STB library rework and update.

Moved imgui STB dependencies to STB library.
Cleaned up leftover SystemUI namespaces.

Source/Atomic/Resource/Image.cpp

@@ -40,9 +40,7 @@
 
 // ATOMIC END
 
-#define STB_IMAGE_IMPLEMENTATION
 #include <STB/stb_image.h>
-#define STB_IMAGE_WRITE_IMPLEMENTATION
 #include <STB/stb_image_write.h>
 
 #include "../DebugNew.h"

Source/Atomic/UI/SystemUI/SystemUI.cpp

@@ -19,6 +19,7 @@
 // OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
 // THE SOFTWARE.
 //
+#define STB_TEXTEDIT_IMPLEMENTATION 1
 #include "../../Input/InputEvents.h"
 #include "../../Input/Input.h"
 #include "../../Core/CoreEvents.h"
@@ -32,7 +33,6 @@
 #include "../../UI/SystemUI/Console.h"
 #include "../../UI/SystemUI/DebugHud.h"
 #include <SDL.h>
-#include <imgui_internal.h>
 
 
 using namespace std::placeholders;
@@ -171,7 +171,7 @@ void SystemUI::OnRawEvent(VariantMap& args)
     case SDL_KEYUP:
     case SDL_KEYDOWN:
     case SDL_TEXTINPUT:
-        args[SDLRawInput::P_CONSUMED] = ImGui::GetCurrentContext()->FocusedWindow != 0;
+        args[SDLRawInput::P_CONSUMED] = ImGui::IsAnyWindowFocused();
         break;
     case SDL_MOUSEWHEEL:
     case SDL_MOUSEBUTTONUP:

Source/Atomic/UI/SystemUI/SystemUI.h

@@ -33,20 +33,21 @@
 
 #include <imgui.h>
 
+
 namespace Atomic
 {
 
-class SystemUI : public Atomic::Object
+class SystemUI
+        : public Atomic::Object
 {
-    ATOMIC_OBJECT(SystemUI, Atomic::Object);
+ATOMIC_OBJECT(SystemUI, Atomic::Object);
 public:
     SystemUI(Atomic::Context* context);
     ~SystemUI();
 
     //! Get ui scale.
     //! \return scale of ui.
-    float GetScale() const
-    { return uiScale_; };
+    float GetScale() const { return uiScale_; };
     //! Set ui scale.
     //! \param scale of ui.
     void SetScale(float scale);
@@ -58,8 +59,8 @@ public:
       \param merge set to true if new font should be merged to last active font.
       \return ImFont instance that may be used for setting current font when drawing GUI.
     */
-    ImFont*
-    AddFont(const Atomic::String& font_path, float size = 0, const unsigned short* ranges = 0, bool merge = false);
+    ImFont* AddFont(const Atomic::String& font_path, float size = 0, const unsigned short* ranges = 0,
+                    bool merge = false);
     //! Add font to imgui subsystem.
     /*!
       \param font_path a string pointing to TTF font resource.
@@ -68,8 +69,9 @@ public:
       \param merge set to true if new font should be merged to last active font.
       \return ImFont instance that may be used for setting current font when drawing GUI.
     */
-    ImFont* AddFont(const Atomic::String& font_path, float size = 0,
-                    const std::initializer_list<unsigned short>& ranges = {}, bool merge = false);
+    ImFont* AddFont(
+            const Atomic::String& font_path, float size = 0,
+            const std::initializer_list<unsigned short>& ranges = {}, bool merge = false);
 
     void CreateConsoleAndDebugHud();
 

Source/AtomicApp/Player/IPCPlayerApp.cpp

@@ -225,7 +225,7 @@ namespace Atomic
         brokerActive_ = true;
 
         // If the parent application has a profile mode up, sync 
-        SystemUI::DebugHud* debugHud = GetSubsystem<SystemUI::DebugHud>();
+        DebugHud* debugHud = GetSubsystem<DebugHud>();
         if (debugHud)
         {
             unsigned mode = eventData["debugHudMode"].GetUInt();

Source/AtomicEditor/EditorMode/AEEditorMode.cpp

@@ -67,7 +67,7 @@ EditorMode::~EditorMode()
 void EditorMode::HandleIPCWorkerStarted(StringHash eventType, VariantMap& eventData)
 {
     VariantMap startupData;
-    SystemUI::DebugHud* debugHud = GetSubsystem<SystemUI::DebugHud>();
+    DebugHud* debugHud = GetSubsystem<DebugHud>();
 
     startupData["debugHudMode"] = debugHud ? debugHud->GetMode() : (unsigned) 0;
     startupData["debugHudProfilerMode"] = (unsigned) (debugHud ? debugHud->GetProfilerMode() : DEBUG_HUD_PROFILE_PERFORMANCE);
@@ -272,9 +272,9 @@ bool EditorMode::PlayProjectInternal(const String &addArgs, bool debug)
     {
         // enabled metrics at app start
 
-        SystemUI::DebugHud* debugHud = GetSubsystem<SystemUI::DebugHud>();
+        DebugHud* debugHud = GetSubsystem<DebugHud>();
 
-        if ( debugHud && ( debugHud->GetMode() & Atomic::SystemUI::DEBUGHUD_SHOW_PROFILER) && (debugHud->GetProfilerMode() == DEBUG_HUD_PROFILE_METRICS))
+        if ( debugHud && ( debugHud->GetMode() & Atomic::DEBUGHUD_SHOW_PROFILER) && (debugHud->GetProfilerMode() == DEBUG_HUD_PROFILE_METRICS))
         {
             vargs.Insert(0, "--playermetrics");
         }

Source/AtomicEditor/PlayerMode/AEPlayerMode.cpp

@@ -94,7 +94,7 @@ void PlayerMode::HandleIPCInitialize(StringHash eventType, VariantMap& eventData
 
     // END LICENSE MANAGEMENT
 
-    SystemUI::DebugHud* debugHud = GetSubsystem<SystemUI::DebugHud>();
+    DebugHud* debugHud = GetSubsystem<DebugHud>();
     if (debugHud)
     {
         debugHud->SetMode(eventData["debugHudMode"].GetUInt());
@@ -243,7 +243,7 @@ void PlayerMode::HandleViewRender(StringHash eventType, VariantMap& eventData)
         done = true;
 
         messageBox_ = GetSubsystem<UI>()->ShowSystemMessageBox("3D Module License Required", "A 3D Module License is required to display 3D content.\n\nUpgrade to Atomic Pro for all features and platforms.");
-        SubscribeToEvent(messageBox_, SystemUI::E_MESSAGEACK, ATOMIC_HANDLER(PlayerMode, HandleMessageAck));
+        SubscribeToEvent(messageBox_, E_MESSAGEACK, ATOMIC_HANDLER(PlayerMode, HandleMessageAck));
 
         if (brokerActive_)
         {

Source/AtomicEditor/PlayerMode/AEPlayerMode.h

@@ -63,7 +63,7 @@ private:
 // BEGIN LICENSE MANAGEMENT
     void HandleMessageAck(StringHash eventType, VariantMap& eventData);
     bool licenseModule3D_;
-    SharedPtr<SystemUI::MessageBox> messageBox_;
+    SharedPtr<MessageBox> messageBox_;
 // END LICENSE MANAGEMENT
 
     IPCHandle fd_[2];

Source/ThirdParty/STB/CMakeLists.txt

@@ -1,5 +1,5 @@
 #
-# Copyright (c) 2008-2016 the Urho3D project.
+# Copyright (c) 2017 the Atomic project.
 #
 # Permission is hereby granted, free of charge, to any person obtaining a copy
 # of this software and associated documentation files (the "Software"), to deal
@@ -20,15 +20,6 @@
 # THE SOFTWARE.
 #
 
-# Define target name
-set (TARGET_NAME STB)
-
-# Define source files
-define_source_files (GLOB_CPP_PATTERNS *.c *.cpp *.h)
-
-# ATOMIC BEGIN
-set (INCLUDE_DIRS . ${CMAKE_CURRENT_SOURCE_DIR}/../..)
-# ATOMIC END
-
-# Setup target
-setup_library ()
+add_library(STB STATIC stb_impl.cpp)
+target_compile_definitions(STB PUBLIC -DSTB_VORBIS_HEADER_ONLY=1)
+target_include_directories(STB PUBLIC ${CMAKE_CURRENT_SOURCE_DIR} PRIVATE ../..)

