }
+         // delete entry
+         f->scan[i] = f->scan[--f->page_crc_tests];
+      } else {
+         ++i;
+      }
+   }
+
+   return data_len;
+}
+
+// return value: number of bytes we used
+int stb_vorbis_decode_frame_pushdata(
+         stb_vorbis *f,                 // the file we're decoding
+         uint8 *data, int data_len,     // the memory available for decoding
+         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
+     )
+{
+   int i;
+   int len,right,left;
+
+   if (!IS_PUSH_MODE(f)) return error(f, VORBIS_invalid_api_mixing);
+
+   if (f->page_crc_tests >= 0) {
+      *samples = 0;
+      return vorbis_search_for_page_pushdata(f, data, data_len);
+   }
+
+   f->stream     = data;
+   f->stream_end = data + data_len;
+   f->error      = VORBIS__no_error;
+
+   // check that we have the entire packet in memory
+   if (!is_whole_packet_present(f, FALSE)) {
+      *samples = 0;
+      return 0;
+   }
+
+   if (!vorbis_decode_packet(f, &len, &left, &right)) {
+      // save the actual error we encountered
+      enum STBVorbisError error = f->error;
+      if (error == VORBIS_bad_packet_type) {
+         // flush and resynch
+         f->error = VORBIS__no_error;
+         while (get8_packet(f) != EOP)
+            if (f->eof) break;
+         *samples = 0;
+         return f->stream - data;
+      }
+      if (error == VORBIS_continued_packet_flag_invalid) {
+         if (f->previous_length == 0) {
+            // we may be resynching, in which case it's ok to hit one
+            // of these; just discard the packet
+            f->error = VORBIS__no_error;
+            while (get8_packet(f) != EOP)
+               if (f->eof) break;
+            *samples = 0;
+            return f->stream - data;
+         }
+      }
+      // if we get an error while parsing, what to do?
+      // well, it DEFINITELY won't work to continue from where we are!
+      stb_vorbis_flush_pushdata(f);
+      // restore the error that actually made us bail
+      f->error = error;
+      *samples = 0;
+      return 1;
+   }
+
+   // success!
+   len = vorbis_finish_frame(f, len, left, right);
+   for (i=0; i < f->channels; ++i)
+      f->outputs[i] = f->channel_buffers[i] + left;
+
+   if (channels) *channels = f->channels;
+   *samples = len;
+   *output = f->outputs;
+   return f->stream - data;
+}
+
+stb_vorbis *stb_vorbis_open_pushdata(
+         unsigned char *data, int data_len, // the memory available for decoding
+         int *data_used,              // only defined if result is not NULL
+         int *error, stb_vorbis_alloc *alloc)
+{
+   stb_vorbis *f, p;
+   vorbis_init(&p, alloc);
+   p.stream     = data;
+   p.stream_end = data + data_len;
+   p.push_mode  = TRUE;
+   if (!start_decoder(&p)) {
+      if (p.eof)
+         *error = VORBIS_need_more_data;
+      else
+         *error = p.error;
+      return NULL;
+   }
+   f = vorbis_alloc(&p);
+   if (f) {
+      *f = p;
+      *data_used = f->stream - data;
+      *error = 0;
+      return f;
+   } else {
+      vorbis_deinit(&p);
+      return NULL;
+   }
+}
+#endif // STB_VORBIS_NO_PUSHDATA_API
+
+unsigned int stb_vorbis_get_file_offset(stb_vorbis *f)
+{
+   #ifndef STB_VORBIS_NO_PUSHDATA_API
+   if (f->push_mode) return 0;
+   #endif
+   if (USE_MEMORY(f)) return f->stream - f->stream_start;
+   #ifndef STB_VORBIS_NO_STDIO
+   return ftell(f->f) - f->f_start;
+   #endif
+}
+
+#ifndef STB_VORBIS_NO_PULLDATA_API
+//
+// DATA-PULLING API
+//
+
+static uint32 vorbis_find_page(stb_vorbis *f, uint32 *end, uint32 *last)
+{
+   for(;;) {
+      int n;
+      if (f->eof) return 0;
+      n = get8(f);
+      if (n == 0x4f) { // page header
+         unsigned int retry_loc = stb_vorbis_get_file_offset(f);
+         int i;
+         // check if we're off the end of a file_section stream
+         if (retry_loc - 25 > f->stream_len)
+            return 0;
+         // check the rest of the header
+         for (i=1; i < 4; ++i)
+            if (get8(f) != ogg_page_header[i])
+               break;
+         if (f->eof) return 0;
+         if (i == 4) {
+            uint8 header[27];
+            uint32 i, crc, goal, len;
+            for (i=0; i < 4; ++i)
+               header[i] = ogg_page_header[i];
+            for (; i < 27; ++i)
+               header[i] = get8(f);
+            if (f->eof) return 0;
+            if (header[4] != 0) goto invalid;
+            goal = header[22] + (header[23] << 8) + (header[24]<<16) + (header[25]<<24);
+            for (i=22; i < 26; ++i)
+               header[i] = 0;
+            crc = 0;
+            for (i=0; i < 27; ++i)
+               crc = crc32_update(crc, header[i]);
+            len = 0;
+            for (i=0; i < header[26]; ++i) {
+               int s = get8(f);
+               crc = crc32_update(crc, s);
+               len += s;
+            }
+            if (len && f->eof) return 0;
+            for (i=0; i < len; ++i)
+               crc = crc32_update(crc, get8(f));
+            // finished parsing probable page
+            if (crc == goal) {
+               // we could now check that it's either got the last
+               // page flag set, OR it's followed by the capture
+               // pattern, but I guess TECHNICALLY you could have
+               // a file with garbage between each ogg page and recover
+               // from it automatically? So even though that paranoia
+               // might decrease the chance of an invalid decode by
+               // another 2^32, not worth it since it would hose those
+               // invalid-but-useful files?
+               if (end)
+                  *end = stb_vorbis_get_file_offset(f);
+               if (last) {
+                  if (header[5] & 0x04)
+                     *last = 1;
+                  else
+                     *last = 0;
+               }
+               set_file_offset(f, retry_loc-1);
+               return 1;
+            }
+         }
+        invalid:
+         // not a valid page, so rewind and look for next one
+         set_file_offset(f, retry_loc);
+      }
+   }
+}
+
+// seek is implemented with 'interpolation search'--this is like
+// binary search, but we use the data values to estimate the likely
+// location of the data item (plus a bit of a bias so when the
+// estimation is wrong we don't waste overly much time)
+
+#define SAMPLE_unknown  0xffffffff
+
+
+// ogg vorbis, in its insane infinite wisdom, only provides
+// information about the sample at the END of the page.
+// therefore we COULD have the data we need in the current
+// page, and not know it. we could just use the end location
+// as our only knowledge for bounds, seek back, and eventually
+// the binary search finds it. or we can try to be smart and
+// not waste time trying to locate more pages. we try to be
+// smart, since this data is already in memory anyway, so
+// doing needless I/O would be crazy!
+static int vorbis_analyze_page(stb_vorbis *f, ProbedPage *z)
+{
+   uint8 header[27], lacing[255];
+   uint8 packet_type[255];
+   int num_packet, packet_start;
+   int i,len;
+   uint32 samples;
+
+   // record where the page starts
+   z->page_start = stb_vorbis_get_file_offset(f);
+
+   // parse the header
+   getn(f, header, 27);
+   assert(header[0] == 'O' && header[1] == 'g' && header[2] == 'g' && header[3] == 'S');
+   getn(f, lacing, header[26]);
+
+   // determine the length of the payload
+   len = 0;
+   for (i=0; i < header[26]; ++i)
+      len += lacing[i];
+
+   // this implies where the page ends
+   z->page_end = z->page_start + 27 + header[26] + len;
+
+   // read the last-decoded sample out of the data
+   z->last_decoded_sample = header[6] + (header[7] << 8) + (header[8] << 16) + (header[9] << 16);
+
+   if (header[5] & 4) {
+      // if this is the last page, it's not possible to work
+      // backwards to figure out the first sample! whoops! fuck.
+      z->first_decoded_sample = SAMPLE_unknown;
+      set_file_offset(f, z->page_start);
+      return 1;
+   }
+
+   // scan through the frames to determine the sample-count of each one...
+   // our goal is the sample # of the first fully-decoded sample on the
+   // page, which is the first decoded sample of the 2nd packet
+
+   num_packet=0;
+
+   packet_start = ((header[5] & 1) == 0);
+
+   for (i=0; i < header[26]; ++i) {
+      if (packet_start) {
+         uint8 n,b;
+         if (lacing[i] == 0) goto bail; // trying to read from zero-length packet
+         n = get8(f);
+         // if bottom bit is non-zero, we've got corruption
+         if (n & 1) goto bail;
+         n >>= 1;
+         b = ilog(f->mode_count-1);
+         n &= (1 << b)-1;
+         if (n >= f->mode_count) goto bail;
+         packet_type[num_packet++] = f->mode_config[n].blockflag;
+         skip(f, lacing[i]-1);
+      } else
+         skip(f, lacing[i]);
+      packet_start = (lacing[i] < 255);
+   }
+
+   // now that we know the sizes of all the pages, we can start determining
+   // how much sample data there is.
+
+   samples = 0;
+
+   // for the last packet, we step by its whole length, because the definition
+   // is that we encoded the end sample loc of the 'last packet completed',
+   // where 'completed' refers to packets being split, and we are left to guess
+   // what 'end sample loc' means. we assume it means ignoring the fact that
+   // the last half of the data is useless without windowing against the next
+   // packet... (so it's not REALLY complete in that sense)
+   if (num_packet > 1)
+      samples += f->blocksize[packet_type[num_packet-1]];
+
+   for (i=num_packet-2; i >= 1; --i) {
+      // now, for this packet, how many samples do we have that
+      // do not overlap the following packet?
+      if (packet_type[i] == 1)
+         if (packet_type[i+1] == 1)
+            samples += f->blocksize_1 >> 1;
+         else
+            samples += ((f->blocksize_1 - f->blocksize_0) >> 2) + (f->blocksize_0 >> 1);
+      else
+         samples += f->blocksize_0 >> 1;
+   }
+   // now, at this point, we've rewound to the very beginning of the
+   // _second_ packet. if we entirely discard the first packet after
+   // a seek, this will be exactly the right sample number. HOWEVER!
+   // we can't as easily compute this number for the LAST page. The
+   // only way to get the sample offset of the LAST page is to use
+   // the end loc from the previous page. But what that returns us
+   // is _exactly_ the place where we get our first non-overlapped
+   // sample. (I think. Stupid spec for being ambiguous.) So for
+   // consistency it's better to do that here, too. However, that
+   // will then require us to NOT discard all of the first frame we
+   // decode, in some cases, which means an even weirder frame size
+   // and extra code. what a fucking pain.
+   
+   // we're going to discard the first packet if we
+   // start the seek here, so we don't care about it. (we could actually
+   // do better; if the first packet is long, and the previous packet
+   // is short, there's actually data in the first half of the first
+   // packet that doesn't need discarding... but not worth paying the
+   // effort of tracking that of that here and in the seeking logic)
+   // except crap, if we infer it from the _previous_ packet's end
+   // location, we DO need to use that definition... and we HAVE to
+   // infer the start loc of the LAST packet from the previous packet's
+   // end location. fuck you, ogg vorbis.