Source/ThirdParty/STB/stb_image.c

@@ -1,6077 +0,0 @@
-
-#include "stb_image.h"
-
-#define STB_IMAGE_IMPLEMENTATION
-
-#ifdef STB_IMAGE_IMPLEMENTATION
-
-#if defined(STBI_ONLY_JPEG) || defined(STBI_ONLY_PNG) || defined(STBI_ONLY_BMP) \
-  || defined(STBI_ONLY_TGA) || defined(STBI_ONLY_GIF) || defined(STBI_ONLY_PSD) \
-  || defined(STBI_ONLY_HDR) || defined(STBI_ONLY_PIC) || defined(STBI_ONLY_PNM) \
-  || defined(STBI_ONLY_ZLIB)
-   #ifndef STBI_ONLY_JPEG
-   #define STBI_NO_JPEG
-   #endif
-   #ifndef STBI_ONLY_PNG
-   #define STBI_NO_PNG
-   #endif
-   #ifndef STBI_ONLY_BMP
-   #define STBI_NO_BMP
-   #endif
-   #ifndef STBI_ONLY_PSD
-   #define STBI_NO_PSD
-   #endif
-   #ifndef STBI_ONLY_TGA
-   #define STBI_NO_TGA
-   #endif
-   #ifndef STBI_ONLY_GIF
-   #define STBI_NO_GIF
-   #endif
-   #ifndef STBI_ONLY_HDR
-   #define STBI_NO_HDR
-   #endif
-   #ifndef STBI_ONLY_PIC
-   #define STBI_NO_PIC
-   #endif
-   #ifndef STBI_ONLY_PNM
-   #define STBI_NO_PNM
-   #endif
-#endif
-
-#if defined(STBI_NO_PNG) && !defined(STBI_SUPPORT_ZLIB) && !defined(STBI_NO_ZLIB)
-#define STBI_NO_ZLIB
-#endif
-
-
-#include <stdarg.h>
-#include <stddef.h> // ptrdiff_t on osx
-#include <stdlib.h>
-#include <string.h>
-
-#if !defined(STBI_NO_LINEAR) || !defined(STBI_NO_HDR)
-#include <math.h>  // ldexp
-#endif
-
-#ifndef STBI_NO_STDIO
-#include <stdio.h>
-#endif
-
-#ifndef STBI_ASSERT
-#include <assert.h>
-#define STBI_ASSERT(x) assert(x)
-#endif
-
-
-#ifndef _MSC_VER
-   #ifdef __cplusplus
-   #define stbi_inline inline
-   #else
-   #define stbi_inline
-   #endif
-#else
-   #define stbi_inline __forceinline
-#endif
-
-
-#ifdef _MSC_VER
-typedef unsigned short stbi__uint16;
-typedef   signed short stbi__int16;
-typedef unsigned int   stbi__uint32;
-typedef   signed int   stbi__int32;
-#else
-#include <stdint.h>
-typedef uint16_t stbi__uint16;
-typedef int16_t  stbi__int16;
-typedef uint32_t stbi__uint32;
-typedef int32_t  stbi__int32;
-#endif
-
-// should produce compiler error if size is wrong
-typedef unsigned char validate_uint32[sizeof(stbi__uint32)==4 ? 1 : -1];
-
-#ifdef _MSC_VER
-#define STBI_NOTUSED(v)  (void)(v)
-#else
-#define STBI_NOTUSED(v)  (void)sizeof(v)
-#endif
-
-#ifdef _MSC_VER
-#define STBI_HAS_LROTL
-#endif
-
-#ifdef STBI_HAS_LROTL
-   #define stbi_lrot(x,y)  _lrotl(x,y)
-#else
-   #define stbi_lrot(x,y)  (((x) << (y)) | ((x) >> (32 - (y))))
-#endif
-
-#if defined(STBI_MALLOC) && defined(STBI_FREE) && (defined(STBI_REALLOC) || defined(STBI_REALLOC_SIZED))
-// ok
-#elif !defined(STBI_MALLOC) && !defined(STBI_FREE) && !defined(STBI_REALLOC) && !defined(STBI_REALLOC_SIZED)
-// ok
-#else
-#error "Must define all or none of STBI_MALLOC, STBI_FREE, and STBI_REALLOC (or STBI_REALLOC_SIZED)."
-#endif
-
-#ifndef STBI_MALLOC
-#define STBI_MALLOC(sz)           malloc(sz)
-#define STBI_REALLOC(p,newsz)     realloc(p,newsz)
-#define STBI_FREE(p)              free(p)
-#endif
-
-#ifndef STBI_REALLOC_SIZED
-#define STBI_REALLOC_SIZED(p,oldsz,newsz) STBI_REALLOC(p,newsz)
-#endif
-
-// x86/x64 detection
-#if defined(__x86_64__) || defined(_M_X64)
-#define STBI__X64_TARGET
-#elif defined(__i386) || defined(_M_IX86)
-#define STBI__X86_TARGET
-#endif
-
-#if defined(__GNUC__) && (defined(STBI__X86_TARGET) || defined(STBI__X64_TARGET)) && !defined(__SSE2__) && !defined(STBI_NO_SIMD)
-// NOTE: not clear do we actually need this for the 64-bit path?
-// gcc doesn't support sse2 intrinsics unless you compile with -msse2,
-// (but compiling with -msse2 allows the compiler to use SSE2 everywhere;
-// this is just broken and gcc are jerks for not fixing it properly
-// http://www.virtualdub.org/blog/pivot/entry.php?id=363 )
-#define STBI_NO_SIMD
-#endif
-
-#if defined(__MINGW32__) && defined(STBI__X86_TARGET) && !defined(STBI_MINGW_ENABLE_SSE2) && !defined(STBI_NO_SIMD)
-// Note that __MINGW32__ doesn't actually mean 32-bit, so we have to avoid STBI__X64_TARGET
-//
-// 32-bit MinGW wants ESP to be 16-byte aligned, but this is not in the
-// Windows ABI and VC++ as well as Windows DLLs don't maintain that invariant.
-// As a result, enabling SSE2 on 32-bit MinGW is dangerous when not
-// simultaneously enabling "-mstackrealign".
-//
-// See https://github.com/nothings/stb/issues/81 for more information.
-//
-// So default to no SSE2 on 32-bit MinGW. If you've read this far and added
-// -mstackrealign to your build settings, feel free to #define STBI_MINGW_ENABLE_SSE2.
-#define STBI_NO_SIMD
-#endif
-
-#if !defined(STBI_NO_SIMD) && (defined(STBI__X86_TARGET) || defined(STBI__X64_TARGET))
-#define STBI_SSE2
-#include <emmintrin.h>
-
-#ifdef _MSC_VER
-
-#if _MSC_VER >= 1400  // not VC6
-#include <intrin.h> // __cpuid
-static int stbi__cpuid3(void)
-{
-   int info[4];
-   __cpuid(info,1);
-   return info[3];
-}
-#else
-static int stbi__cpuid3(void)
-{
-   int res;
-   __asm {
-      mov  eax,1
-      cpuid
-      mov  res,edx
-   }
-   return res;
-}
-#endif
-
-#define STBI_SIMD_ALIGN(type, name) __declspec(align(16)) type name
-
-static int stbi__sse2_available()
-{
-   int info3 = stbi__cpuid3();
-   return ((info3 >> 26) & 1) != 0;
-}
-#else // assume GCC-style if not VC++
-#define STBI_SIMD_ALIGN(type, name) type name __attribute__((aligned(16)))
-
-static int stbi__sse2_available()
-{
-#if defined(__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__) >= 408 // GCC 4.8 or later
-   // GCC 4.8+ has a nice way to do this
-   // ATOMIC BEGIN
-   // We only do 64 bit Linux, and all x64 have sse2
-   // this was causing a problem: https://github.com/nothings/stb/issues/280
-   return 1;//__builtin_cpu_supports("sse2");
-   // ATOMIC END
-#else
-   // portable way to do this, preferably without using GCC inline ASM?
-   // just bail for now.
-   return 0;
-#endif
-}
-#endif
-#endif
-
-// ARM NEON
-#if defined(STBI_NO_SIMD) && defined(STBI_NEON)
-#undef STBI_NEON
-#endif
-
-#ifdef STBI_NEON
-#include <arm_neon.h>
-// assume GCC or Clang on ARM targets
-#define STBI_SIMD_ALIGN(type, name) type name __attribute__((aligned(16)))
-#endif
-
-#ifndef STBI_SIMD_ALIGN
-#define STBI_SIMD_ALIGN(type, name) type name
-#endif
-
-///////////////////////////////////////////////
-//
-//  stbi__context struct and start_xxx functions
-
-// stbi__context structure is our basic context used by all images, so it
-// contains all the IO context, plus some basic image information
-typedef struct
-{
-   stbi__uint32 img_x, img_y;
-   int img_n, img_out_n;
-
-   stbi_io_callbacks io;
-   void *io_user_data;
-
-   int read_from_callbacks;
-   int buflen;
-   stbi_uc buffer_start[128];
-
-   stbi_uc *img_buffer, *img_buffer_end;
-   stbi_uc *img_buffer_original, *img_buffer_original_end;
-} stbi__context;
-
-
-static void stbi__refill_buffer(stbi__context *s);
-
-// initialize a memory-decode context
-static void stbi__start_mem(stbi__context *s, stbi_uc const *buffer, int len)
-{
-   s->io.read = NULL;
-   s->read_from_callbacks = 0;
-   s->img_buffer = s->img_buffer_original = (stbi_uc *) buffer;
-   s->img_buffer_end = s->img_buffer_original_end = (stbi_uc *) buffer+len;
-}
-
-// initialize a callback-based context
-static void stbi__start_callbacks(stbi__context *s, stbi_io_callbacks *c, void *user)
-{
-   s->io = *c;
-   s->io_user_data = user;
-   s->buflen = sizeof(s->buffer_start);
-   s->read_from_callbacks = 1;
-   s->img_buffer_original = s->buffer_start;
-   stbi__refill_buffer(s);
-   s->img_buffer_original_end = s->img_buffer_end;
-}
-
-#ifndef STBI_NO_STDIO
-
-static int stbi__stdio_read(void *user, char *data, int size)
-{
-   return (int) fread(data,1,size,(FILE*) user);
-}
-
-static void stbi__stdio_skip(void *user, int n)
-{
-   fseek((FILE*) user, n, SEEK_CUR);
-}
-
-static int stbi__stdio_eof(void *user)
-{
-   return feof((FILE*) user);
-}
-
-static stbi_io_callbacks stbi__stdio_callbacks =
-{
-   stbi__stdio_read,
-   stbi__stdio_skip,
-   stbi__stdio_eof,
-};
-
-static void stbi__start_file(stbi__context *s, FILE *f)
-{
-   stbi__start_callbacks(s, &stbi__stdio_callbacks, (void *) f);
-}
-
-//static void stop_file(stbi__context *s) { }
-
-#endif // !STBI_NO_STDIO
-
-static void stbi__rewind(stbi__context *s)
-{
-   // conceptually rewind SHOULD rewind to the beginning of the stream,
-   // but we just rewind to the beginning of the initial buffer, because
-   // we only use it after doing 'test', which only ever looks at at most 92 bytes
-   s->img_buffer = s->img_buffer_original;
-   s->img_buffer_end = s->img_buffer_original_end;
-}
-
-#ifndef STBI_NO_JPEG
-static int      stbi__jpeg_test(stbi__context *s);
-static stbi_uc *stbi__jpeg_load(stbi__context *s, int *x, int *y, int *comp, int req_comp);
-static int      stbi__jpeg_info(stbi__context *s, int *x, int *y, int *comp);
-#endif
-
-#ifndef STBI_NO_PNG
-static int      stbi__png_test(stbi__context *s);
-static stbi_uc *stbi__png_load(stbi__context *s, int *x, int *y, int *comp, int req_comp);
-static int      stbi__png_info(stbi__context *s, int *x, int *y, int *comp);
-#endif
-
-#ifndef STBI_NO_BMP
-static int      stbi__bmp_test(stbi__context *s);
-static stbi_uc *stbi__bmp_load(stbi__context *s, int *x, int *y, int *comp, int req_comp);
-static int      stbi__bmp_info(stbi__context *s, int *x, int *y, int *comp);
-#endif
-
-#ifndef STBI_NO_TGA
-static int      stbi__tga_test(stbi__context *s);
-static stbi_uc *stbi__tga_load(stbi__context *s, int *x, int *y, int *comp, int req_comp);
-static int      stbi__tga_info(stbi__context *s, int *x, int *y, int *comp);
-#endif
-
-#ifndef STBI_NO_PSD
-static int      stbi__psd_test(stbi__context *s);
-static stbi_uc *stbi__psd_load(stbi__context *s, int *x, int *y, int *comp, int req_comp);
-static int      stbi__psd_info(stbi__context *s, int *x, int *y, int *comp);
-#endif
-
-#ifndef STBI_NO_HDR
-static int      stbi__hdr_test(stbi__context *s);
-static float   *stbi__hdr_load(stbi__context *s, int *x, int *y, int *comp, int req_comp);
-static int      stbi__hdr_info(stbi__context *s, int *x, int *y, int *comp);
-#endif
-
-#ifndef STBI_NO_PIC
-static int      stbi__pic_test(stbi__context *s);
-static stbi_uc *stbi__pic_load(stbi__context *s, int *x, int *y, int *comp, int req_comp);
-static int      stbi__pic_info(stbi__context *s, int *x, int *y, int *comp);
-#endif
-
-#ifndef STBI_NO_GIF
-static int      stbi__gif_test(stbi__context *s);
-static stbi_uc *stbi__gif_load(stbi__context *s, int *x, int *y, int *comp, int req_comp);
-static int      stbi__gif_info(stbi__context *s, int *x, int *y, int *comp);
-#endif
-
-#ifndef STBI_NO_PNM
-static int      stbi__pnm_test(stbi__context *s);
-static stbi_uc *stbi__pnm_load(stbi__context *s, int *x, int *y, int *comp, int req_comp);
-static int      stbi__pnm_info(stbi__context *s, int *x, int *y, int *comp);
-#endif
-
-// this is not threadsafe
-static const char *stbi__g_failure_reason;
-
-STBIDEF const char *stbi_failure_reason(void)
-{
-   return stbi__g_failure_reason;
-}
-
-static int stbi__err(const char *str)
-{
-   stbi__g_failure_reason = str;
-   return 0;
-}
-
-static void *stbi__malloc(size_t size)
-{
-    return STBI_MALLOC(size);
-}
-
-// stbi__err - error
-// stbi__errpf - error returning pointer to float
-// stbi__errpuc - error returning pointer to unsigned char
-
-#ifdef STBI_NO_FAILURE_STRINGS
-   #define stbi__err(x,y)  0
-#elif defined(STBI_FAILURE_USERMSG)
-   #define stbi__err(x,y)  stbi__err(y)
-#else
-   #define stbi__err(x,y)  stbi__err(x)
-#endif
-
-#define stbi__errpf(x,y)   ((float *)(size_t) (stbi__err(x,y)?NULL:NULL))
-#define stbi__errpuc(x,y)  ((unsigned char *)(size_t) (stbi__err(x,y)?NULL:NULL))
-
-STBIDEF void stbi_image_free(void *retval_from_stbi_load)
-{
-   STBI_FREE(retval_from_stbi_load);
-}
-
-#ifndef STBI_NO_LINEAR
-static float   *stbi__ldr_to_hdr(stbi_uc *data, int x, int y, int comp);
-#endif
-
-#ifndef STBI_NO_HDR
-static stbi_uc *stbi__hdr_to_ldr(float   *data, int x, int y, int comp);
-#endif
-
-static int stbi__vertically_flip_on_load = 0;
-
-STBIDEF void stbi_set_flip_vertically_on_load(int flag_true_if_should_flip)
-{
-    stbi__vertically_flip_on_load = flag_true_if_should_flip;
-}
-
-static unsigned char *stbi__load_main(stbi__context *s, int *x, int *y, int *comp, int req_comp)
-{
-   #ifndef STBI_NO_JPEG
-   if (stbi__jpeg_test(s)) return stbi__jpeg_load(s,x,y,comp,req_comp);
-   #endif
-   #ifndef STBI_NO_PNG
-   if (stbi__png_test(s))  return stbi__png_load(s,x,y,comp,req_comp);
-   #endif
-   #ifndef STBI_NO_BMP
-   if (stbi__bmp_test(s))  return stbi__bmp_load(s,x,y,comp,req_comp);
-   #endif
-   #ifndef STBI_NO_GIF
-   if (stbi__gif_test(s))  return stbi__gif_load(s,x,y,comp,req_comp);
-   #endif
-   #ifndef STBI_NO_PSD
-   if (stbi__psd_test(s))  return stbi__psd_load(s,x,y,comp,req_comp);
-   #endif
-   #ifndef STBI_NO_PIC
-   if (stbi__pic_test(s))  return stbi__pic_load(s,x,y,comp,req_comp);
-   #endif
-   #ifndef STBI_NO_PNM
-   if (stbi__pnm_test(s))  return stbi__pnm_load(s,x,y,comp,req_comp);
-   #endif
-
-   #ifndef STBI_NO_HDR
-   if (stbi__hdr_test(s)) {
-      float *hdr = stbi__hdr_load(s, x,y,comp,req_comp);
-      return stbi__hdr_to_ldr(hdr, *x, *y, req_comp ? req_comp : *comp);
-   }
-   #endif
-
-   #ifndef STBI_NO_TGA
-   // test tga last because it's a crappy test!
-   if (stbi__tga_test(s))
-      return stbi__tga_load(s,x,y,comp,req_comp);
-   #endif
-
-   return stbi__errpuc("unknown image type", "Image not of any known type, or corrupt");
-}
-
-static unsigned char *stbi__load_flip(stbi__context *s, int *x, int *y, int *comp, int req_comp)
-{
-   unsigned char *result = stbi__load_main(s, x, y, comp, req_comp);
-
-   if (stbi__vertically_flip_on_load && result != NULL) {
-      int w = *x, h = *y;
-      int depth = req_comp ? req_comp : *comp;
-      int row,col,z;
-      stbi_uc temp;
-
-      // @OPTIMIZE: use a bigger temp buffer and memcpy multiple pixels at once
-      for (row = 0; row < (h>>1); row++) {
-         for (col = 0; col < w; col++) {
-            for (z = 0; z < depth; z++) {
-               temp = result[(row * w + col) * depth + z];
-               result[(row * w + col) * depth + z] = result[((h - row - 1) * w + col) * depth + z];
-               result[((h - row - 1) * w + col) * depth + z] = temp;
-            }
-         }
-      }
-   }
-
-   return result;
-}
-
-#ifndef STBI_NO_HDR
-static void stbi__float_postprocess(float *result, int *x, int *y, int *comp, int req_comp)
-{
-   if (stbi__vertically_flip_on_load && result != NULL) {
-      int w = *x, h = *y;
-      int depth = req_comp ? req_comp : *comp;
-      int row,col,z;
-      float temp;
-
-      // @OPTIMIZE: use a bigger temp buffer and memcpy multiple pixels at once
-      for (row = 0; row < (h>>1); row++) {
-         for (col = 0; col < w; col++) {
-            for (z = 0; z < depth; z++) {
-               temp = result[(row * w + col) * depth + z];
-               result[(row * w + col) * depth + z] = result[((h - row - 1) * w + col) * depth + z];
-               result[((h - row - 1) * w + col) * depth + z] = temp;
-            }
-         }
-      }
-   }
-}
-#endif
-
-#ifndef STBI_NO_STDIO
-
-static FILE *stbi__fopen(char const *filename, char const *mode)
-{
-   FILE *f;
-#if defined(_MSC_VER) && _MSC_VER >= 1400
-   if (0 != fopen_s(&f, filename, mode))
-      f=0;
-#else
-   f = fopen(filename, mode);
-#endif
-   return f;
-}
-
-
-STBIDEF stbi_uc *stbi_load(char const *filename, int *x, int *y, int *comp, int req_comp)
-{
-   FILE *f = stbi__fopen(filename, "rb");
-   unsigned char *result;
-   if (!f) return stbi__errpuc("can't fopen", "Unable to open file");
-   result = stbi_load_from_file(f,x,y,comp,req_comp);
-   fclose(f);
-   return result;
-}
-
-STBIDEF stbi_uc *stbi_load_from_file(FILE *f, int *x, int *y, int *comp, int req_comp)
-{
-   unsigned char *result;
-   stbi__context s;
-   stbi__start_file(&s,f);
-   result = stbi__load_flip(&s,x,y,comp,req_comp);
-   if (result) {
-      // need to 'unget' all the characters in the IO buffer
-      fseek(f, - (int) (s.img_buffer_end - s.img_buffer), SEEK_CUR);
-   }
-   return result;
-}
-#endif //!STBI_NO_STDIO
-
-STBIDEF stbi_uc *stbi_load_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp)
-{
-   stbi__context s;
-   stbi__start_mem(&s,buffer,len);
-   return stbi__load_flip(&s,x,y,comp,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)
-{
-   stbi__context s;
-   stbi__start_callbacks(&s, (stbi_io_callbacks *) clbk, user);
-   return stbi__load_flip(&s,x,y,comp,req_comp);
-}
-
-#ifndef STBI_NO_LINEAR
-static float *stbi__loadf_main(stbi__context *s, int *x, int *y, int *comp, int req_comp)
-{
-   unsigned char *data;
-   #ifndef STBI_NO_HDR
-   if (stbi__hdr_test(s)) {
-      float *hdr_data = stbi__hdr_load(s,x,y,comp,req_comp);
-      if (hdr_data)
-         stbi__float_postprocess(hdr_data,x,y,comp,req_comp);
-      return hdr_data;
-   }
-   #endif
-   data = stbi__load_flip(s, x, y, comp, req_comp);
-   if (data)
-      return stbi__ldr_to_hdr(data, *x, *y, req_comp ? req_comp : *comp);
-   return stbi__errpf("unknown image type", "Image not of any known type, or corrupt");
-}
-
-STBIDEF float *stbi_loadf_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp)
-{
-   stbi__context s;
-   stbi__start_mem(&s,buffer,len);
-   return stbi__loadf_main(&s,x,y,comp,req_comp);
-}
-
-STBIDEF float *stbi_loadf_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp, int req_comp)
-{
-   stbi__context s;
-   stbi__start_callbacks(&s, (stbi_io_callbacks *) clbk, user);
-   return stbi__loadf_main(&s,x,y,comp,req_comp);
-}
-
-#ifndef STBI_NO_STDIO
-STBIDEF float *stbi_loadf(char const *filename, int *x, int *y, int *comp, int req_comp)
-{
-   float *result;
-   FILE *f = stbi__fopen(filename, "rb");
-   if (!f) return stbi__errpf("can't fopen", "Unable to open file");
-   result = stbi_loadf_from_file(f,x,y,comp,req_comp);
-   fclose(f);
-   return result;
-}
-
-STBIDEF float *stbi_loadf_from_file(FILE *f, int *x, int *y, int *comp, int req_comp)
-{
-   stbi__context s;
-   stbi__start_file(&s,f);
-   return stbi__loadf_main(&s,x,y,comp,req_comp);
-}
-#endif // !STBI_NO_STDIO
-
-#endif // !STBI_NO_LINEAR
-
-// these is-hdr-or-not is defined independent of whether STBI_NO_LINEAR is
-// defined, for API simplicity; if STBI_NO_LINEAR is defined, it always
-// reports false!
-
-STBIDEF int stbi_is_hdr_from_memory(stbi_uc const *buffer, int len)
-{
-   #ifndef STBI_NO_HDR
-   stbi__context s;
-   stbi__start_mem(&s,buffer,len);
-   return stbi__hdr_test(&s);
-   #else
-   STBI_NOTUSED(buffer);
-   STBI_NOTUSED(len);
-   return 0;
-   #endif
-}
-
-#ifndef STBI_NO_STDIO
-STBIDEF int      stbi_is_hdr          (char const *filename)
-{
-   FILE *f = stbi__fopen(filename, "rb");
-   int result=0;
-   if (f) {
-      result = stbi_is_hdr_from_file(f);
-      fclose(f);
-   }
-   return result;
-}
-
-STBIDEF int      stbi_is_hdr_from_file(FILE *f)
-{
-   #ifndef STBI_NO_HDR
-   stbi__context s;
-   stbi__start_file(&s,f);
-   return stbi__hdr_test(&s);
-   #else
-   STBI_NOTUSED(f);
-   return 0;
-   #endif
-}
-#endif // !STBI_NO_STDIO
-
-STBIDEF int      stbi_is_hdr_from_callbacks(stbi_io_callbacks const *clbk, void *user)
-{
-   #ifndef STBI_NO_HDR
-   stbi__context s;
-   stbi__start_callbacks(&s, (stbi_io_callbacks *) clbk, user);
-   return stbi__hdr_test(&s);
-   #else
-   STBI_NOTUSED(clbk);
-   STBI_NOTUSED(user);
-   return 0;
-   #endif
-}
-
-#ifndef STBI_NO_LINEAR
-static float stbi__l2h_gamma=2.2f, stbi__l2h_scale=1.0f;
-
-STBIDEF void   stbi_ldr_to_hdr_gamma(float gamma) { stbi__l2h_gamma = gamma; }
-STBIDEF void   stbi_ldr_to_hdr_scale(float scale) { stbi__l2h_scale = scale; }
-#endif
-
-static float stbi__h2l_gamma_i=1.0f/2.2f, stbi__h2l_scale_i=1.0f;
-
-STBIDEF void   stbi_hdr_to_ldr_gamma(float gamma) { stbi__h2l_gamma_i = 1/gamma; }
-STBIDEF void   stbi_hdr_to_ldr_scale(float scale) { stbi__h2l_scale_i = 1/scale; }
-
-
-//////////////////////////////////////////////////////////////////////////////
-//
-// Common code used by all image loaders
-//
-
-enum
-{
-   STBI__SCAN_load=0,
-   STBI__SCAN_type,
-   STBI__SCAN_header
-};
-
-static void stbi__refill_buffer(stbi__context *s)
-{
-   int n = (s->io.read)(s->io_user_data,(char*)s->buffer_start,s->buflen);
-   if (n == 0) {
-      // at end of file, treat same as if from memory, but need to handle case
-      // where s->img_buffer isn't pointing to safe memory, e.g. 0-byte file
-      s->read_from_callbacks = 0;
-      s->img_buffer = s->buffer_start;
-      s->img_buffer_end = s->buffer_start+1;
-      *s->img_buffer = 0;
-   } else {
-      s->img_buffer = s->buffer_start;
-      s->img_buffer_end = s->buffer_start + n;
-   }
-}
-
-stbi_inline static stbi_uc stbi__get8(stbi__context *s)
-{
-   if (s->img_buffer < s->img_buffer_end)
-      return *s->img_buffer++;
-   if (s->read_from_callbacks) {
-      stbi__refill_buffer(s);
-      return *s->img_buffer++;
-   }
-   return 0;
-}
-
-stbi_inline static int stbi__at_eof(stbi__context *s)
-{
-   if (s->io.read) {
-      if (!(s->io.eof)(s->io_user_data)) return 0;
-      // if feof() is true, check if buffer = end
-      // special case: we've only got the special 0 character at the end
-      if (s->read_from_callbacks == 0) return 1;
-   }
-
-   return s->img_buffer >= s->img_buffer_end;
-}
-
-static void stbi__skip(stbi__context *s, int n)
-{
-   if (n < 0) {
-      s->img_buffer = s->img_buffer_end;
-      return;
-   }
-   if (s->io.read) {
-      int blen = (int) (s->img_buffer_end - s->img_buffer);
-      if (blen < n) {
-         s->img_buffer = s->img_buffer_end;
-         (s->io.skip)(s->io_user_data, n - blen);
-         return;
-      }
-   }
-   s->img_buffer += n;
-}
-
-static int stbi__getn(stbi__context *s, stbi_uc *buffer, int n)
-{
-   if (s->io.read) {
-      int blen = (int) (s->img_buffer_end - s->img_buffer);
-      if (blen < n) {
-         int res, count;
-
-         memcpy(buffer, s->img_buffer, blen);
-
-         count = (s->io.read)(s->io_user_data, (char*) buffer + blen, n - blen);
-         res = (count == (n-blen));
-         s->img_buffer = s->img_buffer_end;
-         return res;
-      }
-   }
-
-   if (s->img_buffer+n <= s->img_buffer_end) {
-      memcpy(buffer, s->img_buffer, n);
-      s->img_buffer += n;
-      return 1;
-   } else
-      return 0;
-}
-
-static int stbi__get16be(stbi__context *s)
-{
-   int z = stbi__get8(s);
-   return (z << 8) + stbi__get8(s);
-}
-
-static stbi__uint32 stbi__get32be(stbi__context *s)
-{
-   stbi__uint32 z = stbi__get16be(s);
-   return (z << 16) + stbi__get16be(s);
-}
-
-#if defined(STBI_NO_BMP) && defined(STBI_NO_TGA) && defined(STBI_NO_GIF)
-// nothing
-#else
-static int stbi__get16le(stbi__context *s)
-{
-   int z = stbi__get8(s);
-   return z + (stbi__get8(s) << 8);
-}
-#endif
-
-#ifndef STBI_NO_BMP
-static stbi__uint32 stbi__get32le(stbi__context *s)
-{
-   stbi__uint32 z = stbi__get16le(s);
-   return z + (stbi__get16le(s) << 16);
-}
-#endif
-
-#define STBI__BYTECAST(x)  ((stbi_uc) ((x) & 255))  // truncate int to byte without warnings
-
-
-//////////////////////////////////////////////////////////////////////////////
-//
-//  generic converter from built-in img_n to req_comp
-//    individual types do this automatically as much as possible (e.g. jpeg
-//    does all cases internally since it needs to colorspace convert anyway,
-//    and it never has alpha, so very few cases ). png can automatically
-//    interleave an alpha=255 channel, but falls back to this for other cases
-//
-//  assume data buffer is malloced, so malloc a new one and free that one
-//  only failure mode is malloc failing
-
-static stbi_uc stbi__compute_y(int r, int g, int b)
-{
-   return (stbi_uc) (((r*77) + (g*150) +  (29*b)) >> 8);
-}
-
-static unsigned char *stbi__convert_format(unsigned char *data, int img_n, int req_comp, unsigned int x, unsigned int y)
-{
-   int i,j;
-   unsigned char *good;
-
-   if (req_comp == img_n) return data;
-   STBI_ASSERT(req_comp >= 1 && req_comp <= 4);
-
-   good = (unsigned char *) stbi__malloc(req_comp * x * y);
-   if (good == NULL) {
-      STBI_FREE(data);
-      return stbi__errpuc("outofmem", "Out of memory");
-   }
-
-   for (j=0; j < (int) y; ++j) {
-      unsigned char *src  = data + j * x * img_n   ;
-      unsigned char *dest = good + j * x * req_comp;
-
-      #define COMBO(a,b)  ((a)*8+(b))
-      #define CASE(a,b)   case COMBO(a,b): for(i=x-1; i >= 0; --i, src += a, dest += b)
-      // convert source image with img_n components to one with req_comp components;
-      // avoid switch per pixel, so use switch per scanline and massive macros
-      switch (COMBO(img_n, req_comp)) {
-         CASE(1,2) dest[0]=src[0], dest[1]=255; break;
-         CASE(1,3) dest[0]=dest[1]=dest[2]=src[0]; break;
-         CASE(1,4) dest[0]=dest[1]=dest[2]=src[0], dest[3]=255; break;
-         CASE(2,1) dest[0]=src[0]; break;
-         CASE(2,3) dest[0]=dest[1]=dest[2]=src[0]; break;
-         CASE(2,4) dest[0]=dest[1]=dest[2]=src[0], dest[3]=src[1]; break;
-         CASE(3,4) dest[0]=src[0],dest[1]=src[1],dest[2]=src[2],dest[3]=255; break;
-         CASE(3,1) dest[0]=stbi__compute_y(src[0],src[1],src[2]); break;
-         CASE(3,2) dest[0]=stbi__compute_y(src[0],src[1],src[2]), dest[1] = 255; break;
-         CASE(4,1) dest[0]=stbi__compute_y(src[0],src[1],src[2]); break;
-         CASE(4,2) dest[0]=stbi__compute_y(src[0],src[1],src[2]), dest[1] = src[3]; break;
-         CASE(4,3) dest[0]=src[0],dest[1]=src[1],dest[2]=src[2]; break;
-         default: STBI_ASSERT(0);
-      }
-      #undef CASE
-   }
-
-   STBI_FREE(data);
-   return good;
-}
-
-#ifndef STBI_NO_LINEAR
-static float   *stbi__ldr_to_hdr(stbi_uc *data, int x, int y, int comp)
-{
-   int i,k,n;
-   float *output = (float *) stbi__malloc(x * y * comp * sizeof(float));
-   if (output == NULL) { STBI_FREE(data); return stbi__errpf("outofmem", "Out of memory"); }
-   // compute number of non-alpha components
-   if (comp & 1) n = comp; else n = comp-1;
-   for (i=0; i < x*y; ++i) {
-      for (k=0; k < n; ++k) {
-         output[i*comp + k] = (float) (pow(data[i*comp+k]/255.0f, stbi__l2h_gamma) * stbi__l2h_scale);
-      }
-      if (k < comp) output[i*comp + k] = data[i*comp+k]/255.0f;
-   }
-   STBI_FREE(data);
-   return output;
-}
-#endif
-
-#ifndef STBI_NO_HDR
-#define stbi__float2int(x)   ((int) (x))
-static stbi_uc *stbi__hdr_to_ldr(float   *data, int x, int y, int comp)
-{
-   int i,k,n;
-   stbi_uc *output = (stbi_uc *) stbi__malloc(x * y * comp);
-   if (output == NULL) { STBI_FREE(data); return stbi__errpuc("outofmem", "Out of memory"); }
-   // compute number of non-alpha components
-   if (comp & 1) n = comp; else n = comp-1;
-   for (i=0; i < x*y; ++i) {
-      for (k=0; k < n; ++k) {
-         float z = (float) pow(data[i*comp+k]*stbi__h2l_scale_i, stbi__h2l_gamma_i) * 255 + 0.5f;
-         if (z < 0) z = 0;
-         if (z > 255) z = 255;
-         output[i*comp + k] = (stbi_uc) stbi__float2int(z);
-      }
-      if (k < comp) {
-         float z = data[i*comp+k] * 255 + 0.5f;
-         if (z < 0) z = 0;
-         if (z > 255) z = 255;
-         output[i*comp + k] = (stbi_uc) stbi__float2int(z);
-      }
-   }
-   STBI_FREE(data);
-   return output;
-}
-#endif
-
-//////////////////////////////////////////////////////////////////////////////
-//
-//  "baseline" JPEG/JFIF decoder
-//
-//    simple implementation
-//      - doesn't support delayed output of y-dimension
-//      - simple interface (only one output format: 8-bit interleaved RGB)
-//      - doesn't try to recover corrupt jpegs
-//      - doesn't allow partial loading, loading multiple at once
-//      - still fast on x86 (copying globals into locals doesn't help x86)
-//      - allocates lots of intermediate memory (full size of all components)
-//        - non-interleaved case requires this anyway
-//        - allows good upsampling (see next)
-//    high-quality
-//      - upsampled channels are bilinearly interpolated, even across blocks
-//      - quality integer IDCT derived from IJG's 'slow'
-//    performance
-//      - fast huffman; reasonable integer IDCT
-//      - some SIMD kernels for common paths on targets with SSE2/NEON
-//      - uses a lot of intermediate memory, could cache poorly
-
-#ifndef STBI_NO_JPEG
-
-// huffman decoding acceleration
-#define FAST_BITS   9  // larger handles more cases; smaller stomps less cache
-
-typedef struct
-{
-   stbi_uc  fast[1 << FAST_BITS];
-   // weirdly, repacking this into AoS is a 10% speed loss, instead of a win
-   stbi__uint16 code[256];
-   stbi_uc  values[256];
-   stbi_uc  size[257];
-   unsigned int maxcode[18];
-   int    delta[17];   // old 'firstsymbol' - old 'firstcode'
-} stbi__huffman;
-
-typedef struct
-{
-   stbi__context *s;
-   stbi__huffman huff_dc[4];
-   stbi__huffman huff_ac[4];
-   stbi_uc dequant[4][64];
-   stbi__int16 fast_ac[4][1 << FAST_BITS];
-
-// sizes for components, interleaved MCUs
-   int img_h_max, img_v_max;
-   int img_mcu_x, img_mcu_y;
-   int img_mcu_w, img_mcu_h;
-
-// definition of jpeg image component
-   struct
-   {
-      int id;
-      int h,v;
-      int tq;
-      int hd,ha;
-      int dc_pred;
-
-      int x,y,w2,h2;
-      stbi_uc *data;
-      void *raw_data, *raw_coeff;
-      stbi_uc *linebuf;
-      short   *coeff;   // progressive only
-      int      coeff_w, coeff_h; // number of 8x8 coefficient blocks
-   } img_comp[4];
-
-   stbi__uint32   code_buffer; // jpeg entropy-coded buffer
-   int            code_bits;   // number of valid bits
-   unsigned char  marker;      // marker seen while filling entropy buffer
-   int            nomore;      // flag if we saw a marker so must stop
-
-   int            progressive;
-   int            spec_start;
-   int            spec_end;
-   int            succ_high;
-   int            succ_low;
-   int            eob_run;
-   int            rgb;
-
-   int scan_n, order[4];
-   int restart_interval, todo;
-
-// kernels
-   void (*idct_block_kernel)(stbi_uc *out, int out_stride, short data[64]);
-   void (*YCbCr_to_RGB_kernel)(stbi_uc *out, const stbi_uc *y, const stbi_uc *pcb, const stbi_uc *pcr, int count, int step);
-   stbi_uc *(*resample_row_hv_2_kernel)(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs);
-} stbi__jpeg;
-
-static int stbi__build_huffman(stbi__huffman *h, int *count)
-{
-   int i,j,k=0,code;
-   // build size list for each symbol (from JPEG spec)
-   for (i=0; i < 16; ++i)
-      for (j=0; j < count[i]; ++j)
-         h->size[k++] = (stbi_uc) (i+1);
-   h->size[k] = 0;
-
-   // compute actual symbols (from jpeg spec)
-   code = 0;
-   k = 0;
-   for(j=1; j <= 16; ++j) {
-      // compute delta to add to code to compute symbol id
-      h->delta[j] = k - code;
-      if (h->size[k] == j) {
-         while (h->size[k] == j)
-            h->code[k++] = (stbi__uint16) (code++);
-         if (code-1 >= (1 << j)) return stbi__err("bad code lengths","Corrupt JPEG");
-      }
-      // compute largest code + 1 for this size, preshifted as needed later
-      h->maxcode[j] = code << (16-j);
-      code <<= 1;
-   }
-   h->maxcode[j] = 0xffffffff;
-
-   // build non-spec acceleration table; 255 is flag for not-accelerated
-   memset(h->fast, 255, 1 << FAST_BITS);
-   for (i=0; i < k; ++i) {
-      int s = h->size[i];
-      if (s <= FAST_BITS) {
-         int c = h->code[i] << (FAST_BITS-s);
-         int m = 1 << (FAST_BITS-s);
-         for (j=0; j < m; ++j) {
-            h->fast[c+j] = (stbi_uc) i;
-         }
-      }
-   }
-   return 1;
-}
-
-// build a table that decodes both magnitude and value of small ACs in
-// one go.
-static void stbi__build_fast_ac(stbi__int16 *fast_ac, stbi__huffman *h)
-{
-   int i;
-   for (i=0; i < (1 << FAST_BITS); ++i) {
-      stbi_uc fast = h->fast[i];
-      fast_ac[i] = 0;
-      if (fast < 255) {
-         int rs = h->values[fast];
-         int run = (rs >> 4) & 15;
-         int magbits = rs & 15;
-         int len = h->size[fast];
-
-         if (magbits && len + magbits <= FAST_BITS) {
-            // magnitude code followed by receive_extend code
-            int k = ((i << len) & ((1 << FAST_BITS) - 1)) >> (FAST_BITS - magbits);
-            int m = 1 << (magbits - 1);
-            if (k < m) k += (-1 << magbits) + 1;
-            // if the result is small enough, we can fit it in fast_ac table
-            if (k >= -128 && k <= 127)
-               fast_ac[i] = (stbi__int16) ((k << 8) + (run << 4) + (len + magbits));
-         }
-      }
-   }
-}
-
-static void stbi__grow_buffer_unsafe(stbi__jpeg *j)
-{
-   do {
-      int b = j->nomore ? 0 : stbi__get8(j->s);
-      if (b == 0xff) {
-         int c = stbi__get8(j->s);
-         if (c != 0) {
-            j->marker = (unsigned char) c;
-            j->nomore = 1;
-            return;
-         }
-      }
-      j->code_buffer |= b << (24 - j->code_bits);
-      j->code_bits += 8;
-   } while (j->code_bits <= 24);
-}
-
-// (1 << n) - 1
-static stbi__uint32 stbi__bmask[17]={0,1,3,7,15,31,63,127,255,511,1023,2047,4095,8191,16383,32767,65535};
-
-// decode a jpeg huffman value from the bitstream
-stbi_inline static int stbi__jpeg_huff_decode(stbi__jpeg *j, stbi__huffman *h)
-{
-   unsigned int temp;
-   int c,k;
-
-   if (j->code_bits < 16) stbi__grow_buffer_unsafe(j);
-
-   // look at the top FAST_BITS and determine what symbol ID it is,
-   // if the code is <= FAST_BITS
-   c = (j->code_buffer >> (32 - FAST_BITS)) & ((1 << FAST_BITS)-1);
-   k = h->fast[c];
-   if (k < 255) {
-      int s = h->size[k];
-      if (s > j->code_bits)
-         return -1;
-      j->code_buffer <<= s;
-      j->code_bits -= s;
-      return h->values[k];
-   }
-
-   // naive test is to shift the code_buffer down so k bits are
-   // valid, then test against maxcode. To speed this up, we've
-   // preshifted maxcode left so that it has (16-k) 0s at the
-   // end; in other words, regardless of the number of bits, it
-   // wants to be compared against something shifted to have 16;
-   // that way we don't need to shift inside the loop.
-   temp = j->code_buffer >> 16;
-   for (k=FAST_BITS+1 ; ; ++k)
-      if (temp < h->maxcode[k])
-         break;
-   if (k == 17) {
-      // error! code not found
-      j->code_bits -= 16;
-      return -1;
-   }
-
-   if (k > j->code_bits)
-      return -1;
-
-   // convert the huffman code to the symbol id
-   c = ((j->code_buffer >> (32 - k)) & stbi__bmask[k]) + h->delta[k];
-   STBI_ASSERT((((j->code_buffer) >> (32 - h->size[c])) & stbi__bmask[h->size[c]]) == h->code[c]);
-
-   // convert the id to a symbol
-   j->code_bits -= k;
-   j->code_buffer <<= k;
-   return h->values[c];
-}
-
-// bias[n] = (-1<<n) + 1
-static int const stbi__jbias[16] = {0,-1,-3,-7,-15,-31,-63,-127,-255,-511,-1023,-2047,-4095,-8191,-16383,-32767};
-
-// combined JPEG 'receive' and JPEG 'extend', since baseline
-// always extends everything it receives.
-stbi_inline static int stbi__extend_receive(stbi__jpeg *j, int n)
-{
-   unsigned int k;
-   int sgn;
-   if (j->code_bits < n) stbi__grow_buffer_unsafe(j);
-
-   sgn = (stbi__int32)j->code_buffer >> 31; // sign bit is always in MSB
-   k = stbi_lrot(j->code_buffer, n);
-   STBI_ASSERT(n >= 0 && n < (int) (sizeof(stbi__bmask)/sizeof(*stbi__bmask)));
-   j->code_buffer = k & ~stbi__bmask[n];
-   k &= stbi__bmask[n];
-   j->code_bits -= n;
-   return k + (stbi__jbias[n] & ~sgn);
-}
-
-// get some unsigned bits
-stbi_inline static int stbi__jpeg_get_bits(stbi__jpeg *j, int n)
-{
-   unsigned int k;
-   if (j->code_bits < n) stbi__grow_buffer_unsafe(j);
-   k = stbi_lrot(j->code_buffer, n);
-   j->code_buffer = k & ~stbi__bmask[n];
-   k &= stbi__bmask[n];
-   j->code_bits -= n;
-   return k;
-}
-
-stbi_inline static int stbi__jpeg_get_bit(stbi__jpeg *j)
-{
-   unsigned int k;
-   if (j->code_bits < 1) stbi__grow_buffer_unsafe(j);
-   k = j->code_buffer;
-   j->code_buffer <<= 1;
-   --j->code_bits;
-   return k & 0x80000000;
-}
-
-// given a value that's at position X in the zigzag stream,
-// where does it appear in the 8x8 matrix coded as row-major?
-static stbi_uc stbi__jpeg_dezigzag[64+15] =
-{
-    0,  1,  8, 16,  9,  2,  3, 10,
-   17, 24, 32, 25, 18, 11,  4,  5,
-   12, 19, 26, 33, 40, 48, 41, 34,
-   27, 20, 13,  6,  7, 14, 21, 28,
-   35, 42, 49, 56, 57, 50, 43, 36,
-   29, 22, 15, 23, 30, 37, 44, 51,
-   58, 59, 52, 45, 38, 31, 39, 46,
-   53, 60, 61, 54, 47, 55, 62, 63,
-   // let corrupt input sample past end
-   63, 63, 63, 63, 63, 63, 63, 63,
-   63, 63, 63, 63, 63, 63, 63
-};
-
-// decode one 64-entry block--
-static int stbi__jpeg_decode_block(stbi__jpeg *j, short data[64], stbi__huffman *hdc, stbi__huffman *hac, stbi__int16 *fac, int b, stbi_uc *dequant)
-{
-   int diff,dc,k;
-   int t;
-
-   if (j->code_bits < 16) stbi__grow_buffer_unsafe(j);
-   t = stbi__jpeg_huff_decode(j, hdc);
-   if (t < 0) return stbi__err("bad huffman code","Corrupt JPEG");
-
-   // 0 all the ac values now so we can do it 32-bits at a time
-   memset(data,0,64*sizeof(data[0]));
-
-   diff = t ? stbi__extend_receive(j, t) : 0;
-   dc = j->img_comp[b].dc_pred + diff;
-   j->img_comp[b].dc_pred = dc;
-   data[0] = (short) (dc * dequant[0]);
-
-   // decode AC components, see JPEG spec
-   k = 1;
-   do {
-      unsigned int zig;
-      int c,r,s;
-      if (j->code_bits < 16) stbi__grow_buffer_unsafe(j);
-      c = (j->code_buffer >> (32 - FAST_BITS)) & ((1 << FAST_BITS)-1);
-      r = fac[c];
-      if (r) { // fast-AC path
-         k += (r >> 4) & 15; // run
-         s = r & 15; // combined length
-         j->code_buffer <<= s;
-         j->code_bits -= s;
-         // decode into unzigzag'd location
-         zig = stbi__jpeg_dezigzag[k++];
-         data[zig] = (short) ((r >> 8) * dequant[zig]);
-      } else {
-         int rs = stbi__jpeg_huff_decode(j, hac);
-         if (rs < 0) return stbi__err("bad huffman code","Corrupt JPEG");
-         s = rs & 15;
-         r = rs >> 4;
-         if (s == 0) {
-            if (rs != 0xf0) break; // end block
-            k += 16;
-         } else {
-            k += r;
-            // decode into unzigzag'd location
-            zig = stbi__jpeg_dezigzag[k++];
-            data[zig] = (short) (stbi__extend_receive(j,s) * dequant[zig]);
-         }
-      }
-   } while (k < 64);
-   return 1;
-}
-
-static int stbi__jpeg_decode_block_prog_dc(stbi__jpeg *j, short data[64], stbi__huffman *hdc, int b)
-{
-   int diff,dc;
-   int t;
-   if (j->spec_end != 0) return stbi__err("can't merge dc and ac", "Corrupt JPEG");
-
-   if (j->code_bits < 16) stbi__grow_buffer_unsafe(j);
-
-   if (j->succ_high == 0) {
-      // first scan for DC coefficient, must be first
-      memset(data,0,64*sizeof(data[0])); // 0 all the ac values now
-      t = stbi__jpeg_huff_decode(j, hdc);
-      diff = t ? stbi__extend_receive(j, t) : 0;
-
-      dc = j->img_comp[b].dc_pred + diff;
-      j->img_comp[b].dc_pred = dc;
-      data[0] = (short) (dc << j->succ_low);
-   } else {
-      // refinement scan for DC coefficient
-      if (stbi__jpeg_get_bit(j))
-         data[0] += (short) (1 << j->succ_low);
-   }
-   return 1;
-}
-
-// @OPTIMIZE: store non-zigzagged during the decode passes,
-// and only de-zigzag when dequantizing
-static int stbi__jpeg_decode_block_prog_ac(stbi__jpeg *j, short data[64], stbi__huffman *hac, stbi__int16 *fac)
-{
-   int k;
-   if (j->spec_start == 0) return stbi__err("can't merge dc and ac", "Corrupt JPEG");
-
-   if (j->succ_high == 0) {
-      int shift = j->succ_low;
-
-      if (j->eob_run) {
-         --j->eob_run;
-         return 1;
-      }
-
-      k = j->spec_start;
-      do {
-         unsigned int zig;
-         int c,r,s;
-         if (j->code_bits < 16) stbi__grow_buffer_unsafe(j);
-         c = (j->code_buffer >> (32 - FAST_BITS)) & ((1 << FAST_BITS)-1);
-         r = fac[c];
-         if (r) { // fast-AC path
-            k += (r >> 4) & 15; // run
-            s = r & 15; // combined length
-            j->code_buffer <<= s;
-            j->code_bits -= s;
-            zig = stbi__jpeg_dezigzag[k++];
-            data[zig] = (short) ((r >> 8) << shift);
-         } else {
-            int rs = stbi__jpeg_huff_decode(j, hac);
-            if (rs < 0) return stbi__err("bad huffman code","Corrupt JPEG");
-            s = rs & 15;
-            r = rs >> 4;
-            if (s == 0) {
-               if (r < 15) {
-                  j->eob_run = (1 << r);
-                  if (r)
-                     j->eob_run += stbi__jpeg_get_bits(j, r);
-                  --j->eob_run;
-                  break;
-               }
-               k += 16;
-            } else {
-               k += r;
-               zig = stbi__jpeg_dezigzag[k++];
-               data[zig] = (short) (stbi__extend_receive(j,s) << shift);
-            }
-         }
-      } while (k <= j->spec_end);
-   } else {
-      // refinement scan for these AC coefficients
-
-      short bit = (short) (1 << j->succ_low);
-
-      if (j->eob_run) {
-         --j->eob_run;
-         for (k = j->spec_start; k <= j->spec_end; ++k) {
-            short *p = &data[stbi__jpeg_dezigzag[k]];
-            if (*p != 0)
-               if (stbi__jpeg_get_bit(j))
-                  if ((*p & bit)==0) {
-                     if (*p > 0)
-                        *p += bit;
-                     else
-                        *p -= bit;
-                  }
-         }
-      } else {
-         k = j->spec_start;
-         do {
-            int r,s;
-            int rs = stbi__jpeg_huff_decode(j, hac); // @OPTIMIZE see if we can use the fast path here, advance-by-r is so slow, eh
-            if (rs < 0) return stbi__err("bad huffman code","Corrupt JPEG");
-            s = rs & 15;
-            r = rs >> 4;
-            if (s == 0) {
-               if (r < 15) {
-                  j->eob_run = (1 << r) - 1;
-                  if (r)
-                     j->eob_run += stbi__jpeg_get_bits(j, r);
-                  r = 64; // force end of block
-               } else {
-                  // r=15 s=0 should write 16 0s, so we just do
-                  // a run of 15 0s and then write s (which is 0),
-                  // so we don't have to do anything special here
-               }
-            } else {
-               if (s != 1) return stbi__err("bad huffman code", "Corrupt JPEG");
-               // sign bit
-               if (stbi__jpeg_get_bit(j))
-                  s = bit;
-               else
-                  s = -bit;
-            }
-
-            // advance by r
-            while (k <= j->spec_end) {
-               short *p = &data[stbi__jpeg_dezigzag[k++]];
-               if (*p != 0) {
-                  if (stbi__jpeg_get_bit(j))
-                     if ((*p & bit)==0) {
-                        if (*p > 0)
-                           *p += bit;
-                        else
-                           *p -= bit;
-                     }
-               } else {
-                  if (r == 0) {
-                     *p = (short) s;
-                     break;
-                  }
-                  --r;
-               }
-            }
-         } while (k <= j->spec_end);
-      }
-   }
-   return 1;
-}
-
-// take a -128..127 value and stbi__clamp it and convert to 0..255
-stbi_inline static stbi_uc stbi__clamp(int x)
-{
-   // trick to use a single test to catch both cases
-   if ((unsigned int) x > 255) {
-      if (x < 0) return 0;
-      if (x > 255) return 255;
-   }
-   return (stbi_uc) x;
-}
-
-#define stbi__f2f(x)  ((int) (((x) * 4096 + 0.5)))
-#define stbi__fsh(x)  ((x) << 12)
-
-// derived from jidctint -- DCT_ISLOW
-#define STBI__IDCT_1D(s0,s1,s2,s3,s4,s5,s6,s7) \
-   int t0,t1,t2,t3,p1,p2,p3,p4,p5,x0,x1,x2,x3; \
-   p2 = s2;                                    \
-   p3 = s6;                                    \
-   p1 = (p2+p3) * stbi__f2f(0.5411961f);       \
-   t2 = p1 + p3*stbi__f2f(-1.847759065f);      \
-   t3 = p1 + p2*stbi__f2f( 0.765366865f);      \
-   p2 = s0;                                    \
-   p3 = s4;                                    \
-   t0 = stbi__fsh(p2+p3);                      \
-   t1 = stbi__fsh(p2-p3);                      \
-   x0 = t0+t3;                                 \
-   x3 = t0-t3;                                 \
-   x1 = t1+t2;                                 \
-   x2 = t1-t2;                                 \
-   t0 = s7;                                    \
-   t1 = s5;                                    \
-   t2 = s3;                                    \
-   t3 = s1;                                    \
-   p3 = t0+t2;                                 \
-   p4 = t1+t3;                                 \
-   p1 = t0+t3;                                 \
-   p2 = t1+t2;                                 \
-   p5 = (p3+p4)*stbi__f2f( 1.175875602f);      \
-   t0 = t0*stbi__f2f( 0.298631336f);           \
-   t1 = t1*stbi__f2f( 2.053119869f);           \
-   t2 = t2*stbi__f2f( 3.072711026f);           \
-   t3 = t3*stbi__f2f( 1.501321110f);           \
-   p1 = p5 + p1*stbi__f2f(-0.899976223f);      \
-   p2 = p5 + p2*stbi__f2f(-2.562915447f);      \
-   p3 = p3*stbi__f2f(-1.961570560f);           \
-   p4 = p4*stbi__f2f(-0.390180644f);           \
-   t3 += p1+p4;                                \
-   t2 += p2+p3;                                \
-   t1 += p2+p4;                                \
-   t0 += p1+p3;
-
-static void stbi__idct_block(stbi_uc *out, int out_stride, short data[64])
-{
-   int i,val[64],*v=val;
-   stbi_uc *o;
-   short *d = data;
-
-   // columns
-   for (i=0; i < 8; ++i,++d, ++v) {
-      // if all zeroes, shortcut -- this avoids dequantizing 0s and IDCTing
-      if (d[ 8]==0 && d[16]==0 && d[24]==0 && d[32]==0
-           && d[40]==0 && d[48]==0 && d[56]==0) {
-         //    no shortcut                 0     seconds
-         //    (1|2|3|4|5|6|7)==0          0     seconds
-         //    all separate               -0.047 seconds
-         //    1 && 2|3 && 4|5 && 6|7:    -0.047 seconds
-         int dcterm = d[0] << 2;
-         v[0] = v[8] = v[16] = v[24] = v[32] = v[40] = v[48] = v[56] = dcterm;
-      } else {
-         STBI__IDCT_1D(d[ 0],d[ 8],d[16],d[24],d[32],d[40],d[48],d[56])
-         // constants scaled things up by 1<<12; let's bring them back
-         // down, but keep 2 extra bits of precision
-         x0 += 512; x1 += 512; x2 += 512; x3 += 512;
-         v[ 0] = (x0+t3) >> 10;
-         v[56] = (x0-t3) >> 10;
-         v[ 8] = (x1+t2) >> 10;
-         v[48] = (x1-t2) >> 10;
-         v[16] = (x2+t1) >> 10;
-         v[40] = (x2-t1) >> 10;
-         v[24] = (x3+t0) >> 10;
-         v[32] = (x3-t0) >> 10;
-      }
-   }
-
-   for (i=0, v=val, o=out; i < 8; ++i,v+=8,o+=out_stride) {
-      // no fast case since the first 1D IDCT spread components out
-      STBI__IDCT_1D(v[0],v[1],v[2],v[3],v[4],v[5],v[6],v[7])
-      // constants scaled things up by 1<<12, plus we had 1<<2 from first
-      // loop, plus horizontal and vertical each scale by sqrt(8) so together
-      // we've got an extra 1<<3, so 1<<17 total we need to remove.
-      // so we want to round that, which means adding 0.5 * 1<<17,
-      // aka 65536. Also, we'll end up with -128 to 127 that we want
-      // to encode as 0..255 by adding 128, so we'll add that before the shift
-      x0 += 65536 + (128<<17);
-      x1 += 65536 + (128<<17);
-      x2 += 65536 + (128<<17);
-      x3 += 65536 + (128<<17);
-      // tried computing the shifts into temps, or'ing the temps to see
-      // if any were out of range, but that was slower
-      o[0] = stbi__clamp((x0+t3) >> 17);
-      o[7] = stbi__clamp((x0-t3) >> 17);
-      o[1] = stbi__clamp((x1+t2) >> 17);
-      o[6] = stbi__clamp((x1-t2) >> 17);
-      o[2] = stbi__clamp((x2+t1) >> 17);
-      o[5] = stbi__clamp((x2-t1) >> 17);
-      o[3] = stbi__clamp((x3+t0) >> 17);
-      o[4] = stbi__clamp((x3-t0) >> 17);
-   }
-}
-
-#ifdef STBI_SSE2
-// sse2 integer IDCT. not the fastest possible implementation but it
-// produces bit-identical results to the generic C version so it's
-// fully "transparent".
-static void stbi__idct_simd(stbi_uc *out, int out_stride, short data[64])
-{
-   // This is constructed to match our regular (generic) integer IDCT exactly.
-   __m128i row0, row1, row2, row3, row4, row5, row6, row7;
-   __m128i tmp;
-
-   // dot product constant: even elems=x, odd elems=y
-   #define dct_const(x,y)  _mm_setr_epi16((x),(y),(x),(y),(x),(y),(x),(y))
-
-   // out(0) = c0[even]*x + c0[odd]*y   (c0, x, y 16-bit, out 32-bit)
-   // out(1) = c1[even]*x + c1[odd]*y
-   #define dct_rot(out0,out1, x,y,c0,c1) \
-      __m128i c0##lo = _mm_unpacklo_epi16((x),(y)); \
-      __m128i c0##hi = _mm_unpackhi_epi16((x),(y)); \
-      __m128i out0##_l = _mm_madd_epi16(c0##lo, c0); \
-      __m128i out0##_h = _mm_madd_epi16(c0##hi, c0); \
-      __m128i out1##_l = _mm_madd_epi16(c0##lo, c1); \
-      __m128i out1##_h = _mm_madd_epi16(c0##hi, c1)
-
-   // out = in << 12  (in 16-bit, out 32-bit)
-   #define dct_widen(out, in) \
-      __m128i out##_l = _mm_srai_epi32(_mm_unpacklo_epi16(_mm_setzero_si128(), (in)), 4); \
-      __m128i out##_h = _mm_srai_epi32(_mm_unpackhi_epi16(_mm_setzero_si128(), (in)), 4)
-
-   // wide add
-   #define dct_wadd(out, a, b) \
-      __m128i out##_l = _mm_add_epi32(a##_l, b##_l); \
-      __m128i out##_h = _mm_add_epi32(a##_h, b##_h)
-
-   // wide sub
-   #define dct_wsub(out, a, b) \
-      __m128i out##_l = _mm_sub_epi32(a##_l, b##_l); \
-      __m128i out##_h = _mm_sub_epi32(a##_h, b##_h)
-
-   // butterfly a/b, add bias, then shift by "s" and pack
-   #define dct_bfly32o(out0, out1, a,b,bias,s) \
-      { \
-         __m128i abiased_l = _mm_add_epi32(a##_l, bias); \
-         __m128i abiased_h = _mm_add_epi32(a##_h, bias); \
-         dct_wadd(sum, abiased, b); \
-         dct_wsub(dif, abiased, b); \
-         out0 = _mm_packs_epi32(_mm_srai_epi32(sum_l, s), _mm_srai_epi32(sum_h, s)); \
-         out1 = _mm_packs_epi32(_mm_srai_epi32(dif_l, s), _mm_srai_epi32(dif_h, s)); \
-      }
-
-   // 8-bit interleave step (for transposes)
-   #define dct_interleave8(a, b) \
-      tmp = a; \
-      a = _mm_unpacklo_epi8(a, b); \
-      b = _mm_unpackhi_epi8(tmp, b)
-
-   // 16-bit interleave step (for transposes)
-   #define dct_interleave16(a, b) \
-      tmp = a; \
-      a = _mm_unpacklo_epi16(a, b); \
-      b = _mm_unpackhi_epi16(tmp, b)
-
-   #define dct_pass(bias,shift) \
-      { \
-         /* even part */ \
-         dct_rot(t2e,t3e, row2,row6, rot0_0,rot0_1); \
-         __m128i sum04 = _mm_add_epi16(row0, row4); \
-         __m128i dif04 = _mm_sub_epi16(row0, row4); \
-         dct_widen(t0e, sum04); \
-         dct_widen(t1e, dif04); \
-         dct_wadd(x0, t0e, t3e); \
-         dct_wsub(x3, t0e, t3e); \
-         dct_wadd(x1, t1e, t2e); \
-         dct_wsub(x2, t1e, t2e); \
-         /* odd part */ \
-         dct_rot(y0o,y2o, row7,row3, rot2_0,rot2_1); \
-         dct_rot(y1o,y3o, row5,row1, rot3_0,rot3_1); \
-         __m128i sum17 = _mm_add_epi16(row1, row7); \
-         __m128i sum35 = _mm_add_epi16(row3, row5); \
-         dct_rot(y4o,y5o, sum17,sum35, rot1_0,rot1_1); \
-         dct_wadd(x4, y0o, y4o); \
-         dct_wadd(x5, y1o, y5o); \
-         dct_wadd(x6, y2o, y5o); \
-         dct_wadd(x7, y3o, y4o); \
-         dct_bfly32o(row0,row7, x0,x7,bias,shift); \
-         dct_bfly32o(row1,row6, x1,x6,bias,shift); \
-         dct_bfly32o(row2,row5, x2,x5,bias,shift); \
-         dct_bfly32o(row3,row4, x3,x4,bias,shift); \
-      }
-
-   __m128i rot0_0 = dct_const(stbi__f2f(0.5411961f), stbi__f2f(0.5411961f) + stbi__f2f(-1.847759065f));
-   __m128i rot0_1 = dct_const(stbi__f2f(0.5411961f) + stbi__f2f( 0.765366865f), stbi__f2f(0.5411961f));