+
+   z->first_decoded_sample = z->last_decoded_sample - samples;
+
+   // restore file state to where we were
+   set_file_offset(f, z->page_start);
+   return 1;
+
+   // restore file state to where we were
+  bail:
+   set_file_offset(f, z->page_start);
+   return 0;
+}
+
+static int vorbis_seek_frame_from_page(stb_vorbis *f, uint32 page_start, uint32 first_sample, uint32 target_sample, int fine)
+{
+   int left_start, left_end, right_start, right_end, mode,i;
+   int frame=0;
+   uint32 frame_start;
+   int frames_to_skip, data_to_skip;
+
+   // first_sample is the sample # of the first sample that doesn't
+   // overlap the previous page... note that this requires us to
+   // _partially_ discard the first packet! bleh.
+   set_file_offset(f, page_start);
+
+   f->next_seg = -1;  // force page resync
+
+   frame_start = first_sample;
+   // frame start is where the previous packet's last decoded sample
+   // was, which corresponds to left_end... EXCEPT if the previous
+   // packet was long and this packet is short? Probably a bug here.
+
+
+   // now, we can start decoding frames... we'll only FAKE decode them,
+   // until we find the frame that contains our sample; then we'll rewind,
+   // and try again
+   for (;;) {
+      int start;
+
+      if (!vorbis_decode_initial(f, &left_start, &left_end, &right_start, &right_end, &mode))
+         return error(f, VORBIS_seek_failed);
+
+      if (frame == 0)
+         start = left_end;
+      else
+         start = left_start;
+
+      // the window starts at left_start; the last valid sample we generate
+      // before the next frame's window start is right_start-1
+      if (target_sample < frame_start + right_start-start)
+         break;
+
+      flush_packet(f);
+      if (f->eof)
+         return error(f, VORBIS_seek_failed);
+
+      frame_start += right_start - start;
+
+      ++frame;
+   }
+
+   // ok, at this point, the sample we want is contained in frame #'frame'
+
+   // to decode frame #'frame' normally, we have to decode the
+   // previous frame first... but if it's the FIRST frame of the page
+   // we can't. if it's the first frame, it means it falls in the part
+   // of the first frame that doesn't overlap either of the other frames.
+   // so, if we have to handle that case for the first frame, we might
+   // as well handle it for all of them, so:
+   if (target_sample > frame_start + (left_end - left_start)) {
+      // so what we want to do is go ahead and just immediately decode
+      // this frame, but then make it so the next get_frame_float() uses
+      // this already-decoded data? or do we want to go ahead and rewind,
+      // and leave a flag saying to skip the first N data? let's do that
+      frames_to_skip = frame;  // if this is frame #1, skip 1 frame (#0)
+      data_to_skip = left_end - left_start;
+   } else {
+      // otherwise, we want to skip frames 0, 1, 2, ... frame-2
+      // (which means frame-2+1 total frames) then decode frame-1,
+      // then leave frame pending
+      frames_to_skip = frame - 1;
+      assert(frames_to_skip >= 0);
+      data_to_skip = -1;      
+   }
+
+   set_file_offset(f, page_start);
+   f->next_seg = - 1; // force page resync
+
+   for (i=0; i < frames_to_skip; ++i) {
+      maybe_start_packet(f);
+      flush_packet(f);
+   }
+
+   if (data_to_skip >= 0) {
+      int i,j,n = f->blocksize_0 >> 1;
+      f->discard_samples_deferred = data_to_skip;
+      for (i=0; i < f->channels; ++i)
+         for (j=0; j < n; ++j)
+            f->previous_window[i][j] = 0;
+      f->previous_length = n;
+      frame_start += data_to_skip;
+   } else {
+      f->previous_length = 0;
+      vorbis_pump_first_frame(f);
+   }
+
+   // at this point, the NEXT decoded frame will generate the desired sample
+   if (fine) {
+      // so if we're doing sample accurate streaming, we want to go ahead and decode it!
+      if (target_sample != frame_start) {
+         int n;
+         stb_vorbis_get_frame_float(f, &n, NULL);
+         assert(target_sample > frame_start);
+         assert(f->channel_buffer_start + (int) (target_sample-frame_start) < f->channel_buffer_end);
+         f->channel_buffer_start += (target_sample - frame_start);
+      }
+   }
+
+   return 0;
+}
+
+static int vorbis_seek_base(stb_vorbis *f, unsigned int sample_number, int fine)
+{
+   ProbedPage p[2],q;
+   if (IS_PUSH_MODE(f)) return error(f, VORBIS_invalid_api_mixing);
+
+   // do we know the location of the last page?
+   if (f->p_last.page_start == 0) {
+      uint32 z = stb_vorbis_stream_length_in_samples(f);
+      if (z == 0) return error(f, VORBIS_cant_find_last_page);
+   }
+
+   p[0] = f->p_first;
+   p[1] = f->p_last;
+
+   if (sample_number >= f->p_last.last_decoded_sample)
+      sample_number = f->p_last.last_decoded_sample-1;
+
+   if (sample_number < f->p_first.last_decoded_sample) {
+      vorbis_seek_frame_from_page(f, p[0].page_start, 0, sample_number, fine);
+      return 0;
+   } else {
+      int attempts=0;
+      while (p[0].page_end < p[1].page_start) {
+         uint32 probe;
+         uint32 start_offset, end_offset;
+         uint32 start_sample, end_sample;
+
+         // copy these into local variables so we can tweak them
+         // if any are unknown
+         start_offset = p[0].page_end;
+         end_offset   = p[1].after_previous_page_start; // an address known to seek to page p[1]
+         start_sample = p[0].last_decoded_sample;
+         end_sample   = p[1].last_decoded_sample;
+
+         // currently there is no such tweaking logic needed/possible?
+         if (start_sample == SAMPLE_unknown || end_sample == SAMPLE_unknown)
+            return error(f, VORBIS_seek_failed);
+
+         // now we want to lerp between these for the target samples...
+      
+         // step 1: we need to bias towards the page start...