-   __m128i rot1_0 = dct_const(stbi__f2f(1.175875602f) + stbi__f2f(-0.899976223f), stbi__f2f(1.175875602f));
-   __m128i rot1_1 = dct_const(stbi__f2f(1.175875602f), stbi__f2f(1.175875602f) + stbi__f2f(-2.562915447f));
-   __m128i rot2_0 = dct_const(stbi__f2f(-1.961570560f) + stbi__f2f( 0.298631336f), stbi__f2f(-1.961570560f));
-   __m128i rot2_1 = dct_const(stbi__f2f(-1.961570560f), stbi__f2f(-1.961570560f) + stbi__f2f( 3.072711026f));
-   __m128i rot3_0 = dct_const(stbi__f2f(-0.390180644f) + stbi__f2f( 2.053119869f), stbi__f2f(-0.390180644f));
-   __m128i rot3_1 = dct_const(stbi__f2f(-0.390180644f), stbi__f2f(-0.390180644f) + stbi__f2f( 1.501321110f));
-
-   // rounding biases in column/row passes, see stbi__idct_block for explanation.
-   __m128i bias_0 = _mm_set1_epi32(512);
-   __m128i bias_1 = _mm_set1_epi32(65536 + (128<<17));
-
-   // load
-   row0 = _mm_load_si128((const __m128i *) (data + 0*8));
-   row1 = _mm_load_si128((const __m128i *) (data + 1*8));
-   row2 = _mm_load_si128((const __m128i *) (data + 2*8));
-   row3 = _mm_load_si128((const __m128i *) (data + 3*8));
-   row4 = _mm_load_si128((const __m128i *) (data + 4*8));
-   row5 = _mm_load_si128((const __m128i *) (data + 5*8));
-   row6 = _mm_load_si128((const __m128i *) (data + 6*8));
-   row7 = _mm_load_si128((const __m128i *) (data + 7*8));
-
-   // column pass
-   dct_pass(bias_0, 10);
-
-   {
-      // 16bit 8x8 transpose pass 1
-      dct_interleave16(row0, row4);
-      dct_interleave16(row1, row5);
-      dct_interleave16(row2, row6);
-      dct_interleave16(row3, row7);
-
-      // transpose pass 2
-      dct_interleave16(row0, row2);
-      dct_interleave16(row1, row3);
-      dct_interleave16(row4, row6);
-      dct_interleave16(row5, row7);
-
-      // transpose pass 3
-      dct_interleave16(row0, row1);
-      dct_interleave16(row2, row3);
-      dct_interleave16(row4, row5);
-      dct_interleave16(row6, row7);
-   }
-
-   // row pass
-   dct_pass(bias_1, 17);
-
-   {
-      // pack
-      __m128i p0 = _mm_packus_epi16(row0, row1); // a0a1a2a3...a7b0b1b2b3...b7
-      __m128i p1 = _mm_packus_epi16(row2, row3);
-      __m128i p2 = _mm_packus_epi16(row4, row5);
-      __m128i p3 = _mm_packus_epi16(row6, row7);
-
-      // 8bit 8x8 transpose pass 1
-      dct_interleave8(p0, p2); // a0e0a1e1...
-      dct_interleave8(p1, p3); // c0g0c1g1...
-
-      // transpose pass 2
-      dct_interleave8(p0, p1); // a0c0e0g0...
-      dct_interleave8(p2, p3); // b0d0f0h0...
-
-      // transpose pass 3
-      dct_interleave8(p0, p2); // a0b0c0d0...
-      dct_interleave8(p1, p3); // a4b4c4d4...
-
-      // store
-      _mm_storel_epi64((__m128i *) out, p0); out += out_stride;
-      _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p0, 0x4e)); out += out_stride;
-      _mm_storel_epi64((__m128i *) out, p2); out += out_stride;
-      _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p2, 0x4e)); out += out_stride;
-      _mm_storel_epi64((__m128i *) out, p1); out += out_stride;
-      _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p1, 0x4e)); out += out_stride;
-      _mm_storel_epi64((__m128i *) out, p3); out += out_stride;
-      _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p3, 0x4e));
-   }
-
-#undef dct_const
-#undef dct_rot
-#undef dct_widen
-#undef dct_wadd
-#undef dct_wsub
-#undef dct_bfly32o
-#undef dct_interleave8
-#undef dct_interleave16
-#undef dct_pass
-}
-
-#endif // STBI_SSE2
-
-#ifdef STBI_NEON
-
-// NEON integer IDCT. should produce bit-identical
-// results to the generic C version.
-static void stbi__idct_simd(stbi_uc *out, int out_stride, short data[64])
-{
-   int16x8_t row0, row1, row2, row3, row4, row5, row6, row7;
-
-   int16x4_t rot0_0 = vdup_n_s16(stbi__f2f(0.5411961f));
-   int16x4_t rot0_1 = vdup_n_s16(stbi__f2f(-1.847759065f));
-   int16x4_t rot0_2 = vdup_n_s16(stbi__f2f( 0.765366865f));
-   int16x4_t rot1_0 = vdup_n_s16(stbi__f2f( 1.175875602f));
-   int16x4_t rot1_1 = vdup_n_s16(stbi__f2f(-0.899976223f));
-   int16x4_t rot1_2 = vdup_n_s16(stbi__f2f(-2.562915447f));
-   int16x4_t rot2_0 = vdup_n_s16(stbi__f2f(-1.961570560f));
-   int16x4_t rot2_1 = vdup_n_s16(stbi__f2f(-0.390180644f));
-   int16x4_t rot3_0 = vdup_n_s16(stbi__f2f( 0.298631336f));
-   int16x4_t rot3_1 = vdup_n_s16(stbi__f2f( 2.053119869f));
-   int16x4_t rot3_2 = vdup_n_s16(stbi__f2f( 3.072711026f));
-   int16x4_t rot3_3 = vdup_n_s16(stbi__f2f( 1.501321110f));
-
-#define dct_long_mul(out, inq, coeff) \
-   int32x4_t out##_l = vmull_s16(vget_low_s16(inq), coeff); \
-   int32x4_t out##_h = vmull_s16(vget_high_s16(inq), coeff)
-
-#define dct_long_mac(out, acc, inq, coeff) \
-   int32x4_t out##_l = vmlal_s16(acc##_l, vget_low_s16(inq), coeff); \
-   int32x4_t out##_h = vmlal_s16(acc##_h, vget_high_s16(inq), coeff)
-
-#define dct_widen(out, inq) \
-   int32x4_t out##_l = vshll_n_s16(vget_low_s16(inq), 12); \
-   int32x4_t out##_h = vshll_n_s16(vget_high_s16(inq), 12)
-
-// wide add
-#define dct_wadd(out, a, b) \
-   int32x4_t out##_l = vaddq_s32(a##_l, b##_l); \
-   int32x4_t out##_h = vaddq_s32(a##_h, b##_h)
-
-// wide sub
-#define dct_wsub(out, a, b) \
-   int32x4_t out##_l = vsubq_s32(a##_l, b##_l); \
-   int32x4_t out##_h = vsubq_s32(a##_h, b##_h)
-
-// butterfly a/b, then shift using "shiftop" by "s" and pack
-#define dct_bfly32o(out0,out1, a,b,shiftop,s) \
-   { \
-      dct_wadd(sum, a, b); \
-      dct_wsub(dif, a, b); \
-      out0 = vcombine_s16(shiftop(sum_l, s), shiftop(sum_h, s)); \
-      out1 = vcombine_s16(shiftop(dif_l, s), shiftop(dif_h, s)); \
-   }
-
-#define dct_pass(shiftop, shift) \
-   { \
-      /* even part */ \
-      int16x8_t sum26 = vaddq_s16(row2, row6); \
-      dct_long_mul(p1e, sum26, rot0_0); \
-      dct_long_mac(t2e, p1e, row6, rot0_1); \
-      dct_long_mac(t3e, p1e, row2, rot0_2); \
-      int16x8_t sum04 = vaddq_s16(row0, row4); \
-      int16x8_t dif04 = vsubq_s16(row0, row4); \
-      dct_widen(t0e, sum04); \
-      dct_widen(t1e, dif04); \
-      dct_wadd(x0, t0e, t3e); \
-      dct_wsub(x3, t0e, t3e); \
-      dct_wadd(x1, t1e, t2e); \
-      dct_wsub(x2, t1e, t2e); \
-      /* odd part */ \
-      int16x8_t sum15 = vaddq_s16(row1, row5); \
-      int16x8_t sum17 = vaddq_s16(row1, row7); \
-      int16x8_t sum35 = vaddq_s16(row3, row5); \
-      int16x8_t sum37 = vaddq_s16(row3, row7); \
-      int16x8_t sumodd = vaddq_s16(sum17, sum35); \
-      dct_long_mul(p5o, sumodd, rot1_0); \
-      dct_long_mac(p1o, p5o, sum17, rot1_1); \
-      dct_long_mac(p2o, p5o, sum35, rot1_2); \
-      dct_long_mul(p3o, sum37, rot2_0); \
-      dct_long_mul(p4o, sum15, rot2_1); \
-      dct_wadd(sump13o, p1o, p3o); \
-      dct_wadd(sump24o, p2o, p4o); \
-      dct_wadd(sump23o, p2o, p3o); \
-      dct_wadd(sump14o, p1o, p4o); \
-      dct_long_mac(x4, sump13o, row7, rot3_0); \
-      dct_long_mac(x5, sump24o, row5, rot3_1); \
-      dct_long_mac(x6, sump23o, row3, rot3_2); \
-      dct_long_mac(x7, sump14o, row1, rot3_3); \
-      dct_bfly32o(row0,row7, x0,x7,shiftop,shift); \
-      dct_bfly32o(row1,row6, x1,x6,shiftop,shift); \
-      dct_bfly32o(row2,row5, x2,x5,shiftop,shift); \
-      dct_bfly32o(row3,row4, x3,x4,shiftop,shift); \
-   }
-
-   // load
-   row0 = vld1q_s16(data + 0*8);
-   row1 = vld1q_s16(data + 1*8);
-   row2 = vld1q_s16(data + 2*8);
-   row3 = vld1q_s16(data + 3*8);
-   row4 = vld1q_s16(data + 4*8);
-   row5 = vld1q_s16(data + 5*8);
-   row6 = vld1q_s16(data + 6*8);
-   row7 = vld1q_s16(data + 7*8);
-
-   // add DC bias
-   row0 = vaddq_s16(row0, vsetq_lane_s16(1024, vdupq_n_s16(0), 0));
-
-   // column pass
-   dct_pass(vrshrn_n_s32, 10);
-
-   // 16bit 8x8 transpose
-   {
-// these three map to a single VTRN.16, VTRN.32, and VSWP, respectively.
-// whether compilers actually get this is another story, sadly.
-#define dct_trn16(x, y) { int16x8x2_t t = vtrnq_s16(x, y); x = t.val[0]; y = t.val[1]; }
-#define dct_trn32(x, y) { int32x4x2_t t = vtrnq_s32(vreinterpretq_s32_s16(x), vreinterpretq_s32_s16(y)); x = vreinterpretq_s16_s32(t.val[0]); y = vreinterpretq_s16_s32(t.val[1]); }
-#define dct_trn64(x, y) { int16x8_t x0 = x; int16x8_t y0 = y; x = vcombine_s16(vget_low_s16(x0), vget_low_s16(y0)); y = vcombine_s16(vget_high_s16(x0), vget_high_s16(y0)); }
-
-      // pass 1
-      dct_trn16(row0, row1); // a0b0a2b2a4b4a6b6
-      dct_trn16(row2, row3);
-      dct_trn16(row4, row5);
-      dct_trn16(row6, row7);
-
-      // pass 2
-      dct_trn32(row0, row2); // a0b0c0d0a4b4c4d4
-      dct_trn32(row1, row3);
-      dct_trn32(row4, row6);
-      dct_trn32(row5, row7);
-
-      // pass 3
-      dct_trn64(row0, row4); // a0b0c0d0e0f0g0h0
-      dct_trn64(row1, row5);
-      dct_trn64(row2, row6);
-      dct_trn64(row3, row7);
-
-#undef dct_trn16
-#undef dct_trn32
-#undef dct_trn64
-   }
-
-   // row pass
-   // vrshrn_n_s32 only supports shifts up to 16, we need
-   // 17. so do a non-rounding shift of 16 first then follow
-   // up with a rounding shift by 1.
-   dct_pass(vshrn_n_s32, 16);
-
-   {
-      // pack and round
-      uint8x8_t p0 = vqrshrun_n_s16(row0, 1);
-      uint8x8_t p1 = vqrshrun_n_s16(row1, 1);
-      uint8x8_t p2 = vqrshrun_n_s16(row2, 1);
-      uint8x8_t p3 = vqrshrun_n_s16(row3, 1);
-      uint8x8_t p4 = vqrshrun_n_s16(row4, 1);
-      uint8x8_t p5 = vqrshrun_n_s16(row5, 1);
-      uint8x8_t p6 = vqrshrun_n_s16(row6, 1);
-      uint8x8_t p7 = vqrshrun_n_s16(row7, 1);
-
-      // again, these can translate into one instruction, but often don't.
-#define dct_trn8_8(x, y) { uint8x8x2_t t = vtrn_u8(x, y); x = t.val[0]; y = t.val[1]; }
-#define dct_trn8_16(x, y) { uint16x4x2_t t = vtrn_u16(vreinterpret_u16_u8(x), vreinterpret_u16_u8(y)); x = vreinterpret_u8_u16(t.val[0]); y = vreinterpret_u8_u16(t.val[1]); }
-#define dct_trn8_32(x, y) { uint32x2x2_t t = vtrn_u32(vreinterpret_u32_u8(x), vreinterpret_u32_u8(y)); x = vreinterpret_u8_u32(t.val[0]); y = vreinterpret_u8_u32(t.val[1]); }
-
-      // sadly can't use interleaved stores here since we only write
-      // 8 bytes to each scan line!
-
-      // 8x8 8-bit transpose pass 1
-      dct_trn8_8(p0, p1);
-      dct_trn8_8(p2, p3);
-      dct_trn8_8(p4, p5);
-      dct_trn8_8(p6, p7);
-
-      // pass 2
-      dct_trn8_16(p0, p2);
-      dct_trn8_16(p1, p3);
-      dct_trn8_16(p4, p6);
-      dct_trn8_16(p5, p7);
-
-      // pass 3
-      dct_trn8_32(p0, p4);
-      dct_trn8_32(p1, p5);
-      dct_trn8_32(p2, p6);
-      dct_trn8_32(p3, p7);
-
-      // store
-      vst1_u8(out, p0); out += out_stride;
-      vst1_u8(out, p1); out += out_stride;
-      vst1_u8(out, p2); out += out_stride;
-      vst1_u8(out, p3); out += out_stride;
-      vst1_u8(out, p4); out += out_stride;
-      vst1_u8(out, p5); out += out_stride;
-      vst1_u8(out, p6); out += out_stride;
-      vst1_u8(out, p7);
-
-#undef dct_trn8_8
-#undef dct_trn8_16
-#undef dct_trn8_32
-   }
-
-#undef dct_long_mul
-#undef dct_long_mac
-#undef dct_widen
-#undef dct_wadd
-#undef dct_wsub
-#undef dct_bfly32o
-#undef dct_pass
-}
-
-#endif // STBI_NEON
-
-#define STBI__MARKER_none  0xff
-// if there's a pending marker from the entropy stream, return that
-// otherwise, fetch from the stream and get a marker. if there's no
-// marker, return 0xff, which is never a valid marker value
-static stbi_uc stbi__get_marker(stbi__jpeg *j)
-{
-   stbi_uc x;
-   if (j->marker != STBI__MARKER_none) { x = j->marker; j->marker = STBI__MARKER_none; return x; }
-   x = stbi__get8(j->s);
-   if (x != 0xff) return STBI__MARKER_none;
-   while (x == 0xff)
-      x = stbi__get8(j->s);
-   return x;
-}
-
-// in each scan, we'll have scan_n components, and the order
-// of the components is specified by order[]
-#define STBI__RESTART(x)     ((x) >= 0xd0 && (x) <= 0xd7)
-
-// after a restart interval, stbi__jpeg_reset the entropy decoder and
-// the dc prediction
-static void stbi__jpeg_reset(stbi__jpeg *j)
-{
-   j->code_bits = 0;
-   j->code_buffer = 0;
-   j->nomore = 0;
-   j->img_comp[0].dc_pred = j->img_comp[1].dc_pred = j->img_comp[2].dc_pred = 0;
-   j->marker = STBI__MARKER_none;
-   j->todo = j->restart_interval ? j->restart_interval : 0x7fffffff;
-   j->eob_run = 0;
-   // no more than 1<<31 MCUs if no restart_interal? that's plenty safe,
-   // since we don't even allow 1<<30 pixels
-}
-
-static int stbi__parse_entropy_coded_data(stbi__jpeg *z)
-{
-   stbi__jpeg_reset(z);
-   if (!z->progressive) {
-      if (z->scan_n == 1) {
-         int i,j;
-         STBI_SIMD_ALIGN(short, data[64]);
-         int n = z->order[0];
-         // non-interleaved data, we just need to process one block at a time,
-         // in trivial scanline order
-         // number of blocks to do just depends on how many actual "pixels" this
-         // component has, independent of interleaved MCU blocking and such
-         int w = (z->img_comp[n].x+7) >> 3;
-         int h = (z->img_comp[n].y+7) >> 3;
-         for (j=0; j < h; ++j) {
-            for (i=0; i < w; ++i) {
-               int ha = z->img_comp[n].ha;
-               if (!stbi__jpeg_decode_block(z, data, z->huff_dc+z->img_comp[n].hd, z->huff_ac+ha, z->fast_ac[ha], n, z->dequant[z->img_comp[n].tq])) return 0;
-               z->idct_block_kernel(z->img_comp[n].data+z->img_comp[n].w2*j*8+i*8, z->img_comp[n].w2, data);
-               // every data block is an MCU, so countdown the restart interval
-               if (--z->todo <= 0) {
-                  if (z->code_bits < 24) stbi__grow_buffer_unsafe(z);
-                  // if it's NOT a restart, then just bail, so we get corrupt data
-                  // rather than no data
-                  if (!STBI__RESTART(z->marker)) return 1;
-                  stbi__jpeg_reset(z);
-               }
-            }
-         }
-         return 1;
-      } else { // interleaved
-         int i,j,k,x,y;
-         STBI_SIMD_ALIGN(short, data[64]);
-         for (j=0; j < z->img_mcu_y; ++j) {
-            for (i=0; i < z->img_mcu_x; ++i) {
-               // scan an interleaved mcu... process scan_n components in order
-               for (k=0; k < z->scan_n; ++k) {
-                  int n = z->order[k];
-                  // scan out an mcu's worth of this component; that's just determined
-                  // by the basic H and V specified for the component
-                  for (y=0; y < z->img_comp[n].v; ++y) {
-                     for (x=0; x < z->img_comp[n].h; ++x) {
-                        int x2 = (i*z->img_comp[n].h + x)*8;
-                        int y2 = (j*z->img_comp[n].v + y)*8;
-                        int ha = z->img_comp[n].ha;
-                        if (!stbi__jpeg_decode_block(z, data, z->huff_dc+z->img_comp[n].hd, z->huff_ac+ha, z->fast_ac[ha], n, z->dequant[z->img_comp[n].tq])) return 0;
-                        z->idct_block_kernel(z->img_comp[n].data+z->img_comp[n].w2*y2+x2, z->img_comp[n].w2, data);
-                     }
-                  }
-               }
-               // after all interleaved components, that's an interleaved MCU,
-               // so now count down the restart interval
-               if (--z->todo <= 0) {
-                  if (z->code_bits < 24) stbi__grow_buffer_unsafe(z);
-                  if (!STBI__RESTART(z->marker)) return 1;
-                  stbi__jpeg_reset(z);
-               }
-            }
-         }
-         return 1;
-      }
-   } else {
-      if (z->scan_n == 1) {
-         int i,j;
-         int n = z->order[0];
-         // non-interleaved data, we just need to process one block at a time,
-         // in trivial scanline order
-         // number of blocks to do just depends on how many actual "pixels" this
-         // component has, independent of interleaved MCU blocking and such
-         int w = (z->img_comp[n].x+7) >> 3;
-         int h = (z->img_comp[n].y+7) >> 3;
-         for (j=0; j < h; ++j) {
-            for (i=0; i < w; ++i) {
-               short *data = z->img_comp[n].coeff + 64 * (i + j * z->img_comp[n].coeff_w);
-               if (z->spec_start == 0) {
-                  if (!stbi__jpeg_decode_block_prog_dc(z, data, &z->huff_dc[z->img_comp[n].hd], n))
-                     return 0;
-               } else {
-                  int ha = z->img_comp[n].ha;
-                  if (!stbi__jpeg_decode_block_prog_ac(z, data, &z->huff_ac[ha], z->fast_ac[ha]))
-                     return 0;
-               }
-               // every data block is an MCU, so countdown the restart interval
-               if (--z->todo <= 0) {
-                  if (z->code_bits < 24) stbi__grow_buffer_unsafe(z);
-                  if (!STBI__RESTART(z->marker)) return 1;
-                  stbi__jpeg_reset(z);
-               }
-            }
-         }
-         return 1;
-      } else { // interleaved
-         int i,j,k,x,y;
-         for (j=0; j < z->img_mcu_y; ++j) {
-            for (i=0; i < z->img_mcu_x; ++i) {
-               // scan an interleaved mcu... process scan_n components in order
-               for (k=0; k < z->scan_n; ++k) {
-                  int n = z->order[k];
-                  // scan out an mcu's worth of this component; that's just determined
-                  // by the basic H and V specified for the component
-                  for (y=0; y < z->img_comp[n].v; ++y) {
-                     for (x=0; x < z->img_comp[n].h; ++x) {
-                        int x2 = (i*z->img_comp[n].h + x);
-                        int y2 = (j*z->img_comp[n].v + y);
-                        short *data = z->img_comp[n].coeff + 64 * (x2 + y2 * z->img_comp[n].coeff_w);
-                        if (!stbi__jpeg_decode_block_prog_dc(z, data, &z->huff_dc[z->img_comp[n].hd], n))
-                           return 0;
-                     }
-                  }
-               }
-               // after all interleaved components, that's an interleaved MCU,
-               // so now count down the restart interval
-               if (--z->todo <= 0) {
-                  if (z->code_bits < 24) stbi__grow_buffer_unsafe(z);
-                  if (!STBI__RESTART(z->marker)) return 1;
-                  stbi__jpeg_reset(z);
-               }
-            }
-         }
-         return 1;
-      }
-   }
-}
-
-static void stbi__jpeg_dequantize(short *data, stbi_uc *dequant)
-{
-   int i;
-   for (i=0; i < 64; ++i)
-      data[i] *= dequant[i];
-}
-
-static void stbi__jpeg_finish(stbi__jpeg *z)
-{
-   if (z->progressive) {
-      // dequantize and idct the data
-      int i,j,n;
-      for (n=0; n < z->s->img_n; ++n) {
-         int w = (z->img_comp[n].x+7) >> 3;
-         int h = (z->img_comp[n].y+7) >> 3;
-         for (j=0; j < h; ++j) {
-            for (i=0; i < w; ++i) {
-               short *data = z->img_comp[n].coeff + 64 * (i + j * z->img_comp[n].coeff_w);
-               stbi__jpeg_dequantize(data, z->dequant[z->img_comp[n].tq]);
-               z->idct_block_kernel(z->img_comp[n].data+z->img_comp[n].w2*j*8+i*8, z->img_comp[n].w2, data);
-            }
-         }
-      }
-   }
-}
-
-static int stbi__process_marker(stbi__jpeg *z, int m)
-{
-   int L;
-   switch (m) {
-      case STBI__MARKER_none: // no marker found
-         return stbi__err("expected marker","Corrupt JPEG");
-
-      case 0xDD: // DRI - specify restart interval
-         if (stbi__get16be(z->s) != 4) return stbi__err("bad DRI len","Corrupt JPEG");
-         z->restart_interval = stbi__get16be(z->s);
-         return 1;
-
-      case 0xDB: // DQT - define quantization table
-         L = stbi__get16be(z->s)-2;
-         while (L > 0) {
-            int q = stbi__get8(z->s);
-            int p = q >> 4;
-            int t = q & 15,i;
-            if (p != 0) return stbi__err("bad DQT type","Corrupt JPEG");
-            if (t > 3) return stbi__err("bad DQT table","Corrupt JPEG");
-            for (i=0; i < 64; ++i)
-               z->dequant[t][stbi__jpeg_dezigzag[i]] = stbi__get8(z->s);
-            L -= 65;
-         }
-         return L==0;
-
-      case 0xC4: // DHT - define huffman table
-         L = stbi__get16be(z->s)-2;
-         while (L > 0) {
-            stbi_uc *v;
-            int sizes[16],i,n=0;
-            int q = stbi__get8(z->s);
-            int tc = q >> 4;
-            int th = q & 15;
-            if (tc > 1 || th > 3) return stbi__err("bad DHT header","Corrupt JPEG");
-            for (i=0; i < 16; ++i) {
-               sizes[i] = stbi__get8(z->s);
-               n += sizes[i];
-            }
-            L -= 17;
-            if (tc == 0) {
-               if (!stbi__build_huffman(z->huff_dc+th, sizes)) return 0;
-               v = z->huff_dc[th].values;
-            } else {
-               if (!stbi__build_huffman(z->huff_ac+th, sizes)) return 0;
-               v = z->huff_ac[th].values;
-            }
-            for (i=0; i < n; ++i)
-               v[i] = stbi__get8(z->s);
-            if (tc != 0)
-               stbi__build_fast_ac(z->fast_ac[th], z->huff_ac + th);
-            L -= n;
-         }
-         return L==0;
-   }
-   // check for comment block or APP blocks
-   if ((m >= 0xE0 && m <= 0xEF) || m == 0xFE) {
-      stbi__skip(z->s, stbi__get16be(z->s)-2);
-      return 1;
-   }
-   return 0;
-}
-
-// after we see SOS
-static int stbi__process_scan_header(stbi__jpeg *z)
-{
-   int i;
-   int Ls = stbi__get16be(z->s);
-   z->scan_n = stbi__get8(z->s);
-   if (z->scan_n < 1 || z->scan_n > 4 || z->scan_n > (int) z->s->img_n) return stbi__err("bad SOS component count","Corrupt JPEG");
-   if (Ls != 6+2*z->scan_n) return stbi__err("bad SOS len","Corrupt JPEG");
-   for (i=0; i < z->scan_n; ++i) {
-      int id = stbi__get8(z->s), which;
-      int q = stbi__get8(z->s);
-      for (which = 0; which < z->s->img_n; ++which)
-         if (z->img_comp[which].id == id)
-            break;
-      if (which == z->s->img_n) return 0; // no match
-      z->img_comp[which].hd = q >> 4;   if (z->img_comp[which].hd > 3) return stbi__err("bad DC huff","Corrupt JPEG");
-      z->img_comp[which].ha = q & 15;   if (z->img_comp[which].ha > 3) return stbi__err("bad AC huff","Corrupt JPEG");
-      z->order[i] = which;
-   }
-
-   {
-      int aa;
-      z->spec_start = stbi__get8(z->s);
-      z->spec_end   = stbi__get8(z->s); // should be 63, but might be 0
-      aa = stbi__get8(z->s);
-      z->succ_high = (aa >> 4);
-      z->succ_low  = (aa & 15);
-      if (z->progressive) {
-         if (z->spec_start > 63 || z->spec_end > 63  || z->spec_start > z->spec_end || z->succ_high > 13 || z->succ_low > 13)
-            return stbi__err("bad SOS", "Corrupt JPEG");
-      } else {
-         if (z->spec_start != 0) return stbi__err("bad SOS","Corrupt JPEG");
-         if (z->succ_high != 0 || z->succ_low != 0) return stbi__err("bad SOS","Corrupt JPEG");
-         z->spec_end = 63;
-      }
-   }
-
-   return 1;
-}
-
-static int stbi__process_frame_header(stbi__jpeg *z, int scan)
-{
-   stbi__context *s = z->s;
-   int Lf,p,i,q, h_max=1,v_max=1,c;
-   Lf = stbi__get16be(s);         if (Lf < 11) return stbi__err("bad SOF len","Corrupt JPEG"); // JPEG
-   p  = stbi__get8(s);            if (p != 8) return stbi__err("only 8-bit","JPEG format not supported: 8-bit only"); // JPEG baseline
-   s->img_y = stbi__get16be(s);   if (s->img_y == 0) return stbi__err("no header height", "JPEG format not supported: delayed height"); // Legal, but we don't handle it--but neither does IJG
-   s->img_x = stbi__get16be(s);   if (s->img_x == 0) return stbi__err("0 width","Corrupt JPEG"); // JPEG requires
-   c = stbi__get8(s);
-   if (c != 3 && c != 1) return stbi__err("bad component count","Corrupt JPEG");    // JFIF requires
-   s->img_n = c;
-   for (i=0; i < c; ++i) {
-      z->img_comp[i].data = NULL;
-      z->img_comp[i].linebuf = NULL;
-   }
-
-   if (Lf != 8+3*s->img_n) return stbi__err("bad SOF len","Corrupt JPEG");
-
-   z->rgb = 0;
-   for (i=0; i < s->img_n; ++i) {
-      static unsigned char rgb[3] = { 'R', 'G', 'B' };
-      z->img_comp[i].id = stbi__get8(s);
-      if (z->img_comp[i].id != i+1)   // JFIF requires
-         if (z->img_comp[i].id != i) {  // some version of jpegtran outputs non-JFIF-compliant files!
-            // somethings output this (see http://fileformats.archiveteam.org/wiki/JPEG#Color_format)
-            if (z->img_comp[i].id != rgb[i])
-               return stbi__err("bad component ID","Corrupt JPEG");
-            ++z->rgb;
-         }
-      q = stbi__get8(s);
-      z->img_comp[i].h = (q >> 4);  if (!z->img_comp[i].h || z->img_comp[i].h > 4) return stbi__err("bad H","Corrupt JPEG");
-      z->img_comp[i].v = q & 15;    if (!z->img_comp[i].v || z->img_comp[i].v > 4) return stbi__err("bad V","Corrupt JPEG");
-      z->img_comp[i].tq = stbi__get8(s);  if (z->img_comp[i].tq > 3) return stbi__err("bad TQ","Corrupt JPEG");
-   }
-
-   if (scan != STBI__SCAN_load) return 1;
-
-   if ((1 << 30) / s->img_x / s->img_n < s->img_y) return stbi__err("too large", "Image too large to decode");
-
-   for (i=0; i < s->img_n; ++i) {
-      if (z->img_comp[i].h > h_max) h_max = z->img_comp[i].h;
-      if (z->img_comp[i].v > v_max) v_max = z->img_comp[i].v;
-   }
-
-   // compute interleaved mcu info
-   z->img_h_max = h_max;
-   z->img_v_max = v_max;
-   z->img_mcu_w = h_max * 8;
-   z->img_mcu_h = v_max * 8;
-   z->img_mcu_x = (s->img_x + z->img_mcu_w-1) / z->img_mcu_w;
-   z->img_mcu_y = (s->img_y + z->img_mcu_h-1) / z->img_mcu_h;
-
-   for (i=0; i < s->img_n; ++i) {
-      // number of effective pixels (e.g. for non-interleaved MCU)
-      z->img_comp[i].x = (s->img_x * z->img_comp[i].h + h_max-1) / h_max;
-      z->img_comp[i].y = (s->img_y * z->img_comp[i].v + v_max-1) / v_max;
-      // to simplify generation, we'll allocate enough memory to decode
-      // the bogus oversized data from using interleaved MCUs and their
-      // big blocks (e.g. a 16x16 iMCU on an image of width 33); we won't
-      // discard the extra data until colorspace conversion
-      z->img_comp[i].w2 = z->img_mcu_x * z->img_comp[i].h * 8;
-      z->img_comp[i].h2 = z->img_mcu_y * z->img_comp[i].v * 8;
-      z->img_comp[i].raw_data = stbi__malloc(z->img_comp[i].w2 * z->img_comp[i].h2+15);
-
-      if (z->img_comp[i].raw_data == NULL) {
-         for(--i; i >= 0; --i) {
-            STBI_FREE(z->img_comp[i].raw_data);
-            z->img_comp[i].raw_data = NULL;
-         }
-         return stbi__err("outofmem", "Out of memory");
-      }
-      // align blocks for idct using mmx/sse
-      z->img_comp[i].data = (stbi_uc*) (((size_t) z->img_comp[i].raw_data + 15) & ~15);
-      z->img_comp[i].linebuf = NULL;
-      if (z->progressive) {
-         z->img_comp[i].coeff_w = (z->img_comp[i].w2 + 7) >> 3;
-         z->img_comp[i].coeff_h = (z->img_comp[i].h2 + 7) >> 3;
-         z->img_comp[i].raw_coeff = STBI_MALLOC(z->img_comp[i].coeff_w * z->img_comp[i].coeff_h * 64 * sizeof(short) + 15);
-         z->img_comp[i].coeff = (short*) (((size_t) z->img_comp[i].raw_coeff + 15) & ~15);
-      } else {
-         z->img_comp[i].coeff = 0;
-         z->img_comp[i].raw_coeff = 0;
-      }
-   }
-
-   return 1;
-}
-
-// use comparisons since in some cases we handle more than one case (e.g. SOF)
-#define stbi__DNL(x)         ((x) == 0xdc)
-#define stbi__SOI(x)         ((x) == 0xd8)
-#define stbi__EOI(x)         ((x) == 0xd9)
-#define stbi__SOF(x)         ((x) == 0xc0 || (x) == 0xc1 || (x) == 0xc2)
-#define stbi__SOS(x)         ((x) == 0xda)
-
-#define stbi__SOF_progressive(x)   ((x) == 0xc2)
-
-static int stbi__decode_jpeg_header(stbi__jpeg *z, int scan)
-{
-   int m;
-   z->marker = STBI__MARKER_none; // initialize cached marker to empty
-   m = stbi__get_marker(z);
-   if (!stbi__SOI(m)) return stbi__err("no SOI","Corrupt JPEG");
-   if (scan == STBI__SCAN_type) return 1;
-   m = stbi__get_marker(z);
-   while (!stbi__SOF(m)) {
-      if (!stbi__process_marker(z,m)) return 0;
-      m = stbi__get_marker(z);
-      while (m == STBI__MARKER_none) {
-         // some files have extra padding after their blocks, so ok, we'll scan
-         if (stbi__at_eof(z->s)) return stbi__err("no SOF", "Corrupt JPEG");
-         m = stbi__get_marker(z);
-      }
-   }
-   z->progressive = stbi__SOF_progressive(m);
-   if (!stbi__process_frame_header(z, scan)) return 0;
-   return 1;
-}
-
-// decode image to YCbCr format
-static int stbi__decode_jpeg_image(stbi__jpeg *j)
-{
-   int m;
-   for (m = 0; m < 4; m++) {
-      j->img_comp[m].raw_data = NULL;
-      j->img_comp[m].raw_coeff = NULL;
-   }
-   j->restart_interval = 0;
-   if (!stbi__decode_jpeg_header(j, STBI__SCAN_load)) return 0;
-   m = stbi__get_marker(j);
-   while (!stbi__EOI(m)) {
-      if (stbi__SOS(m)) {
-         if (!stbi__process_scan_header(j)) return 0;
-         if (!stbi__parse_entropy_coded_data(j)) return 0;
-         if (j->marker == STBI__MARKER_none ) {
-            // handle 0s at the end of image data from IP Kamera 9060
-            while (!stbi__at_eof(j->s)) {
-               int x = stbi__get8(j->s);
-               if (x == 255) {
-                  j->marker = stbi__get8(j->s);
-                  break;
-               } else if (x != 0) {
-                  return stbi__err("junk before marker", "Corrupt JPEG");
-               }
-            }
-            // if we reach eof without hitting a marker, stbi__get_marker() below will fail and we'll eventually return 0
-         }
-      } else {
-         if (!stbi__process_marker(j, m)) return 0;
-      }
-      m = stbi__get_marker(j);
-   }
-   if (j->progressive)
-      stbi__jpeg_finish(j);
-   return 1;
-}
-
-// static jfif-centered resampling (across block boundaries)
-
-typedef stbi_uc *(*resample_row_func)(stbi_uc *out, stbi_uc *in0, stbi_uc *in1,
-                                    int w, int hs);
-
-#define stbi__div4(x) ((stbi_uc) ((x) >> 2))
-
-static stbi_uc *resample_row_1(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
-{
-   STBI_NOTUSED(out);
-   STBI_NOTUSED(in_far);
-   STBI_NOTUSED(w);
-   STBI_NOTUSED(hs);
-   return in_near;
-}
-
-static stbi_uc* stbi__resample_row_v_2(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
-{
-   // need to generate two samples vertically for every one in input
-   int i;
-   STBI_NOTUSED(hs);
-   for (i=0; i < w; ++i)
-      out[i] = stbi__div4(3*in_near[i] + in_far[i] + 2);
-   return out;
-}
-
-static stbi_uc*  stbi__resample_row_h_2(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
-{
-   // need to generate two samples horizontally for every one in input
-   int i;
-   stbi_uc *input = in_near;
-
-   if (w == 1) {
-      // if only one sample, can't do any interpolation
-      out[0] = out[1] = input[0];
-      return out;
-   }
-
-   out[0] = input[0];
-   out[1] = stbi__div4(input[0]*3 + input[1] + 2);
-   for (i=1; i < w-1; ++i) {
-      int n = 3*input[i]+2;
-      out[i*2+0] = stbi__div4(n+input[i-1]);
-      out[i*2+1] = stbi__div4(n+input[i+1]);
-   }
-   out[i*2+0] = stbi__div4(input[w-2]*3 + input[w-1] + 2);
-   out[i*2+1] = input[w-1];
-
-   STBI_NOTUSED(in_far);
-   STBI_NOTUSED(hs);
-
-   return out;
-}
-
-#define stbi__div16(x) ((stbi_uc) ((x) >> 4))
-
-static stbi_uc *stbi__resample_row_hv_2(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
-{
-   // need to generate 2x2 samples for every one in input
-   int i,t0,t1;
-   if (w == 1) {
-      out[0] = out[1] = stbi__div4(3*in_near[0] + in_far[0] + 2);
-      return out;
-   }
-
-   t1 = 3*in_near[0] + in_far[0];
-   out[0] = stbi__div4(t1+2);
-   for (i=1; i < w; ++i) {
-      t0 = t1;
-      t1 = 3*in_near[i]+in_far[i];
-      out[i*2-1] = stbi__div16(3*t0 + t1 + 8);
-      out[i*2  ] = stbi__div16(3*t1 + t0 + 8);
-   }
-   out[w*2-1] = stbi__div4(t1+2);
-
-   STBI_NOTUSED(hs);
-
-   return out;
-}
-
-#if defined(STBI_SSE2) || defined(STBI_NEON)
-static stbi_uc *stbi__resample_row_hv_2_simd(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
-{
-   // need to generate 2x2 samples for every one in input
-   int i=0,t0,t1;
-
-   if (w == 1) {
-      out[0] = out[1] = stbi__div4(3*in_near[0] + in_far[0] + 2);
-      return out;
-   }
-
-   t1 = 3*in_near[0] + in_far[0];
-   // process groups of 8 pixels for as long as we can.
-   // note we can't handle the last pixel in a row in this loop
-   // because we need to handle the filter boundary conditions.
-   for (; i < ((w-1) & ~7); i += 8) {
-#if defined(STBI_SSE2)
-      // load and perform the vertical filtering pass
-      // this uses 3*x + y = 4*x + (y - x)
-      __m128i zero  = _mm_setzero_si128();
-      __m128i farb  = _mm_loadl_epi64((__m128i *) (in_far + i));
-      __m128i nearb = _mm_loadl_epi64((__m128i *) (in_near + i));
-      __m128i farw  = _mm_unpacklo_epi8(farb, zero);
-      __m128i nearw = _mm_unpacklo_epi8(nearb, zero);
-      __m128i diff  = _mm_sub_epi16(farw, nearw);
-      __m128i nears = _mm_slli_epi16(nearw, 2);
-      __m128i curr  = _mm_add_epi16(nears, diff); // current row
-
-      // horizontal filter works the same based on shifted vers of current
-      // row. "prev" is current row shifted right by 1 pixel; we need to
-      // insert the previous pixel value (from t1).
-      // "next" is current row shifted left by 1 pixel, with first pixel
-      // of next block of 8 pixels added in.
-      __m128i prv0 = _mm_slli_si128(curr, 2);
-      __m128i nxt0 = _mm_srli_si128(curr, 2);
-      __m128i prev = _mm_insert_epi16(prv0, t1, 0);
-      __m128i next = _mm_insert_epi16(nxt0, 3*in_near[i+8] + in_far[i+8], 7);
-
-      // horizontal filter, polyphase implementation since it's convenient:
-      // even pixels = 3*cur + prev = cur*4 + (prev - cur)
-      // odd  pixels = 3*cur + next = cur*4 + (next - cur)
-      // note the shared term.
-      __m128i bias  = _mm_set1_epi16(8);
-      __m128i curs = _mm_slli_epi16(curr, 2);
-      __m128i prvd = _mm_sub_epi16(prev, curr);
-      __m128i nxtd = _mm_sub_epi16(next, curr);
-      __m128i curb = _mm_add_epi16(curs, bias);
-      __m128i even = _mm_add_epi16(prvd, curb);
-      __m128i odd  = _mm_add_epi16(nxtd, curb);
-
-      // interleave even and odd pixels, then undo scaling.
-      __m128i int0 = _mm_unpacklo_epi16(even, odd);
-      __m128i int1 = _mm_unpackhi_epi16(even, odd);
-      __m128i de0  = _mm_srli_epi16(int0, 4);
-      __m128i de1  = _mm_srli_epi16(int1, 4);
-
-      // pack and write output
-      __m128i outv = _mm_packus_epi16(de0, de1);
-      _mm_storeu_si128((__m128i *) (out + i*2), outv);
-#elif defined(STBI_NEON)
-      // load and perform the vertical filtering pass
-      // this uses 3*x + y = 4*x + (y - x)
-      uint8x8_t farb  = vld1_u8(in_far + i);
-      uint8x8_t nearb = vld1_u8(in_near + i);
-      int16x8_t diff  = vreinterpretq_s16_u16(vsubl_u8(farb, nearb));
-      int16x8_t nears = vreinterpretq_s16_u16(vshll_n_u8(nearb, 2));
-      int16x8_t curr  = vaddq_s16(nears, diff); // current row
-
-      // horizontal filter works the same based on shifted vers of current
-      // row. "prev" is current row shifted right by 1 pixel; we need to
-      // insert the previous pixel value (from t1).
-      // "next" is current row shifted left by 1 pixel, with first pixel
-      // of next block of 8 pixels added in.
-      int16x8_t prv0 = vextq_s16(curr, curr, 7);
-      int16x8_t nxt0 = vextq_s16(curr, curr, 1);
-      int16x8_t prev = vsetq_lane_s16(t1, prv0, 0);
-      int16x8_t next = vsetq_lane_s16(3*in_near[i+8] + in_far[i+8], nxt0, 7);
-
-      // horizontal filter, polyphase implementation since it's convenient:
-      // even pixels = 3*cur + prev = cur*4 + (prev - cur)
-      // odd  pixels = 3*cur + next = cur*4 + (next - cur)
-      // note the shared term.
-      int16x8_t curs = vshlq_n_s16(curr, 2);
-      int16x8_t prvd = vsubq_s16(prev, curr);
-      int16x8_t nxtd = vsubq_s16(next, curr);
-      int16x8_t even = vaddq_s16(curs, prvd);
-      int16x8_t odd  = vaddq_s16(curs, nxtd);
-
-      // undo scaling and round, then store with even/odd phases interleaved
-      uint8x8x2_t o;
-      o.val[0] = vqrshrun_n_s16(even, 4);
-      o.val[1] = vqrshrun_n_s16(odd,  4);
-      vst2_u8(out + i*2, o);
-#endif
-
-      // "previous" value for next iter
-      t1 = 3*in_near[i+7] + in_far[i+7];
-   }
-
-   t0 = t1;
-   t1 = 3*in_near[i] + in_far[i];
-   out[i*2] = stbi__div16(3*t1 + t0 + 8);
-
-   for (++i; i < w; ++i) {
-      t0 = t1;
-      t1 = 3*in_near[i]+in_far[i];
-      out[i*2-1] = stbi__div16(3*t0 + t1 + 8);
-      out[i*2  ] = stbi__div16(3*t1 + t0 + 8);
-   }
-   out[w*2-1] = stbi__div4(t1+2);
-
-   STBI_NOTUSED(hs);
-
-   return out;
-}
-#endif
-
-static stbi_uc *stbi__resample_row_generic(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
-{
-   // resample with nearest-neighbor
-   int i,j;
-   STBI_NOTUSED(in_far);
-   for (i=0; i < w; ++i)
-      for (j=0; j < hs; ++j)
-         out[i*hs+j] = in_near[i];
-   return out;
-}
-
-#ifdef STBI_JPEG_OLD
-// this is the same YCbCr-to-RGB calculation that stb_image has used
-// historically before the algorithm changes in 1.49
-#define float2fixed(x)  ((int) ((x) * 65536 + 0.5))
-static void stbi__YCbCr_to_RGB_row(stbi_uc *out, const stbi_uc *y, const stbi_uc *pcb, const stbi_uc *pcr, int count, int step)
-{
-   int i;
-   for (i=0; i < count; ++i) {
-      int y_fixed = (y[i] << 16) + 32768; // rounding
-      int r,g,b;
-      int cr = pcr[i] - 128;
-      int cb = pcb[i] - 128;
-      r = y_fixed + cr*float2fixed(1.40200f);
-      g = y_fixed - cr*float2fixed(0.71414f) - cb*float2fixed(0.34414f);
-      b = y_fixed                            + cb*float2fixed(1.77200f);
-      r >>= 16;
-      g >>= 16;
-      b >>= 16;
-      if ((unsigned) r > 255) { if (r < 0) r = 0; else r = 255; }
-      if ((unsigned) g > 255) { if (g < 0) g = 0; else g = 255; }
-      if ((unsigned) b > 255) { if (b < 0) b = 0; else b = 255; }
-      out[0] = (stbi_uc)r;
-      out[1] = (stbi_uc)g;
-      out[2] = (stbi_uc)b;
-      out[3] = 255;
-      out += step;
-   }
-}
-#else
-// this is a reduced-precision calculation of YCbCr-to-RGB introduced
-// to make sure the code produces the same results in both SIMD and scalar
-#define float2fixed(x)  (((int) ((x) * 4096.0f + 0.5f)) << 8)
-static void stbi__YCbCr_to_RGB_row(stbi_uc *out, const stbi_uc *y, const stbi_uc *pcb, const stbi_uc *pcr, int count, int step)
-{
-   int i;
-   for (i=0; i < count; ++i) {
-      int y_fixed = (y[i] << 20) + (1<<19); // rounding
-      int r,g,b;
-      int cr = pcr[i] - 128;
-      int cb = pcb[i] - 128;
-      r = y_fixed +  cr* float2fixed(1.40200f);
-      g = y_fixed + (cr*-float2fixed(0.71414f)) + ((cb*-float2fixed(0.34414f)) & 0xffff0000);
-      b = y_fixed                               +   cb* float2fixed(1.77200f);
-      r >>= 20;
-      g >>= 20;
-      b >>= 20;
-      if ((unsigned) r > 255) { if (r < 0) r = 0; else r = 255; }
-      if ((unsigned) g > 255) { if (g < 0) g = 0; else g = 255; }
-      if ((unsigned) b > 255) { if (b < 0) b = 0; else b = 255; }
-      out[0] = (stbi_uc)r;
-      out[1] = (stbi_uc)g;
-      out[2] = (stbi_uc)b;
-      out[3] = 255;
-      out += step;
-   }
-}
-#endif
-
-#if defined(STBI_SSE2) || defined(STBI_NEON)
-static void stbi__YCbCr_to_RGB_simd(stbi_uc *out, stbi_uc const *y, stbi_uc const *pcb, stbi_uc const *pcr, int count, int step)
-{
-   int i = 0;
-
-#ifdef STBI_SSE2
-   // step == 3 is pretty ugly on the final interleave, and i'm not convinced
-   // it's useful in practice (you wouldn't use it for textures, for example).
-   // so just accelerate step == 4 case.
-   if (step == 4) {
-      // this is a fairly straightforward implementation and not super-optimized.
-      __m128i signflip  = _mm_set1_epi8(-0x80);
-      __m128i cr_const0 = _mm_set1_epi16(   (short) ( 1.40200f*4096.0f+0.5f));
-      __m128i cr_const1 = _mm_set1_epi16( - (short) ( 0.71414f*4096.0f+0.5f));
-      __m128i cb_const0 = _mm_set1_epi16( - (short) ( 0.34414f*4096.0f+0.5f));
-      __m128i cb_const1 = _mm_set1_epi16(   (short) ( 1.77200f*4096.0f+0.5f));
-      __m128i y_bias = _mm_set1_epi8((char) (unsigned char) 128);
-      __m128i xw = _mm_set1_epi16(255); // alpha channel
-
-      for (; i+7 < count; i += 8) {
-         // load
-         __m128i y_bytes = _mm_loadl_epi64((__m128i *) (y+i));
-         __m128i cr_bytes = _mm_loadl_epi64((__m128i *) (pcr+i));
-         __m128i cb_bytes = _mm_loadl_epi64((__m128i *) (pcb+i));
-         __m128i cr_biased = _mm_xor_si128(cr_bytes, signflip); // -128
-         __m128i cb_biased = _mm_xor_si128(cb_bytes, signflip); // -128
-
-         // unpack to short (and left-shift cr, cb by 8)
-         __m128i yw  = _mm_unpacklo_epi8(y_bias, y_bytes);
-         __m128i crw = _mm_unpacklo_epi8(_mm_setzero_si128(), cr_biased);
-         __m128i cbw = _mm_unpacklo_epi8(_mm_setzero_si128(), cb_biased);
-
-         // color transform
-         __m128i yws = _mm_srli_epi16(yw, 4);
-         __m128i cr0 = _mm_mulhi_epi16(cr_const0, crw);
-         __m128i cb0 = _mm_mulhi_epi16(cb_const0, cbw);
-         __m128i cb1 = _mm_mulhi_epi16(cbw, cb_const1);
-         __m128i cr1 = _mm_mulhi_epi16(crw, cr_const1);
-         __m128i rws = _mm_add_epi16(cr0, yws);
-         __m128i gwt = _mm_add_epi16(cb0, yws);
-         __m128i bws = _mm_add_epi16(yws, cb1);
-         __m128i gws = _mm_add_epi16(gwt, cr1);
-
-         // descale
-         __m128i rw = _mm_srai_epi16(rws, 4);
-         __m128i bw = _mm_srai_epi16(bws, 4);
-         __m128i gw = _mm_srai_epi16(gws, 4);
-
-         // back to byte, set up for transpose
-         __m128i brb = _mm_packus_epi16(rw, bw);
-         __m128i gxb = _mm_packus_epi16(gw, xw);
-
-         // transpose to interleave channels
-         __m128i t0 = _mm_unpacklo_epi8(brb, gxb);
-         __m128i t1 = _mm_unpackhi_epi8(brb, gxb);
-         __m128i o0 = _mm_unpacklo_epi16(t0, t1);
-         __m128i o1 = _mm_unpackhi_epi16(t0, t1);
-
-         // store
-         _mm_storeu_si128((__m128i *) (out + 0), o0);
-         _mm_storeu_si128((__m128i *) (out + 16), o1);
-         out += 32;
-      }
-   }
-#endif
-
-#ifdef STBI_NEON
-   // in this version, step=3 support would be easy to add. but is there demand?
-   if (step == 4) {
-      // this is a fairly straightforward implementation and not super-optimized.
-      uint8x8_t signflip = vdup_n_u8(0x80);
-      int16x8_t cr_const0 = vdupq_n_s16(   (short) ( 1.40200f*4096.0f+0.5f));
-      int16x8_t cr_const1 = vdupq_n_s16( - (short) ( 0.71414f*4096.0f+0.5f));
-      int16x8_t cb_const0 = vdupq_n_s16( - (short) ( 0.34414f*4096.0f+0.5f));
-      int16x8_t cb_const1 = vdupq_n_s16(   (short) ( 1.77200f*4096.0f+0.5f));
-
-      for (; i+7 < count; i += 8) {
-         // load
-         uint8x8_t y_bytes  = vld1_u8(y + i);
-         uint8x8_t cr_bytes = vld1_u8(pcr + i);
-         uint8x8_t cb_bytes = vld1_u8(pcb + i);
-         int8x8_t cr_biased = vreinterpret_s8_u8(vsub_u8(cr_bytes, signflip));
-         int8x8_t cb_biased = vreinterpret_s8_u8(vsub_u8(cb_bytes, signflip));
-
-         // expand to s16
-         int16x8_t yws = vreinterpretq_s16_u16(vshll_n_u8(y_bytes, 4));
-         int16x8_t crw = vshll_n_s8(cr_biased, 7);
-         int16x8_t cbw = vshll_n_s8(cb_biased, 7);
-
-         // color transform
-         int16x8_t cr0 = vqdmulhq_s16(crw, cr_const0);
-         int16x8_t cb0 = vqdmulhq_s16(cbw, cb_const0);
-         int16x8_t cr1 = vqdmulhq_s16(crw, cr_const1);
-         int16x8_t cb1 = vqdmulhq_s16(cbw, cb_const1);
-         int16x8_t rws = vaddq_s16(yws, cr0);
-         int16x8_t gws = vaddq_s16(vaddq_s16(yws, cb0), cr1);
-         int16x8_t bws = vaddq_s16(yws, cb1);
-
-         // undo scaling, round, convert to byte
-         uint8x8x4_t o;
-         o.val[0] = vqrshrun_n_s16(rws, 4);
-         o.val[1] = vqrshrun_n_s16(gws, 4);
-         o.val[2] = vqrshrun_n_s16(bws, 4);
-         o.val[3] = vdup_n_u8(255);
-
-         // store, interleaving r/g/b/a
-         vst4_u8(out, o);
-         out += 8*4;
-      }
-   }
-#endif
-
-   for (; i < count; ++i) {
-      int y_fixed = (y[i] << 20) + (1<<19); // rounding
-      int r,g,b;
-      int cr = pcr[i] - 128;
-      int cb = pcb[i] - 128;
-      r = y_fixed + cr* float2fixed(1.40200f);
-      g = y_fixed + cr*-float2fixed(0.71414f) + ((cb*-float2fixed(0.34414f)) & 0xffff0000);
-      b = y_fixed                             +   cb* float2fixed(1.77200f);
-      r >>= 20;
-      g >>= 20;
-      b >>= 20;
-      if ((unsigned) r > 255) { if (r < 0) r = 0; else r = 255; }
-      if ((unsigned) g > 255) { if (g < 0) g = 0; else g = 255; }
-      if ((unsigned) b > 255) { if (b < 0) b = 0; else b = 255; }
-      out[0] = (stbi_uc)r;
-      out[1] = (stbi_uc)g;
-      out[2] = (stbi_uc)b;
-      out[3] = 255;
-      out += step;
-   }
-}
-#endif
-
-// set up the kernels
-static void stbi__setup_jpeg(stbi__jpeg *j)
-{
-   j->idct_block_kernel = stbi__idct_block;
-   j->YCbCr_to_RGB_kernel = stbi__YCbCr_to_RGB_row;
-   j->resample_row_hv_2_kernel = stbi__resample_row_hv_2;
-
-#ifdef STBI_SSE2
-   if (stbi__sse2_available()) {
-      j->idct_block_kernel = stbi__idct_simd;
-      #ifndef STBI_JPEG_OLD
-      j->YCbCr_to_RGB_kernel = stbi__YCbCr_to_RGB_simd;
-      #endif
-      j->resample_row_hv_2_kernel = stbi__resample_row_hv_2_simd;
-   }
-#endif
-
-#ifdef STBI_NEON
-   j->idct_block_kernel = stbi__idct_simd;
-   #ifndef STBI_JPEG_OLD
-   j->YCbCr_to_RGB_kernel = stbi__YCbCr_to_RGB_simd;
-   #endif
-   j->resample_row_hv_2_kernel = stbi__resample_row_hv_2_simd;
-#endif
-}
-
-// clean up the temporary component buffers
-static void stbi__cleanup_jpeg(stbi__jpeg *j)
-{
-   int i;
-   for (i=0; i < j->s->img_n; ++i) {
-      if (j->img_comp[i].raw_data) {
-         STBI_FREE(j->img_comp[i].raw_data);
-         j->img_comp[i].raw_data = NULL;
-         j->img_comp[i].data = NULL;
-      }
-      if (j->img_comp[i].raw_coeff) {
-         STBI_FREE(j->img_comp[i].raw_coeff);
-         j->img_comp[i].raw_coeff = 0;
-         j->img_comp[i].coeff = 0;
-      }
-      if (j->img_comp[i].linebuf) {
-         STBI_FREE(j->img_comp[i].linebuf);
-         j->img_comp[i].linebuf = NULL;
-      }
-   }
-}
-
-typedef struct
-{
-   resample_row_func resample;
-   stbi_uc *line0,*line1;
-   int hs,vs;   // expansion factor in each axis
-   int w_lores; // horizontal pixels pre-expansion
-   int ystep;   // how far through vertical expansion we are
-   int ypos;    // which pre-expansion row we're on
-} stbi__resample;
-
-static stbi_uc *load_jpeg_image(stbi__jpeg *z, int *out_x, int *out_y, int *comp, int req_comp)
-{
-   int n, decode_n;
-   z->s->img_n = 0; // make stbi__cleanup_jpeg safe
-
-   // validate req_comp
-   if (req_comp < 0 || req_comp > 4) return stbi__errpuc("bad req_comp", "Internal error");
-
-   // load a jpeg image from whichever source, but leave in YCbCr format
-   if (!stbi__decode_jpeg_image(z)) { stbi__cleanup_jpeg(z); return NULL; }
-
-   // determine actual number of components to generate
-   n = req_comp ? req_comp : z->s->img_n;
-
-   if (z->s->img_n == 3 && n < 3)
-      decode_n = 1;
-   else
-      decode_n = z->s->img_n;
-
-   // resample and color-convert
-   {
-      int k;
-      unsigned int i,j;
-      stbi_uc *output;
-      stbi_uc *coutput[4];
-
-      stbi__resample res_comp[4];
-
-      for (k=0; k < decode_n; ++k) {
-         stbi__resample *r = &res_comp[k];
-
-         // allocate line buffer big enough for upsampling off the edges
-         // with upsample factor of 4
-         z->img_comp[k].linebuf = (stbi_uc *) stbi__malloc(z->s->img_x + 3);
-         if (!z->img_comp[k].linebuf) { stbi__cleanup_jpeg(z); return stbi__errpuc("outofmem", "Out of memory"); }
-
-         r->hs      = z->img_h_max / z->img_comp[k].h;
-         r->vs      = z->img_v_max / z->img_comp[k].v;
-         r->ystep   = r->vs >> 1;
-         r->w_lores = (z->s->img_x + r->hs-1) / r->hs;
-         r->ypos    = 0;
-         r->line0   = r->line1 = z->img_comp[k].data;
-
-         if      (r->hs == 1 && r->vs == 1) r->resample = resample_row_1;
-         else if (r->hs == 1 && r->vs == 2) r->resample = stbi__resample_row_v_2;
-         else if (r->hs == 2 && r->vs == 1) r->resample = stbi__resample_row_h_2;
-         else if (r->hs == 2 && r->vs == 2) r->resample = z->resample_row_hv_2_kernel;
-         else                               r->resample = stbi__resample_row_generic;
-      }
-
-      // can't error after this so, this is safe
-      output = (stbi_uc *) stbi__malloc(n * z->s->img_x * z->s->img_y + 1);
-      if (!output) { stbi__cleanup_jpeg(z); return stbi__errpuc("outofmem", "Out of memory"); }
-
-      // now go ahead and resample
-      for (j=0; j < z->s->img_y; ++j) {
-         stbi_uc *out = output + n * z->s->img_x * j;
-         for (k=0; k < decode_n; ++k) {
-            stbi__resample *r = &res_comp[k];
-            int y_bot = r->ystep >= (r->vs >> 1);
-            coutput[k] = r->resample(z->img_comp[k].linebuf,
-                                     y_bot ? r->line1 : r->line0,
-                                     y_bot ? r->line0 : r->line1,
-                                     r->w_lores, r->hs);
-            if (++r->ystep >= r->vs) {
-               r->ystep = 0;
-               r->line0 = r->line1;
-               if (++r->ypos < z->img_comp[k].y)
-                  r->line1 += z->img_comp[k].w2;
-            }
-         }
-         if (n >= 3) {
-            stbi_uc *y = coutput[0];
-            if (z->s->img_n == 3) {
-               if (z->rgb == 3) {
-                  for (i=0; i < z->s->img_x; ++i) {
-                     out[0] = y[i];
-                     out[1] = coutput[1][i];
-                     out[2] = coutput[2][i];
-                     out[3] = 255;
-                     out += n;
-                  }
-               } else {
-                  z->YCbCr_to_RGB_kernel(out, y, coutput[1], coutput[2], z->s->img_x, n);
-               }
-            } else
-               for (i=0; i < z->s->img_x; ++i) {
-                  out[0] = out[1] = out[2] = y[i];
-                  out[3] = 255; // not used if n==3
-                  out += n;
-               }
-         } else {
-            stbi_uc *y = coutput[0];
-            if (n == 1)
-               for (i=0; i < z->s->img_x; ++i) out[i] = y[i];
-            else
-               for (i=0; i < z->s->img_x; ++i) *out++ = y[i], *out++ = 255;
-         }
-      }
-      stbi__cleanup_jpeg(z);
-      *out_x = z->s->img_x;
-      *out_y = z->s->img_y;
-      if (comp) *comp  = z->s->img_n; // report original components, not output
-      return output;
-   }
-}
-
-static unsigned char *stbi__jpeg_load(stbi__context *s, int *x, int *y, int *comp, int req_comp)
-{
-   unsigned char* result;
-   stbi__jpeg* j = (stbi__jpeg*) stbi__malloc(sizeof(stbi__jpeg));
-   j->s = s;
-   stbi__setup_jpeg(j);
-   result = load_jpeg_image(j, x,y,comp,req_comp);
-   STBI_FREE(j);
-   return result;
-}
-
-static int stbi__jpeg_test(stbi__context *s)
-{
-   int r;
-   stbi__jpeg j;
-   j.s = s;
-   stbi__setup_jpeg(&j);
-   r = stbi__decode_jpeg_header(&j, STBI__SCAN_type);
-   stbi__rewind(s);
-   return r;
-}
-
-static int stbi__jpeg_info_raw(stbi__jpeg *j, int *x, int *y, int *comp)
-{
-   if (!stbi__decode_jpeg_header(j, STBI__SCAN_header)) {
-      stbi__rewind( j->s );
-      return 0;
-   }
-   if (x) *x = j->s->img_x;
-   if (y) *y = j->s->img_y;
-   if (comp) *comp = j->s->img_n;
-   return 1;
-}
-
-static int stbi__jpeg_info(stbi__context *s, int *x, int *y, int *comp)
-{
-   int result;
-   stbi__jpeg* j = (stbi__jpeg*) (stbi__malloc(sizeof(stbi__jpeg)));
-   j->s = s;
-   result = stbi__jpeg_info_raw(j, x, y, comp);
-   STBI_FREE(j);
-   return result;
-}
-#endif
-
-// public domain zlib decode    v0.2  Sean Barrett 2006-11-18
-//    simple implementation
-//      - all input must be provided in an upfront buffer
-//      - all output is written to a single output buffer (can malloc/realloc)