+         if (start_offset + 4000 < end_offset)
+            end_offset -= 4000;
+
+         // now compute an interpolated search loc
+         probe = start_offset + (int) floor((float) (end_offset - start_offset) / (end_sample - start_sample) * (sample_number - start_sample));
+
+         // next we need to bias towards binary search...
+         // code is a little wonky to allow for full 32-bit unsigned values
+         if (attempts >= 4) {
+            uint32 probe2 = start_offset + ((end_offset - start_offset) >> 1);
+            if (attempts >= 8)
+               probe = probe2;
+            else if (probe < probe2)
+               probe = probe + ((probe2 - probe) >> 1);
+            else
+               probe = probe2 + ((probe - probe2) >> 1);
+         }
+         ++attempts;
+
+         set_file_offset(f, probe);
+         if (!vorbis_find_page(f, NULL, NULL))   return error(f, VORBIS_seek_failed);
+         if (!vorbis_analyze_page(f, &q))        return error(f, VORBIS_seek_failed);
+         q.after_previous_page_start = probe;
+
+         // it's possible we've just found the last page again
+         if (q.page_start == p[1].page_start) {
+            p[1] = q;
+            continue;
+         }
+
+         if (sample_number < q.last_decoded_sample)
+            p[1] = q;
+         else
+            p[0] = q;
+      }
+
+      if (p[0].last_decoded_sample <= sample_number && sample_number < p[1].last_decoded_sample) {
+         vorbis_seek_frame_from_page(f, p[1].page_start, p[0].last_decoded_sample, sample_number, fine);
+         return 0;
+      }
+      return error(f, VORBIS_seek_failed);
+   }
+}
+
+int stb_vorbis_seek_frame(stb_vorbis *f, unsigned int sample_number)
+{
+   return vorbis_seek_base(f, sample_number, FALSE);
+}
+
+int stb_vorbis_seek(stb_vorbis *f, unsigned int sample_number)
+{
+   return vorbis_seek_base(f, sample_number, TRUE);
+}
+
+void stb_vorbis_seek_start(stb_vorbis *f)
+{
+   if (IS_PUSH_MODE(f)) { error(f, VORBIS_invalid_api_mixing); return; }
+   set_file_offset(f, f->first_audio_page_offset);
+   f->previous_length = 0;
+   f->first_decode = TRUE;
+   f->next_seg = -1;
+   vorbis_pump_first_frame(f);
+}
+
+unsigned int stb_vorbis_stream_length_in_samples(stb_vorbis *f)
+{
+   unsigned int restore_offset, previous_safe;
+   unsigned int end, last_page_loc;
+
+   if (IS_PUSH_MODE(f)) return error(f, VORBIS_invalid_api_mixing);
+   if (!f->total_samples) {
+      unsigned int last;
+      uint32 lo,hi;
+      char header[6];
+
+      // first, store the current decode position so we can restore it
+      restore_offset = stb_vorbis_get_file_offset(f);
+
+      // now we want to seek back 64K from the end (the last page must
+      // be at most a little less than 64K, but let's allow a little slop)
+      if (f->stream_len >= 65536 && f->stream_len-65536 >= f->first_audio_page_offset)
+         previous_safe = f->stream_len - 65536;
+      else
+         previous_safe = f->first_audio_page_offset;
+
+      set_file_offset(f, previous_safe);
+      // previous_safe is now our candidate 'earliest known place that seeking
+      // to will lead to the final page'
+
+      if (!vorbis_find_page(f, &end, &last)) {
+         // if we can't find a page, we're hosed!
+         f->error = VORBIS_cant_find_last_page;
+         f->total_samples = 0xffffffff;
+         goto done;
+      }
+
+      // check if there are more pages
+      last_page_loc = stb_vorbis_get_file_offset(f);
+
+      // stop when the last_page flag is set, not when we reach eof;
+      // this allows us to stop short of a 'file_section' end without
+      // explicitly checking the length of the section
+      while (!last) {
+         set_file_offset(f, end);
+         if (!vorbis_find_page(f, &end, &last)) {
+            // the last page we found didn't have the 'last page' flag
+            // set. whoops!
+            break;
+         }
+         previous_safe = last_page_loc+1;
+         last_page_loc = stb_vorbis_get_file_offset(f);
+      }
+
+      set_file_offset(f, last_page_loc);
+
+      // parse the header
+      getn(f, (unsigned char *)header, 6);
+      // extract the absolute granule position
+      lo = get32(f);
+      hi = get32(f);
+      if (lo == 0xffffffff && hi == 0xffffffff) {
+         f->error = VORBIS_cant_find_last_page;
+         f->total_samples = SAMPLE_unknown;
+         goto done;
+      }
+      if (hi)
+         lo = 0xfffffffe; // saturate
+      f->total_samples = lo;
+
+      f->p_last.page_start = last_page_loc;
+      f->p_last.page_end   = end;
+      f->p_last.last_decoded_sample = lo;
+      f->p_last.first_decoded_sample = SAMPLE_unknown;
+      f->p_last.after_previous_page_start = previous_safe;
+
+     done:
+      set_file_offset(f, restore_offset);
+   }
+   return f->total_samples == SAMPLE_unknown ? 0 : f->total_samples;
+}
+
+float stb_vorbis_stream_length_in_seconds(stb_vorbis *f)
+{
+   return stb_vorbis_stream_length_in_samples(f) / (float) f->sample_rate;
+}
+
+
+
+int stb_vorbis_get_frame_float(stb_vorbis *f, int *channels, float ***output)
+{
+   int len, right,left,i;