-//    performance
-//      - fast huffman
-
-#ifndef STBI_NO_ZLIB
-
-// fast-way is faster to check than jpeg huffman, but slow way is slower
-#define STBI__ZFAST_BITS  9 // accelerate all cases in default tables
-#define STBI__ZFAST_MASK  ((1 << STBI__ZFAST_BITS) - 1)
-
-// zlib-style huffman encoding
-// (jpegs packs from left, zlib from right, so can't share code)
-typedef struct
-{
-   stbi__uint16 fast[1 << STBI__ZFAST_BITS];
-   stbi__uint16 firstcode[16];
-   int maxcode[17];
-   stbi__uint16 firstsymbol[16];
-   stbi_uc  size[288];
-   stbi__uint16 value[288];
-} stbi__zhuffman;
-
-stbi_inline static int stbi__bitreverse16(int n)
-{
-  n = ((n & 0xAAAA) >>  1) | ((n & 0x5555) << 1);
-  n = ((n & 0xCCCC) >>  2) | ((n & 0x3333) << 2);
-  n = ((n & 0xF0F0) >>  4) | ((n & 0x0F0F) << 4);
-  n = ((n & 0xFF00) >>  8) | ((n & 0x00FF) << 8);
-  return n;
-}
-
-stbi_inline static int stbi__bit_reverse(int v, int bits)
-{
-   STBI_ASSERT(bits <= 16);
-   // to bit reverse n bits, reverse 16 and shift
-   // e.g. 11 bits, bit reverse and shift away 5
-   return stbi__bitreverse16(v) >> (16-bits);
-}
-
-static int stbi__zbuild_huffman(stbi__zhuffman *z, stbi_uc *sizelist, int num)
-{
-   int i,k=0;
-   int code, next_code[16], sizes[17];
-
-   // DEFLATE spec for generating codes
-   memset(sizes, 0, sizeof(sizes));
-   memset(z->fast, 0, sizeof(z->fast));
-   for (i=0; i < num; ++i)
-      ++sizes[sizelist[i]];
-   sizes[0] = 0;
-   for (i=1; i < 16; ++i)
-      if (sizes[i] > (1 << i))
-         return stbi__err("bad sizes", "Corrupt PNG");
-   code = 0;
-   for (i=1; i < 16; ++i) {
-      next_code[i] = code;
-      z->firstcode[i] = (stbi__uint16) code;
-      z->firstsymbol[i] = (stbi__uint16) k;
-      code = (code + sizes[i]);
-      if (sizes[i])
-         if (code-1 >= (1 << i)) return stbi__err("bad codelengths","Corrupt PNG");
-      z->maxcode[i] = code << (16-i); // preshift for inner loop
-      code <<= 1;
-      k += sizes[i];
-   }
-   z->maxcode[16] = 0x10000; // sentinel
-   for (i=0; i < num; ++i) {
-      int s = sizelist[i];
-      if (s) {
-         int c = next_code[s] - z->firstcode[s] + z->firstsymbol[s];
-         stbi__uint16 fastv = (stbi__uint16) ((s << 9) | i);
-         z->size [c] = (stbi_uc     ) s;
-         z->value[c] = (stbi__uint16) i;
-         if (s <= STBI__ZFAST_BITS) {
-            int j = stbi__bit_reverse(next_code[s],s);
-            while (j < (1 << STBI__ZFAST_BITS)) {
-               z->fast[j] = fastv;
-               j += (1 << s);
-            }
-         }
-         ++next_code[s];
-      }
-   }
-   return 1;
-}
-
-// zlib-from-memory implementation for PNG reading
-//    because PNG allows splitting the zlib stream arbitrarily,
-//    and it's annoying structurally to have PNG call ZLIB call PNG,
-//    we require PNG read all the IDATs and combine them into a single
-//    memory buffer
-
-typedef struct
-{
-   stbi_uc *zbuffer, *zbuffer_end;
-   int num_bits;
-   stbi__uint32 code_buffer;
-
-   char *zout;
-   char *zout_start;
-   char *zout_end;
-   int   z_expandable;
-
-   stbi__zhuffman z_length, z_distance;
-} stbi__zbuf;
-
-stbi_inline static stbi_uc stbi__zget8(stbi__zbuf *z)
-{
-   if (z->zbuffer >= z->zbuffer_end) return 0;
-   return *z->zbuffer++;
-}
-
-static void stbi__fill_bits(stbi__zbuf *z)
-{
-   do {
-      STBI_ASSERT(z->code_buffer < (1U << z->num_bits));
-      z->code_buffer |= (unsigned int) stbi__zget8(z) << z->num_bits;
-      z->num_bits += 8;
-   } while (z->num_bits <= 24);
-}
-
-stbi_inline static unsigned int stbi__zreceive(stbi__zbuf *z, int n)
-{
-   unsigned int k;
-   if (z->num_bits < n) stbi__fill_bits(z);
-   k = z->code_buffer & ((1 << n) - 1);
-   z->code_buffer >>= n;
-   z->num_bits -= n;
-   return k;
-}
-
-static int stbi__zhuffman_decode_slowpath(stbi__zbuf *a, stbi__zhuffman *z)
-{
-   int b,s,k;
-   // not resolved by fast table, so compute it the slow way
-   // use jpeg approach, which requires MSbits at top
-   k = stbi__bit_reverse(a->code_buffer, 16);
-   for (s=STBI__ZFAST_BITS+1; ; ++s)
-      if (k < z->maxcode[s])
-         break;
-   if (s == 16) return -1; // invalid code!
-   // code size is s, so:
-   b = (k >> (16-s)) - z->firstcode[s] + z->firstsymbol[s];
-   STBI_ASSERT(z->size[b] == s);
-   a->code_buffer >>= s;
-   a->num_bits -= s;
-   return z->value[b];
-}
-
-stbi_inline static int stbi__zhuffman_decode(stbi__zbuf *a, stbi__zhuffman *z)
-{
-   int b,s;
-   if (a->num_bits < 16) stbi__fill_bits(a);
-   b = z->fast[a->code_buffer & STBI__ZFAST_MASK];
-   if (b) {
-      s = b >> 9;
-      a->code_buffer >>= s;
-      a->num_bits -= s;
-      return b & 511;
-   }
-   return stbi__zhuffman_decode_slowpath(a, z);
-}
-
-static int stbi__zexpand(stbi__zbuf *z, char *zout, int n)  // need to make room for n bytes
-{
-   char *q;
-   int cur, limit, old_limit;
-   z->zout = zout;
-   if (!z->z_expandable) return stbi__err("output buffer limit","Corrupt PNG");
-   cur   = (int) (z->zout     - z->zout_start);
-   limit = old_limit = (int) (z->zout_end - z->zout_start);
-   while (cur + n > limit)
-      limit *= 2;
-   q = (char *) STBI_REALLOC_SIZED(z->zout_start, old_limit, limit);
-   STBI_NOTUSED(old_limit);
-   if (q == NULL) return stbi__err("outofmem", "Out of memory");
-   z->zout_start = q;
-   z->zout       = q + cur;
-   z->zout_end   = q + limit;
-   return 1;
-}
-
-static int stbi__zlength_base[31] = {
-   3,4,5,6,7,8,9,10,11,13,
-   15,17,19,23,27,31,35,43,51,59,
-   67,83,99,115,131,163,195,227,258,0,0 };
-
-static int stbi__zlength_extra[31]=
-{ 0,0,0,0,0,0,0,0,1,1,1,1,2,2,2,2,3,3,3,3,4,4,4,4,5,5,5,5,0,0,0 };
-
-static int stbi__zdist_base[32] = { 1,2,3,4,5,7,9,13,17,25,33,49,65,97,129,193,
-257,385,513,769,1025,1537,2049,3073,4097,6145,8193,12289,16385,24577,0,0};
-
-static int stbi__zdist_extra[32] =
-{ 0,0,0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,8,8,9,9,10,10,11,11,12,12,13,13};
-
-static int stbi__parse_huffman_block(stbi__zbuf *a)
-{
-   char *zout = a->zout;
-   for(;;) {
-      int z = stbi__zhuffman_decode(a, &a->z_length);
-      if (z < 256) {
-         if (z < 0) return stbi__err("bad huffman code","Corrupt PNG"); // error in huffman codes
-         if (zout >= a->zout_end) {
-            if (!stbi__zexpand(a, zout, 1)) return 0;
-            zout = a->zout;
-         }
-         *zout++ = (char) z;
-      } else {
-         stbi_uc *p;
-         int len,dist;
-         if (z == 256) {
-            a->zout = zout;
-            return 1;
-         }
-         z -= 257;
-         len = stbi__zlength_base[z];
-         if (stbi__zlength_extra[z]) len += stbi__zreceive(a, stbi__zlength_extra[z]);
-         z = stbi__zhuffman_decode(a, &a->z_distance);
-         if (z < 0) return stbi__err("bad huffman code","Corrupt PNG");
-         dist = stbi__zdist_base[z];
-         if (stbi__zdist_extra[z]) dist += stbi__zreceive(a, stbi__zdist_extra[z]);
-         if (zout - a->zout_start < dist) return stbi__err("bad dist","Corrupt PNG");
-         if (zout + len > a->zout_end) {
-            if (!stbi__zexpand(a, zout, len)) return 0;
-            zout = a->zout;
-         }
-         p = (stbi_uc *) (zout - dist);
-         if (dist == 1) { // run of one byte; common in images.
-            stbi_uc v = *p;
-            if (len) { do *zout++ = v; while (--len); }
-         } else {
-            if (len) { do *zout++ = *p++; while (--len); }
-         }
-      }
-   }
-}
-
-static int stbi__compute_huffman_codes(stbi__zbuf *a)
-{
-   static stbi_uc length_dezigzag[19] = { 16,17,18,0,8,7,9,6,10,5,11,4,12,3,13,2,14,1,15 };
-   stbi__zhuffman z_codelength;
-   stbi_uc lencodes[286+32+137];//padding for maximum single op
-   stbi_uc codelength_sizes[19];
-   int i,n;
-
-   int hlit  = stbi__zreceive(a,5) + 257;
-   int hdist = stbi__zreceive(a,5) + 1;
-   int hclen = stbi__zreceive(a,4) + 4;
-
-   memset(codelength_sizes, 0, sizeof(codelength_sizes));
-   for (i=0; i < hclen; ++i) {
-      int s = stbi__zreceive(a,3);
-      codelength_sizes[length_dezigzag[i]] = (stbi_uc) s;
-   }
-   if (!stbi__zbuild_huffman(&z_codelength, codelength_sizes, 19)) return 0;
-
-   n = 0;
-   while (n < hlit + hdist) {
-      int c = stbi__zhuffman_decode(a, &z_codelength);
-      if (c < 0 || c >= 19) return stbi__err("bad codelengths", "Corrupt PNG");
-      if (c < 16)
-         lencodes[n++] = (stbi_uc) c;
-      else if (c == 16) {
-         c = stbi__zreceive(a,2)+3;
-         memset(lencodes+n, lencodes[n-1], c);
-         n += c;
-      } else if (c == 17) {
-         c = stbi__zreceive(a,3)+3;
-         memset(lencodes+n, 0, c);
-         n += c;
-      } else {
-         STBI_ASSERT(c == 18);
-         c = stbi__zreceive(a,7)+11;
-         memset(lencodes+n, 0, c);
-         n += c;
-      }
-   }
-   if (n != hlit+hdist) return stbi__err("bad codelengths","Corrupt PNG");
-   if (!stbi__zbuild_huffman(&a->z_length, lencodes, hlit)) return 0;
-   if (!stbi__zbuild_huffman(&a->z_distance, lencodes+hlit, hdist)) return 0;
-   return 1;
-}
-
-static int stbi__parse_uncompressed_block(stbi__zbuf *a)
-{
-   stbi_uc header[4];
-   int len,nlen,k;
-   if (a->num_bits & 7)
-      stbi__zreceive(a, a->num_bits & 7); // discard
-   // drain the bit-packed data into header
-   k = 0;
-   while (a->num_bits > 0) {
-      header[k++] = (stbi_uc) (a->code_buffer & 255); // suppress MSVC run-time check
-      a->code_buffer >>= 8;
-      a->num_bits -= 8;
-   }
-   STBI_ASSERT(a->num_bits == 0);
-   // now fill header the normal way
-   while (k < 4)
-      header[k++] = stbi__zget8(a);
-   len  = header[1] * 256 + header[0];
-   nlen = header[3] * 256 + header[2];
-   if (nlen != (len ^ 0xffff)) return stbi__err("zlib corrupt","Corrupt PNG");
-   if (a->zbuffer + len > a->zbuffer_end) return stbi__err("read past buffer","Corrupt PNG");
-   if (a->zout + len > a->zout_end)
-      if (!stbi__zexpand(a, a->zout, len)) return 0;
-   memcpy(a->zout, a->zbuffer, len);
-   a->zbuffer += len;
-   a->zout += len;
-   return 1;
-}
-
-static int stbi__parse_zlib_header(stbi__zbuf *a)
-{
-   int cmf   = stbi__zget8(a);
-   int cm    = cmf & 15;
-   /* int cinfo = cmf >> 4; */
-   int flg   = stbi__zget8(a);
-   if ((cmf*256+flg) % 31 != 0) return stbi__err("bad zlib header","Corrupt PNG"); // zlib spec
-   if (flg & 32) return stbi__err("no preset dict","Corrupt PNG"); // preset dictionary not allowed in png
-   if (cm != 8) return stbi__err("bad compression","Corrupt PNG"); // DEFLATE required for png
-   // window = 1 << (8 + cinfo)... but who cares, we fully buffer output
-   return 1;
-}
-
-// @TODO: should statically initialize these for optimal thread safety
-static stbi_uc stbi__zdefault_length[288], stbi__zdefault_distance[32];
-static void stbi__init_zdefaults(void)
-{
-   int i;   // use <= to match clearly with spec
-   for (i=0; i <= 143; ++i)     stbi__zdefault_length[i]   = 8;
-   for (   ; i <= 255; ++i)     stbi__zdefault_length[i]   = 9;
-   for (   ; i <= 279; ++i)     stbi__zdefault_length[i]   = 7;
-   for (   ; i <= 287; ++i)     stbi__zdefault_length[i]   = 8;
-
-   for (i=0; i <=  31; ++i)     stbi__zdefault_distance[i] = 5;
-}
-
-static int stbi__parse_zlib(stbi__zbuf *a, int parse_header)
-{
-   int final, type;
-   if (parse_header)
-      if (!stbi__parse_zlib_header(a)) return 0;
-   a->num_bits = 0;
-   a->code_buffer = 0;
-   do {
-      final = stbi__zreceive(a,1);
-      type = stbi__zreceive(a,2);
-      if (type == 0) {
-         if (!stbi__parse_uncompressed_block(a)) return 0;
-      } else if (type == 3) {
-         return 0;
-      } else {
-         if (type == 1) {
-            // use fixed code lengths
-            if (!stbi__zdefault_distance[31]) stbi__init_zdefaults();
-            if (!stbi__zbuild_huffman(&a->z_length  , stbi__zdefault_length  , 288)) return 0;
-            if (!stbi__zbuild_huffman(&a->z_distance, stbi__zdefault_distance,  32)) return 0;
-         } else {
-            if (!stbi__compute_huffman_codes(a)) return 0;
-         }
-         if (!stbi__parse_huffman_block(a)) return 0;
-      }
-   } while (!final);
-   return 1;
-}
-
-static int stbi__do_zlib(stbi__zbuf *a, char *obuf, int olen, int exp, int parse_header)
-{
-   a->zout_start = obuf;
-   a->zout       = obuf;
-   a->zout_end   = obuf + olen;
-   a->z_expandable = exp;
-
-   return stbi__parse_zlib(a, parse_header);
-}
-
-STBIDEF char *stbi_zlib_decode_malloc_guesssize(const char *buffer, int len, int initial_size, int *outlen)
-{
-   stbi__zbuf a;
-   char *p = (char *) stbi__malloc(initial_size);
-   if (p == NULL) return NULL;
-   a.zbuffer = (stbi_uc *) buffer;
-   a.zbuffer_end = (stbi_uc *) buffer + len;
-   if (stbi__do_zlib(&a, p, initial_size, 1, 1)) {
-      if (outlen) *outlen = (int) (a.zout - a.zout_start);
-      return a.zout_start;
-   } else {
-      STBI_FREE(a.zout_start);
-      return NULL;
-   }
-}
-
-STBIDEF char *stbi_zlib_decode_malloc(char const *buffer, int len, int *outlen)
-{
-   return stbi_zlib_decode_malloc_guesssize(buffer, len, 16384, outlen);
-}
-
-STBIDEF char *stbi_zlib_decode_malloc_guesssize_headerflag(const char *buffer, int len, int initial_size, int *outlen, int parse_header)
-{
-   stbi__zbuf a;
-   char *p = (char *) stbi__malloc(initial_size);
-   if (p == NULL) return NULL;
-   a.zbuffer = (stbi_uc *) buffer;
-   a.zbuffer_end = (stbi_uc *) buffer + len;
-   if (stbi__do_zlib(&a, p, initial_size, 1, parse_header)) {
-      if (outlen) *outlen = (int) (a.zout - a.zout_start);
-      return a.zout_start;
-   } else {
-      STBI_FREE(a.zout_start);
-      return NULL;
-   }
-}
-
-STBIDEF int stbi_zlib_decode_buffer(char *obuffer, int olen, char const *ibuffer, int ilen)
-{
-   stbi__zbuf a;
-   a.zbuffer = (stbi_uc *) ibuffer;
-   a.zbuffer_end = (stbi_uc *) ibuffer + ilen;
-   if (stbi__do_zlib(&a, obuffer, olen, 0, 1))
-      return (int) (a.zout - a.zout_start);
-   else
-      return -1;
-}
-
-STBIDEF char *stbi_zlib_decode_noheader_malloc(char const *buffer, int len, int *outlen)
-{
-   stbi__zbuf a;
-   char *p = (char *) stbi__malloc(16384);
-   if (p == NULL) return NULL;
-   a.zbuffer = (stbi_uc *) buffer;
-   a.zbuffer_end = (stbi_uc *) buffer+len;
-   if (stbi__do_zlib(&a, p, 16384, 1, 0)) {
-      if (outlen) *outlen = (int) (a.zout - a.zout_start);
-      return a.zout_start;
-   } else {
-      STBI_FREE(a.zout_start);
-      return NULL;
-   }
-}
-
-STBIDEF int stbi_zlib_decode_noheader_buffer(char *obuffer, int olen, const char *ibuffer, int ilen)
-{
-   stbi__zbuf a;
-   a.zbuffer = (stbi_uc *) ibuffer;
-   a.zbuffer_end = (stbi_uc *) ibuffer + ilen;
-   if (stbi__do_zlib(&a, obuffer, olen, 0, 0))
-      return (int) (a.zout - a.zout_start);
-   else
-      return -1;
-}
-#endif
-
-// public domain "baseline" PNG decoder   v0.10  Sean Barrett 2006-11-18
-//    simple implementation
-//      - only 8-bit samples
-//      - no CRC checking
-//      - allocates lots of intermediate memory
-//        - avoids problem of streaming data between subsystems
-//        - avoids explicit window management
-//    performance
-//      - uses stb_zlib, a PD zlib implementation with fast huffman decoding
-
-#ifndef STBI_NO_PNG
-typedef struct
-{
-   stbi__uint32 length;
-   stbi__uint32 type;
-} stbi__pngchunk;
-
-static stbi__pngchunk stbi__get_chunk_header(stbi__context *s)
-{
-   stbi__pngchunk c;
-   c.length = stbi__get32be(s);
-   c.type   = stbi__get32be(s);
-   return c;
-}
-
-static int stbi__check_png_header(stbi__context *s)
-{
-   static stbi_uc png_sig[8] = { 137,80,78,71,13,10,26,10 };
-   int i;
-   for (i=0; i < 8; ++i)
-      if (stbi__get8(s) != png_sig[i]) return stbi__err("bad png sig","Not a PNG");
-   return 1;
-}
-
-typedef struct
-{
-   stbi__context *s;
-   stbi_uc *idata, *expanded, *out;
-   int depth;
-} stbi__png;
-
-
-enum {
-   STBI__F_none=0,
-   STBI__F_sub=1,
-   STBI__F_up=2,
-   STBI__F_avg=3,
-   STBI__F_paeth=4,
-   // synthetic filters used for first scanline to avoid needing a dummy row of 0s
-   STBI__F_avg_first,
-   STBI__F_paeth_first
-};
-
-static stbi_uc first_row_filter[5] =
-{
-   STBI__F_none,
-   STBI__F_sub,
-   STBI__F_none,
-   STBI__F_avg_first,
-   STBI__F_paeth_first
-};
-
-static int stbi__paeth(int a, int b, int c)
-{
-   int p = a + b - c;
-   int pa = abs(p-a);
-   int pb = abs(p-b);
-   int pc = abs(p-c);
-   if (pa <= pb && pa <= pc) return a;
-   if (pb <= pc) return b;
-   return c;
-}
-
-static stbi_uc stbi__depth_scale_table[9] = { 0, 0xff, 0x55, 0, 0x11, 0,0,0, 0x01 };
-
-// create the png data from post-deflated data
-static int stbi__create_png_image_raw(stbi__png *a, stbi_uc *raw, stbi__uint32 raw_len, int out_n, stbi__uint32 x, stbi__uint32 y, int depth, int color)
-{
-   int bytes = (depth == 16? 2 : 1);
-   stbi__context *s = a->s;
-   stbi__uint32 i,j,stride = x*out_n*bytes;
-   stbi__uint32 img_len, img_width_bytes;
-   int k;
-   int img_n = s->img_n; // copy it into a local for later
-
-   int output_bytes = out_n*bytes;
-   int filter_bytes = img_n*bytes;
-   int width = x;
-
-   STBI_ASSERT(out_n == s->img_n || out_n == s->img_n+1);
-   a->out = (stbi_uc *) stbi__malloc(x * y * output_bytes); // extra bytes to write off the end into
-   if (!a->out) return stbi__err("outofmem", "Out of memory");
-
-   img_width_bytes = (((img_n * x * depth) + 7) >> 3);
-   img_len = (img_width_bytes + 1) * y;
-   if (s->img_x == x && s->img_y == y) {
-      if (raw_len != img_len) return stbi__err("not enough pixels","Corrupt PNG");
-   } else { // interlaced:
-      if (raw_len < img_len) return stbi__err("not enough pixels","Corrupt PNG");
-   }
-
-   for (j=0; j < y; ++j) {
-      stbi_uc *cur = a->out + stride*j;
-      stbi_uc *prior = cur - stride;
-      int filter = *raw++;
-
-      if (filter > 4)
-         return stbi__err("invalid filter","Corrupt PNG");
-
-      if (depth < 8) {
-         STBI_ASSERT(img_width_bytes <= x);
-         cur += x*out_n - img_width_bytes; // store output to the rightmost img_len bytes, so we can decode in place
-         filter_bytes = 1;
-         width = img_width_bytes;
-      }
-
-      // if first row, use special filter that doesn't sample previous row
-      if (j == 0) filter = first_row_filter[filter];
-
-      // handle first byte explicitly
-      for (k=0; k < filter_bytes; ++k) {
-         switch (filter) {
-            case STBI__F_none       : cur[k] = raw[k]; break;
-            case STBI__F_sub        : cur[k] = raw[k]; break;
-            case STBI__F_up         : cur[k] = STBI__BYTECAST(raw[k] + prior[k]); break;
-            case STBI__F_avg        : cur[k] = STBI__BYTECAST(raw[k] + (prior[k]>>1)); break;
-            case STBI__F_paeth      : cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(0,prior[k],0)); break;
-            case STBI__F_avg_first  : cur[k] = raw[k]; break;
-            case STBI__F_paeth_first: cur[k] = raw[k]; break;
-         }
-      }
-
-      if (depth == 8) {
-         if (img_n != out_n)
-            cur[img_n] = 255; // first pixel
-         raw += img_n;
-         cur += out_n;
-         prior += out_n;
-      } else if (depth == 16) {
-         if (img_n != out_n) {
-            cur[filter_bytes]   = 255; // first pixel top byte
-            cur[filter_bytes+1] = 255; // first pixel bottom byte
-         }
-         raw += filter_bytes;
-         cur += output_bytes;
-         prior += output_bytes;
-      } else {
-         raw += 1;
-         cur += 1;
-         prior += 1;
-      }
-
-      // this is a little gross, so that we don't switch per-pixel or per-component
-      if (depth < 8 || img_n == out_n) {
-         int nk = (width - 1)*filter_bytes;
-         #define CASE(f) \
-             case f:     \
-                for (k=0; k < nk; ++k)
-         switch (filter) {
-            // "none" filter turns into a memcpy here; make that explicit.
-            case STBI__F_none:         memcpy(cur, raw, nk); break;
-            CASE(STBI__F_sub)          cur[k] = STBI__BYTECAST(raw[k] + cur[k-filter_bytes]); break;
-            CASE(STBI__F_up)           cur[k] = STBI__BYTECAST(raw[k] + prior[k]); break;
-            CASE(STBI__F_avg)          cur[k] = STBI__BYTECAST(raw[k] + ((prior[k] + cur[k-filter_bytes])>>1)); break;
-            CASE(STBI__F_paeth)        cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k-filter_bytes],prior[k],prior[k-filter_bytes])); break;
-            CASE(STBI__F_avg_first)    cur[k] = STBI__BYTECAST(raw[k] + (cur[k-filter_bytes] >> 1)); break;
-            CASE(STBI__F_paeth_first)  cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k-filter_bytes],0,0)); break;
-         }
-         #undef CASE
-         raw += nk;
-      } else {
-         STBI_ASSERT(img_n+1 == out_n);
-         #define CASE(f) \
-             case f:     \
-                for (i=x-1; i >= 1; --i, cur[filter_bytes]=255,raw+=filter_bytes,cur+=output_bytes,prior+=output_bytes) \
-                   for (k=0; k < filter_bytes; ++k)
-         switch (filter) {
-            CASE(STBI__F_none)         cur[k] = raw[k]; break;
-            CASE(STBI__F_sub)          cur[k] = STBI__BYTECAST(raw[k] + cur[k- output_bytes]); break;
-            CASE(STBI__F_up)           cur[k] = STBI__BYTECAST(raw[k] + prior[k]); break;
-            CASE(STBI__F_avg)          cur[k] = STBI__BYTECAST(raw[k] + ((prior[k] + cur[k- output_bytes])>>1)); break;
-            CASE(STBI__F_paeth)        cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k- output_bytes],prior[k],prior[k- output_bytes])); break;
-            CASE(STBI__F_avg_first)    cur[k] = STBI__BYTECAST(raw[k] + (cur[k- output_bytes] >> 1)); break;
-            CASE(STBI__F_paeth_first)  cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k- output_bytes],0,0)); break;
-         }
-         #undef CASE
-
-         // the loop above sets the high byte of the pixels' alpha, but for
-         // 16 bit png files we also need the low byte set. we'll do that here.
-         if (depth == 16) {
-            cur = a->out + stride*j; // start at the beginning of the row again
-            for (i=0; i < x; ++i,cur+=output_bytes) {
-               cur[filter_bytes+1] = 255;
-            }
-         }
-      }
-   }
-
-   // we make a separate pass to expand bits to pixels; for performance,
-   // this could run two scanlines behind the above code, so it won't
-   // intefere with filtering but will still be in the cache.
-   if (depth < 8) {
-      for (j=0; j < y; ++j) {
-         stbi_uc *cur = a->out + stride*j;
-         stbi_uc *in  = a->out + stride*j + x*out_n - img_width_bytes;
-         // unpack 1/2/4-bit into a 8-bit buffer. allows us to keep the common 8-bit path optimal at minimal cost for 1/2/4-bit
-         // png guarante byte alignment, if width is not multiple of 8/4/2 we'll decode dummy trailing data that will be skipped in the later loop
-         stbi_uc scale = (color == 0) ? stbi__depth_scale_table[depth] : 1; // scale grayscale values to 0..255 range
-
-         // note that the final byte might overshoot and write more data than desired.
-         // we can allocate enough data that this never writes out of memory, but it
-         // could also overwrite the next scanline. can it overwrite non-empty data
-         // on the next scanline? yes, consider 1-pixel-wide scanlines with 1-bit-per-pixel.
-         // so we need to explicitly clamp the final ones
-
-         if (depth == 4) {
-            for (k=x*img_n; k >= 2; k-=2, ++in) {
-               *cur++ = scale * ((*in >> 4)       );
-               *cur++ = scale * ((*in     ) & 0x0f);
-            }
-            if (k > 0) *cur++ = scale * ((*in >> 4)       );
-         } else if (depth == 2) {
-            for (k=x*img_n; k >= 4; k-=4, ++in) {
-               *cur++ = scale * ((*in >> 6)       );
-               *cur++ = scale * ((*in >> 4) & 0x03);
-               *cur++ = scale * ((*in >> 2) & 0x03);
-               *cur++ = scale * ((*in     ) & 0x03);
-            }
-            if (k > 0) *cur++ = scale * ((*in >> 6)       );
-            if (k > 1) *cur++ = scale * ((*in >> 4) & 0x03);
-            if (k > 2) *cur++ = scale * ((*in >> 2) & 0x03);
-         } else if (depth == 1) {
-            for (k=x*img_n; k >= 8; k-=8, ++in) {
-               *cur++ = scale * ((*in >> 7)       );
-               *cur++ = scale * ((*in >> 6) & 0x01);
-               *cur++ = scale * ((*in >> 5) & 0x01);
-               *cur++ = scale * ((*in >> 4) & 0x01);
-               *cur++ = scale * ((*in >> 3) & 0x01);
-               *cur++ = scale * ((*in >> 2) & 0x01);
-               *cur++ = scale * ((*in >> 1) & 0x01);
-               *cur++ = scale * ((*in     ) & 0x01);
-            }
-            if (k > 0) *cur++ = scale * ((*in >> 7)       );
-            if (k > 1) *cur++ = scale * ((*in >> 6) & 0x01);
-            if (k > 2) *cur++ = scale * ((*in >> 5) & 0x01);
-            if (k > 3) *cur++ = scale * ((*in >> 4) & 0x01);
-            if (k > 4) *cur++ = scale * ((*in >> 3) & 0x01);
-            if (k > 5) *cur++ = scale * ((*in >> 2) & 0x01);
-            if (k > 6) *cur++ = scale * ((*in >> 1) & 0x01);
-         }
-         if (img_n != out_n) {
-            int q;
-            // insert alpha = 255
-            cur = a->out + stride*j;
-            if (img_n == 1) {
-               for (q=x-1; q >= 0; --q) {
-                  cur[q*2+1] = 255;
-                  cur[q*2+0] = cur[q];
-               }
-            } else {
-               STBI_ASSERT(img_n == 3);
-               for (q=x-1; q >= 0; --q) {
-                  cur[q*4+3] = 255;
-                  cur[q*4+2] = cur[q*3+2];
-                  cur[q*4+1] = cur[q*3+1];
-                  cur[q*4+0] = cur[q*3+0];
-               }
-            }
-         }
-      }
-   } else if (depth == 16) {
-      // force the image data from big-endian to platform-native.
-      // this is done in a separate pass due to the decoding relying
-      // on the data being untouched, but could probably be done
-      // per-line during decode if care is taken.
-      stbi_uc *cur = a->out;
-      stbi__uint16 *cur16 = (stbi__uint16*)cur;
-
-      for(i=0; i < x*y*out_n; ++i,cur16++,cur+=2) {
-         *cur16 = (cur[0] << 8) | cur[1];
-      }
-   }
-
-   return 1;
-}
-
-static int stbi__create_png_image(stbi__png *a, stbi_uc *image_data, stbi__uint32 image_data_len, int out_n, int depth, int color, int interlaced)
-{
-   stbi_uc *final;
-   int p;
-   if (!interlaced)
-      return stbi__create_png_image_raw(a, image_data, image_data_len, out_n, a->s->img_x, a->s->img_y, depth, color);
-
-   // de-interlacing
-   final = (stbi_uc *) stbi__malloc(a->s->img_x * a->s->img_y * out_n);
-   for (p=0; p < 7; ++p) {
-      int xorig[] = { 0,4,0,2,0,1,0 };
-      int yorig[] = { 0,0,4,0,2,0,1 };
-      int xspc[]  = { 8,8,4,4,2,2,1 };
-      int yspc[]  = { 8,8,8,4,4,2,2 };
-      int i,j,x,y;
-      // pass1_x[4] = 0, pass1_x[5] = 1, pass1_x[12] = 1
-      x = (a->s->img_x - xorig[p] + xspc[p]-1) / xspc[p];
-      y = (a->s->img_y - yorig[p] + yspc[p]-1) / yspc[p];
-      if (x && y) {
-         stbi__uint32 img_len = ((((a->s->img_n * x * depth) + 7) >> 3) + 1) * y;
-         if (!stbi__create_png_image_raw(a, image_data, image_data_len, out_n, x, y, depth, color)) {
-            STBI_FREE(final);
-            return 0;
-         }
-         for (j=0; j < y; ++j) {
-            for (i=0; i < x; ++i) {
-               int out_y = j*yspc[p]+yorig[p];
-               int out_x = i*xspc[p]+xorig[p];
-               memcpy(final + out_y*a->s->img_x*out_n + out_x*out_n,
-                      a->out + (j*x+i)*out_n, out_n);
-            }
-         }
-         STBI_FREE(a->out);
-         image_data += img_len;
-         image_data_len -= img_len;
-      }
-   }
-   a->out = final;
-
-   return 1;
-}
-
-static int stbi__compute_transparency(stbi__png *z, stbi_uc tc[3], int out_n)
-{
-   stbi__context *s = z->s;
-   stbi__uint32 i, pixel_count = s->img_x * s->img_y;
-   stbi_uc *p = z->out;
-
-   // compute color-based transparency, assuming we've
-   // already got 255 as the alpha value in the output
-   STBI_ASSERT(out_n == 2 || out_n == 4);
-
-   if (out_n == 2) {
-      for (i=0; i < pixel_count; ++i) {
-         p[1] = (p[0] == tc[0] ? 0 : 255);
-         p += 2;
-      }
-   } else {
-      for (i=0; i < pixel_count; ++i) {
-         if (p[0] == tc[0] && p[1] == tc[1] && p[2] == tc[2])
-            p[3] = 0;
-         p += 4;
-      }
-   }
-   return 1;
-}
-
-static int stbi__compute_transparency16(stbi__png *z, stbi__uint16 tc[3], int out_n)
-{
-   stbi__context *s = z->s;
-   stbi__uint32 i, pixel_count = s->img_x * s->img_y;
-   stbi__uint16 *p = (stbi__uint16*) z->out;
-
-   // compute color-based transparency, assuming we've
-   // already got 65535 as the alpha value in the output
-   STBI_ASSERT(out_n == 2 || out_n == 4);
-
-   if (out_n == 2) {
-      for (i = 0; i < pixel_count; ++i) {
-         p[1] = (p[0] == tc[0] ? 0 : 65535);
-         p += 2;
-      }
-   } else {
-      for (i = 0; i < pixel_count; ++i) {
-         if (p[0] == tc[0] && p[1] == tc[1] && p[2] == tc[2])
-            p[3] = 0;
-         p += 4;
-      }
-   }
-   return 1;
-}
-
-static int stbi__expand_png_palette(stbi__png *a, stbi_uc *palette, int len, int pal_img_n)
-{
-   stbi__uint32 i, pixel_count = a->s->img_x * a->s->img_y;
-   stbi_uc *p, *temp_out, *orig = a->out;
-
-   p = (stbi_uc *) stbi__malloc(pixel_count * pal_img_n);
-   if (p == NULL) return stbi__err("outofmem", "Out of memory");
-
-   // between here and free(out) below, exitting would leak
-   temp_out = p;
-
-   if (pal_img_n == 3) {
-      for (i=0; i < pixel_count; ++i) {
-         int n = orig[i]*4;
-         p[0] = palette[n  ];
-         p[1] = palette[n+1];
-         p[2] = palette[n+2];
-         p += 3;
-      }
-   } else {
-      for (i=0; i < pixel_count; ++i) {
-         int n = orig[i]*4;
-         p[0] = palette[n  ];
-         p[1] = palette[n+1];
-         p[2] = palette[n+2];
-         p[3] = palette[n+3];
-         p += 4;
-      }
-   }
-   STBI_FREE(a->out);
-   a->out = temp_out;
-
-   STBI_NOTUSED(len);
-
-   return 1;
-}
-
-static int stbi__reduce_png(stbi__png *p)
-{
-   int i;
-   int img_len = p->s->img_x * p->s->img_y * p->s->img_out_n;
-   stbi_uc *reduced;
-   stbi__uint16 *orig = (stbi__uint16*)p->out;
-
-   if (p->depth != 16) return 1; // don't need to do anything if not 16-bit data
-
-   reduced = (stbi_uc *)stbi__malloc(img_len);
-   if (p == NULL) return stbi__err("outofmem", "Out of memory");
-
-   for (i = 0; i < img_len; ++i) reduced[i] = (stbi_uc)((orig[i] >> 8) & 0xFF); // top half of each byte is a decent approx of 16->8 bit scaling
-
-   p->out = reduced;
-   STBI_FREE(orig);
-
-   return 1;
-}
-
-static int stbi__unpremultiply_on_load = 0;
-static int stbi__de_iphone_flag = 0;
-
-STBIDEF void stbi_set_unpremultiply_on_load(int flag_true_if_should_unpremultiply)
-{
-   stbi__unpremultiply_on_load = flag_true_if_should_unpremultiply;
-}
-
-STBIDEF void stbi_convert_iphone_png_to_rgb(int flag_true_if_should_convert)
-{
-   stbi__de_iphone_flag = flag_true_if_should_convert;
-}
-
-static void stbi__de_iphone(stbi__png *z)
-{
-   stbi__context *s = z->s;
-   stbi__uint32 i, pixel_count = s->img_x * s->img_y;
-   stbi_uc *p = z->out;
-
-   if (s->img_out_n == 3) {  // convert bgr to rgb
-      for (i=0; i < pixel_count; ++i) {
-         stbi_uc t = p[0];
-         p[0] = p[2];
-         p[2] = t;
-         p += 3;
-      }
-   } else {
-      STBI_ASSERT(s->img_out_n == 4);
-      if (stbi__unpremultiply_on_load) {
-         // convert bgr to rgb and unpremultiply
-         for (i=0; i < pixel_count; ++i) {
-            stbi_uc a = p[3];
-            stbi_uc t = p[0];
-            if (a) {
-               p[0] = p[2] * 255 / a;
-               p[1] = p[1] * 255 / a;
-               p[2] =  t   * 255 / a;
-            } else {
-               p[0] = p[2];
-               p[2] = t;
-            }
-            p += 4;
-         }
-      } else {
-         // convert bgr to rgb
-         for (i=0; i < pixel_count; ++i) {
-            stbi_uc t = p[0];
-            p[0] = p[2];
-            p[2] = t;
-            p += 4;
-         }
-      }
-   }
-}
-
-#define STBI__PNG_TYPE(a,b,c,d)  (((a) << 24) + ((b) << 16) + ((c) << 8) + (d))
-
-static int stbi__parse_png_file(stbi__png *z, int scan, int req_comp)
-{
-   stbi_uc palette[1024], pal_img_n=0;
-   stbi_uc has_trans=0, tc[3];
-   stbi__uint16 tc16[3];
-   stbi__uint32 ioff=0, idata_limit=0, i, pal_len=0;
-   int first=1,k,interlace=0, color=0, is_iphone=0;
-   stbi__context *s = z->s;
-
-   z->expanded = NULL;
-   z->idata = NULL;
-   z->out = NULL;
-
-   if (!stbi__check_png_header(s)) return 0;
-
-   if (scan == STBI__SCAN_type) return 1;
-
-   for (;;) {
-      stbi__pngchunk c = stbi__get_chunk_header(s);
-      switch (c.type) {
-         case STBI__PNG_TYPE('C','g','B','I'):
-            is_iphone = 1;
-            stbi__skip(s, c.length);
-            break;
-         case STBI__PNG_TYPE('I','H','D','R'): {
-            int comp,filter;
-            if (!first) return stbi__err("multiple IHDR","Corrupt PNG");
-            first = 0;
-            if (c.length != 13) return stbi__err("bad IHDR len","Corrupt PNG");
-            s->img_x = stbi__get32be(s); if (s->img_x > (1 << 24)) return stbi__err("too large","Very large image (corrupt?)");
-            s->img_y = stbi__get32be(s); if (s->img_y > (1 << 24)) return stbi__err("too large","Very large image (corrupt?)");
-            z->depth = stbi__get8(s);  if (z->depth != 1 && z->depth != 2 && z->depth != 4 && z->depth != 8 && z->depth != 16)  return stbi__err("1/2/4/8/16-bit only","PNG not supported: 1/2/4/8/16-bit only");
-            color = stbi__get8(s);  if (color > 6)         return stbi__err("bad ctype","Corrupt PNG");
-			if (color == 3 && z->depth == 16)                  return stbi__err("bad ctype","Corrupt PNG");
-            if (color == 3) pal_img_n = 3; else if (color & 1) return stbi__err("bad ctype","Corrupt PNG");
-            comp  = stbi__get8(s);  if (comp) return stbi__err("bad comp method","Corrupt PNG");
-            filter= stbi__get8(s);  if (filter) return stbi__err("bad filter method","Corrupt PNG");
-            interlace = stbi__get8(s); if (interlace>1) return stbi__err("bad interlace method","Corrupt PNG");
-            if (!s->img_x || !s->img_y) return stbi__err("0-pixel image","Corrupt PNG");
-            if (!pal_img_n) {
-               s->img_n = (color & 2 ? 3 : 1) + (color & 4 ? 1 : 0);
-               if ((1 << 30) / s->img_x / s->img_n < s->img_y) return stbi__err("too large", "Image too large to decode");
-               if (scan == STBI__SCAN_header) return 1;
-            } else {
-               // if paletted, then pal_n is our final components, and
-               // img_n is # components to decompress/filter.
-               s->img_n = 1;
-               if ((1 << 30) / s->img_x / 4 < s->img_y) return stbi__err("too large","Corrupt PNG");
-               // if SCAN_header, have to scan to see if we have a tRNS
-            }
-            break;
-         }
-
-         case STBI__PNG_TYPE('P','L','T','E'):  {
-            if (first) return stbi__err("first not IHDR", "Corrupt PNG");
-            if (c.length > 256*3) return stbi__err("invalid PLTE","Corrupt PNG");
-            pal_len = c.length / 3;
-            if (pal_len * 3 != c.length) return stbi__err("invalid PLTE","Corrupt PNG");
-            for (i=0; i < pal_len; ++i) {
-               palette[i*4+0] = stbi__get8(s);
-               palette[i*4+1] = stbi__get8(s);
-               palette[i*4+2] = stbi__get8(s);
-               palette[i*4+3] = 255;
-            }
-            break;
-         }
-
-         case STBI__PNG_TYPE('t','R','N','S'): {
-            if (first) return stbi__err("first not IHDR", "Corrupt PNG");
-            if (z->idata) return stbi__err("tRNS after IDAT","Corrupt PNG");
-            if (pal_img_n) {
-               if (scan == STBI__SCAN_header) { s->img_n = 4; return 1; }
-               if (pal_len == 0) return stbi__err("tRNS before PLTE","Corrupt PNG");
-               if (c.length > pal_len) return stbi__err("bad tRNS len","Corrupt PNG");
-               pal_img_n = 4;
-               for (i=0; i < c.length; ++i)
-                  palette[i*4+3] = stbi__get8(s);
-            } else {
-               if (!(s->img_n & 1)) return stbi__err("tRNS with alpha","Corrupt PNG");
-               if (c.length != (stbi__uint32) s->img_n*2) return stbi__err("bad tRNS len","Corrupt PNG");
-               has_trans = 1;
-               if (z->depth == 16) {
-                  for (k = 0; k < s->img_n; ++k) tc16[k] = stbi__get16be(s); // copy the values as-is
-               } else {
-                  for (k = 0; k < s->img_n; ++k) tc[k] = (stbi_uc)(stbi__get16be(s) & 255) * stbi__depth_scale_table[z->depth]; // non 8-bit images will be larger
-               }
-            }
-            break;
-         }
-
-         case STBI__PNG_TYPE('I','D','A','T'): {
-            if (first) return stbi__err("first not IHDR", "Corrupt PNG");
-            if (pal_img_n && !pal_len) return stbi__err("no PLTE","Corrupt PNG");
-            if (scan == STBI__SCAN_header) { s->img_n = pal_img_n; return 1; }
-            if ((int)(ioff + c.length) < (int)ioff) return 0;
-            if (ioff + c.length > idata_limit) {
-               stbi__uint32 idata_limit_old = idata_limit;
-               stbi_uc *p;
-               if (idata_limit == 0) idata_limit = c.length > 4096 ? c.length : 4096;
-               while (ioff + c.length > idata_limit)
-                  idata_limit *= 2;
-               STBI_NOTUSED(idata_limit_old);
-               p = (stbi_uc *) STBI_REALLOC_SIZED(z->idata, idata_limit_old, idata_limit); if (p == NULL) return stbi__err("outofmem", "Out of memory");
-               z->idata = p;
-            }
-            if (!stbi__getn(s, z->idata+ioff,c.length)) return stbi__err("outofdata","Corrupt PNG");
-            ioff += c.length;
-            break;
-         }
-
-         case STBI__PNG_TYPE('I','E','N','D'): {
-            stbi__uint32 raw_len, bpl;
-            if (first) return stbi__err("first not IHDR", "Corrupt PNG");
-            if (scan != STBI__SCAN_load) return 1;
-            if (z->idata == NULL) return stbi__err("no IDAT","Corrupt PNG");
-            // initial guess for decoded data size to avoid unnecessary reallocs
-            bpl = (s->img_x * z->depth + 7) / 8; // bytes per line, per component
-            raw_len = bpl * s->img_y * s->img_n /* pixels */ + s->img_y /* filter mode per row */;
-            z->expanded = (stbi_uc *) stbi_zlib_decode_malloc_guesssize_headerflag((char *) z->idata, ioff, raw_len, (int *) &raw_len, !is_iphone);
-            if (z->expanded == NULL) return 0; // zlib should set error
-            STBI_FREE(z->idata); z->idata = NULL;
-            if ((req_comp == s->img_n+1 && req_comp != 3 && !pal_img_n) || has_trans)
-               s->img_out_n = s->img_n+1;
-            else
-               s->img_out_n = s->img_n;
-            if (!stbi__create_png_image(z, z->expanded, raw_len, s->img_out_n, z->depth, color, interlace)) return 0;
-            if (has_trans) {
-               if (z->depth == 16) {
-                  if (!stbi__compute_transparency16(z, tc16, s->img_out_n)) return 0;
-               } else {
-                  if (!stbi__compute_transparency(z, tc, s->img_out_n)) return 0;
-               }
-            }
-            if (is_iphone && stbi__de_iphone_flag && s->img_out_n > 2)
-               stbi__de_iphone(z);
-            if (pal_img_n) {
-               // pal_img_n == 3 or 4
-               s->img_n = pal_img_n; // record the actual colors we had
-               s->img_out_n = pal_img_n;
-               if (req_comp >= 3) s->img_out_n = req_comp;
-               if (!stbi__expand_png_palette(z, palette, pal_len, s->img_out_n))
-                  return 0;
-            }
-            STBI_FREE(z->expanded); z->expanded = NULL;
-            return 1;
-         }
-
-         default:
-            // if critical, fail
-            if (first) return stbi__err("first not IHDR", "Corrupt PNG");
-            if ((c.type & (1 << 29)) == 0) {
-               #ifndef STBI_NO_FAILURE_STRINGS
-               // not threadsafe
-               static char invalid_chunk[] = "XXXX PNG chunk not known";
-               invalid_chunk[0] = STBI__BYTECAST(c.type >> 24);
-               invalid_chunk[1] = STBI__BYTECAST(c.type >> 16);
-               invalid_chunk[2] = STBI__BYTECAST(c.type >>  8);
-               invalid_chunk[3] = STBI__BYTECAST(c.type >>  0);
-               #endif
-               return stbi__err(invalid_chunk, "PNG not supported: unknown PNG chunk type");
-            }
-            stbi__skip(s, c.length);
-            break;
-      }
-      // end of PNG chunk, read and skip CRC
-      stbi__get32be(s);
-   }
-}
-
-static unsigned char *stbi__do_png(stbi__png *p, int *x, int *y, int *n, int req_comp)
-{
-   unsigned char *result=NULL;
-   if (req_comp < 0 || req_comp > 4) return stbi__errpuc("bad req_comp", "Internal error");
-   if (stbi__parse_png_file(p, STBI__SCAN_load, req_comp)) {
-      if (p->depth == 16) {
-         if (!stbi__reduce_png(p)) {
-            return result;
-         }
-      }
-      result = p->out;
-      p->out = NULL;
-      if (req_comp && req_comp != p->s->img_out_n) {
-         result = stbi__convert_format(result, p->s->img_out_n, req_comp, p->s->img_x, p->s->img_y);
-         p->s->img_out_n = req_comp;
-         if (result == NULL) return result;
-      }
-      *x = p->s->img_x;
-      *y = p->s->img_y;
-      if (n) *n = p->s->img_n;
-   }
-   STBI_FREE(p->out);      p->out      = NULL;
-   STBI_FREE(p->expanded); p->expanded = NULL;
-   STBI_FREE(p->idata);    p->idata    = NULL;
-
-   return result;
-}
-
-static unsigned char *stbi__png_load(stbi__context *s, int *x, int *y, int *comp, int req_comp)
-{
-   stbi__png p;
-   p.s = s;
-   return stbi__do_png(&p, x,y,comp,req_comp);
-}
-
-static int stbi__png_test(stbi__context *s)
-{
-   int r;
-   r = stbi__check_png_header(s);
-   stbi__rewind(s);
-   return r;
-}
-
-static int stbi__png_info_raw(stbi__png *p, int *x, int *y, int *comp)
-{
-   if (!stbi__parse_png_file(p, STBI__SCAN_header, 0)) {
-      stbi__rewind( p->s );
-      return 0;
-   }
-   if (x) *x = p->s->img_x;
-   if (y) *y = p->s->img_y;
-   if (comp) *comp = p->s->img_n;
-   return 1;
-}
-
-static int stbi__png_info(stbi__context *s, int *x, int *y, int *comp)
-{
-   stbi__png p;
-   p.s = s;
-   return stbi__png_info_raw(&p, x, y, comp);
-}
-#endif
-
-// Microsoft/Windows BMP image
-
-#ifndef STBI_NO_BMP
-static int stbi__bmp_test_raw(stbi__context *s)
-{
-   int r;
-   int sz;
-   if (stbi__get8(s) != 'B') return 0;
-   if (stbi__get8(s) != 'M') return 0;
-   stbi__get32le(s); // discard filesize
-   stbi__get16le(s); // discard reserved
-   stbi__get16le(s); // discard reserved
-   stbi__get32le(s); // discard data offset
-   sz = stbi__get32le(s);
-   r = (sz == 12 || sz == 40 || sz == 56 || sz == 108 || sz == 124);
-   return r;
-}
-
-static int stbi__bmp_test(stbi__context *s)
-{
-   int r = stbi__bmp_test_raw(s);
-   stbi__rewind(s);
-   return r;
-}
-
-
-// returns 0..31 for the highest set bit
-static int stbi__high_bit(unsigned int z)
-{
-   int n=0;
-   if (z == 0) return -1;
-   if (z >= 0x10000) n += 16, z >>= 16;
-   if (z >= 0x00100) n +=  8, z >>=  8;
-   if (z >= 0x00010) n +=  4, z >>=  4;
-   if (z >= 0x00004) n +=  2, z >>=  2;
-   if (z >= 0x00002) n +=  1, z >>=  1;
-   return n;
-}
-
-static int stbi__bitcount(unsigned int a)
-{
-   a = (a & 0x55555555) + ((a >>  1) & 0x55555555); // max 2
-   a = (a & 0x33333333) + ((a >>  2) & 0x33333333); // max 4
-   a = (a + (a >> 4)) & 0x0f0f0f0f; // max 8 per 4, now 8 bits
-   a = (a + (a >> 8)); // max 16 per 8 bits
-   a = (a + (a >> 16)); // max 32 per 8 bits
-   return a & 0xff;
-}
-
-static int stbi__shiftsigned(int v, int shift, int bits)
-{
-   int result;
-   int z=0;
-
-   if (shift < 0) v <<= -shift;
-   else v >>= shift;
-   result = v;
-
-   z = bits;
-   while (z < 8) {
-      result += v >> z;
-      z += bits;
-   }
-   return result;
-}
-
-typedef struct
-{
-   int bpp, offset, hsz;
-   unsigned int mr,mg,mb,ma, all_a;
-} stbi__bmp_data;
-
-static void *stbi__bmp_parse_header(stbi__context *s, stbi__bmp_data *info)
-{
-   int hsz;
-   if (stbi__get8(s) != 'B' || stbi__get8(s) != 'M') return stbi__errpuc("not BMP", "Corrupt BMP");
-   stbi__get32le(s); // discard filesize
-   stbi__get16le(s); // discard reserved
-   stbi__get16le(s); // discard reserved
-   info->offset = stbi__get32le(s);
-   info->hsz = hsz = stbi__get32le(s);
-   info->mr = info->mg = info->mb = info->ma = 0;
-
-   if (hsz != 12 && hsz != 40 && hsz != 56 && hsz != 108 && hsz != 124) return stbi__errpuc("unknown BMP", "BMP type not supported: unknown");
-   if (hsz == 12) {
-      s->img_x = stbi__get16le(s);
-      s->img_y = stbi__get16le(s);
-   } else {
-      s->img_x = stbi__get32le(s);
-      s->img_y = stbi__get32le(s);
-   }
-   if (stbi__get16le(s) != 1) return stbi__errpuc("bad BMP", "bad BMP");
-   info->bpp = stbi__get16le(s);
-   if (info->bpp == 1) return stbi__errpuc("monochrome", "BMP type not supported: 1-bit");
-   if (hsz != 12) {
-      int compress = stbi__get32le(s);
-      if (compress == 1 || compress == 2) return stbi__errpuc("BMP RLE", "BMP type not supported: RLE");
-      stbi__get32le(s); // discard sizeof
-      stbi__get32le(s); // discard hres
-      stbi__get32le(s); // discard vres
-      stbi__get32le(s); // discard colorsused
-      stbi__get32le(s); // discard max important
-      if (hsz == 40 || hsz == 56) {
-         if (hsz == 56) {
-            stbi__get32le(s);
-            stbi__get32le(s);
-            stbi__get32le(s);
-            stbi__get32le(s);
-         }
-         if (info->bpp == 16 || info->bpp == 32) {
-            if (compress == 0) {
-               if (info->bpp == 32) {
-                  info->mr = 0xffu << 16;
-                  info->mg = 0xffu <<  8;
-                  info->mb = 0xffu <<  0;
-                  info->ma = 0xffu << 24;
-                  info->all_a = 0; // if all_a is 0 at end, then we loaded alpha channel but it was all 0
-               } else {
-                  info->mr = 31u << 10;
-                  info->mg = 31u <<  5;
-                  info->mb = 31u <<  0;
-               }
-            } else if (compress == 3) {
-               info->mr = stbi__get32le(s);
-               info->mg = stbi__get32le(s);
-               info->mb = stbi__get32le(s);
-               // not documented, but generated by photoshop and handled by mspaint
-               if (info->mr == info->mg && info->mg == info->mb) {
-                  // ?!?!?
-                  return stbi__errpuc("bad BMP", "bad BMP");
-               }
-            } else
-               return stbi__errpuc("bad BMP", "bad BMP");
-         }
-      } else {
-         int i;
-         if (hsz != 108 && hsz != 124)
-            return stbi__errpuc("bad BMP", "bad BMP");
-         info->mr = stbi__get32le(s);
-         info->mg = stbi__get32le(s);
-         info->mb = stbi__get32le(s);
-         info->ma = stbi__get32le(s);
-         stbi__get32le(s); // discard color space
-         for (i=0; i < 12; ++i)
-            stbi__get32le(s); // discard color space parameters
-         if (hsz == 124) {
-            stbi__get32le(s); // discard rendering intent
-            stbi__get32le(s); // discard offset of profile data
-            stbi__get32le(s); // discard size of profile data
-            stbi__get32le(s); // discard reserved
-         }
-      }
-   }
-   return (void *) 1;
-}
-
-
-static stbi_uc *stbi__bmp_load(stbi__context *s, int *x, int *y, int *comp, int req_comp)
-{
-   stbi_uc *out;
-   unsigned int mr=0,mg=0,mb=0,ma=0, all_a;
-   stbi_uc pal[256][4];
-   int psize=0,i,j,width;
-   int flip_vertically, pad, target;
-   stbi__bmp_data info;
-
-   info.all_a = 255;
-   if (stbi__bmp_parse_header(s, &info) == NULL)
-      return NULL; // error code already set
-
-   flip_vertically = ((int) s->img_y) > 0;
-   s->img_y = abs((int) s->img_y);
-
-   mr = info.mr;
-   mg = info.mg;
-   mb = info.mb;
-   ma = info.ma;
-   all_a = info.all_a;
-
-   if (info.hsz == 12) {
-      if (info.bpp < 24)
-         psize = (info.offset - 14 - 24) / 3;
-   } else {
-      if (info.bpp < 16)
-         psize = (info.offset - 14 - info.hsz) >> 2;
-   }
-
-   s->img_n = ma ? 4 : 3;
-   if (req_comp && req_comp >= 3) // we can directly decode 3 or 4
-      target = req_comp;
-   else
-      target = s->img_n; // if they want monochrome, we'll post-convert
-
-   out = (stbi_uc *) stbi__malloc(target * s->img_x * s->img_y);
-   if (!out) return stbi__errpuc("outofmem", "Out of memory");
-   if (info.bpp < 16) {
-      int z=0;
-      if (psize == 0 || psize > 256) { STBI_FREE(out); return stbi__errpuc("invalid", "Corrupt BMP"); }
-      for (i=0; i < psize; ++i) {
-         pal[i][2] = stbi__get8(s);
-         pal[i][1] = stbi__get8(s);
-         pal[i][0] = stbi__get8(s);
-         if (info.hsz != 12) stbi__get8(s);
-         pal[i][3] = 255;
-      }
-      stbi__skip(s, info.offset - 14 - info.hsz - psize * (info.hsz == 12 ? 3 : 4));
-      if (info.bpp == 4) width = (s->img_x + 1) >> 1;
-      else if (info.bpp == 8) width = s->img_x;
-      else { STBI_FREE(out); return stbi__errpuc("bad bpp", "Corrupt BMP"); }
-      pad = (-width)&3;
-      for (j=0; j < (int) s->img_y; ++j) {
-         for (i=0; i < (int) s->img_x; i += 2) {
-            int v=stbi__get8(s),v2=0;
-            if (info.bpp == 4) {
-               v2 = v & 15;
-               v >>= 4;
-            }
-            out[z++] = pal[v][0];
-            out[z++] = pal[v][1];
-            out[z++] = pal[v][2];
-            if (target == 4) out[z++] = 255;
-            if (i+1 == (int) s->img_x) break;
-            v = (info.bpp == 8) ? stbi__get8(s) : v2;
-            out[z++] = pal[v][0];
-            out[z++] = pal[v][1];
-            out[z++] = pal[v][2];
-            if (target == 4) out[z++] = 255;
-         }
-         stbi__skip(s, pad);
-      }
-   } else {
-      int rshift=0,gshift=0,bshift=0,ashift=0,rcount=0,gcount=0,bcount=0,acount=0;
-      int z = 0;
-      int easy=0;
-      stbi__skip(s, info.offset - 14 - info.hsz);
-      if (info.bpp == 24) width = 3 * s->img_x;
-      else if (info.bpp == 16) width = 2*s->img_x;
-      else /* bpp = 32 and pad = 0 */ width=0;
-      pad = (-width) & 3;
-      if (info.bpp == 24) {
-         easy = 1;
-      } else if (info.bpp == 32) {
-         if (mb == 0xff && mg == 0xff00 && mr == 0x00ff0000 && ma == 0xff000000)
-            easy = 2;
-      }
-      if (!easy) {
-         if (!mr || !mg || !mb) { STBI_FREE(out); return stbi__errpuc("bad masks", "Corrupt BMP"); }
-         // right shift amt to put high bit in position #7
-         rshift = stbi__high_bit(mr)-7; rcount = stbi__bitcount(mr);
-         gshift = stbi__high_bit(mg)-7; gcount = stbi__bitcount(mg);
-         bshift = stbi__high_bit(mb)-7; bcount = stbi__bitcount(mb);
-         ashift = stbi__high_bit(ma)-7; acount = stbi__bitcount(ma);
-      }
-      for (j=0; j < (int) s->img_y; ++j) {
-         if (easy) {
-            for (i=0; i < (int) s->img_x; ++i) {
-               unsigned char a;
-               out[z+2] = stbi__get8(s);
-               out[z+1] = stbi__get8(s);
-               out[z+0] = stbi__get8(s);
-               z += 3;
-               a = (easy == 2 ? stbi__get8(s) : 255);
-               all_a |= a;
-               if (target == 4) out[z++] = a;
-            }
-         } else {
-            int bpp = info.bpp;
-            for (i=0; i < (int) s->img_x; ++i) {
-               stbi__uint32 v = (bpp == 16 ? (stbi__uint32) stbi__get16le(s) : stbi__get32le(s));
-               int a;
-               out[z++] = STBI__BYTECAST(stbi__shiftsigned(v & mr, rshift, rcount));
-               out[z++] = STBI__BYTECAST(stbi__shiftsigned(v & mg, gshift, gcount));
-               out[z++] = STBI__BYTECAST(stbi__shiftsigned(v & mb, bshift, bcount));
-               a = (ma ? stbi__shiftsigned(v & ma, ashift, acount) : 255);
-               all_a |= a;
-               if (target == 4) out[z++] = STBI__BYTECAST(a);
-            }
-         }
-         stbi__skip(s, pad);
-      }
-   }
-
-   // if alpha channel is all 0s, replace with all 255s
-   if (target == 4 && all_a == 0)
-      for (i=4*s->img_x*s->img_y-1; i >= 0; i -= 4)
-         out[i] = 255;
-
-   if (flip_vertically) {
-      stbi_uc t;
-      for (j=0; j < (int) s->img_y>>1; ++j) {
-         stbi_uc *p1 = out +      j     *s->img_x*target;
-         stbi_uc *p2 = out + (s->img_y-1-j)*s->img_x*target;
-         for (i=0; i < (int) s->img_x*target; ++i) {
-            t = p1[i], p1[i] = p2[i], p2[i] = t;
-         }
-      }
-   }
-
-   if (req_comp && req_comp != target) {
-      out = stbi__convert_format(out, target, req_comp, s->img_x, s->img_y);
-      if (out == NULL) return out; // stbi__convert_format frees input on failure
-   }
-
-   *x = s->img_x;
-   *y = s->img_y;
-   if (comp) *comp = s->img_n;
-   return out;
-}
-#endif
-
-// Targa Truevision - TGA
-// by Jonathan Dummer
-#ifndef STBI_NO_TGA
-// returns STBI_rgb or whatever, 0 on error
-static int stbi__tga_get_comp(int bits_per_pixel, int is_grey, int* is_rgb16)
-{
-   // only RGB or RGBA (incl. 16bit) or grey allowed
-   if(is_rgb16) *is_rgb16 = 0;
-   switch(bits_per_pixel) {
-      case 8:  return STBI_grey;
-      case 16: if(is_grey) return STBI_grey_alpha;
-            // else: fall-through
-      case 15: if(is_rgb16) *is_rgb16 = 1;
-            return STBI_rgb;
-      case 24: // fall-through
-      case 32: return bits_per_pixel/8;
-      default: return 0;
-   }
-}
-
-static int stbi__tga_info(stbi__context *s, int *x, int *y, int *comp)
-{
-    int tga_w, tga_h, tga_comp, tga_image_type, tga_bits_per_pixel, tga_colormap_bpp;
-    int sz, tga_colormap_type;
-    stbi__get8(s);                   // discard Offset
-    tga_colormap_type = stbi__get8(s); // colormap type
-    if( tga_colormap_type > 1 ) {
-        stbi__rewind(s);
-        return 0;      // only RGB or indexed allowed
-    }
-    tga_image_type = stbi__get8(s); // image type
-    if ( tga_colormap_type == 1 ) { // colormapped (paletted) image
-        if (tga_image_type != 1 && tga_image_type != 9) {
-            stbi__rewind(s);
-            return 0;
-        }
-        stbi__skip(s,4);       // skip index of first colormap entry and number of entries
-        sz = stbi__get8(s);    //   check bits per palette color entry
-        if ( (sz != 8) && (sz != 15) && (sz != 16) && (sz != 24) && (sz != 32) ) {
-            stbi__rewind(s);
-            return 0;
-        }
-        stbi__skip(s,4);       // skip image x and y origin
-        tga_colormap_bpp = sz;
-    } else { // "normal" image w/o colormap - only RGB or grey allowed, +/- RLE
-        if ( (tga_image_type != 2) && (tga_image_type != 3) && (tga_image_type != 10) && (tga_image_type != 11) ) {
-            stbi__rewind(s);
-            return 0; // only RGB or grey allowed, +/- RLE
-        }
-        stbi__skip(s,9); // skip colormap specification and image x/y origin
-        tga_colormap_bpp = 0;
-    }
-    tga_w = stbi__get16le(s);
-    if( tga_w < 1 ) {
-        stbi__rewind(s);
-        return 0;   // test width
-    }
-    tga_h = stbi__get16le(s);
-    if( tga_h < 1 ) {
-        stbi__rewind(s);
-        return 0;   // test height
-    }
-    tga_bits_per_pixel = stbi__get8(s); // bits per pixel
-    stbi__get8(s); // ignore alpha bits
-    if (tga_colormap_bpp != 0) {
-        if((tga_bits_per_pixel != 8) && (tga_bits_per_pixel != 16)) {
-            // when using a colormap, tga_bits_per_pixel is the size of the indexes
-            // I don't think anything but 8 or 16bit indexes makes sense
-            stbi__rewind(s);
-            return 0;
-        }
-        tga_comp = stbi__tga_get_comp(tga_colormap_bpp, 0, NULL);
-    } else {
-        tga_comp = stbi__tga_get_comp(tga_bits_per_pixel, (tga_image_type == 3) || (tga_image_type == 11), NULL);
-    }
-    if(!tga_comp) {
-      stbi__rewind(s);
-      return 0;
-    }
-    if (x) *x = tga_w;
-    if (y) *y = tga_h;
-    if (comp) *comp = tga_comp;
-    return 1;                   // seems to have passed everything
-}
-
-static int stbi__tga_test(stbi__context *s)
-{
-   int res = 0;
-   int sz, tga_color_type;
-   stbi__get8(s);      //   discard Offset
-   tga_color_type = stbi__get8(s);   //   color type
-   if ( tga_color_type > 1 ) goto errorEnd;   //   only RGB or indexed allowed
-   sz = stbi__get8(s);   //   image type
-   if ( tga_color_type == 1 ) { // colormapped (paletted) image
-      if (sz != 1 && sz != 9) goto errorEnd; // colortype 1 demands image type 1 or 9
-      stbi__skip(s,4);       // skip index of first colormap entry and number of entries
-      sz = stbi__get8(s);    //   check bits per palette color entry
-      if ( (sz != 8) && (sz != 15) && (sz != 16) && (sz != 24) && (sz != 32) ) goto errorEnd;
-      stbi__skip(s,4);       // skip image x and y origin
-   } else { // "normal" image w/o colormap
-      if ( (sz != 2) && (sz != 3) && (sz != 10) && (sz != 11) ) goto errorEnd; // only RGB or grey allowed, +/- RLE
-      stbi__skip(s,9); // skip colormap specification and image x/y origin
-   }
-   if ( stbi__get16le(s) < 1 ) goto errorEnd;      //   test width
-   if ( stbi__get16le(s) < 1 ) goto errorEnd;      //   test height
-   sz = stbi__get8(s);   //   bits per pixel
-   if ( (tga_color_type == 1) && (sz != 8) && (sz != 16) ) goto errorEnd; // for colormapped images, bpp is size of an index
-   if ( (sz != 8) && (sz != 15) && (sz != 16) && (sz != 24) && (sz != 32) ) goto errorEnd;
-
-   res = 1; // if we got this far, everything's good and we can return 1 instead of 0
-
-errorEnd:
-   stbi__rewind(s);
-   return res;
-}
-
-// read 16bit value and convert to 24bit RGB
-void stbi__tga_read_rgb16(stbi__context *s, stbi_uc* out)
-{
-   stbi__uint16 px = stbi__get16le(s);
-   stbi__uint16 fiveBitMask = 31;
-   // we have 3 channels with 5bits each
-   int r = (px >> 10) & fiveBitMask;
-   int g = (px >> 5) & fiveBitMask;
-   int b = px & fiveBitMask;
-   // Note that this saves the data in RGB(A) order, so it doesn't need to be swapped later
-   out[0] = (r * 255)/31;
-   out[1] = (g * 255)/31;
-   out[2] = (b * 255)/31;
-
-   // some people claim that the most significant bit might be used for alpha
-   // (possibly if an alpha-bit is set in the "image descriptor byte")
-   // but that only made 16bit test images completely translucent..
-   // so let's treat all 15 and 16bit TGAs as RGB with no alpha.
-}
-
-static stbi_uc *stbi__tga_load(stbi__context *s, int *x, int *y, int *comp, int req_comp)
-{
-   //   read in the TGA header stuff
-   int tga_offset = stbi__get8(s);
-   int tga_indexed = stbi__get8(s);
-   int tga_image_type = stbi__get8(s);
-   int tga_is_RLE = 0;
-   int tga_palette_start = stbi__get16le(s);
-   int tga_palette_len = stbi__get16le(s);
-   int tga_palette_bits = stbi__get8(s);
-   int tga_x_origin = stbi__get16le(s);
-   int tga_y_origin = stbi__get16le(s);
-   int tga_width = stbi__get16le(s);
-   int tga_height = stbi__get16le(s);
-   int tga_bits_per_pixel = stbi__get8(s);
-   int tga_comp, tga_rgb16=0;
-   int tga_inverted = stbi__get8(s);
-   // int tga_alpha_bits = tga_inverted & 15; // the 4 lowest bits - unused (useless?)
-   //   image data
-   unsigned char *tga_data;
-   unsigned char *tga_palette = NULL;
-   int i, j;
-   unsigned char raw_data[4];
-   int RLE_count = 0;
-   int RLE_repeating = 0;
-   int read_next_pixel = 1;
-
-   //   do a tiny bit of precessing
-   if ( tga_image_type >= 8 )
-   {
-      tga_image_type -= 8;
-      tga_is_RLE = 1;
-   }
-   tga_inverted = 1 - ((tga_inverted >> 5) & 1);
-
-   //   If I'm paletted, then I'll use the number of bits from the palette
-   if ( tga_indexed ) tga_comp = stbi__tga_get_comp(tga_palette_bits, 0, &tga_rgb16);
-   else tga_comp = stbi__tga_get_comp(tga_bits_per_pixel, (tga_image_type == 3), &tga_rgb16);
-
-   if(!tga_comp) // shouldn't really happen, stbi__tga_test() should have ensured basic consistency
-      return stbi__errpuc("bad format", "Can't find out TGA pixelformat");
-
-   //   tga info
-   *x = tga_width;
-   *y = tga_height;
-   if (comp) *comp = tga_comp;
-
-   tga_data = (unsigned char*)stbi__malloc( (size_t)tga_width * tga_height * tga_comp );
-   if (!tga_data) return stbi__errpuc("outofmem", "Out of memory");
-
-   // skip to the data's starting position (offset usually = 0)
-   stbi__skip(s, tga_offset );
-
-   if ( !tga_indexed && !tga_is_RLE && !tga_rgb16 ) {
-      for (i=0; i < tga_height; ++i) {
-         int row = tga_inverted ? tga_height -i - 1 : i;
-         stbi_uc *tga_row = tga_data + row*tga_width*tga_comp;
-         stbi__getn(s, tga_row, tga_width * tga_comp);
-      }
-   } else  {
-      //   do I need to load a palette?
-      if ( tga_indexed)
-      {
-         //   any data to skip? (offset usually = 0)
-         stbi__skip(s, tga_palette_start );
-         //   load the palette
-         tga_palette = (unsigned char*)stbi__malloc( tga_palette_len * tga_comp );
-         if (!tga_palette) {
-            STBI_FREE(tga_data);
-            return stbi__errpuc("outofmem", "Out of memory");
-         }
-         if (tga_rgb16) {
-            stbi_uc *pal_entry = tga_palette;
-            STBI_ASSERT(tga_comp == STBI_rgb);
-            for (i=0; i < tga_palette_len; ++i) {
-               stbi__tga_read_rgb16(s, pal_entry);
-               pal_entry += tga_comp;
-            }
-         } else if (!stbi__getn(s, tga_palette, tga_palette_len * tga_comp)) {
-               STBI_FREE(tga_data);
-               STBI_FREE(tga_palette);
-               return stbi__errpuc("bad palette", "Corrupt TGA");
-         }
-      }
-      //   load the data
-      for (i=0; i < tga_width * tga_height; ++i)
-      {
-         //   if I'm in RLE mode, do I need to get a RLE stbi__pngchunk?
-         if ( tga_is_RLE )
-         {
-            if ( RLE_count == 0 )
-            {
-               //   yep, get the next byte as a RLE command
-               int RLE_cmd = stbi__get8(s);
-               RLE_count = 1 + (RLE_cmd & 127);
-               RLE_repeating = RLE_cmd >> 7;
-               read_next_pixel = 1;
-            } else if ( !RLE_repeating )
-            {
-               read_next_pixel = 1;
-            }
-         } else
-         {
-            read_next_pixel = 1;
-         }
-         //   OK, if I need to read a pixel, do it now
-         if ( read_next_pixel )
-         {
-            //   load however much data we did have
-            if ( tga_indexed )
-            {
-               // read in index, then perform the lookup
-               int pal_idx = (tga_bits_per_pixel == 8) ? stbi__get8(s) : stbi__get16le(s);
-               if ( pal_idx >= tga_palette_len ) {
-                  // invalid index
-                  pal_idx = 0;
-               }
-               pal_idx *= tga_comp;
-               for (j = 0; j < tga_comp; ++j) {
-                  raw_data[j] = tga_palette[pal_idx+j];
-               }
-            } else if(tga_rgb16) {
-               STBI_ASSERT(tga_comp == STBI_rgb);
-               stbi__tga_read_rgb16(s, raw_data);
-            } else {
-               //   read in the data raw
-               for (j = 0; j < tga_comp; ++j) {
-                  raw_data[j] = stbi__get8(s);
-               }
-            }
-            //   clear the reading flag for the next pixel
-            read_next_pixel = 0;
-         } // end of reading a pixel
-
-         // copy data
-         for (j = 0; j < tga_comp; ++j)
-           tga_data[i*tga_comp+j] = raw_data[j];
-
-         //   in case we're in RLE mode, keep counting down
-         --RLE_count;
-      }
-      //   do I need to invert the image?
-      if ( tga_inverted )
-      {
-         for (j = 0; j*2 < tga_height; ++j)
-         {
-            int index1 = j * tga_width * tga_comp;
-            int index2 = (tga_height - 1 - j) * tga_width * tga_comp;
-            for (i = tga_width * tga_comp; i > 0; --i)
-            {
-               unsigned char temp = tga_data[index1];
-               tga_data[index1] = tga_data[index2];
-               tga_data[index2] = temp;
-               ++index1;
-               ++index2;
-            }
-         }
-      }
-      //   clear my palette, if I had one
-      if ( tga_palette != NULL )
-      {
-         STBI_FREE( tga_palette );
-      }
-   }
-
-   // swap RGB - if the source data was RGB16, it already is in the right order
-   if (tga_comp >= 3 && !tga_rgb16)
-   {
-      unsigned char* tga_pixel = tga_data;
-      for (i=0; i < tga_width * tga_height; ++i)
-      {
-         unsigned char temp = tga_pixel[0];
-         tga_pixel[0] = tga_pixel[2];
-         tga_pixel[2] = temp;
-         tga_pixel += tga_comp;
-      }
-   }
-
-   // convert to target component count
-   if (req_comp && req_comp != tga_comp)
-      tga_data = stbi__convert_format(tga_data, tga_comp, req_comp, tga_width, tga_height);
-
-   //   the things I do to get rid of an error message, and yet keep
-   //   Microsoft's C compilers happy... [8^(
-   tga_palette_start = tga_palette_len = tga_palette_bits =
-         tga_x_origin = tga_y_origin = 0;
-   //   OK, done
-   return tga_data;
-}
-#endif
-
-// *************************************************************************************************
-// Photoshop PSD loader -- PD by Thatcher Ulrich, integration by Nicolas Schulz, tweaked by STB
-
-#ifndef STBI_NO_PSD
-static int stbi__psd_test(stbi__context *s)
-{
-   int r = (stbi__get32be(s) == 0x38425053);
-   stbi__rewind(s);
-   return r;
-}
-
-static stbi_uc *stbi__psd_load(stbi__context *s, int *x, int *y, int *comp, int req_comp)
-{
-   int   pixelCount;
-   int channelCount, compression;
-   int channel, i, count, len;
-   int bitdepth;
-   int w,h;
-   stbi_uc *out;
-
-   // Check identifier
-   if (stbi__get32be(s) != 0x38425053)   // "8BPS"
-      return stbi__errpuc("not PSD", "Corrupt PSD image");
-
-   // Check file type version.
-   if (stbi__get16be(s) != 1)
-      return stbi__errpuc("wrong version", "Unsupported version of PSD image");
-
-   // Skip 6 reserved bytes.
-   stbi__skip(s, 6 );
-
-   // Read the number of channels (R, G, B, A, etc).
-   channelCount = stbi__get16be(s);
-   if (channelCount < 0 || channelCount > 16)
-      return stbi__errpuc("wrong channel count", "Unsupported number of channels in PSD image");
-
-   // Read the rows and columns of the image.
-   h = stbi__get32be(s);
-   w = stbi__get32be(s);
-
-   // Make sure the depth is 8 bits.
-   bitdepth = stbi__get16be(s);
-   if (bitdepth != 8 && bitdepth != 16)
-      return stbi__errpuc("unsupported bit depth", "PSD bit depth is not 8 or 16 bit");
-
-   // Make sure the color mode is RGB.
-   // Valid options are:
-   //   0: Bitmap
-   //   1: Grayscale
-   //   2: Indexed color
-   //   3: RGB color
-   //   4: CMYK color
-   //   7: Multichannel
-   //   8: Duotone
-   //   9: Lab color
-   if (stbi__get16be(s) != 3)
-      return stbi__errpuc("wrong color format", "PSD is not in RGB color format");
-
-   // Skip the Mode Data.  (It's the palette for indexed color; other info for other modes.)
-   stbi__skip(s,stbi__get32be(s) );
-
-   // Skip the image resources.  (resolution, pen tool paths, etc)
-   stbi__skip(s, stbi__get32be(s) );
-
-   // Skip the reserved data.
-   stbi__skip(s, stbi__get32be(s) );
-
-   // Find out if the data is compressed.
-   // Known values:
-   //   0: no compression
-   //   1: RLE compressed
-   compression = stbi__get16be(s);
-   if (compression > 1)
-      return stbi__errpuc("bad compression", "PSD has an unknown compression format");
-
-   // Create the destination image.
-   out = (stbi_uc *) stbi__malloc(4 * w*h);
-   if (!out) return stbi__errpuc("outofmem", "Out of memory");
-   pixelCount = w*h;
-
-   // Initialize the data to zero.
-   //memset( out, 0, pixelCount * 4 );
-
-   // Finally, the image data.
-   if (compression) {
-      // RLE as used by .PSD and .TIFF
-      // Loop until you get the number of unpacked bytes you are expecting:
-      //     Read the next source byte into n.
-      //     If n is between 0 and 127 inclusive, copy the next n+1 bytes literally.
-      //     Else if n is between -127 and -1 inclusive, copy the next byte -n+1 times.
-      //     Else if n is 128, noop.
-      // Endloop
-
-      // The RLE-compressed data is preceeded by a 2-byte data count for each row in the data,
-      // which we're going to just skip.
-      stbi__skip(s, h * channelCount * 2 );
-
-      // Read the RLE data by channel.
-      for (channel = 0; channel < 4; channel++) {
-         stbi_uc *p;
-
-         p = out+channel;
-         if (channel >= channelCount) {
-            // Fill this channel with default data.
-            for (i = 0; i < pixelCount; i++, p += 4)
-               *p = (channel == 3 ? 255 : 0);
-         } else {
-            // Read the RLE data.
-            count = 0;
-            while (count < pixelCount) {
-               len = stbi__get8(s);
-               if (len == 128) {
-                  // No-op.
-               } else if (len < 128) {
-                  // Copy next len+1 bytes literally.
-                  len++;
-                  count += len;
-                  while (len) {
-                     *p = stbi__get8(s);
-                     p += 4;
-                     len--;
-                  }
-               } else if (len > 128) {
-                  stbi_uc   val;
-                  // Next -len+1 bytes in the dest are replicated from next source byte.
-                  // (Interpret len as a negative 8-bit int.)
-                  len ^= 0x0FF;
-                  len += 2;
-                  val = stbi__get8(s);
-                  count += len;
-                  while (len) {
-                     *p = val;
-                     p += 4;
-                     len--;
-                  }
-               }
-            }
-         }
-      }
-
-   } else {
-      // We're at the raw image data.  It's each channel in order (Red, Green, Blue, Alpha, ...)
-      // where each channel consists of an 8-bit value for each pixel in the image.
-
-      // Read the data by channel.
-      for (channel = 0; channel < 4; channel++) {
-         stbi_uc *p;
-
-         p = out + channel;
-         if (channel >= channelCount) {
-            // Fill this channel with default data.
-            stbi_uc val = channel == 3 ? 255 : 0;
-            for (i = 0; i < pixelCount; i++, p += 4)
-               *p = val;
-         } else {
-            // Read the data.
-            if (bitdepth == 16) {
-               for (i = 0; i < pixelCount; i++, p += 4)
-                  *p = (stbi_uc) (stbi__get16be(s) >> 8);
-            } else {
-               for (i = 0; i < pixelCount; i++, p += 4)
-                  *p = stbi__get8(s);
-            }
-         }
-      }
-   }
-
-   if (channelCount >= 4) {
-      for (i=0; i < w*h; ++i) {
-         unsigned char *pixel = out + 4*i;
-         if (pixel[3] != 0 && pixel[3] != 255) {
-            // remove weird white matte from PSD
-            float a = pixel[3] / 255.0f;
-            float ra = 1.0f / a;
-            float inv_a = 255.0f * (1 - ra);
-            pixel[0] = (unsigned char) (pixel[0]*ra + inv_a);
-            pixel[1] = (unsigned char) (pixel[1]*ra + inv_a);
-            pixel[2] = (unsigned char) (pixel[2]*ra + inv_a);
-         }
-      }
-   }
-
-   if (req_comp && req_comp != 4) {
-      out = stbi__convert_format(out, 4, req_comp, w, h);
-      if (out == NULL) return out; // stbi__convert_format frees input on failure
-   }
-
-   if (comp) *comp = 4;
-   *y = h;
-   *x = w;
-
-   return out;
-}
-#endif
-
-// *************************************************************************************************
-// Softimage PIC loader
-// by Tom Seddon
-//
-// See http://softimage.wiki.softimage.com/index.php/INFO:_PIC_file_format
-// See http://ozviz.wasp.uwa.edu.au/~pbourke/dataformats/softimagepic/
-
-#ifndef STBI_NO_PIC
-static int stbi__pic_is4(stbi__context *s,const char *str)
-{
-   int i;
-   for (i=0; i<4; ++i)
-      if (stbi__get8(s) != (stbi_uc)str[i])
-         return 0;
-
-   return 1;
-}
-
-static int stbi__pic_test_core(stbi__context *s)
-{
-   int i;
-
-   if (!stbi__pic_is4(s,"\x53\x80\xF6\x34"))
-      return 0;
-
-   for(i=0;i<84;++i)
-      stbi__get8(s);
-
-   if (!stbi__pic_is4(s,"PICT"))
-      return 0;
-
-   return 1;
-}
-
-typedef struct
-{
-   stbi_uc size,type,channel;
-} stbi__pic_packet;
-
-static stbi_uc *stbi__readval(stbi__context *s, int channel, stbi_uc *dest)
-{
-   int mask=0x80, i;
-
-   for (i=0; i<4; ++i, mask>>=1) {
-      if (channel & mask) {
-         if (stbi__at_eof(s)) return stbi__errpuc("bad file","PIC file too short");
-         dest[i]=stbi__get8(s);
-      }
-   }
-
-   return dest;
-}
-
-static void stbi__copyval(int channel,stbi_uc *dest,const stbi_uc *src)
-{
-   int mask=0x80,i;
-
-   for (i=0;i<4; ++i, mask>>=1)
-      if (channel&mask)
-         dest[i]=src[i];
-}
-
-static stbi_uc *stbi__pic_load_core(stbi__context *s,int width,int height,int *comp, stbi_uc *result)
-{
-   int act_comp=0,num_packets=0,y,chained;
-   stbi__pic_packet packets[10];
-
-   // this will (should...) cater for even some bizarre stuff like having data
-    // for the same channel in multiple packets.
-   do {
-      stbi__pic_packet *packet;
-
-      if (num_packets==sizeof(packets)/sizeof(packets[0]))
-         return stbi__errpuc("bad format","too many packets");
-
-      packet = &packets[num_packets++];
-
-      chained = stbi__get8(s);
-      packet->size    = stbi__get8(s);
-      packet->type    = stbi__get8(s);
-      packet->channel = stbi__get8(s);
-
-      act_comp |= packet->channel;
-
-      if (stbi__at_eof(s))          return stbi__errpuc("bad file","file too short (reading packets)");
-      if (packet->size != 8)  return stbi__errpuc("bad format","packet isn't 8bpp");
-   } while (chained);
-
-   *comp = (act_comp & 0x10 ? 4 : 3); // has alpha channel?
-
-   for(y=0; y<height; ++y) {
-      int packet_idx;
-
-      for(packet_idx=0; packet_idx < num_packets; ++packet_idx) {
-         stbi__pic_packet *packet = &packets[packet_idx];
-         stbi_uc *dest = result+y*width*4;
-
-         switch (packet->type) {
-            default:
-               return stbi__errpuc("bad format","packet has bad compression type");
-
-            case 0: {//uncompressed
-               int x;
-
-               for(x=0;x<width;++x, dest+=4)
-                  if (!stbi__readval(s,packet->channel,dest))
-                     return 0;
-               break;
-            }
-
-            case 1://Pure RLE
-               {
-                  int left=width, i;
-
-                  while (left>0) {
-                     stbi_uc count,value[4];
-
-                     count=stbi__get8(s);
-                     if (stbi__at_eof(s))   return stbi__errpuc("bad file","file too short (pure read count)");
-
-                     if (count > left)
-                        count = (stbi_uc) left;
-
-                     if (!stbi__readval(s,packet->channel,value))  return 0;
-
-                     for(i=0; i<count; ++i,dest+=4)
-                        stbi__copyval(packet->channel,dest,value);
-                     left -= count;
-                  }
-               }
-               break;
-
-            case 2: {//Mixed RLE
-               int left=width;
-               while (left>0) {
-                  int count = stbi__get8(s), i;
-                  if (stbi__at_eof(s))  return stbi__errpuc("bad file","file too short (mixed read count)");
-
-                  if (count >= 128) { // Repeated
-                     stbi_uc value[4];
-
-                     if (count==128)
-                        count = stbi__get16be(s);
-                     else
-                        count -= 127;
-                     if (count > left)
-                        return stbi__errpuc("bad file","scanline overrun");
-
-                     if (!stbi__readval(s,packet->channel,value))
-                        return 0;
-
-                     for(i=0;i<count;++i, dest += 4)
-                        stbi__copyval(packet->channel,dest,value);
-                  } else { // Raw
-                     ++count;
-                     if (count>left) return stbi__errpuc("bad file","scanline overrun");
-
-                     for(i=0;i<count;++i, dest+=4)
-                        if (!stbi__readval(s,packet->channel,dest))
-                           return 0;
-                  }
-                  left-=count;
-               }
-               break;
-            }
-         }
-      }
-   }
-
-   return result;
-}
-
-static stbi_uc *stbi__pic_load(stbi__context *s,int *px,int *py,int *comp,int req_comp)
-{
-   stbi_uc *result;
-   int i, x,y;
-
-   for (i=0; i<92; ++i)
-      stbi__get8(s);
-
-   x = stbi__get16be(s);
-   y = stbi__get16be(s);
-   if (stbi__at_eof(s))  return stbi__errpuc("bad file","file too short (pic header)");
-   if ((1 << 28) / x < y) return stbi__errpuc("too large", "Image too large to decode");
-
-   stbi__get32be(s); //skip `ratio'
-   stbi__get16be(s); //skip `fields'
-   stbi__get16be(s); //skip `pad'
-
-   // intermediate buffer is RGBA
-   result = (stbi_uc *) stbi__malloc(x*y*4);
-   memset(result, 0xff, x*y*4);
-
-   if (!stbi__pic_load_core(s,x,y,comp, result)) {
-      STBI_FREE(result);
-      result=0;
-   }
-   *px = x;
-   *py = y;
-   if (req_comp == 0) req_comp = *comp;
-   result=stbi__convert_format(result,4,req_comp,x,y);
-
-   return result;
-}
-
-static int stbi__pic_test(stbi__context *s)
-{
-   int r = stbi__pic_test_core(s);
-   stbi__rewind(s);
-   return r;
-}
-#endif
-
-// *************************************************************************************************
-// GIF loader -- public domain by Jean-Marc Lienher -- simplified/shrunk by stb
-
-#ifndef STBI_NO_GIF
-typedef struct
-{
-   stbi__int16 prefix;
-   stbi_uc first;
-   stbi_uc suffix;
-} stbi__gif_lzw;
-
-typedef struct
-{
-   int w,h;
-   stbi_uc *out, *old_out;             // output buffer (always 4 components)
-   int flags, bgindex, ratio, transparent, eflags, delay;
-   stbi_uc  pal[256][4];
-   stbi_uc lpal[256][4];
-   stbi__gif_lzw codes[4096];
-   stbi_uc *color_table;
-   int parse, step;
-   int lflags;
-   int start_x, start_y;
-   int max_x, max_y;
-   int cur_x, cur_y;
-   int line_size;
-} stbi__gif;
-
-static int stbi__gif_test_raw(stbi__context *s)
-{
-   int sz;
-   if (stbi__get8(s) != 'G' || stbi__get8(s) != 'I' || stbi__get8(s) != 'F' || stbi__get8(s) != '8') return 0;
-   sz = stbi__get8(s);
-   if (sz != '9' && sz != '7') return 0;
-   if (stbi__get8(s) != 'a') return 0;
-   return 1;
-}
-
-static int stbi__gif_test(stbi__context *s)
-{
-   int r = stbi__gif_test_raw(s);
-   stbi__rewind(s);
-   return r;
-}
-
-static void stbi__gif_parse_colortable(stbi__context *s, stbi_uc pal[256][4], int num_entries, int transp)
-{
-   int i;
-   for (i=0; i < num_entries; ++i) {
-      pal[i][2] = stbi__get8(s);
-      pal[i][1] = stbi__get8(s);
-      pal[i][0] = stbi__get8(s);
-      pal[i][3] = transp == i ? 0 : 255;
-   }
-}
-
-static int stbi__gif_header(stbi__context *s, stbi__gif *g, int *comp, int is_info)
-{
-   stbi_uc version;
-   if (stbi__get8(s) != 'G' || stbi__get8(s) != 'I' || stbi__get8(s) != 'F' || stbi__get8(s) != '8')
-      return stbi__err("not GIF", "Corrupt GIF");
-
-   version = stbi__get8(s);
-   if (version != '7' && version != '9')    return stbi__err("not GIF", "Corrupt GIF");
-   if (stbi__get8(s) != 'a')                return stbi__err("not GIF", "Corrupt GIF");
-
-   stbi__g_failure_reason = "";
-   g->w = stbi__get16le(s);
-   g->h = stbi__get16le(s);
-   g->flags = stbi__get8(s);
-   g->bgindex = stbi__get8(s);
-   g->ratio = stbi__get8(s);
-   g->transparent = -1;
-
-   if (comp != 0) *comp = 4;  // can't actually tell whether it's 3 or 4 until we parse the comments
-
-   if (is_info) return 1;
-
-   if (g->flags & 0x80)
-      stbi__gif_parse_colortable(s,g->pal, 2 << (g->flags & 7), -1);
-
-   return 1;
-}
-
-static int stbi__gif_info_raw(stbi__context *s, int *x, int *y, int *comp)
-{
-   stbi__gif* g = (stbi__gif*) stbi__malloc(sizeof(stbi__gif));
-   if (!stbi__gif_header(s, g, comp, 1)) {
-      STBI_FREE(g);
-      stbi__rewind( s );
-      return 0;
-   }
-   if (x) *x = g->w;
-   if (y) *y = g->h;
-   STBI_FREE(g);
-   return 1;
-}
-
-static void stbi__out_gif_code(stbi__gif *g, stbi__uint16 code)
-{
-   stbi_uc *p, *c;
-
-   // recurse to decode the prefixes, since the linked-list is backwards,
-   // and working backwards through an interleaved image would be nasty
-   if (g->codes[code].prefix >= 0)
-      stbi__out_gif_code(g, g->codes[code].prefix);
-
-   if (g->cur_y >= g->max_y) return;
-
-   p = &g->out[g->cur_x + g->cur_y];
-   c = &g->color_table[g->codes[code].suffix * 4];
-
-   if (c[3] >= 128) {
-      p[0] = c[2];
-      p[1] = c[1];
-      p[2] = c[0];
-      p[3] = c[3];
-   }
-   g->cur_x += 4;
-
-   if (g->cur_x >= g->max_x) {
-      g->cur_x = g->start_x;
-      g->cur_y += g->step;
-
-      while (g->cur_y >= g->max_y && g->parse > 0) {
-         g->step = (1 << g->parse) * g->line_size;
-         g->cur_y = g->start_y + (g->step >> 1);
-         --g->parse;
-      }
-   }
-}
-
-static stbi_uc *stbi__process_gif_raster(stbi__context *s, stbi__gif *g)
-{
-   stbi_uc lzw_cs;
-   stbi__int32 len, init_code;
-   stbi__uint32 first;
-   stbi__int32 codesize, codemask, avail, oldcode, bits, valid_bits, clear;
-   stbi__gif_lzw *p;
-
-   lzw_cs = stbi__get8(s);
-   if (lzw_cs > 12) return NULL;
-   clear = 1 << lzw_cs;
-   first = 1;
-   codesize = lzw_cs + 1;
-   codemask = (1 << codesize) - 1;
-   bits = 0;
-   valid_bits = 0;
-   for (init_code = 0; init_code < clear; init_code++) {
-      g->codes[init_code].prefix = -1;
-      g->codes[init_code].first = (stbi_uc) init_code;
-      g->codes[init_code].suffix = (stbi_uc) init_code;
-   }
-
-   // support no starting clear code
-   avail = clear+2;
-   oldcode = -1;
-
-   len = 0;
-   for(;;) {
-      if (valid_bits < codesize) {
-         if (len == 0) {
-            len = stbi__get8(s); // start new block
-            if (len == 0)
-               return g->out;
-         }
-         --len;
-         bits |= (stbi__int32) stbi__get8(s) << valid_bits;
-         valid_bits += 8;
-      } else {
-         stbi__int32 code = bits & codemask;
-         bits >>= codesize;
-         valid_bits -= codesize;
-         // @OPTIMIZE: is there some way we can accelerate the non-clear path?
-         if (code == clear) {  // clear code
-            codesize = lzw_cs + 1;
-            codemask = (1 << codesize) - 1;
-            avail = clear + 2;
-            oldcode = -1;
-            first = 0;
-         } else if (code == clear + 1) { // end of stream code
-            stbi__skip(s, len);
-            while ((len = stbi__get8(s)) > 0)
-               stbi__skip(s,len);
-            return g->out;
-         } else if (code <= avail) {
-            if (first) return stbi__errpuc("no clear code", "Corrupt GIF");
-
-            if (oldcode >= 0) {
-               p = &g->codes[avail++];
-               if (avail > 4096)        return stbi__errpuc("too many codes", "Corrupt GIF");
-               p->prefix = (stbi__int16) oldcode;
-               p->first = g->codes[oldcode].first;
-               p->suffix = (code == avail) ? p->first : g->codes[code].first;
-            } else if (code == avail)
-               return stbi__errpuc("illegal code in raster", "Corrupt GIF");
-
-            stbi__out_gif_code(g, (stbi__uint16) code);
-
-            if ((avail & codemask) == 0 && avail <= 0x0FFF) {
-               codesize++;
-               codemask = (1 << codesize) - 1;
-            }
-
-            oldcode = code;
-         } else {
-            return stbi__errpuc("illegal code in raster", "Corrupt GIF");
-         }
-      }
-   }
-}
-
-static void stbi__fill_gif_background(stbi__gif *g, int x0, int y0, int x1, int y1)
-{
-   int x, y;
-   stbi_uc *c = g->pal[g->bgindex];
-   for (y = y0; y < y1; y += 4 * g->w) {
-      for (x = x0; x < x1; x += 4) {
-         stbi_uc *p  = &g->out[y + x];
-         p[0] = c[2];
-         p[1] = c[1];
-         p[2] = c[0];
-         p[3] = 0;
-      }
-   }
-}
-
-// this function is designed to support animated gifs, although stb_image doesn't support it
-static stbi_uc *stbi__gif_load_next(stbi__context *s, stbi__gif *g, int *comp, int req_comp)
-{
-   int i;
-   stbi_uc *prev_out = 0;
-
-   if (g->out == 0 && !stbi__gif_header(s, g, comp,0))
-      return 0; // stbi__g_failure_reason set by stbi__gif_header
-
-   prev_out = g->out;
-   g->out = (stbi_uc *) stbi__malloc(4 * g->w * g->h);
-   if (g->out == 0) return stbi__errpuc("outofmem", "Out of memory");
-
-   switch ((g->eflags & 0x1C) >> 2) {
-      case 0: // unspecified (also always used on 1st frame)
-         stbi__fill_gif_background(g, 0, 0, 4 * g->w, 4 * g->w * g->h);
-         break;
-      case 1: // do not dispose
-         if (prev_out) memcpy(g->out, prev_out, 4 * g->w * g->h);
-         g->old_out = prev_out;
-         break;
-      case 2: // dispose to background
-         if (prev_out) memcpy(g->out, prev_out, 4 * g->w * g->h);
-         stbi__fill_gif_background(g, g->start_x, g->start_y, g->max_x, g->max_y);
-         break;
-      case 3: // dispose to previous
-         if (g->old_out) {
-            for (i = g->start_y; i < g->max_y; i += 4 * g->w)
-               memcpy(&g->out[i + g->start_x], &g->old_out[i + g->start_x], g->max_x - g->start_x);
-         }
-         break;
-   }
-
-   for (;;) {
-      switch (stbi__get8(s)) {
-         case 0x2C: /* Image Descriptor */
-         {
-            int prev_trans = -1;
-            stbi__int32 x, y, w, h;
-            stbi_uc *o;
-
-            x = stbi__get16le(s);
-            y = stbi__get16le(s);
-            w = stbi__get16le(s);
-            h = stbi__get16le(s);
-            if (((x + w) > (g->w)) || ((y + h) > (g->h)))
-               return stbi__errpuc("bad Image Descriptor", "Corrupt GIF");
-
-            g->line_size = g->w * 4;
-            g->start_x = x * 4;
-            g->start_y = y * g->line_size;
-            g->max_x   = g->start_x + w * 4;
-            g->max_y   = g->start_y + h * g->line_size;
-            g->cur_x   = g->start_x;
-            g->cur_y   = g->start_y;
-
-            g->lflags = stbi__get8(s);
-
-            if (g->lflags & 0x40) {
-               g->step = 8 * g->line_size; // first interlaced spacing
-               g->parse = 3;
-            } else {
-               g->step = g->line_size;
-               g->parse = 0;
-            }
-
-            if (g->lflags & 0x80) {
-               stbi__gif_parse_colortable(s,g->lpal, 2 << (g->lflags & 7), g->eflags & 0x01 ? g->transparent : -1);
-               g->color_table = (stbi_uc *) g->lpal;
-            } else if (g->flags & 0x80) {
-               if (g->transparent >= 0 && (g->eflags & 0x01)) {
-                  prev_trans = g->pal[g->transparent][3];
-                  g->pal[g->transparent][3] = 0;
-               }
-               g->color_table = (stbi_uc *) g->pal;
-            } else
-               return stbi__errpuc("missing color table", "Corrupt GIF");
-
-            o = stbi__process_gif_raster(s, g);
-            if (o == NULL) return NULL;
-
-            if (prev_trans != -1)
-               g->pal[g->transparent][3] = (stbi_uc) prev_trans;
-
-            return o;
-         }
-
-         case 0x21: // Comment Extension.
-         {
-            int len;
-            if (stbi__get8(s) == 0xF9) { // Graphic Control Extension.
-               len = stbi__get8(s);
-               if (len == 4) {
-                  g->eflags = stbi__get8(s);
-                  g->delay = stbi__get16le(s);
-                  g->transparent = stbi__get8(s);
-               } else {
-                  stbi__skip(s, len);
-                  break;
-               }
-            }
-            while ((len = stbi__get8(s)) != 0)
-               stbi__skip(s, len);
-            break;
-         }
-
-         case 0x3B: // gif stream termination code
-            return (stbi_uc *) s; // using '1' causes warning on some compilers
-
-         default:
-            return stbi__errpuc("unknown code", "Corrupt GIF");
-      }
-   }
-
-   STBI_NOTUSED(req_comp);
-}
-
-static stbi_uc *stbi__gif_load(stbi__context *s, int *x, int *y, int *comp, int req_comp)
-{
-   stbi_uc *u = 0;
-   stbi__gif* g = (stbi__gif*) stbi__malloc(sizeof(stbi__gif));
-   memset(g, 0, sizeof(*g));
-
-   u = stbi__gif_load_next(s, g, comp, req_comp);
-   if (u == (stbi_uc *) s) u = 0;  // end of animated gif marker
-   if (u) {
-      *x = g->w;
-      *y = g->h;
-      if (req_comp && req_comp != 4)
-         u = stbi__convert_format(u, 4, req_comp, g->w, g->h);
-   }
-   else if (g->out)
-      STBI_FREE(g->out);
-   STBI_FREE(g);
-   return u;
-}
-
-static int stbi__gif_info(stbi__context *s, int *x, int *y, int *comp)
-{
-   return stbi__gif_info_raw(s,x,y,comp);
-}
-#endif
-
-// *************************************************************************************************
-// Radiance RGBE HDR loader
-// originally by Nicolas Schulz
-#ifndef STBI_NO_HDR
-static int stbi__hdr_test_core(stbi__context *s)
-{
-   const char *signature = "#?RADIANCE\n";
-   int i;
-   for (i=0; signature[i]; ++i)
-      if (stbi__get8(s) != signature[i])
-         return 0;
-   return 1;
-}
-
-static int stbi__hdr_test(stbi__context* s)
-{
-   int r = stbi__hdr_test_core(s);
-   stbi__rewind(s);
-   return r;
-}
-
-#define STBI__HDR_BUFLEN  1024
-static char *stbi__hdr_gettoken(stbi__context *z, char *buffer)
-{
-   int len=0;
-   char c = '\0';
-
-   c = (char) stbi__get8(z);
-
-   while (!stbi__at_eof(z) && c != '\n') {
-      buffer[len++] = c;
-      if (len == STBI__HDR_BUFLEN-1) {
-         // flush to end of line
-         while (!stbi__at_eof(z) && stbi__get8(z) != '\n')
-            ;
-         break;
-      }
-      c = (char) stbi__get8(z);
-   }
-
-   buffer[len] = 0;
-   return buffer;
-}
-
-static void stbi__hdr_convert(float *output, stbi_uc *input, int req_comp)
-{
-   if ( input[3] != 0 ) {
-      float f1;
-      // Exponent
-      f1 = (float) ldexp(1.0f, input[3] - (int)(128 + 8));
-      if (req_comp <= 2)
-         output[0] = (input[0] + input[1] + input[2]) * f1 / 3;
-      else {
-         output[0] = input[0] * f1;
-         output[1] = input[1] * f1;
-         output[2] = input[2] * f1;
-      }
-      if (req_comp == 2) output[1] = 1;
-      if (req_comp == 4) output[3] = 1;
-   } else {
-      switch (req_comp) {
-         case 4: output[3] = 1; /* fallthrough */
-         case 3: output[0] = output[1] = output[2] = 0;
-                 break;
-         case 2: output[1] = 1; /* fallthrough */
-         case 1: output[0] = 0;
-                 break;
-      }
-   }
-}
-
-static float *stbi__hdr_load(stbi__context *s, int *x, int *y, int *comp, int req_comp)
-{
-   char buffer[STBI__HDR_BUFLEN];
-   char *token;
-   int valid = 0;
-   int width, height;
-   stbi_uc *scanline;
-   float *hdr_data;
-   int len;
-   unsigned char count, value;
-   int i, j, k, c1,c2, z;
-
-
-   // Check identifier
-   if (strcmp(stbi__hdr_gettoken(s,buffer), "#?RADIANCE") != 0)
-      return stbi__errpf("not HDR", "Corrupt HDR image");
-
-   // Parse header
-   for(;;) {
-      token = stbi__hdr_gettoken(s,buffer);
-      if (token[0] == 0) break;
-      if (strcmp(token, "FORMAT=32-bit_rle_rgbe") == 0) valid = 1;
-   }
-
-   if (!valid)    return stbi__errpf("unsupported format", "Unsupported HDR format");
-
-   // Parse width and height
-   // can't use sscanf() if we're not using stdio!
-   token = stbi__hdr_gettoken(s,buffer);
-   if (strncmp(token, "-Y ", 3))  return stbi__errpf("unsupported data layout", "Unsupported HDR format");
-   token += 3;
-   height = (int) strtol(token, &token, 10);
-   while (*token == ' ') ++token;
-   if (strncmp(token, "+X ", 3))  return stbi__errpf("unsupported data layout", "Unsupported HDR format");
-   token += 3;
-   width = (int) strtol(token, NULL, 10);
-
-   *x = width;
-   *y = height;
-
-   if (comp) *comp = 3;
-   if (req_comp == 0) req_comp = 3;
-
-   // Read data
-   hdr_data = (float *) stbi__malloc(height * width * req_comp * sizeof(float));
-
-   // Load image data
-   // image data is stored as some number of sca
-   if ( width < 8 || width >= 32768) {
-      // Read flat data
-      for (j=0; j < height; ++j) {
-         for (i=0; i < width; ++i) {
-            stbi_uc rgbe[4];
-           main_decode_loop:
-            stbi__getn(s, rgbe, 4);
-            stbi__hdr_convert(hdr_data + j * width * req_comp + i * req_comp, rgbe, req_comp);
-         }
-      }
-   } else {
-      // Read RLE-encoded data
-      scanline = NULL;
-
-      for (j = 0; j < height; ++j) {
-         c1 = stbi__get8(s);
-         c2 = stbi__get8(s);
-         len = stbi__get8(s);
-         if (c1 != 2 || c2 != 2 || (len & 0x80)) {
-            // not run-length encoded, so we have to actually use THIS data as a decoded
-            // pixel (note this can't be a valid pixel--one of RGB must be >= 128)
-            stbi_uc rgbe[4];
-            rgbe[0] = (stbi_uc) c1;
-            rgbe[1] = (stbi_uc) c2;
-            rgbe[2] = (stbi_uc) len;
-            rgbe[3] = (stbi_uc) stbi__get8(s);
-            stbi__hdr_convert(hdr_data, rgbe, req_comp);
-            i = 1;
-            j = 0;
-            STBI_FREE(scanline);
-            goto main_decode_loop; // yes, this makes no sense
-         }
-         len <<= 8;
-         len |= stbi__get8(s);
-         if (len != width) { STBI_FREE(hdr_data); STBI_FREE(scanline); return stbi__errpf("invalid decoded scanline length", "corrupt HDR"); }
-         if (scanline == NULL) scanline = (stbi_uc *) stbi__malloc(width * 4);
-
-         for (k = 0; k < 4; ++k) {
-            i = 0;
-            while (i < width) {
-               count = stbi__get8(s);
-               if (count > 128) {
-                  // Run
-                  value = stbi__get8(s);
-                  count -= 128;
-                  for (z = 0; z < count; ++z)
-                     scanline[i++ * 4 + k] = value;
-               } else {
-                  // Dump
-                  for (z = 0; z < count; ++z)
-                     scanline[i++ * 4 + k] = stbi__get8(s);
-               }
-            }
-         }
-         for (i=0; i < width; ++i)
-            stbi__hdr_convert(hdr_data+(j*width + i)*req_comp, scanline + i*4, req_comp);
-      }
-      STBI_FREE(scanline);
-   }
-
-   return hdr_data;
-}
-
-static int stbi__hdr_info(stbi__context *s, int *x, int *y, int *comp)
-{
-   char buffer[STBI__HDR_BUFLEN];
-   char *token;
-   int valid = 0;
-
-   if (stbi__hdr_test(s) == 0) {
-       stbi__rewind( s );
-       return 0;
-   }
-
-   for(;;) {
-      token = stbi__hdr_gettoken(s,buffer);
-      if (token[0] == 0) break;
-      if (strcmp(token, "FORMAT=32-bit_rle_rgbe") == 0) valid = 1;
-   }
-
-   if (!valid) {
-       stbi__rewind( s );
-       return 0;
-   }
-   token = stbi__hdr_gettoken(s,buffer);
-   if (strncmp(token, "-Y ", 3)) {
-       stbi__rewind( s );
-       return 0;
-   }
-   token += 3;
-   *y = (int) strtol(token, &token, 10);
-   while (*token == ' ') ++token;
-   if (strncmp(token, "+X ", 3)) {
-       stbi__rewind( s );
-       return 0;
-   }
-   token += 3;
-   *x = (int) strtol(token, NULL, 10);
-   *comp = 3;
-   return 1;
-}
-#endif // STBI_NO_HDR
-
-#ifndef STBI_NO_BMP
-static int stbi__bmp_info(stbi__context *s, int *x, int *y, int *comp)
-{
-   void *p;
-   stbi__bmp_data info;
-
-   info.all_a = 255;
-   p = stbi__bmp_parse_header(s, &info);
-   stbi__rewind( s );
-   if (p == NULL)
-      return 0;
-   *x = s->img_x;
-   *y = s->img_y;
-   *comp = info.ma ? 4 : 3;
-   return 1;
-}
-#endif
-
-#ifndef STBI_NO_PSD
-static int stbi__psd_info(stbi__context *s, int *x, int *y, int *comp)
-{
-   int channelCount;
-   if (stbi__get32be(s) != 0x38425053) {
-       stbi__rewind( s );
-       return 0;
-   }
-   if (stbi__get16be(s) != 1) {
-       stbi__rewind( s );
-       return 0;
-   }
-   stbi__skip(s, 6);
-   channelCount = stbi__get16be(s);
-   if (channelCount < 0 || channelCount > 16) {
-       stbi__rewind( s );
-       return 0;
-   }
-   *y = stbi__get32be(s);
-   *x = stbi__get32be(s);
-   if (stbi__get16be(s) != 8) {
-       stbi__rewind( s );
-       return 0;
-   }
-   if (stbi__get16be(s) != 3) {
-       stbi__rewind( s );
-       return 0;
-   }
-   *comp = 4;
-   return 1;
-}
-#endif
-
-#ifndef STBI_NO_PIC
-static int stbi__pic_info(stbi__context *s, int *x, int *y, int *comp)
-{
-   int act_comp=0,num_packets=0,chained;
-   stbi__pic_packet packets[10];
-
-   if (!stbi__pic_is4(s,"\x53\x80\xF6\x34")) {
-      stbi__rewind(s);
-      return 0;
-   }
-
-   stbi__skip(s, 88);
-
-   *x = stbi__get16be(s);
-   *y = stbi__get16be(s);
-   if (stbi__at_eof(s)) {
-      stbi__rewind( s);
-      return 0;
-   }
-   if ( (*x) != 0 && (1 << 28) / (*x) < (*y)) {
-      stbi__rewind( s );
-      return 0;
-   }
-
-   stbi__skip(s, 8);
-
-   do {
-      stbi__pic_packet *packet;
-
-      if (num_packets==sizeof(packets)/sizeof(packets[0]))
-         return 0;
-
-      packet = &packets[num_packets++];
-      chained = stbi__get8(s);
-      packet->size    = stbi__get8(s);
-      packet->type    = stbi__get8(s);
-      packet->channel = stbi__get8(s);
-      act_comp |= packet->channel;
-
-      if (stbi__at_eof(s)) {
-          stbi__rewind( s );
-          return 0;
-      }
-      if (packet->size != 8) {
-          stbi__rewind( s );
-          return 0;
-      }
-   } while (chained);
-
-   *comp = (act_comp & 0x10 ? 4 : 3);
-
-   return 1;
-}
-#endif
-
-// *************************************************************************************************
-// Portable Gray Map and Portable Pixel Map loader
-// by Ken Miller
-//
-// PGM: http://netpbm.sourceforge.net/doc/pgm.html
-// PPM: http://netpbm.sourceforge.net/doc/ppm.html
-//
-// Known limitations:
-//    Does not support comments in the header section
-//    Does not support ASCII image data (formats P2 and P3)
-//    Does not support 16-bit-per-channel
-
-#ifndef STBI_NO_PNM
-
-static int      stbi__pnm_test(stbi__context *s)
-{
-   char p, t;
-   p = (char) stbi__get8(s);
-   t = (char) stbi__get8(s);
-   if (p != 'P' || (t != '5' && t != '6')) {
-       stbi__rewind( s );
-       return 0;
-   }
-   return 1;
-}
-
-static stbi_uc *stbi__pnm_load(stbi__context *s, int *x, int *y, int *comp, int req_comp)
-{
-   stbi_uc *out;
-   if (!stbi__pnm_info(s, (int *)&s->img_x, (int *)&s->img_y, (int *)&s->img_n))
-      return 0;
-   *x = s->img_x;
-   *y = s->img_y;
-   *comp = s->img_n;
-
-   out = (stbi_uc *) stbi__malloc(s->img_n * s->img_x * s->img_y);
-   if (!out) return stbi__errpuc("outofmem", "Out of memory");
-   stbi__getn(s, out, s->img_n * s->img_x * s->img_y);
-
-   if (req_comp && req_comp != s->img_n) {
-      out = stbi__convert_format(out, s->img_n, req_comp, s->img_x, s->img_y);
-      if (out == NULL) return out; // stbi__convert_format frees input on failure
-   }
-   return out;
-}
-
-static int      stbi__pnm_isspace(char c)
-{
-   return c == ' ' || c == '\t' || c == '\n' || c == '\v' || c == '\f' || c == '\r';
-}
-
-static void     stbi__pnm_skip_whitespace(stbi__context *s, char *c)
-{
-   for (;;) {
-      while (!stbi__at_eof(s) && stbi__pnm_isspace(*c))
-         *c = (char) stbi__get8(s);
-
-      if (stbi__at_eof(s) || *c != '#')
-         break;
-
-      while (!stbi__at_eof(s) && *c != '\n' && *c != '\r' )
-         *c = (char) stbi__get8(s);
-   }
-}
-
-static int      stbi__pnm_isdigit(char c)
-{
-   return c >= '0' && c <= '9';
-}
-
-static int      stbi__pnm_getinteger(stbi__context *s, char *c)
-{
-   int value = 0;
-
-   while (!stbi__at_eof(s) && stbi__pnm_isdigit(*c)) {
-      value = value*10 + (*c - '0');
-      *c = (char) stbi__get8(s);
-   }
-
-   return value;
-}
-
-static int      stbi__pnm_info(stbi__context *s, int *x, int *y, int *comp)
-{
-   int maxv;
-   char c, p, t;
-
-   stbi__rewind( s );
-
-   // Get identifier
-   p = (char) stbi__get8(s);
-   t = (char) stbi__get8(s);
-   if (p != 'P' || (t != '5' && t != '6')) {
-       stbi__rewind( s );
-       return 0;
-   }
-
-   *comp = (t == '6') ? 3 : 1;  // '5' is 1-component .pgm; '6' is 3-component .ppm
-
-   c = (char) stbi__get8(s);
-   stbi__pnm_skip_whitespace(s, &c);
-
-   *x = stbi__pnm_getinteger(s, &c); // read width
-   stbi__pnm_skip_whitespace(s, &c);
-
-   *y = stbi__pnm_getinteger(s, &c); // read height
-   stbi__pnm_skip_whitespace(s, &c);
-
-   maxv = stbi__pnm_getinteger(s, &c);  // read max value
-
-   if (maxv > 255)
-      return stbi__err("max value > 255", "PPM image not 8-bit");
-   else
-      return 1;
-}
-#endif
-
-static int stbi__info_main(stbi__context *s, int *x, int *y, int *comp)
-{
-   #ifndef STBI_NO_JPEG
-   if (stbi__jpeg_info(s, x, y, comp)) return 1;
-   #endif
-
-   #ifndef STBI_NO_PNG
-   if (stbi__png_info(s, x, y, comp))  return 1;
-   #endif
-
-   #ifndef STBI_NO_GIF
-   if (stbi__gif_info(s, x, y, comp))  return 1;
-   #endif
-
-   #ifndef STBI_NO_BMP
-   if (stbi__bmp_info(s, x, y, comp))  return 1;
-   #endif
-
-   #ifndef STBI_NO_PSD
-   if (stbi__psd_info(s, x, y, comp))  return 1;
-   #endif
-
-   #ifndef STBI_NO_PIC
-   if (stbi__pic_info(s, x, y, comp))  return 1;
-   #endif
-
-   #ifndef STBI_NO_PNM
-   if (stbi__pnm_info(s, x, y, comp))  return 1;
-   #endif
-
-   #ifndef STBI_NO_HDR
-   if (stbi__hdr_info(s, x, y, comp))  return 1;
-   #endif
-
-   // test tga last because it's a crappy test!
-   #ifndef STBI_NO_TGA
-   if (stbi__tga_info(s, x, y, comp))
-       return 1;
-   #endif
-   return stbi__err("unknown image type", "Image not of any known type, or corrupt");
-}
-
-#ifndef STBI_NO_STDIO
-STBIDEF int stbi_info(char const *filename, int *x, int *y, int *comp)
-{
-    FILE *f = stbi__fopen(filename, "rb");
-    int result;
-    if (!f) return stbi__err("can't fopen", "Unable to open file");
-    result = stbi_info_from_file(f, x, y, comp);
-    fclose(f);
-    return result;
-}
-
-STBIDEF int stbi_info_from_file(FILE *f, int *x, int *y, int *comp)
-{
-   int r;
-   stbi__context s;
-   long pos = ftell(f);
-   stbi__start_file(&s, f);
-   r = stbi__info_main(&s,x,y,comp);
-   fseek(f,pos,SEEK_SET);
-   return r;
-}
-#endif // !STBI_NO_STDIO
-
-STBIDEF int stbi_info_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp)
-{
-   stbi__context s;
-   stbi__start_mem(&s,buffer,len);
-   return stbi__info_main(&s,x,y,comp);
-}
-
-STBIDEF int stbi_info_from_callbacks(stbi_io_callbacks const *c, void *user, int *x, int *y, int *comp)
-{
-   stbi__context s;
-   stbi__start_callbacks(&s, (stbi_io_callbacks *) c, user);
-   return stbi__info_main(&s,x,y,comp);
-}
-
-#endif // STB_IMAGE_IMPLEMENTATION