+   if (IS_PUSH_MODE(f)) return error(f, VORBIS_invalid_api_mixing);
+
+   if (!vorbis_decode_packet(f, &len, &left, &right)) {
+      f->channel_buffer_start = f->channel_buffer_end = 0;
+      return 0;
+   }
+
+   len = vorbis_finish_frame(f, len, left, right);
+   for (i=0; i < f->channels; ++i)
+      f->outputs[i] = f->channel_buffers[i] + left;
+
+   f->channel_buffer_start = left;
+   f->channel_buffer_end   = left+len;
+
+   if (channels) *channels = f->channels;
+   if (output)   *output = f->outputs;
+   return len;
+}
+
+#ifndef STB_VORBIS_NO_STDIO
+
+stb_vorbis * stb_vorbis_open_file_section(FILE *file, int close_on_free, int *error, stb_vorbis_alloc *alloc, unsigned int length)
+{
+   stb_vorbis *f, p;
+   vorbis_init(&p, alloc);
+   p.f = file;
+   p.f_start = ftell(file);
+   p.stream_len   = length;
+   p.close_on_free = close_on_free;
+   if (start_decoder(&p)) {
+      f = vorbis_alloc(&p);
+      if (f) {
+         *f = p;
+         vorbis_pump_first_frame(f);
+         return f;
+      }
+   }
+   if (error) *error = p.error;
+   vorbis_deinit(&p);
+   return NULL;
+}
+
+stb_vorbis * stb_vorbis_open_file(FILE *file, int close_on_free, int *error, stb_vorbis_alloc *alloc)
+{
+   unsigned int len, start;
+   start = ftell(file);
+   fseek(file, 0, SEEK_END);
+   len = ftell(file) - start;
+   fseek(file, start, SEEK_SET);
+   return stb_vorbis_open_file_section(file, close_on_free, error, alloc, len);
+}
+
+stb_vorbis * stb_vorbis_open_filename(const char *filename, int *error, stb_vorbis_alloc *alloc)
+{
+   FILE *f = fopen(filename, "rb");
+   if (f) 
+      return stb_vorbis_open_file(f, TRUE, error, alloc);
+   if (error) *error = VORBIS_file_open_failure;
+   return NULL;
+}
+#endif // STB_VORBIS_NO_STDIO
+
+stb_vorbis * stb_vorbis_open_memory(const unsigned char *data, int len, int *error, stb_vorbis_alloc *alloc)
+{
+   stb_vorbis *f, p;
+   if (data == NULL) return NULL;
+   vorbis_init(&p, alloc);
+   p.stream = (uint8 *) data;
+   p.stream_end = (uint8 *) data + len;
+   p.stream_start = (uint8 *) p.stream;
+   p.stream_len = len;
+   p.push_mode = FALSE;
+   if (start_decoder(&p)) {
+      f = vorbis_alloc(&p);
+      if (f) {
+         *f = p;
+         vorbis_pump_first_frame(f);
+         return f;
+      }
+   }
+   if (error) *error = p.error;
+   vorbis_deinit(&p);
+   return NULL;
+}
+
+#ifndef STB_VORBIS_NO_INTEGER_CONVERSION
+#define PLAYBACK_MONO     1
+#define PLAYBACK_LEFT     2
+#define PLAYBACK_RIGHT    4
+
+#define L  (PLAYBACK_LEFT  | PLAYBACK_MONO)
+#define C  (PLAYBACK_LEFT  | PLAYBACK_RIGHT | PLAYBACK_MONO)
+#define R  (PLAYBACK_RIGHT | PLAYBACK_MONO)
+
+static int8 channel_position[7][6] =
+{
+   { 0 },
+   { C },
+   { L, R },
+   { L, C, R },
+   { L, R, L, R },
+   { L, C, R, L, R },
+   { L, C, R, L, R, C },
+};
+
+
+#ifndef STB_VORBIS_NO_FAST_SCALED_FLOAT
+   typedef union {
+      float f;
+      int i;
+   } float_conv;
+   typedef char stb_vorbis_float_size_test[sizeof(float)==4 && sizeof(int) == 4];
+   #define FASTDEF(x) float_conv x
+   // add (1<<23) to convert to int, then divide by 2^SHIFT, then add 0.5/2^SHIFT to round
+   #define MAGIC(SHIFT) (1.5f * (1 << (23-SHIFT)) + 0.5f/(1 << SHIFT))
+   #define ADDEND(SHIFT) (((150-SHIFT) << 23) + (1 << 22))
+   #define FAST_SCALED_FLOAT_TO_INT(temp,x,s) (temp.f = (x) + MAGIC(s), temp.i - ADDEND(s))
+   #define check_endianness()  
+#else
+   #define FAST_SCALED_FLOAT_TO_INT(temp,x,s) ((int) ((x) * (1 << (s))))
+   #define check_endianness()
+   #define FASTDEF(x)
+#endif
+
+static void copy_samples(short *dest, float *src, int len)
+{
+   int i;
+   check_endianness();
+   for (i=0; i < len; ++i) {
+      FASTDEF(temp);
+      int v = FAST_SCALED_FLOAT_TO_INT(temp, src[i],15);
+      if ((unsigned int) (v + 32768) > 65535)
+         v = v < 0 ? -32768 : 32767;
+      dest[i] = v;
+   }
+}
+
+static void compute_samples(int mask, short *output, int num_c, float **data, int d_offset, int len)
+{
+   #define BUFFER_SIZE  32
+   float buffer[BUFFER_SIZE];
+   int i,j,o,n = BUFFER_SIZE;
+   check_endianness();
+   for (o = 0; o < len; o += BUFFER_SIZE) {
+      memset(buffer, 0, sizeof(buffer));
+      if (o + n > len) n = len - o;
+      for (j=0; j < num_c; ++j) {
+         if (channel_position[num_c][j] & mask) {
+            for (i=0; i < n; ++i)
+               buffer[i] += data[j][d_offset+o+i];
+         }
+      }
+      for (i=0; i < n; ++i) {
+         FASTDEF(temp);
+         int v = FAST_SCALED_FLOAT_TO_INT(temp,buffer[i],15);
+         if ((unsigned int) (v + 32768) > 65535)
+            v = v < 0 ? -32768 : 32767;
+         output[o+i] = v;
+      }
+   }
+}
+
+static void compute_stereo_samples(short *output, int num_c, float **data, int d_offset, int len)
+{
+   #define BUFFER_SIZE  32
+   float buffer[BUFFER_SIZE];
+   int i,j,o,n = BUFFER_SIZE >> 1;
+   // o is the offset in the source data
+   check_endianness();
+   for (o = 0; o < len; o += BUFFER_SIZE >> 1) {
+      // o2 is the offset in the output data
+      int o2 = o << 1;
+      memset(buffer, 0, sizeof(buffer));
+      if (o + n > len) n = len - o;
+      for (j=0; j < num_c; ++j) {
+         int m = channel_position[num_c][j] & (PLAYBACK_LEFT | PLAYBACK_RIGHT);
+         if (m == (PLAYBACK_LEFT | PLAYBACK_RIGHT)) {
+            for (i=0; i < n; ++i) {
+               buffer[i*2+0] += data[j][d_offset+o+i];
+               buffer[i*2+1] += data[j][d_offset+o+i];
+            }
+         } else if (m == PLAYBACK_LEFT) {
+            for (i=0; i < n; ++i) {
+               buffer[i*2+0] += data[j][d_offset+o+i];
+            }
+         } else if (m == PLAYBACK_RIGHT) {
+            for (i=0; i < n; ++i) {
+               buffer[i*2+1] += data[j][d_offset+o+i];
+            }
+         }
+      }
+      for (i=0; i < (n<<1); ++i) {
+         FASTDEF(temp);
+         int v = FAST_SCALED_FLOAT_TO_INT(temp,buffer[i],15);
+         if ((unsigned int) (v + 32768) > 65535)
+            v = v < 0 ? -32768 : 32767;
+         output[o2+i] = v;
+      }
+   }
+}
+
+static void convert_samples_short(int buf_c, short **buffer, int b_offset, int data_c, float **data, int d_offset, int samples)
+{
+   int i;
+   if (buf_c != data_c && buf_c <= 2 && data_c <= 6) {
+      static int channel_selector[3][2] = { {0}, {PLAYBACK_MONO}, {PLAYBACK_LEFT, PLAYBACK_RIGHT} };
+      for (i=0; i < buf_c; ++i)
+         compute_samples(channel_selector[buf_c][i], buffer[i]+b_offset, data_c, data, d_offset, samples);
+   } else {
+      int limit = buf_c < data_c ? buf_c : data_c;
+      for (i=0; i < limit; ++i)
+         copy_samples(buffer[i]+b_offset, data[i]+d_offset, samples);
+      for (   ; i < buf_c; ++i)
+         memset(buffer[i]+b_offset, 0, sizeof(short) * samples);
+   }
+}
+
+int stb_vorbis_get_frame_short(stb_vorbis *f, int num_c, short **buffer, int num_samples)
+{
+   float **output;
+   int len = stb_vorbis_get_frame_float(f, NULL, &output);
+   if (len > num_samples) len = num_samples;
+   if (len)
+      convert_samples_short(num_c, buffer, 0, f->channels, output, 0, len);
+   return len;
+}
+
+static void convert_channels_short_interleaved(int buf_c, short *buffer, int data_c, float **data, int d_offset, int len)
+{
+   int i;
+   check_endianness();
+   if (buf_c != data_c && buf_c <= 2 && data_c <= 6) {
+      assert(buf_c == 2);
+      for (i=0; i < buf_c; ++i)
+         compute_stereo_samples(buffer, data_c, data, d_offset, len);
+   } else {
+      int limit = buf_c < data_c ? buf_c : data_c;
+      int j;
+      for (j=0; j < len; ++j) {
+         for (i=0; i < limit; ++i) {
+            FASTDEF(temp);
+            float f = data[i][d_offset+j];
+            int v = FAST_SCALED_FLOAT_TO_INT(temp, f,15);//data[i][d_offset+j],15);
+            if ((unsigned int) (v + 32768) > 65535)
+               v = v < 0 ? -32768 : 32767;
+            *buffer++ = v;
+         }
+         for (   ; i < buf_c; ++i)
+            *buffer++ = 0;
+      }
+   }
+}
+
+int stb_vorbis_get_frame_short_interleaved(stb_vorbis *f, int num_c, short *buffer, int num_shorts)
+{
+   float **output;
+   int len;
+   if (num_c == 1) return stb_vorbis_get_frame_short(f,num_c,&buffer, num_shorts);
+   len = stb_vorbis_get_frame_float(f, NULL, &output);
+   if (len) {
+      if (len*num_c > num_shorts) len = num_shorts / num_c;
+      convert_channels_short_interleaved(num_c, buffer, f->channels, output, 0, len);
+   }
+   return len;
+}
+
+int stb_vorbis_get_samples_short_interleaved(stb_vorbis *f, int channels, short *buffer, int num_shorts)
+{
+   float **outputs;
+   int len = num_shorts / channels;
+   int n=0;
+   int z = f->channels;
+   if (z > channels) z = channels;
+   while (n < len) {
+      int k = f->channel_buffer_end - f->channel_buffer_start;
+      if (n+k >= len) k = len - n;
+      if (k)
+         convert_channels_short_interleaved(channels, buffer, f->channels, f->channel_buffers, f->channel_buffer_start, k);
+      buffer += k*channels;
+      n += k;
+      f->channel_buffer_start += k;
+      if (n == len) break;
+      if (!stb_vorbis_get_frame_float(f, NULL, &outputs)) break;
+   }
+   return n;
+}
+
+int stb_vorbis_get_samples_short(stb_vorbis *f, int channels, short **buffer, int len)
+{
+   float **outputs;
+   int n=0;
+   int z = f->channels;
+   if (z > channels) z = channels;
+   while (n < len) {
+      int k = f->channel_buffer_end - f->channel_buffer_start;
+      if (n+k >= len) k = len - n;
+      if (k)
+         convert_samples_short(channels, buffer, n, f->channels, f->channel_buffers, f->channel_buffer_start, k);
+      n += k;
+      f->channel_buffer_start += k;
+      if (n == len) break;
+      if (!stb_vorbis_get_frame_float(f, NULL, &outputs)) break;
+   }
+   return n;
+}
+
+#ifndef STB_VORBIS_NO_STDIO
+int stb_vorbis_decode_filename(const char *filename, int *channels, int *sample_rate, short **output)
+{
+   int data_len, offset, total, limit, error;
+   short *data;
+   stb_vorbis *v = stb_vorbis_open_filename(filename, &error, NULL);
+   if (v == NULL) return -1;
+   limit = v->channels * 4096;
+   *channels = v->channels;
+   if (sample_rate)
+      *sample_rate = v->sample_rate;
+   offset = data_len = 0;
+   total = limit;
+   data = (short *) malloc(total * sizeof(*data));
+   if (data == NULL) {