Source/ThirdParty/STB/stb_image.h

@@ -1,4 +1,4 @@
-/* stb_image - v2.12 - public domain image loader - http://nothings.org/stb_image.h
+/* stb_image - v2.15 - public domain image loader - http://nothings.org/stb_image.h
                                      no warranty implied; use at your own risk
 
    Do this:
@@ -21,7 +21,7 @@
           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)
+      PNG 1/2/4/8/16-bit-per-channel
 
       TGA (not sure what subset, if a subset)
       BMP non-1bpp, non-RLE
@@ -42,110 +42,15 @@
    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:
+LICENSE
+
+  See end of file for license information.
+
+RECENT REVISION HISTORY:
+
+      2.15  (2017-03-18) fix png-1,2,4; all Imagenet JPGs; no runtime SSE detection on GCC
+      2.14  (2017-03-03) remove deprecated STBI_JPEG_OLD; fixes for Imagenet JPGs
+      2.13  (2016-12-04) experimental 16-bit API, only for PNG so far; fixes
       2.12  (2016-04-02) fix typo in 2.11 PSD fix that caused crashes
       2.11  (2016-04-02) 16-bit PNGS; enable SSE2 in non-gcc x64
                          RGB-format JPEG; remove white matting in PSD;
@@ -157,21 +62,6 @@
       2.07  (2015-09-13) partial animated GIF support
                          limited 16-bit PSD support
                          minor bugs, code cleanup, and compiler warnings
-      2.06  (2015-04-19) fix bug where PSD returns wrong '*comp' value
-      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
 
    See end of file for full revision history.
 
@@ -186,33 +76,28 @@
     Tom Seddon (pic)                       Omar Cornut (1/2/4-bit PNG)
     Thatcher Ulrich (psd)                  Nicolas Guillemot (vertical flip)
     Ken Miller (pgm, ppm)                  Richard Mitton (16-bit PSD)
-    urraka@github (animated gif)           Junggon Kim (PNM comments)
+    github:urraka (animated gif)           Junggon Kim (PNM comments)
                                            Daniel Gibson (16-bit TGA)
-
+                                           socks-the-fox (16-bit PNG)
+                                           Jeremy Sawicki (handle all ImageNet JPGs)
  Optimizations & bugfixes
     Fabian "ryg" Giesen
     Arseny Kapoulkine
 
  Bug & warning fixes
     Marc LeBlanc            David Woo          Guillaume George   Martins Mozeiko
-    Christpher Lloyd        Martin Golini      Jerry Jansson      Joseph Thomson
-    Dave Moore              Roy Eltham         Hayaki Saito       Phil Jordan
-    Won Chun                Luke Graham        Johan Duparc       Nathan Reed
-    the Horde3D community   Thomas Ruf         Ronny Chevalier    Nick Verigakis
-    Janez Zemva             John Bartholomew   Michal Cichon      svdijk@github
-    Jonathan Blow           Ken Hamada         Tero Hanninen      Baldur Karlsson
-    Laurent Gomila          Cort Stratton      Sergio Gonzalez    romigrou@github
-    Aruelien Pocheville     Thibault Reuille   Cass Everitt       Matthew Gregan
-    Ryamond Barbiero        Paul Du Bois       Engin Manap        snagar@github
-    Michaelangel007@github  Oriol Ferrer Mesia socks-the-fox
-    Blazej Dariusz Roszkowski
-
-
-LICENSE
-
-This software is dual-licensed to the public domain and under the following
-license: you are granted a perpetual, irrevocable license to copy, modify,
-publish, and distribute this file as you see fit.
+    Christpher Lloyd        Jerry Jansson      Joseph Thomson     Phil Jordan
+    Dave Moore              Roy Eltham         Hayaki Saito       Nathan Reed
+    Won Chun                Luke Graham        Johan Duparc       Nick Verigakis
+    the Horde3D community   Thomas Ruf         Ronny Chevalier    Baldur Karlsson
+    Janez Zemva             John Bartholomew   Michal Cichon      github:rlyeh
+    Jonathan Blow           Ken Hamada         Tero Hanninen      github:romigrou
+    Laurent Gomila          Cort Stratton      Sergio Gonzalez    github:svdijk
+    Aruelien Pocheville     Thibault Reuille   Cass Everitt       github:snagar
+    Ryamond Barbiero        Paul Du Bois       Engin Manap        github:Zelex
+    Michaelangel007@github  Philipp Wiesemann  Dale Weiler        github:grim210
+    Oriol Ferrer Mesia      Josh Tobin         Matthew Gregan     github:sammyhw
+    Blazej Dariusz Roszkowski                  Gregory Mullen     github:phprus
 
 */
 
@@ -238,10 +123,10 @@ publish, and distribute this file as you see fit.
 //    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
+//    int *x                 -- outputs image width in pixels
+//    int *y                 -- outputs image height in pixels
+//    int *channels_in_file  -- outputs # of image components in image file
+//    int desired_channels   -- 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
@@ -287,13 +172,13 @@ publish, and distribute this file as you see fit.
 // 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.
+// 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")
+//    - Small source code footprint ("easy to maintain")
 //    - No dependencies ("ease of use")
 //
 // ===========================================================================
@@ -325,13 +210,6 @@ publish, and distribute this file as you see fit.
 // (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.
@@ -387,6 +265,41 @@ publish, and distribute this file as you see fit.
 // says there's premultiplied data (currently only happens in iPhone images,
 // and only if iPhone convert-to-rgb processing is on).
 //
+// ===========================================================================
+//
+// ADDITIONAL CONFIGURATION
+//
+//  - 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)
+//
+//   - If you use STBI_NO_PNG (or _ONLY_ without PNG), and you still
+//     want the zlib decoder to be available, #define STBI_SUPPORT_ZLIB
+//
 
 
 #ifndef STBI_NO_STDIO
@@ -406,6 +319,7 @@ enum
 };
 
 typedef unsigned char stbi_uc;
+typedef unsigned short stbi_us;
 
 #ifdef __cplusplus
 extern "C" {
@@ -433,22 +347,42 @@ typedef struct
    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);
+////////////////////////////////////
+//
+// 8-bits-per-channel interface
+//
+
+STBIDEF stbi_uc *stbi_load               (char              const *filename,           int *x, int *y, int *channels_in_file, int desired_channels);
+STBIDEF stbi_uc *stbi_load_from_memory   (stbi_uc           const *buffer, int len   , int *x, int *y, int *channels_in_file, int desired_channels);
+STBIDEF stbi_uc *stbi_load_from_callbacks(stbi_io_callbacks const *clbk  , void *user, int *x, int *y, int *channels_in_file, int desired_channels);
 
 #ifndef STBI_NO_STDIO
-STBIDEF stbi_uc *stbi_load_from_file  (FILE *f,                  int *x, int *y, int *comp, int req_comp);
+STBIDEF stbi_uc *stbi_load_from_file   (FILE *f, int *x, int *y, int *channels_in_file, int desired_channels);
 // for stbi_load_from_file, file pointer is left pointing immediately after image
 #endif
 
+////////////////////////////////////
+//
+// 16-bits-per-channel interface
+//
+
+STBIDEF stbi_us *stbi_load_16(char const *filename, int *x, int *y, int *channels_in_file, int desired_channels);
+#ifndef STBI_NO_STDIO
+STBIDEF stbi_us *stbi_load_from_file_16(FILE *f, int *x, int *y, int *channels_in_file, int desired_channels);
+#endif
+// @TODO the other variants
+
+////////////////////////////////////
+//
+// float-per-channel interface
+//
 #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);
+   STBIDEF float *stbi_loadf                 (char const *filename,           int *x, int *y, int *channels_in_file, int desired_channels);
+   STBIDEF float *stbi_loadf_from_memory     (stbi_uc const *buffer, int len, int *x, int *y, int *channels_in_file, int desired_channels);
+   STBIDEF float *stbi_loadf_from_callbacks  (stbi_io_callbacks const *clbk, void *user, int *x, int *y,  int *channels_in_file, int desired_channels);
 
    #ifndef STBI_NO_STDIO
-   STBIDEF float *stbi_loadf_from_file  (FILE *f,                int *x, int *y, int *comp, int req_comp);
+   STBIDEF float *stbi_loadf_from_file  (FILE *f, int *x, int *y, int *channels_in_file, int desired_channels);
    #endif
 #endif
 