+      stb_vorbis_close(v);
+      return -2;
+   }
+   for (;;) {
+      int n = stb_vorbis_get_frame_short_interleaved(v, v->channels, data+offset, total-offset);
+      if (n == 0) break;
+      data_len += n;
+      offset += n * v->channels;
+      if (offset + limit > total) {
+         short *data2;
+         total *= 2;
+         data2 = (short *) realloc(data, total * sizeof(*data));
+         if (data2 == NULL) {
+            free(data);
+            stb_vorbis_close(v);
+            return -2;
+         }
+         data = data2;
+      }
+   }
+   *output = data;
+   stb_vorbis_close(v);
+   return data_len;
+}
+#endif // NO_STDIO
+
+int stb_vorbis_decode_memory(const uint8 *mem, int len, int *channels, int *sample_rate, short **output)
+{
+   int data_len, offset, total, limit, error;
+   short *data;
+   stb_vorbis *v = stb_vorbis_open_memory(mem, len, &error, NULL);
+   if (v == NULL) return -1;
+   limit = v->channels * 4096;
+   *channels = v->channels;
+   if (sample_rate)
+      *sample_rate = v->sample_rate;
+   offset = data_len = 0;
+   total = limit;
+   data = (short *) malloc(total * sizeof(*data));
+   if (data == NULL) {
+      stb_vorbis_close(v);
+      return -2;
+   }
+   for (;;) {
+      int n = stb_vorbis_get_frame_short_interleaved(v, v->channels, data+offset, total-offset);
+      if (n == 0) break;
+      data_len += n;
+      offset += n * v->channels;
+      if (offset + limit > total) {
+         short *data2;
+         total *= 2;
+         data2 = (short *) realloc(data, total * sizeof(*data));
+         if (data2 == NULL) {
+            free(data);
+            stb_vorbis_close(v);
+            return -2;
+         }
+         data = data2;
+      }
+   }
+   *output = data;
+   stb_vorbis_close(v);
+   return data_len;
+}
+#endif // STB_VORBIS_NO_INTEGER_CONVERSION
+
+int stb_vorbis_get_samples_float_interleaved(stb_vorbis *f, int channels, float *buffer, int num_floats)
+{
+   float **outputs;
+   int len = num_floats / channels;
+   int n=0;
+   int z = f->channels;
+   if (z > channels) z = channels;
+   while (n < len) {
+      int i,j;
+      int k = f->channel_buffer_end - f->channel_buffer_start;
+      if (n+k >= len) k = len - n;
+      for (j=0; j < k; ++j) {
+         for (i=0; i < z; ++i)
+            *buffer++ = f->channel_buffers[i][f->channel_buffer_start+j];
+         for (   ; i < channels; ++i)
+            *buffer++ = 0;
+      }
+      n += k;
+      f->channel_buffer_start += k;
+      if (n == len)
+         break;
+      if (!stb_vorbis_get_frame_float(f, NULL, &outputs))
+         break;
+   }
+   return n;
+}
+
+int stb_vorbis_get_samples_float(stb_vorbis *f, int channels, float **buffer, int num_samples)
+{
+   float **outputs;
+   int n=0;
+   int z = f->channels;
+   if (z > channels) z = channels;
+   while (n < num_samples) {
+      int i;
+      int k = f->channel_buffer_end - f->channel_buffer_start;
+      if (n+k >= num_samples) k = num_samples - n;
+      if (k) {
+         for (i=0; i < z; ++i)
+            memcpy(buffer[i]+n, f->channel_buffers[i]+f->channel_buffer_start, sizeof(float)*k);
+         for (   ; i < channels; ++i)
+            memset(buffer[i]+n, 0, sizeof(float) * k);
+      }
+      n += k;
+      f->channel_buffer_start += k;
+      if (n == num_samples)
+         break;
+      if (!stb_vorbis_get_frame_float(f, NULL, &outputs))
+         break;
+   }
+   return n;
+}
+#endif // STB_VORBIS_NO_PULLDATA_API
+
+/* Version history
+    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 compare function _cdecl on windows
+    1.01    - 2014/06/18 - fix stb_vorbis_get_samples_float
+    1.0     - 2014/05/26 - fix memory leaks; fix warnings; fix bugs in multichannel
+                           (API change) report sample rate for decode-full-file funcs
+    0.99996 - bracket #include <malloc.h> for macintosh compilation by Laurent Gomila
+    0.99995 - use union instead of pointer-cast for fast-float-to-int to avoid alias-optimization problem
+    0.99994 - change fast-float-to-int to work in single-precision FPU mode, remove endian-dependence
+    0.99993 - remove assert that fired on legal files with empty tables
+    0.99992 - rewind-to-start
+    0.99991 - bugfix to stb_vorbis_get_samples_short by Bernhard Wodo
+    0.9999 - (should have been 0.99990) fix no-CRT support, compiling as C++
+    0.9998 - add a full-decode function with a memory source
+    0.9997 - fix a bug in the read-from-FILE case in 0.9996 addition
+    0.9996 - query length of vorbis stream in samples/seconds
+    0.9995 - bugfix to another optimization that only happened in certain files
+    0.9994 - bugfix to one of the optimizations that caused significant (but inaudible?) errors
+    0.9993 - performance improvements; runs in 99% to 104% of time of reference implementation
+    0.9992 - performance improvement of IMDCT; now performs close to reference implementation
+    0.9991 - performance improvement of IMDCT
+    0.999 - (should have been 0.9990) performance improvement of IMDCT
+    0.998 - no-CRT support from Casey Muratori
+    0.997 - bugfixes for bugs found by Terje Mathisen
+    0.996 - bugfix: fast-huffman decode initialized incorrectly for sparse codebooks; fixing gives 10% speedup - found by Terje Mathisen
+    0.995 - bugfix: fix to 'effective' overrun detection - found by Terje Mathisen
+    0.994 - bugfix: garbage decode on final VQ symbol of a non-multiple - found by Terje Mathisen
+    0.993 - bugfix: pushdata API required 1 extra byte for empty page (failed to consume final page if empty) - found by Terje Mathisen
+    0.992 - fixes for MinGW warning
+    0.991 - turn fast-float-conversion on by default
+    0.990 - fix push-mode seek recovery if you seek into the headers
+    0.98b - fix to bad release of 0.98
+    0.98 - fix push-mode seek recovery; robustify float-to-int and support non-fast mode
+    0.97 - builds under c++ (typecasting, don't use 'class' keyword)
+    0.96 - somehow MY 0.95 was right, but the web one was wrong, so here's my 0.95 rereleased as 0.96, fixes a typo in the clamping code
+    0.95 - clamping code for 16-bit functions
+    0.94 - not publically released
+    0.93 - fixed all-zero-floor case (was decoding garbage)
+    0.92 - fixed a memory leak
+    0.91 - conditional compiles to omit parts of the API and the infrastructure to support them: STB_VORBIS_NO_PULLDATA_API, STB_VORBIS_NO_PUSHDATA_API, STB_VORBIS_NO_STDIO, STB_VORBIS_NO_INTEGER_CONVERSION
+    0.90 - first public release
+*/
+
+#endif // STB_VORBIS_HEADER_ONLY

Source/ThirdParty/STB/stb_vorbis.h

@@ -1,30 +1,46 @@
-// Ogg Vorbis I audio decoder  -- version 0.99996
+// Ogg Vorbis audio decoder - v1.05 - public domain
+// http://nothings.org/stb_vorbis/
 //
-// Written in April 2007 by Sean Barrett, sponsored by RAD Game Tools.
+// Written by Sean Barrett in 2007, last updated in 2014
+// Sponsored by RAD Game Tools.
 //
-// Placed in the public domain April 2007 by the author: no copyright is
-// claimed, and you may use it for any purpose you like.
+// Placed in the public domain April 2007 by the author: no copyright
+// is claimed, and you may use it for any purpose you like.
 //
 // No warranty for any purpose is expressed or implied by the author (nor
 // by RAD Game Tools). Report bugs and send enhancements to the author.
 //
-// Get the latest version and other information at:
-//     http://nothings.org/stb_vorbis/
-
-
-// Todo:
-//
-//   - seeking (note you can seek yourself using the pushdata API)
-//
 // Limitations:
 //
-//   - floor 0 not supported (used in old ogg vorbis files)
+//   - seeking not supported except manually via PUSHDATA api
+//   - 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)
-// 
-// All of these limitations may be removed in future versions.
+//
+// Bugfix/warning contributors:
+//    Terje Mathisen     Niklas Frykholm     Andy Hill
+//    Casey Muratori     John Bolton         Gargaj
+//    Laurent Gomila     Marc LeBlanc        Ronny Chevalier
+//    Bernhard Wodo      Evan Balster			"alxprd"@github
+//    Tom Beaumont       Ingo Leitgeb        Nicolas Guillemot
+// (If you reported a bug but do not appear in this list, it is because
+// someone else reported the bug before you. There were too many of you to
+// list them all because I was lax about updating for a long time, sorry.)
+//
+// Partial history:
+//    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
+//    0.99996 -            - bracket #include <malloc.h> for macintosh compilation
+//    0.99995 -            - avoid alias-optimization issue in float-to-int conversion
+//
+// See end of file for full version history.
 
 
 //////////////////////////////////////////////////////////////////////////////
@@ -153,6 +169,7 @@ extern int stb_vorbis_decode_frame_pushdata(
 // 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)
@@ -197,21 +214,23 @@ extern void stb_vorbis_flush_pushdata(stb_vorbis *f);
 // 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(char *filename, int *channels, short **output);
+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
-extern int stb_vorbis_decode_memory(unsigned char *mem, int len, int *channels, short **output);
 // 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(unsigned char *data, int len,
+extern stb_vorbis * stb_vorbis_open_memory(const unsigned char *data, int len,
                                   int *error, 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(char *filename,
+extern stb_vorbis * stb_vorbis_open_filename(const char *filename,
                                   int *error, 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).
@@ -351,4 +370,7 @@ enum STBVorbisError
 #endif
 
 #endif // STB_VORBIS_INCLUDE_STB_VORBIS_H
-
+//
+//  HEADER ENDS HERE
+//
+//////////////////////////////////////////////////////////////////////////////

Source/ThirdParty/kNet/src/Clock.cpp

@@ -15,6 +15,11 @@
 /** @file Clock.cpp
 	@brief */
 
+// Modified by Lasse Oorni for Urho3D
+
+// Urho3D: ensure that kNetBuildConfig.h is included for WinXP compatibility
+#include "kNetBuildConfig.h"
+
 #if defined(__unix__) || defined(__native_client__) || defined(EMSCRIPTEN) || defined(ANDROID) || defined(__APPLE__) || defined (__CYGWIN__)
 #include <time.h>
 #include <errno.h>