@@ -521,165 +455,6723 @@ STBIDEF int   stbi_zlib_decode_noheader_buffer(char *obuffer, int olen, const ch
 //
 ////   end header file   /////////////////////////////////////////////////////
 #endif // STBI_INCLUDE_STB_IMAGE_H
-/*
-   revision history:
-      2.12  (2016-04-02) fix typo in 2.11 PSD fix that caused crashes
-      2.11  (2016-04-02) allocate large structures on the stack
-                         remove white matting for transparent PSD
-                         fix reported channel count for PNG & BMP
-                         re-enable SSE2 in non-gcc 64-bit
-                         support RGB-formatted JPEG
-                         read 16-bit PNGs (only as 8-bit)
-      2.10  (2016-01-22) avoid warning introduced in 2.09 by STBI_REALLOC_SIZED
-      2.09  (2016-01-16) allow comments in PNM files
-                         16-bit-per-pixel TGA (not bit-per-component)
-                         info() for TGA could break due to .hdr handling
-                         info() for BMP to shares code instead of sloppy parse
-                         can use STBI_REALLOC_SIZED if allocator doesn't support realloc
-                         code cleanup
-      2.08  (2015-09-13) fix to 2.07 cleanup, reading RGB PSD as RGBA
-      2.07  (2015-09-13) fix compiler warnings
-                         partial animated GIF support
-                         limited 16-bpc PSD support
-                         #ifdef unused functions
-                         bug with < 92 byte PIC,PNM,HDR,TGA
-      2.06  (2015-04-19) fix bug where PSD returns wrong '*comp' value
-      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) extra corruption checking (mmozeiko)
-                         stbi_set_flip_vertically_on_load (nguillemot)
-                         fix NEON support; fix mingw support
-      2.02  (2015-01-19) fix incorrect assert, fix warning
-      2.01  (2015-01-17) fix various warnings; suppress SIMD on gcc 32-bit without -msse2
-      2.00b (2014-12-25) fix STBI_MALLOC in progressive JPEG
-      2.00  (2014-12-25) optimize JPG, including x86 SSE2 & NEON SIMD (ryg)
-                         progressive JPEG (stb)
-                         PGM/PPM support (Ken Miller)
-                         STBI_MALLOC,STBI_REALLOC,STBI_FREE
-                         GIF bugfix -- seemingly never worked
-                         STBI_NO_*, STBI_ONLY_*
-      1.48  (2014-12-14) fix incorrectly-named assert()
-      1.47  (2014-12-14) 1/2/4-bit PNG support, both direct and paletted (Omar Cornut & stb)
-                         optimize PNG (ryg)
-                         fix bug in interlaced PNG with user-specified channel count (stb)
-      1.46  (2014-08-26)
-              fix broken tRNS chunk (colorkey-style transparency) in non-paletted PNG
-      1.45  (2014-08-16)
-              fix MSVC-ARM internal compiler error by wrapping malloc
-      1.44  (2014-08-07)
-              various warning fixes from Ronny Chevalier
-      1.43  (2014-07-15)
-              fix MSVC-only compiler problem in code changed in 1.42
-      1.42  (2014-07-09)
-              don't define _CRT_SECURE_NO_WARNINGS (affects user code)
-              fixes to stbi__cleanup_jpeg path
-              added STBI_ASSERT to avoid requiring assert.h
-      1.41  (2014-06-25)
-              fix search&replace from 1.36 that messed up comments/error messages
-      1.40  (2014-06-22)
-              fix gcc struct-initialization warning
-      1.39  (2014-06-15)
-              fix to TGA optimization when req_comp != number of components in TGA;
-              fix to GIF loading because BMP wasn't rewinding (whoops, no GIFs in my test suite)
-              add support for BMP version 5 (more ignored fields)
-      1.38  (2014-06-06)
-              suppress MSVC warnings on integer casts truncating values
-              fix accidental rename of 'skip' field of I/O
-      1.37  (2014-06-04)
-              remove duplicate typedef
-      1.36  (2014-06-03)
-              convert to header file single-file library
-              if de-iphone isn't set, load iphone images color-swapped instead of returning NULL
-      1.35  (2014-05-27)
-              various warnings
-              fix broken STBI_SIMD path
-              fix bug where stbi_load_from_file no longer left file pointer in correct place
-              fix broken non-easy path for 32-bit BMP (possibly never used)
-              TGA optimization by Arseny Kapoulkine
-      1.34  (unknown)
-              use STBI_NOTUSED in stbi__resample_row_generic(), fix one more leak in tga failure case
-      1.33  (2011-07-14)
-              make stbi_is_hdr work in STBI_NO_HDR (as specified), minor compiler-friendly improvements
-      1.32  (2011-07-13)
-              support for "info" function for all supported filetypes (SpartanJ)
-      1.31  (2011-06-20)
-              a few more leak fixes, bug in PNG handling (SpartanJ)
-      1.30  (2011-06-11)
-              added ability to load files via callbacks to accomidate custom input streams (Ben Wenger)
-              removed deprecated format-specific test/load functions
-              removed support for installable file formats (stbi_loader) -- would have been broken for IO callbacks anyway
-              error cases in bmp and tga give messages and don't leak (Raymond Barbiero, grisha)
-              fix inefficiency in decoding 32-bit BMP (David Woo)
-      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-stbi_uc to fix warnings
-      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 on platforms with lock-heavy fgetc()
-              minor perf improvements for jpeg
-              deprecated type-specific functions so we'll get feedback if they're needed
-              attempt to fix trans_data warning (Won Chun)
-      1.23    fixed bug in iPhone support
-      1.22  (2010-07-10)
-              removed image *writing* support
-              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 (i.stbi__err. Janez (U+017D)emva)
-      1.21    fix use of 'stbi_uc' in header (reported by jon blow)
-      1.20    added support for Softimage PIC, by Tom Seddon
-      1.19    bug in interlaced PNG corruption check (found by ryg)
-      1.18  (2008-08-02)
-              fix a threading bug (local mutable static)
-      1.17    support interlaced PNG
-      1.16    major bugfix - stbi__convert_format converted one too many pixels
-      1.15    initialize some fields for thread safety
-      1.14    fix threadsafe conversion bug
-              header-file-only version (#define STBI_HEADER_FILE_ONLY before including)
-      1.13    threadsafe
-      1.12    const qualifiers in the API
-      1.11    Support installable IDCT, colorspace conversion routines
-      1.10    Fixes for 64-bit (don't use "unsigned long")
-              optimized upsampling by Fabian "ryg" Giesen
-      1.09    Fix format-conversion for PSD code (bad global variables!)
-      1.08    Thatcher Ulrich's PSD code integrated by Nicolas Schulz
-      1.07    attempt to fix C++ warning/errors again
-      1.06    attempt to fix C++ warning/errors again
-      1.05    fix TGA loading to return correct *comp and use good luminance calc
-      1.04    default float alpha is 1, not 255; use 'void *' for stbi_image_free
-      1.03    bugfixes to STBI_NO_STDIO, STBI_NO_HDR
-      1.02    support for (subset of) HDR files, float interface for preferred access to them
-      1.01    fix bug: possible bug in handling right-side up bmps... not sure
-              fix bug: the stbi__bmp_load() and stbi__tga_load() functions didn't work at all
-      1.00    interface to zlib that skips zlib header
-      0.99    correct handling of alpha in palette
-      0.98    TGA loader by lonesock; dynamically add loaders (untested)
-      0.97    jpeg errors on too large a file; also catch another malloc failure
-      0.96    fix detection of invalid v value - particleman@mollyrocket forum
-      0.95    during header scan, seek to markers in case of padding
-      0.94    STBI_NO_STDIO to disable stdio usage; rename all #defines the same
-      0.93    handle jpegtran output; verbose errors
-      0.92    read 4,8,16,24,32-bit BMP files of several formats
-      0.91    output 24-bit Windows 3.0 BMP files
-      0.90    fix a few more warnings; bump version number to approach 1.0
-      0.61    bugfixes due to Marc LeBlanc, Christopher Lloyd
-      0.60    fix compiling as c++
-      0.59    fix warnings: merge Dave Moore's -Wall fixes
-      0.58    fix bug: zlib uncompressed mode len/nlen was wrong endian
-      0.57    fix bug: jpg last huffman symbol before marker was >9 bits but less than 16 available
-      0.56    fix bug: zlib uncompressed mode len vs. nlen
-      0.55    fix bug: restart_interval not initialized to 0
-      0.54    allow NULL for 'int *comp'
-      0.53    fix bug in png 3->4; speedup png decoding
-      0.52    png handles req_comp=3,4 directly; minor cleanup; jpeg comments
-      0.51    obey req_comp requests, 1-component jpegs return as 1-component,
-              on 'test' only check type, not whether we support this variant
-      0.50  (2006-11-19)
-              first released version
+
+#ifdef STB_IMAGE_IMPLEMENTATION
+
+#if defined(STBI_ONLY_JPEG) || defined(STBI_ONLY_PNG) || defined(STBI_ONLY_BMP) \
+  || defined(STBI_ONLY_TGA) || defined(STBI_ONLY_GIF) || defined(STBI_ONLY_PSD) \
+  || defined(STBI_ONLY_HDR) || defined(STBI_ONLY_PIC) || defined(STBI_ONLY_PNM) \
+  || defined(STBI_ONLY_ZLIB)
+   #ifndef STBI_ONLY_JPEG
+   #define STBI_NO_JPEG
+   #endif
+   #ifndef STBI_ONLY_PNG
+   #define STBI_NO_PNG
+   #endif
+   #ifndef STBI_ONLY_BMP
+   #define STBI_NO_BMP
+   #endif
+   #ifndef STBI_ONLY_PSD
+   #define STBI_NO_PSD
+   #endif
+   #ifndef STBI_ONLY_TGA
+   #define STBI_NO_TGA
+   #endif
+   #ifndef STBI_ONLY_GIF
+   #define STBI_NO_GIF
+   #endif
+   #ifndef STBI_ONLY_HDR
+   #define STBI_NO_HDR
+   #endif
+   #ifndef STBI_ONLY_PIC
+   #define STBI_NO_PIC
+   #endif
+   #ifndef STBI_ONLY_PNM
+   #define STBI_NO_PNM
+   #endif
+#endif
+
+#if defined(STBI_NO_PNG) && !defined(STBI_SUPPORT_ZLIB) && !defined(STBI_NO_ZLIB)
+#define STBI_NO_ZLIB
+#endif
+
+
+#include <stdarg.h>
+#include <stddef.h> // ptrdiff_t on osx
+#include <stdlib.h>
+#include <string.h>
+#include <limits.h>
+
+#if !defined(STBI_NO_LINEAR) || !defined(STBI_NO_HDR)
+#include <math.h>  // ldexp
+#endif
+
+#ifndef STBI_NO_STDIO
+#include <stdio.h>
+#endif
+
+#ifndef STBI_ASSERT
+#include <assert.h>
+#define STBI_ASSERT(x) assert(x)
+#endif
+
+
+#ifndef _MSC_VER
+   #ifdef __cplusplus
+   #define stbi_inline inline
+   #else
+   #define stbi_inline
+   #endif
+#else
+   #define stbi_inline __forceinline
+#endif
+
+
+#ifdef _MSC_VER
+typedef unsigned short stbi__uint16;
+typedef   signed short stbi__int16;
+typedef unsigned int   stbi__uint32;
+typedef   signed int   stbi__int32;
+#else
+#include <stdint.h>
+typedef uint16_t stbi__uint16;
+typedef int16_t  stbi__int16;
+typedef uint32_t stbi__uint32;
+typedef int32_t  stbi__int32;
+#endif
+
+// should produce compiler error if size is wrong
+typedef unsigned char validate_uint32[sizeof(stbi__uint32)==4 ? 1 : -1];
+
+#ifdef _MSC_VER
+#define STBI_NOTUSED(v)  (void)(v)
+#else
+#define STBI_NOTUSED(v)  (void)sizeof(v)
+#endif
+
+#ifdef _MSC_VER
+#define STBI_HAS_LROTL
+#endif
+
+#ifdef STBI_HAS_LROTL
+   #define stbi_lrot(x,y)  _lrotl(x,y)
+#else
+   #define stbi_lrot(x,y)  (((x) << (y)) | ((x) >> (32 - (y))))
+#endif
+
+#if defined(STBI_MALLOC) && defined(STBI_FREE) && (defined(STBI_REALLOC) || defined(STBI_REALLOC_SIZED))
+// ok
+#elif !defined(STBI_MALLOC) && !defined(STBI_FREE) && !defined(STBI_REALLOC) && !defined(STBI_REALLOC_SIZED)
+// ok
+#else
+#error "Must define all or none of STBI_MALLOC, STBI_FREE, and STBI_REALLOC (or STBI_REALLOC_SIZED)."
+#endif
+
+#ifndef STBI_MALLOC
+#define STBI_MALLOC(sz)           malloc(sz)
+#define STBI_REALLOC(p,newsz)     realloc(p,newsz)
+#define STBI_FREE(p)              free(p)
+#endif
+
+#ifndef STBI_REALLOC_SIZED
+#define STBI_REALLOC_SIZED(p,oldsz,newsz) STBI_REALLOC(p,newsz)
+#endif
+
+// x86/x64 detection
+#if defined(__x86_64__) || defined(_M_X64)
+#define STBI__X64_TARGET
+#elif defined(__i386) || defined(_M_IX86)
+#define STBI__X86_TARGET
+#endif
+
+#if defined(__GNUC__) && defined(STBI__X86_TARGET) && !defined(__SSE2__) && !defined(STBI_NO_SIMD)
+// gcc doesn't support sse2 intrinsics unless you compile with -msse2,
+// which in turn means it gets to use SSE2 everywhere. This is unfortunate,
+// but previous attempts to provide the SSE2 functions with runtime
+// detection caused numerous issues. The way architecture extensions are
+// exposed in GCC/Clang is, sadly, not really suited for one-file libs.
+// New behavior: if compiled with -msse2, we use SSE2 without any
+// detection; if not, we don't use it at all.
+#define STBI_NO_SIMD
+#endif
+
+#if defined(__MINGW32__) && defined(STBI__X86_TARGET) && !defined(STBI_MINGW_ENABLE_SSE2) && !defined(STBI_NO_SIMD)
+// Note that __MINGW32__ doesn't actually mean 32-bit, so we have to avoid STBI__X64_TARGET
+//
+// 32-bit MinGW wants ESP to be 16-byte aligned, but this is not in the
+// Windows ABI and VC++ as well as Windows DLLs don't maintain that invariant.
+// As a result, enabling SSE2 on 32-bit MinGW is dangerous when not
+// simultaneously enabling "-mstackrealign".
+//
+// See https://github.com/nothings/stb/issues/81 for more information.
+//
+// So default to no SSE2 on 32-bit MinGW. If you've read this far and added
+// -mstackrealign to your build settings, feel free to #define STBI_MINGW_ENABLE_SSE2.
+#define STBI_NO_SIMD
+#endif
+
+#if !defined(STBI_NO_SIMD) && (defined(STBI__X86_TARGET) || defined(STBI__X64_TARGET))
+#define STBI_SSE2
+#include <emmintrin.h>
+
+#ifdef _MSC_VER
+
+#if _MSC_VER >= 1400  // not VC6
+#include <intrin.h> // __cpuid
+static int stbi__cpuid3(void)
+{
+   int info[4];
+   __cpuid(info,1);
+   return info[3];
+}
+#else
+static int stbi__cpuid3(void)
+{
+   int res;
+   __asm {
+      mov  eax,1
+      cpuid
+      mov  res,edx
+   }
+   return res;
+}
+#endif
+
+#define STBI_SIMD_ALIGN(type, name) __declspec(align(16)) type name
+
+static int stbi__sse2_available()
+{
+   int info3 = stbi__cpuid3();
+   return ((info3 >> 26) & 1) != 0;
+}
+#else // assume GCC-style if not VC++
+#define STBI_SIMD_ALIGN(type, name) type name __attribute__((aligned(16)))
+
+static int stbi__sse2_available()
+{
+   // If we're even attempting to compile this on GCC/Clang, that means
+   // -msse2 is on, which means the compiler is allowed to use SSE2
+   // instructions at will, and so are we.
+   return 1;
+}
+#endif
+#endif
+
+// ARM NEON
+#if defined(STBI_NO_SIMD) && defined(STBI_NEON)
+#undef STBI_NEON
+#endif
+
+#ifdef STBI_NEON
+#include <arm_neon.h>
+// assume GCC or Clang on ARM targets
+#define STBI_SIMD_ALIGN(type, name) type name __attribute__((aligned(16)))
+#endif
+
+#ifndef STBI_SIMD_ALIGN
+#define STBI_SIMD_ALIGN(type, name) type name
+#endif
+
+///////////////////////////////////////////////
+//
+//  stbi__context struct and start_xxx functions
+
+// stbi__context structure is our basic context used by all images, so it
+// contains all the IO context, plus some basic image information
+typedef struct
+{
+   stbi__uint32 img_x, img_y;
+   int img_n, img_out_n;
+
+   stbi_io_callbacks io;
+   void *io_user_data;
+
+   int read_from_callbacks;
+   int buflen;
+   stbi_uc buffer_start[128];
+
+   stbi_uc *img_buffer, *img_buffer_end;
+   stbi_uc *img_buffer_original, *img_buffer_original_end;
+} stbi__context;
+
+
+static void stbi__refill_buffer(stbi__context *s);
+
+// initialize a memory-decode context
+static void stbi__start_mem(stbi__context *s, stbi_uc const *buffer, int len)
+{
+   s->io.read = NULL;
+   s->read_from_callbacks = 0;
+   s->img_buffer = s->img_buffer_original = (stbi_uc *) buffer;
+   s->img_buffer_end = s->img_buffer_original_end = (stbi_uc *) buffer+len;
+}
+
+// initialize a callback-based context
+static void stbi__start_callbacks(stbi__context *s, stbi_io_callbacks *c, void *user)
+{
+   s->io = *c;
+   s->io_user_data = user;
+   s->buflen = sizeof(s->buffer_start);
+   s->read_from_callbacks = 1;
+   s->img_buffer_original = s->buffer_start;
+   stbi__refill_buffer(s);
+   s->img_buffer_original_end = s->img_buffer_end;
+}
+
+#ifndef STBI_NO_STDIO
+
+static int stbi__stdio_read(void *user, char *data, int size)
+{
+   return (int) fread(data,1,size,(FILE*) user);
+}
+
+static void stbi__stdio_skip(void *user, int n)
+{
+   fseek((FILE*) user, n, SEEK_CUR);
+}
+
+static int stbi__stdio_eof(void *user)
+{
+   return feof((FILE*) user);
+}
+
+static stbi_io_callbacks stbi__stdio_callbacks =
+{
+   stbi__stdio_read,
+   stbi__stdio_skip,
+   stbi__stdio_eof,
+};
+
+static void stbi__start_file(stbi__context *s, FILE *f)
+{
+   stbi__start_callbacks(s, &stbi__stdio_callbacks, (void *) f);
+}
+
+//static void stop_file(stbi__context *s) { }
+
+#endif // !STBI_NO_STDIO
+
+static void stbi__rewind(stbi__context *s)
+{
+   // conceptually rewind SHOULD rewind to the beginning of the stream,
+   // but we just rewind to the beginning of the initial buffer, because
+   // we only use it after doing 'test', which only ever looks at at most 92 bytes
+   s->img_buffer = s->img_buffer_original;
+   s->img_buffer_end = s->img_buffer_original_end;
+}
+
+enum
+{
+   STBI_ORDER_RGB,
+   STBI_ORDER_BGR
+};
+
+typedef struct
+{
+   int bits_per_channel;
+   int num_channels;
+   int channel_order;
+} stbi__result_info;
+
+#ifndef STBI_NO_JPEG
+static int      stbi__jpeg_test(stbi__context *s);
+static void    *stbi__jpeg_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
+static int      stbi__jpeg_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+#ifndef STBI_NO_PNG
+static int      stbi__png_test(stbi__context *s);
+static void    *stbi__png_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
+static int      stbi__png_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+#ifndef STBI_NO_BMP
+static int      stbi__bmp_test(stbi__context *s);
+static void    *stbi__bmp_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
+static int      stbi__bmp_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+#ifndef STBI_NO_TGA
+static int      stbi__tga_test(stbi__context *s);
+static void    *stbi__tga_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
+static int      stbi__tga_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+#ifndef STBI_NO_PSD
+static int      stbi__psd_test(stbi__context *s);
+static void    *stbi__psd_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri, int bpc);
+static int      stbi__psd_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+#ifndef STBI_NO_HDR
+static int      stbi__hdr_test(stbi__context *s);
+static float   *stbi__hdr_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
+static int      stbi__hdr_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+#ifndef STBI_NO_PIC
+static int      stbi__pic_test(stbi__context *s);
+static void    *stbi__pic_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
+static int      stbi__pic_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+#ifndef STBI_NO_GIF
+static int      stbi__gif_test(stbi__context *s);
+static void    *stbi__gif_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
+static int      stbi__gif_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+#ifndef STBI_NO_PNM
+static int      stbi__pnm_test(stbi__context *s);
+static void    *stbi__pnm_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri);
+static int      stbi__pnm_info(stbi__context *s, int *x, int *y, int *comp);
+#endif
+
+// this is not threadsafe
+static const char *stbi__g_failure_reason;
+
+STBIDEF const char *stbi_failure_reason(void)
+{
+   return stbi__g_failure_reason;
+}
+
+static int stbi__err(const char *str)
+{
+   stbi__g_failure_reason = str;
+   return 0;
+}
+
+static void *stbi__malloc(size_t size)
+{
+    return STBI_MALLOC(size);
+}
+
+// stb_image uses ints pervasively, including for offset calculations.
+// therefore the largest decoded image size we can support with the
+// current code, even on 64-bit targets, is INT_MAX. this is not a
+// significant limitation for the intended use case.
+//
+// we do, however, need to make sure our size calculations don't
+// overflow. hence a few helper functions for size calculations that
+// multiply integers together, making sure that they're non-negative
+// and no overflow occurs.
+
+// return 1 if the sum is valid, 0 on overflow.
+// negative terms are considered invalid.
+static int stbi__addsizes_valid(int a, int b)
+{
+   if (b < 0) return 0;
+   // now 0 <= b <= INT_MAX, hence also
+   // 0 <= INT_MAX - b <= INTMAX.
+   // And "a + b <= INT_MAX" (which might overflow) is the
+   // same as a <= INT_MAX - b (no overflow)
+   return a <= INT_MAX - b;
+}
+
+// returns 1 if the product is valid, 0 on overflow.
+// negative factors are considered invalid.
+static int stbi__mul2sizes_valid(int a, int b)
+{
+   if (a < 0 || b < 0) return 0;
+   if (b == 0) return 1; // mul-by-0 is always safe
+   // portable way to check for no overflows in a*b
+   return a <= INT_MAX/b;
+}
+
+// returns 1 if "a*b + add" has no negative terms/factors and doesn't overflow
+static int stbi__mad2sizes_valid(int a, int b, int add)
+{
+   return stbi__mul2sizes_valid(a, b) && stbi__addsizes_valid(a*b, add);
+}
+
+// returns 1 if "a*b*c + add" has no negative terms/factors and doesn't overflow
+static int stbi__mad3sizes_valid(int a, int b, int c, int add)
+{
+   return stbi__mul2sizes_valid(a, b) && stbi__mul2sizes_valid(a*b, c) &&
+      stbi__addsizes_valid(a*b*c, add);
+}
+
+// returns 1 if "a*b*c*d + add" has no negative terms/factors and doesn't overflow
+static int stbi__mad4sizes_valid(int a, int b, int c, int d, int add)
+{
+   return stbi__mul2sizes_valid(a, b) && stbi__mul2sizes_valid(a*b, c) &&
+      stbi__mul2sizes_valid(a*b*c, d) && stbi__addsizes_valid(a*b*c*d, add);
+}
+
+// mallocs with size overflow checking
+static void *stbi__malloc_mad2(int a, int b, int add)
+{
+   if (!stbi__mad2sizes_valid(a, b, add)) return NULL;
+   return stbi__malloc(a*b + add);
+}
+
+static void *stbi__malloc_mad3(int a, int b, int c, int add)
+{
+   if (!stbi__mad3sizes_valid(a, b, c, add)) return NULL;
+   return stbi__malloc(a*b*c + add);
+}
+
+static void *stbi__malloc_mad4(int a, int b, int c, int d, int add)
+{
+   if (!stbi__mad4sizes_valid(a, b, c, d, add)) return NULL;
+   return stbi__malloc(a*b*c*d + add);
+}
+
+// stbi__err - error
+// stbi__errpf - error returning pointer to float
+// stbi__errpuc - error returning pointer to unsigned char
+
+#ifdef STBI_NO_FAILURE_STRINGS
+   #define stbi__err(x,y)  0
+#elif defined(STBI_FAILURE_USERMSG)
+   #define stbi__err(x,y)  stbi__err(y)
+#else
+   #define stbi__err(x,y)  stbi__err(x)
+#endif
+
+#define stbi__errpf(x,y)   ((float *)(size_t) (stbi__err(x,y)?NULL:NULL))
+#define stbi__errpuc(x,y)  ((unsigned char *)(size_t) (stbi__err(x,y)?NULL:NULL))
+
+STBIDEF void stbi_image_free(void *retval_from_stbi_load)
+{
+   STBI_FREE(retval_from_stbi_load);
+}
+
+#ifndef STBI_NO_LINEAR
+static float   *stbi__ldr_to_hdr(stbi_uc *data, int x, int y, int comp);
+#endif
+
+#ifndef STBI_NO_HDR
+static stbi_uc *stbi__hdr_to_ldr(float   *data, int x, int y, int comp);
+#endif
+
+static int stbi__vertically_flip_on_load = 0;
+
+STBIDEF void stbi_set_flip_vertically_on_load(int flag_true_if_should_flip)
+{
+    stbi__vertically_flip_on_load = flag_true_if_should_flip;
+}
+
+static void *stbi__load_main(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri, int bpc)
+{
+   memset(ri, 0, sizeof(*ri)); // make sure it's initialized if we add new fields
+   ri->bits_per_channel = 8; // default is 8 so most paths don't have to be changed
+   ri->channel_order = STBI_ORDER_RGB; // all current input & output are this, but this is here so we can add BGR order
+   ri->num_channels = 0;
+
+   #ifndef STBI_NO_JPEG
+   if (stbi__jpeg_test(s)) return stbi__jpeg_load(s,x,y,comp,req_comp, ri);
+   #endif
+   #ifndef STBI_NO_PNG
+   if (stbi__png_test(s))  return stbi__png_load(s,x,y,comp,req_comp, ri);
+   #endif
+   #ifndef STBI_NO_BMP
+   if (stbi__bmp_test(s))  return stbi__bmp_load(s,x,y,comp,req_comp, ri);
+   #endif
+   #ifndef STBI_NO_GIF
+   if (stbi__gif_test(s))  return stbi__gif_load(s,x,y,comp,req_comp, ri);
+   #endif
+   #ifndef STBI_NO_PSD
+   if (stbi__psd_test(s))  return stbi__psd_load(s,x,y,comp,req_comp, ri, bpc);
+   #endif
+   #ifndef STBI_NO_PIC
+   if (stbi__pic_test(s))  return stbi__pic_load(s,x,y,comp,req_comp, ri);
+   #endif
+   #ifndef STBI_NO_PNM
+   if (stbi__pnm_test(s))  return stbi__pnm_load(s,x,y,comp,req_comp, ri);
+   #endif
+
+   #ifndef STBI_NO_HDR
+   if (stbi__hdr_test(s)) {
+      float *hdr = stbi__hdr_load(s, x,y,comp,req_comp, ri);
+      return stbi__hdr_to_ldr(hdr, *x, *y, req_comp ? req_comp : *comp);
+   }
+   #endif
+
+   #ifndef STBI_NO_TGA
+   // test tga last because it's a crappy test!
+   if (stbi__tga_test(s))
+      return stbi__tga_load(s,x,y,comp,req_comp, ri);
+   #endif
+
+   return stbi__errpuc("unknown image type", "Image not of any known type, or corrupt");
+}
+
+static stbi_uc *stbi__convert_16_to_8(stbi__uint16 *orig, int w, int h, int channels)
+{
+   int i;
+   int img_len = w * h * channels;
+   stbi_uc *reduced;
+
+   reduced = (stbi_uc *) stbi__malloc(img_len);
+   if (reduced == NULL) return stbi__errpuc("outofmem", "Out of memory");
+
+   for (i = 0; i < img_len; ++i)
+      reduced[i] = (stbi_uc)((orig[i] >> 8) & 0xFF); // top half of each byte is sufficient approx of 16->8 bit scaling
+
+   STBI_FREE(orig);
+   return reduced;
+}
+
+static stbi__uint16 *stbi__convert_8_to_16(stbi_uc *orig, int w, int h, int channels)
+{
+   int i;
+   int img_len = w * h * channels;
+   stbi__uint16 *enlarged;
+
+   enlarged = (stbi__uint16 *) stbi__malloc(img_len*2);
+   if (enlarged == NULL) return (stbi__uint16 *) stbi__errpuc("outofmem", "Out of memory");
+
+   for (i = 0; i < img_len; ++i)
+      enlarged[i] = (stbi__uint16)((orig[i] << 8) + orig[i]); // replicate to high and low byte, maps 0->0, 255->0xffff
+
+   STBI_FREE(orig);
+   return enlarged;
+}
+
+static unsigned char *stbi__load_and_postprocess_8bit(stbi__context *s, int *x, int *y, int *comp, int req_comp)
+{
+   stbi__result_info ri;
+   void *result = stbi__load_main(s, x, y, comp, req_comp, &ri, 8);
+
+   if (result == NULL)
+      return NULL;
+
+   if (ri.bits_per_channel != 8) {
+      STBI_ASSERT(ri.bits_per_channel == 16);
+      result = stbi__convert_16_to_8((stbi__uint16 *) result, *x, *y, req_comp == 0 ? *comp : req_comp);
+      ri.bits_per_channel = 8;
+   }
+
+   // @TODO: move stbi__convert_format to here
+
+   if (stbi__vertically_flip_on_load) {
+      int w = *x, h = *y;
+      int channels = req_comp ? req_comp : *comp;
+      int row,col,z;
+      stbi_uc *image = (stbi_uc *) result;
+
+      // @OPTIMIZE: use a bigger temp buffer and memcpy multiple pixels at once
+      for (row = 0; row < (h>>1); row++) {
+         for (col = 0; col < w; col++) {
+            for (z = 0; z < channels; z++) {
+               stbi_uc temp = image[(row * w + col) * channels + z];
+               image[(row * w + col) * channels + z] = image[((h - row - 1) * w + col) * channels + z];
+               image[((h - row - 1) * w + col) * channels + z] = temp;
+            }
+         }
+      }
+   }
+
+   return (unsigned char *) result;
+}
+
+static stbi__uint16 *stbi__load_and_postprocess_16bit(stbi__context *s, int *x, int *y, int *comp, int req_comp)
+{
+   stbi__result_info ri;
+   void *result = stbi__load_main(s, x, y, comp, req_comp, &ri, 16);
+
+   if (result == NULL)
+      return NULL;
+
+   if (ri.bits_per_channel != 16) {
+      STBI_ASSERT(ri.bits_per_channel == 8);
+      result = stbi__convert_8_to_16((stbi_uc *) result, *x, *y, req_comp == 0 ? *comp : req_comp);
+      ri.bits_per_channel = 16;
+   }
+
+   // @TODO: move stbi__convert_format16 to here
+   // @TODO: special case RGB-to-Y (and RGBA-to-YA) for 8-bit-to-16-bit case to keep more precision
+
+   if (stbi__vertically_flip_on_load) {
+      int w = *x, h = *y;
+      int channels = req_comp ? req_comp : *comp;
+      int row,col,z;
+      stbi__uint16 *image = (stbi__uint16 *) result;
+
+      // @OPTIMIZE: use a bigger temp buffer and memcpy multiple pixels at once
+      for (row = 0; row < (h>>1); row++) {
+         for (col = 0; col < w; col++) {
+            for (z = 0; z < channels; z++) {
+               stbi__uint16 temp = image[(row * w + col) * channels + z];
+               image[(row * w + col) * channels + z] = image[((h - row - 1) * w + col) * channels + z];
+               image[((h - row - 1) * w + col) * channels + z] = temp;
+            }
+         }
+      }
+   }
+
+   return (stbi__uint16 *) result;
+}
+
+#ifndef STBI_NO_HDR
+static void stbi__float_postprocess(float *result, int *x, int *y, int *comp, int req_comp)
+{
+   if (stbi__vertically_flip_on_load && result != NULL) {
+      int w = *x, h = *y;
+      int depth = req_comp ? req_comp : *comp;
+      int row,col,z;
+      float temp;
+
+      // @OPTIMIZE: use a bigger temp buffer and memcpy multiple pixels at once
+      for (row = 0; row < (h>>1); row++) {
+         for (col = 0; col < w; col++) {
+            for (z = 0; z < depth; z++) {
+               temp = result[(row * w + col) * depth + z];
+               result[(row * w + col) * depth + z] = result[((h - row - 1) * w + col) * depth + z];
+               result[((h - row - 1) * w + col) * depth + z] = temp;
+            }
+         }
+      }
+   }
+}
+#endif
+
+#ifndef STBI_NO_STDIO
+
+static FILE *stbi__fopen(char const *filename, char const *mode)
+{
+   FILE *f;
+#if defined(_MSC_VER) && _MSC_VER >= 1400
+   if (0 != fopen_s(&f, filename, mode))
+      f=0;
+#else
+   f = fopen(filename, mode);
+#endif
+   return f;
+}
+
+
+STBIDEF stbi_uc *stbi_load(char const *filename, int *x, int *y, int *comp, int req_comp)
+{
+   FILE *f = stbi__fopen(filename, "rb");
+   unsigned char *result;
+   if (!f) return stbi__errpuc("can't fopen", "Unable to open file");
+   result = stbi_load_from_file(f,x,y,comp,req_comp);
+   fclose(f);
+   return result;
+}
+
+STBIDEF stbi_uc *stbi_load_from_file(FILE *f, int *x, int *y, int *comp, int req_comp)
+{
+   unsigned char *result;
+   stbi__context s;
+   stbi__start_file(&s,f);
+   result = stbi__load_and_postprocess_8bit(&s,x,y,comp,req_comp);
+   if (result) {
+      // need to 'unget' all the characters in the IO buffer
+      fseek(f, - (int) (s.img_buffer_end - s.img_buffer), SEEK_CUR);
+   }
+   return result;
+}
+
+STBIDEF stbi__uint16 *stbi_load_from_file_16(FILE *f, int *x, int *y, int *comp, int req_comp)
+{
+   stbi__uint16 *result;
+   stbi__context s;
+   stbi__start_file(&s,f);
+   result = stbi__load_and_postprocess_16bit(&s,x,y,comp,req_comp);
+   if (result) {
+      // need to 'unget' all the characters in the IO buffer
+      fseek(f, - (int) (s.img_buffer_end - s.img_buffer), SEEK_CUR);
+   }
+   return result;
+}
+
+STBIDEF stbi_us *stbi_load_16(char const *filename, int *x, int *y, int *comp, int req_comp)
+{
+   FILE *f = stbi__fopen(filename, "rb");
+   stbi__uint16 *result;
+   if (!f) return (stbi_us *) stbi__errpuc("can't fopen", "Unable to open file");
+   result = stbi_load_from_file_16(f,x,y,comp,req_comp);
+   fclose(f);
+   return result;
+}
+
+
+#endif //!STBI_NO_STDIO
+
+STBIDEF stbi_uc *stbi_load_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp)
+{
+   stbi__context s;
+   stbi__start_mem(&s,buffer,len);
+   return stbi__load_and_postprocess_8bit(&s,x,y,comp,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)
+{
+   stbi__context s;
+   stbi__start_callbacks(&s, (stbi_io_callbacks *) clbk, user);
+   return stbi__load_and_postprocess_8bit(&s,x,y,comp,req_comp);
+}
+
+#ifndef STBI_NO_LINEAR
+static float *stbi__loadf_main(stbi__context *s, int *x, int *y, int *comp, int req_comp)
+{
+   unsigned char *data;
+   #ifndef STBI_NO_HDR
+   if (stbi__hdr_test(s)) {
+      stbi__result_info ri;
+      float *hdr_data = stbi__hdr_load(s,x,y,comp,req_comp, &ri);
+      if (hdr_data)
+         stbi__float_postprocess(hdr_data,x,y,comp,req_comp);
+      return hdr_data;
+   }
+   #endif
+   data = stbi__load_and_postprocess_8bit(s, x, y, comp, req_comp);
+   if (data)
+      return stbi__ldr_to_hdr(data, *x, *y, req_comp ? req_comp : *comp);
+   return stbi__errpf("unknown image type", "Image not of any known type, or corrupt");
+}
+
+STBIDEF float *stbi_loadf_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp)
+{
+   stbi__context s;
+   stbi__start_mem(&s,buffer,len);
+   return stbi__loadf_main(&s,x,y,comp,req_comp);
+}
+
+STBIDEF float *stbi_loadf_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp, int req_comp)
+{
+   stbi__context s;
+   stbi__start_callbacks(&s, (stbi_io_callbacks *) clbk, user);
+   return stbi__loadf_main(&s,x,y,comp,req_comp);
+}
+
+#ifndef STBI_NO_STDIO
+STBIDEF float *stbi_loadf(char const *filename, int *x, int *y, int *comp, int req_comp)
+{
+   float *result;
+   FILE *f = stbi__fopen(filename, "rb");
+   if (!f) return stbi__errpf("can't fopen", "Unable to open file");
+   result = stbi_loadf_from_file(f,x,y,comp,req_comp);
+   fclose(f);
+   return result;
+}
+
+STBIDEF float *stbi_loadf_from_file(FILE *f, int *x, int *y, int *comp, int req_comp)
+{
+   stbi__context s;
+   stbi__start_file(&s,f);
+   return stbi__loadf_main(&s,x,y,comp,req_comp);
+}
+#endif // !STBI_NO_STDIO
+
+#endif // !STBI_NO_LINEAR
+
+// these is-hdr-or-not is defined independent of whether STBI_NO_LINEAR is
+// defined, for API simplicity; if STBI_NO_LINEAR is defined, it always
+// reports false!
+
+STBIDEF int stbi_is_hdr_from_memory(stbi_uc const *buffer, int len)
+{
+   #ifndef STBI_NO_HDR
+   stbi__context s;
+   stbi__start_mem(&s,buffer,len);
+   return stbi__hdr_test(&s);
+   #else
+   STBI_NOTUSED(buffer);
+   STBI_NOTUSED(len);
+   return 0;
+   #endif
+}
+
+#ifndef STBI_NO_STDIO
+STBIDEF int      stbi_is_hdr          (char const *filename)
+{
+   FILE *f = stbi__fopen(filename, "rb");
+   int result=0;
+   if (f) {
+      result = stbi_is_hdr_from_file(f);
+      fclose(f);
+   }
+   return result;
+}
+
+STBIDEF int      stbi_is_hdr_from_file(FILE *f)
+{
+   #ifndef STBI_NO_HDR
+   stbi__context s;
+   stbi__start_file(&s,f);
+   return stbi__hdr_test(&s);
+   #else
+   STBI_NOTUSED(f);
+   return 0;
+   #endif
+}
+#endif // !STBI_NO_STDIO
+
+STBIDEF int      stbi_is_hdr_from_callbacks(stbi_io_callbacks const *clbk, void *user)
+{
+   #ifndef STBI_NO_HDR
+   stbi__context s;
+   stbi__start_callbacks(&s, (stbi_io_callbacks *) clbk, user);
+   return stbi__hdr_test(&s);
+   #else
+   STBI_NOTUSED(clbk);
+   STBI_NOTUSED(user);
+   return 0;
+   #endif
+}
+
+#ifndef STBI_NO_LINEAR
+static float stbi__l2h_gamma=2.2f, stbi__l2h_scale=1.0f;
+
+STBIDEF void   stbi_ldr_to_hdr_gamma(float gamma) { stbi__l2h_gamma = gamma; }
+STBIDEF void   stbi_ldr_to_hdr_scale(float scale) { stbi__l2h_scale = scale; }
+#endif
+
+static float stbi__h2l_gamma_i=1.0f/2.2f, stbi__h2l_scale_i=1.0f;
+
+STBIDEF void   stbi_hdr_to_ldr_gamma(float gamma) { stbi__h2l_gamma_i = 1/gamma; }
+STBIDEF void   stbi_hdr_to_ldr_scale(float scale) { stbi__h2l_scale_i = 1/scale; }
+
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// Common code used by all image loaders
+//
+
+enum
+{
+   STBI__SCAN_load=0,
+   STBI__SCAN_type,
+   STBI__SCAN_header
+};
+
+static void stbi__refill_buffer(stbi__context *s)
+{
+   int n = (s->io.read)(s->io_user_data,(char*)s->buffer_start,s->buflen);
+   if (n == 0) {
+      // at end of file, treat same as if from memory, but need to handle case
+      // where s->img_buffer isn't pointing to safe memory, e.g. 0-byte file
+      s->read_from_callbacks = 0;
+      s->img_buffer = s->buffer_start;
+      s->img_buffer_end = s->buffer_start+1;
+      *s->img_buffer = 0;
+   } else {
+      s->img_buffer = s->buffer_start;
+      s->img_buffer_end = s->buffer_start + n;
+   }
+}
+
+stbi_inline static stbi_uc stbi__get8(stbi__context *s)
+{
+   if (s->img_buffer < s->img_buffer_end)
+      return *s->img_buffer++;
+   if (s->read_from_callbacks) {
+      stbi__refill_buffer(s);
+      return *s->img_buffer++;
+   }
+   return 0;
+}
+
+stbi_inline static int stbi__at_eof(stbi__context *s)
+{
+   if (s->io.read) {
+      if (!(s->io.eof)(s->io_user_data)) return 0;
+      // if feof() is true, check if buffer = end
+      // special case: we've only got the special 0 character at the end
+      if (s->read_from_callbacks == 0) return 1;
+   }
+
+   return s->img_buffer >= s->img_buffer_end;
+}
+
+static void stbi__skip(stbi__context *s, int n)
+{
+   if (n < 0) {
+      s->img_buffer = s->img_buffer_end;
+      return;
+   }
+   if (s->io.read) {
+      int blen = (int) (s->img_buffer_end - s->img_buffer);
+      if (blen < n) {
+         s->img_buffer = s->img_buffer_end;
+         (s->io.skip)(s->io_user_data, n - blen);
+         return;
+      }
+   }
+   s->img_buffer += n;
+}
+
+static int stbi__getn(stbi__context *s, stbi_uc *buffer, int n)
+{
+   if (s->io.read) {
+      int blen = (int) (s->img_buffer_end - s->img_buffer);
+      if (blen < n) {
+         int res, count;
+
+         memcpy(buffer, s->img_buffer, blen);
+
+         count = (s->io.read)(s->io_user_data, (char*) buffer + blen, n - blen);
+         res = (count == (n-blen));
+         s->img_buffer = s->img_buffer_end;
+         return res;
+      }
+   }
+
+   if (s->img_buffer+n <= s->img_buffer_end) {
+      memcpy(buffer, s->img_buffer, n);
+      s->img_buffer += n;
+      return 1;
+   } else
+      return 0;
+}
+
+static int stbi__get16be(stbi__context *s)
+{
+   int z = stbi__get8(s);
+   return (z << 8) + stbi__get8(s);
+}
+
+static stbi__uint32 stbi__get32be(stbi__context *s)
+{
+   stbi__uint32 z = stbi__get16be(s);
+   return (z << 16) + stbi__get16be(s);
+}
+
+#if defined(STBI_NO_BMP) && defined(STBI_NO_TGA) && defined(STBI_NO_GIF)
+// nothing
+#else
+static int stbi__get16le(stbi__context *s)
+{
+   int z = stbi__get8(s);
+   return z + (stbi__get8(s) << 8);
+}
+#endif
+
+#ifndef STBI_NO_BMP
+static stbi__uint32 stbi__get32le(stbi__context *s)
+{
+   stbi__uint32 z = stbi__get16le(s);
+   return z + (stbi__get16le(s) << 16);
+}
+#endif
+
+#define STBI__BYTECAST(x)  ((stbi_uc) ((x) & 255))  // truncate int to byte without warnings
+
+
+//////////////////////////////////////////////////////////////////////////////
+//
+//  generic converter from built-in img_n to req_comp
+//    individual types do this automatically as much as possible (e.g. jpeg
+//    does all cases internally since it needs to colorspace convert anyway,
+//    and it never has alpha, so very few cases ). png can automatically
+//    interleave an alpha=255 channel, but falls back to this for other cases
+//
+//  assume data buffer is malloced, so malloc a new one and free that one
+//  only failure mode is malloc failing
+
+static stbi_uc stbi__compute_y(int r, int g, int b)
+{
+   return (stbi_uc) (((r*77) + (g*150) +  (29*b)) >> 8);
+}
+
+static unsigned char *stbi__convert_format(unsigned char *data, int img_n, int req_comp, unsigned int x, unsigned int y)
+{
+   int i,j;
+   unsigned char *good;
+
+   if (req_comp == img_n) return data;
+   STBI_ASSERT(req_comp >= 1 && req_comp <= 4);
+
+   good = (unsigned char *) stbi__malloc_mad3(req_comp, x, y, 0);
+   if (good == NULL) {
+      STBI_FREE(data);
+      return stbi__errpuc("outofmem", "Out of memory");
+   }
+
+   for (j=0; j < (int) y; ++j) {
+      unsigned char *src  = data + j * x * img_n   ;
+      unsigned char *dest = good + j * x * req_comp;
+
+      #define STBI__COMBO(a,b)  ((a)*8+(b))
+      #define STBI__CASE(a,b)   case STBI__COMBO(a,b): for(i=x-1; i >= 0; --i, src += a, dest += b)
+      // convert source image with img_n components to one with req_comp components;
+      // avoid switch per pixel, so use switch per scanline and massive macros
+      switch (STBI__COMBO(img_n, req_comp)) {
+         STBI__CASE(1,2) { dest[0]=src[0], dest[1]=255;                                     } break;
+         STBI__CASE(1,3) { dest[0]=dest[1]=dest[2]=src[0];                                  } break;
+         STBI__CASE(1,4) { dest[0]=dest[1]=dest[2]=src[0], dest[3]=255;                     } break;
+         STBI__CASE(2,1) { dest[0]=src[0];                                                  } break;
+         STBI__CASE(2,3) { dest[0]=dest[1]=dest[2]=src[0];                                  } break;
+         STBI__CASE(2,4) { dest[0]=dest[1]=dest[2]=src[0], dest[3]=src[1];                  } break;
+         STBI__CASE(3,4) { dest[0]=src[0],dest[1]=src[1],dest[2]=src[2],dest[3]=255;        } break;
+         STBI__CASE(3,1) { dest[0]=stbi__compute_y(src[0],src[1],src[2]);                   } break;
+         STBI__CASE(3,2) { dest[0]=stbi__compute_y(src[0],src[1],src[2]), dest[1] = 255;    } break;
+         STBI__CASE(4,1) { dest[0]=stbi__compute_y(src[0],src[1],src[2]);                   } break;
+         STBI__CASE(4,2) { dest[0]=stbi__compute_y(src[0],src[1],src[2]), dest[1] = src[3]; } break;
+         STBI__CASE(4,3) { dest[0]=src[0],dest[1]=src[1],dest[2]=src[2];                    } break;
+         default: STBI_ASSERT(0);
+      }
+      #undef STBI__CASE
+   }
+
+   STBI_FREE(data);
+   return good;
+}
+
+static stbi__uint16 stbi__compute_y_16(int r, int g, int b)
+{
+   return (stbi__uint16) (((r*77) + (g*150) +  (29*b)) >> 8);
+}
+
+static stbi__uint16 *stbi__convert_format16(stbi__uint16 *data, int img_n, int req_comp, unsigned int x, unsigned int y)
+{
+   int i,j;
+   stbi__uint16 *good;
+
+   if (req_comp == img_n) return data;
+   STBI_ASSERT(req_comp >= 1 && req_comp <= 4);
+
+   good = (stbi__uint16 *) stbi__malloc(req_comp * x * y * 2);
+   if (good == NULL) {
+      STBI_FREE(data);
+      return (stbi__uint16 *) stbi__errpuc("outofmem", "Out of memory");
+   }
+
+   for (j=0; j < (int) y; ++j) {
+      stbi__uint16 *src  = data + j * x * img_n   ;
+      stbi__uint16 *dest = good + j * x * req_comp;
+
+      #define STBI__COMBO(a,b)  ((a)*8+(b))
+      #define STBI__CASE(a,b)   case STBI__COMBO(a,b): for(i=x-1; i >= 0; --i, src += a, dest += b)
+      // convert source image with img_n components to one with req_comp components;
+      // avoid switch per pixel, so use switch per scanline and massive macros
+      switch (STBI__COMBO(img_n, req_comp)) {
+         STBI__CASE(1,2) { dest[0]=src[0], dest[1]=0xffff;                                     } break;
+         STBI__CASE(1,3) { dest[0]=dest[1]=dest[2]=src[0];                                     } break;
+         STBI__CASE(1,4) { dest[0]=dest[1]=dest[2]=src[0], dest[3]=0xffff;                     } break;
+         STBI__CASE(2,1) { dest[0]=src[0];                                                     } break;
+         STBI__CASE(2,3) { dest[0]=dest[1]=dest[2]=src[0];                                     } break;
+         STBI__CASE(2,4) { dest[0]=dest[1]=dest[2]=src[0], dest[3]=src[1];                     } break;
+         STBI__CASE(3,4) { dest[0]=src[0],dest[1]=src[1],dest[2]=src[2],dest[3]=0xffff;        } break;
+         STBI__CASE(3,1) { dest[0]=stbi__compute_y_16(src[0],src[1],src[2]);                   } break;
+         STBI__CASE(3,2) { dest[0]=stbi__compute_y_16(src[0],src[1],src[2]), dest[1] = 0xffff; } break;
+         STBI__CASE(4,1) { dest[0]=stbi__compute_y_16(src[0],src[1],src[2]);                   } break;
+         STBI__CASE(4,2) { dest[0]=stbi__compute_y_16(src[0],src[1],src[2]), dest[1] = src[3]; } break;
+         STBI__CASE(4,3) { dest[0]=src[0],dest[1]=src[1],dest[2]=src[2];                       } break;
+         default: STBI_ASSERT(0);
+      }
+      #undef STBI__CASE
+   }
+
+   STBI_FREE(data);
+   return good;
+}
+
+#ifndef STBI_NO_LINEAR
+static float   *stbi__ldr_to_hdr(stbi_uc *data, int x, int y, int comp)
+{
+   int i,k,n;
+   float *output;
+   if (!data) return NULL;
+   output = (float *) stbi__malloc_mad4(x, y, comp, sizeof(float), 0);
+   if (output == NULL) { STBI_FREE(data); return stbi__errpf("outofmem", "Out of memory"); }
+   // compute number of non-alpha components
+   if (comp & 1) n = comp; else n = comp-1;
+   for (i=0; i < x*y; ++i) {
+      for (k=0; k < n; ++k) {
+         output[i*comp + k] = (float) (pow(data[i*comp+k]/255.0f, stbi__l2h_gamma) * stbi__l2h_scale);
+      }
+      if (k < comp) output[i*comp + k] = data[i*comp+k]/255.0f;
+   }
+   STBI_FREE(data);
+   return output;
+}
+#endif
+
+#ifndef STBI_NO_HDR
+#define stbi__float2int(x)   ((int) (x))
+static stbi_uc *stbi__hdr_to_ldr(float   *data, int x, int y, int comp)
+{
+   int i,k,n;
+   stbi_uc *output;
+   if (!data) return NULL;
+   output = (stbi_uc *) stbi__malloc_mad3(x, y, comp, 0);
+   if (output == NULL) { STBI_FREE(data); return stbi__errpuc("outofmem", "Out of memory"); }
+   // compute number of non-alpha components
+   if (comp & 1) n = comp; else n = comp-1;
+   for (i=0; i < x*y; ++i) {
+      for (k=0; k < n; ++k) {
+         float z = (float) pow(data[i*comp+k]*stbi__h2l_scale_i, stbi__h2l_gamma_i) * 255 + 0.5f;
+         if (z < 0) z = 0;
+         if (z > 255) z = 255;
+         output[i*comp + k] = (stbi_uc) stbi__float2int(z);
+      }
+      if (k < comp) {
+         float z = data[i*comp+k] * 255 + 0.5f;
+         if (z < 0) z = 0;
+         if (z > 255) z = 255;
+         output[i*comp + k] = (stbi_uc) stbi__float2int(z);
+      }
+   }
+   STBI_FREE(data);
+   return output;
+}
+#endif
+
+//////////////////////////////////////////////////////////////////////////////
+//
+//  "baseline" JPEG/JFIF decoder
+//
+//    simple implementation
+//      - doesn't support delayed output of y-dimension
+//      - simple interface (only one output format: 8-bit interleaved RGB)
+//      - doesn't try to recover corrupt jpegs
+//      - doesn't allow partial loading, loading multiple at once
+//      - still fast on x86 (copying globals into locals doesn't help x86)
+//      - allocates lots of intermediate memory (full size of all components)
+//        - non-interleaved case requires this anyway
+//        - allows good upsampling (see next)
+//    high-quality
+//      - upsampled channels are bilinearly interpolated, even across blocks
+//      - quality integer IDCT derived from IJG's 'slow'
+//    performance
+//      - fast huffman; reasonable integer IDCT
+//      - some SIMD kernels for common paths on targets with SSE2/NEON
+//      - uses a lot of intermediate memory, could cache poorly
+
+#ifndef STBI_NO_JPEG
+
+// huffman decoding acceleration
+#define FAST_BITS   9  // larger handles more cases; smaller stomps less cache
+
+typedef struct
+{
+   stbi_uc  fast[1 << FAST_BITS];
+   // weirdly, repacking this into AoS is a 10% speed loss, instead of a win
+   stbi__uint16 code[256];
+   stbi_uc  values[256];
+   stbi_uc  size[257];
+   unsigned int maxcode[18];
+   int    delta[17];   // old 'firstsymbol' - old 'firstcode'
+} stbi__huffman;
+
+typedef struct
+{
+   stbi__context *s;
+   stbi__huffman huff_dc[4];
+   stbi__huffman huff_ac[4];
+   stbi__uint16 dequant[4][64];
+   stbi__int16 fast_ac[4][1 << FAST_BITS];
+
+// sizes for components, interleaved MCUs
+   int img_h_max, img_v_max;
+   int img_mcu_x, img_mcu_y;
+   int img_mcu_w, img_mcu_h;
+
+// definition of jpeg image component
+   struct
+   {
+      int id;
+      int h,v;
+      int tq;
+      int hd,ha;
+      int dc_pred;
+
+      int x,y,w2,h2;
+      stbi_uc *data;
+      void *raw_data, *raw_coeff;
+      stbi_uc *linebuf;
+      short   *coeff;   // progressive only
+      int      coeff_w, coeff_h; // number of 8x8 coefficient blocks
+   } img_comp[4];
+
+   stbi__uint32   code_buffer; // jpeg entropy-coded buffer
+   int            code_bits;   // number of valid bits
+   unsigned char  marker;      // marker seen while filling entropy buffer
+   int            nomore;      // flag if we saw a marker so must stop
+
+   int            progressive;
+   int            spec_start;
+   int            spec_end;
+   int            succ_high;
+   int            succ_low;
+   int            eob_run;
+   int            jfif;
+   int            app14_color_transform; // Adobe APP14 tag
+   int            rgb;
+
+   int scan_n, order[4];
+   int restart_interval, todo;
+
+// kernels
+   void (*idct_block_kernel)(stbi_uc *out, int out_stride, short data[64]);
+   void (*YCbCr_to_RGB_kernel)(stbi_uc *out, const stbi_uc *y, const stbi_uc *pcb, const stbi_uc *pcr, int count, int step);
+   stbi_uc *(*resample_row_hv_2_kernel)(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs);
+} stbi__jpeg;
+
+static int stbi__build_huffman(stbi__huffman *h, int *count)
+{
+   int i,j,k=0,code;
+   // build size list for each symbol (from JPEG spec)
+   for (i=0; i < 16; ++i)
+      for (j=0; j < count[i]; ++j)
+         h->size[k++] = (stbi_uc) (i+1);
+   h->size[k] = 0;
+
+   // compute actual symbols (from jpeg spec)
+   code = 0;
+   k = 0;
+   for(j=1; j <= 16; ++j) {
+      // compute delta to add to code to compute symbol id
+      h->delta[j] = k - code;
+      if (h->size[k] == j) {
+         while (h->size[k] == j)
+            h->code[k++] = (stbi__uint16) (code++);
+         if (code-1 >= (1 << j)) return stbi__err("bad code lengths","Corrupt JPEG");
+      }
+      // compute largest code + 1 for this size, preshifted as needed later
+      h->maxcode[j] = code << (16-j);
+      code <<= 1;
+   }
+   h->maxcode[j] = 0xffffffff;
+
+   // build non-spec acceleration table; 255 is flag for not-accelerated
+   memset(h->fast, 255, 1 << FAST_BITS);
+   for (i=0; i < k; ++i) {
+      int s = h->size[i];
+      if (s <= FAST_BITS) {
+         int c = h->code[i] << (FAST_BITS-s);
+         int m = 1 << (FAST_BITS-s);
+         for (j=0; j < m; ++j) {
+            h->fast[c+j] = (stbi_uc) i;
+         }
+      }
+   }
+   return 1;
+}
+
+// build a table that decodes both magnitude and value of small ACs in
+// one go.
+static void stbi__build_fast_ac(stbi__int16 *fast_ac, stbi__huffman *h)
+{
+   int i;
+   for (i=0; i < (1 << FAST_BITS); ++i) {
+      stbi_uc fast = h->fast[i];
+      fast_ac[i] = 0;
+      if (fast < 255) {
+         int rs = h->values[fast];
+         int run = (rs >> 4) & 15;
+         int magbits = rs & 15;
+         int len = h->size[fast];
+
+         if (magbits && len + magbits <= FAST_BITS) {
+            // magnitude code followed by receive_extend code
+            int k = ((i << len) & ((1 << FAST_BITS) - 1)) >> (FAST_BITS - magbits);
+            int m = 1 << (magbits - 1);
+            if (k < m) k += (~0U << magbits) + 1;
+            // if the result is small enough, we can fit it in fast_ac table
+            if (k >= -128 && k <= 127)
+               fast_ac[i] = (stbi__int16) ((k << 8) + (run << 4) + (len + magbits));
+         }
+      }
+   }
+}
+
+static void stbi__grow_buffer_unsafe(stbi__jpeg *j)
+{
+   do {
+      int b = j->nomore ? 0 : stbi__get8(j->s);
+      if (b == 0xff) {
+         int c = stbi__get8(j->s);
+         while (c == 0xff) c = stbi__get8(j->s); // consume fill bytes
+         if (c != 0) {
+            j->marker = (unsigned char) c;
+            j->nomore = 1;
+            return;
+         }
+      }
+      j->code_buffer |= b << (24 - j->code_bits);
+      j->code_bits += 8;
+   } while (j->code_bits <= 24);
+}
+
+// (1 << n) - 1
+static stbi__uint32 stbi__bmask[17]={0,1,3,7,15,31,63,127,255,511,1023,2047,4095,8191,16383,32767,65535};
+
+// decode a jpeg huffman value from the bitstream
+stbi_inline static int stbi__jpeg_huff_decode(stbi__jpeg *j, stbi__huffman *h)
+{
+   unsigned int temp;
+   int c,k;
+
+   if (j->code_bits < 16) stbi__grow_buffer_unsafe(j);
+
+   // look at the top FAST_BITS and determine what symbol ID it is,
+   // if the code is <= FAST_BITS
+   c = (j->code_buffer >> (32 - FAST_BITS)) & ((1 << FAST_BITS)-1);
+   k = h->fast[c];
+   if (k < 255) {
+      int s = h->size[k];
+      if (s > j->code_bits)
+         return -1;
+      j->code_buffer <<= s;
+      j->code_bits -= s;
+      return h->values[k];
+   }
+
+   // naive test is to shift the code_buffer down so k bits are
+   // valid, then test against maxcode. To speed this up, we've
+   // preshifted maxcode left so that it has (16-k) 0s at the
+   // end; in other words, regardless of the number of bits, it
+   // wants to be compared against something shifted to have 16;
+   // that way we don't need to shift inside the loop.
+   temp = j->code_buffer >> 16;
+   for (k=FAST_BITS+1 ; ; ++k)
+      if (temp < h->maxcode[k])
+         break;
+   if (k == 17) {
+      // error! code not found
+      j->code_bits -= 16;
+      return -1;
+   }
+
+   if (k > j->code_bits)
+      return -1;
+
+   // convert the huffman code to the symbol id
+   c = ((j->code_buffer >> (32 - k)) & stbi__bmask[k]) + h->delta[k];
+   STBI_ASSERT((((j->code_buffer) >> (32 - h->size[c])) & stbi__bmask[h->size[c]]) == h->code[c]);
+
+   // convert the id to a symbol
+   j->code_bits -= k;
+   j->code_buffer <<= k;
+   return h->values[c];
+}
+
+// bias[n] = (-1<<n) + 1
+static int const stbi__jbias[16] = {0,-1,-3,-7,-15,-31,-63,-127,-255,-511,-1023,-2047,-4095,-8191,-16383,-32767};
+
+// combined JPEG 'receive' and JPEG 'extend', since baseline
+// always extends everything it receives.
+stbi_inline static int stbi__extend_receive(stbi__jpeg *j, int n)
+{
+   unsigned int k;
+   int sgn;
+   if (j->code_bits < n) stbi__grow_buffer_unsafe(j);
+
+   sgn = (stbi__int32)j->code_buffer >> 31; // sign bit is always in MSB
+   k = stbi_lrot(j->code_buffer, n);
+   STBI_ASSERT(n >= 0 && n < (int) (sizeof(stbi__bmask)/sizeof(*stbi__bmask)));
+   j->code_buffer = k & ~stbi__bmask[n];
+   k &= stbi__bmask[n];
+   j->code_bits -= n;
+   return k + (stbi__jbias[n] & ~sgn);
+}
+
+// get some unsigned bits
+stbi_inline static int stbi__jpeg_get_bits(stbi__jpeg *j, int n)
+{
+   unsigned int k;
+   if (j->code_bits < n) stbi__grow_buffer_unsafe(j);
+   k = stbi_lrot(j->code_buffer, n);
+   j->code_buffer = k & ~stbi__bmask[n];
+   k &= stbi__bmask[n];
+   j->code_bits -= n;
+   return k;
+}
+
+stbi_inline static int stbi__jpeg_get_bit(stbi__jpeg *j)
+{
+   unsigned int k;
+   if (j->code_bits < 1) stbi__grow_buffer_unsafe(j);
+   k = j->code_buffer;
+   j->code_buffer <<= 1;
+   --j->code_bits;
+   return k & 0x80000000;
+}
+
+// given a value that's at position X in the zigzag stream,
+// where does it appear in the 8x8 matrix coded as row-major?
+static stbi_uc stbi__jpeg_dezigzag[64+15] =
+{
+    0,  1,  8, 16,  9,  2,  3, 10,
+   17, 24, 32, 25, 18, 11,  4,  5,
+   12, 19, 26, 33, 40, 48, 41, 34,
+   27, 20, 13,  6,  7, 14, 21, 28,
+   35, 42, 49, 56, 57, 50, 43, 36,
+   29, 22, 15, 23, 30, 37, 44, 51,
+   58, 59, 52, 45, 38, 31, 39, 46,
+   53, 60, 61, 54, 47, 55, 62, 63,
+   // let corrupt input sample past end
+   63, 63, 63, 63, 63, 63, 63, 63,
+   63, 63, 63, 63, 63, 63, 63
+};
+
+// decode one 64-entry block--
+static int stbi__jpeg_decode_block(stbi__jpeg *j, short data[64], stbi__huffman *hdc, stbi__huffman *hac, stbi__int16 *fac, int b, stbi__uint16 *dequant)
+{
+   int diff,dc,k;
+   int t;
+
+   if (j->code_bits < 16) stbi__grow_buffer_unsafe(j);
+   t = stbi__jpeg_huff_decode(j, hdc);
+   if (t < 0) return stbi__err("bad huffman code","Corrupt JPEG");
+
+   // 0 all the ac values now so we can do it 32-bits at a time
+   memset(data,0,64*sizeof(data[0]));
+
+   diff = t ? stbi__extend_receive(j, t) : 0;
+   dc = j->img_comp[b].dc_pred + diff;
+   j->img_comp[b].dc_pred = dc;
+   data[0] = (short) (dc * dequant[0]);
+
+   // decode AC components, see JPEG spec
+   k = 1;
+   do {
+      unsigned int zig;
+      int c,r,s;
+      if (j->code_bits < 16) stbi__grow_buffer_unsafe(j);
+      c = (j->code_buffer >> (32 - FAST_BITS)) & ((1 << FAST_BITS)-1);
+      r = fac[c];
+      if (r) { // fast-AC path
+         k += (r >> 4) & 15; // run
+         s = r & 15; // combined length
+         j->code_buffer <<= s;
+         j->code_bits -= s;
+         // decode into unzigzag'd location
+         zig = stbi__jpeg_dezigzag[k++];
+         data[zig] = (short) ((r >> 8) * dequant[zig]);
+      } else {
+         int rs = stbi__jpeg_huff_decode(j, hac);
+         if (rs < 0) return stbi__err("bad huffman code","Corrupt JPEG");
+         s = rs & 15;
+         r = rs >> 4;
+         if (s == 0) {
+            if (rs != 0xf0) break; // end block
+            k += 16;
+         } else {
+            k += r;
+            // decode into unzigzag'd location
+            zig = stbi__jpeg_dezigzag[k++];
+            data[zig] = (short) (stbi__extend_receive(j,s) * dequant[zig]);
+         }
+      }
+   } while (k < 64);
+   return 1;
+}
+
+static int stbi__jpeg_decode_block_prog_dc(stbi__jpeg *j, short data[64], stbi__huffman *hdc, int b)
+{
+   int diff,dc;
+   int t;
+   if (j->spec_end != 0) return stbi__err("can't merge dc and ac", "Corrupt JPEG");
+
+   if (j->code_bits < 16) stbi__grow_buffer_unsafe(j);
+
+   if (j->succ_high == 0) {
+      // first scan for DC coefficient, must be first
+      memset(data,0,64*sizeof(data[0])); // 0 all the ac values now
+      t = stbi__jpeg_huff_decode(j, hdc);
+      diff = t ? stbi__extend_receive(j, t) : 0;
+
+      dc = j->img_comp[b].dc_pred + diff;
+      j->img_comp[b].dc_pred = dc;
+      data[0] = (short) (dc << j->succ_low);
+   } else {
+      // refinement scan for DC coefficient
+      if (stbi__jpeg_get_bit(j))
+         data[0] += (short) (1 << j->succ_low);
+   }
+   return 1;
+}
+
+// @OPTIMIZE: store non-zigzagged during the decode passes,
+// and only de-zigzag when dequantizing
+static int stbi__jpeg_decode_block_prog_ac(stbi__jpeg *j, short data[64], stbi__huffman *hac, stbi__int16 *fac)
+{
+   int k;
+   if (j->spec_start == 0) return stbi__err("can't merge dc and ac", "Corrupt JPEG");
+
+   if (j->succ_high == 0) {
+      int shift = j->succ_low;
+
+      if (j->eob_run) {
+         --j->eob_run;
+         return 1;
+      }
+
+      k = j->spec_start;
+      do {
+         unsigned int zig;
+         int c,r,s;
+         if (j->code_bits < 16) stbi__grow_buffer_unsafe(j);
+         c = (j->code_buffer >> (32 - FAST_BITS)) & ((1 << FAST_BITS)-1);
+         r = fac[c];
+         if (r) { // fast-AC path
+            k += (r >> 4) & 15; // run
+            s = r & 15; // combined length
+            j->code_buffer <<= s;
+            j->code_bits -= s;
+            zig = stbi__jpeg_dezigzag[k++];
+            data[zig] = (short) ((r >> 8) << shift);
+         } else {
+            int rs = stbi__jpeg_huff_decode(j, hac);
+            if (rs < 0) return stbi__err("bad huffman code","Corrupt JPEG");
+            s = rs & 15;
+            r = rs >> 4;
+            if (s == 0) {
+               if (r < 15) {
+                  j->eob_run = (1 << r);
+                  if (r)
+                     j->eob_run += stbi__jpeg_get_bits(j, r);
+                  --j->eob_run;
+                  break;
+               }
+               k += 16;
+            } else {
+               k += r;
+               zig = stbi__jpeg_dezigzag[k++];
+               data[zig] = (short) (stbi__extend_receive(j,s) << shift);
+            }
+         }
+      } while (k <= j->spec_end);
+   } else {
+      // refinement scan for these AC coefficients
+
+      short bit = (short) (1 << j->succ_low);
+
+      if (j->eob_run) {
+         --j->eob_run;
+         for (k = j->spec_start; k <= j->spec_end; ++k) {
+            short *p = &data[stbi__jpeg_dezigzag[k]];
+            if (*p != 0)
+               if (stbi__jpeg_get_bit(j))
+                  if ((*p & bit)==0) {
+                     if (*p > 0)
+                        *p += bit;
+                     else
+                        *p -= bit;
+                  }
+         }
+      } else {
+         k = j->spec_start;
+         do {
+            int r,s;
+            int rs = stbi__jpeg_huff_decode(j, hac); // @OPTIMIZE see if we can use the fast path here, advance-by-r is so slow, eh
+            if (rs < 0) return stbi__err("bad huffman code","Corrupt JPEG");
+            s = rs & 15;
+            r = rs >> 4;
+            if (s == 0) {
+               if (r < 15) {
+                  j->eob_run = (1 << r) - 1;
+                  if (r)
+                     j->eob_run += stbi__jpeg_get_bits(j, r);
+                  r = 64; // force end of block
+               } else {
+                  // r=15 s=0 should write 16 0s, so we just do
+                  // a run of 15 0s and then write s (which is 0),
+                  // so we don't have to do anything special here
+               }
+            } else {
+               if (s != 1) return stbi__err("bad huffman code", "Corrupt JPEG");
+               // sign bit
+               if (stbi__jpeg_get_bit(j))
+                  s = bit;
+               else
+                  s = -bit;
+            }
+
+            // advance by r
+            while (k <= j->spec_end) {
+               short *p = &data[stbi__jpeg_dezigzag[k++]];
+               if (*p != 0) {
+                  if (stbi__jpeg_get_bit(j))
+                     if ((*p & bit)==0) {
+                        if (*p > 0)
+                           *p += bit;
+                        else
+                           *p -= bit;
+                     }
+               } else {
+                  if (r == 0) {
+                     *p = (short) s;
+                     break;
+                  }
+                  --r;
+               }
+            }
+         } while (k <= j->spec_end);
+      }
+   }
+   return 1;
+}
+
+// take a -128..127 value and stbi__clamp it and convert to 0..255
+stbi_inline static stbi_uc stbi__clamp(int x)
+{
+   // trick to use a single test to catch both cases
+   if ((unsigned int) x > 255) {
+      if (x < 0) return 0;
+      if (x > 255) return 255;
+   }
+   return (stbi_uc) x;
+}
+
+#define stbi__f2f(x)  ((int) (((x) * 4096 + 0.5)))
+#define stbi__fsh(x)  ((x) << 12)
+
+// derived from jidctint -- DCT_ISLOW
+#define STBI__IDCT_1D(s0,s1,s2,s3,s4,s5,s6,s7) \
+   int t0,t1,t2,t3,p1,p2,p3,p4,p5,x0,x1,x2,x3; \
+   p2 = s2;                                    \
+   p3 = s6;                                    \
+   p1 = (p2+p3) * stbi__f2f(0.5411961f);       \
+   t2 = p1 + p3*stbi__f2f(-1.847759065f);      \
+   t3 = p1 + p2*stbi__f2f( 0.765366865f);      \
+   p2 = s0;                                    \
+   p3 = s4;                                    \
+   t0 = stbi__fsh(p2+p3);                      \
+   t1 = stbi__fsh(p2-p3);                      \
+   x0 = t0+t3;                                 \
+   x3 = t0-t3;                                 \
+   x1 = t1+t2;                                 \
+   x2 = t1-t2;                                 \
+   t0 = s7;                                    \
+   t1 = s5;                                    \
+   t2 = s3;                                    \
+   t3 = s1;                                    \
+   p3 = t0+t2;                                 \
+   p4 = t1+t3;                                 \
+   p1 = t0+t3;                                 \
+   p2 = t1+t2;                                 \
+   p5 = (p3+p4)*stbi__f2f( 1.175875602f);      \
+   t0 = t0*stbi__f2f( 0.298631336f);           \
+   t1 = t1*stbi__f2f( 2.053119869f);           \
+   t2 = t2*stbi__f2f( 3.072711026f);           \
+   t3 = t3*stbi__f2f( 1.501321110f);           \
+   p1 = p5 + p1*stbi__f2f(-0.899976223f);      \
+   p2 = p5 + p2*stbi__f2f(-2.562915447f);      \
+   p3 = p3*stbi__f2f(-1.961570560f);           \
+   p4 = p4*stbi__f2f(-0.390180644f);           \
+   t3 += p1+p4;                                \
+   t2 += p2+p3;                                \
+   t1 += p2+p4;                                \
+   t0 += p1+p3;
+
+static void stbi__idct_block(stbi_uc *out, int out_stride, short data[64])
+{
+   int i,val[64],*v=val;
+   stbi_uc *o;
+   short *d = data;
+
+   // columns
+   for (i=0; i < 8; ++i,++d, ++v) {
+      // if all zeroes, shortcut -- this avoids dequantizing 0s and IDCTing
+      if (d[ 8]==0 && d[16]==0 && d[24]==0 && d[32]==0
+           && d[40]==0 && d[48]==0 && d[56]==0) {
+         //    no shortcut                 0     seconds
+         //    (1|2|3|4|5|6|7)==0          0     seconds
+         //    all separate               -0.047 seconds
+         //    1 && 2|3 && 4|5 && 6|7:    -0.047 seconds
+         int dcterm = d[0] << 2;
+         v[0] = v[8] = v[16] = v[24] = v[32] = v[40] = v[48] = v[56] = dcterm;
+      } else {
+         STBI__IDCT_1D(d[ 0],d[ 8],d[16],d[24],d[32],d[40],d[48],d[56])
+         // constants scaled things up by 1<<12; let's bring them back
+         // down, but keep 2 extra bits of precision
+         x0 += 512; x1 += 512; x2 += 512; x3 += 512;
+         v[ 0] = (x0+t3) >> 10;
+         v[56] = (x0-t3) >> 10;
+         v[ 8] = (x1+t2) >> 10;
+         v[48] = (x1-t2) >> 10;
+         v[16] = (x2+t1) >> 10;
+         v[40] = (x2-t1) >> 10;
+         v[24] = (x3+t0) >> 10;
+         v[32] = (x3-t0) >> 10;
+      }
+   }
+
+   for (i=0, v=val, o=out; i < 8; ++i,v+=8,o+=out_stride) {
+      // no fast case since the first 1D IDCT spread components out
+      STBI__IDCT_1D(v[0],v[1],v[2],v[3],v[4],v[5],v[6],v[7])
+      // constants scaled things up by 1<<12, plus we had 1<<2 from first
+      // loop, plus horizontal and vertical each scale by sqrt(8) so together
+      // we've got an extra 1<<3, so 1<<17 total we need to remove.
+      // so we want to round that, which means adding 0.5 * 1<<17,
+      // aka 65536. Also, we'll end up with -128 to 127 that we want
+      // to encode as 0..255 by adding 128, so we'll add that before the shift
+      x0 += 65536 + (128<<17);
+      x1 += 65536 + (128<<17);
+      x2 += 65536 + (128<<17);
+      x3 += 65536 + (128<<17);
+      // tried computing the shifts into temps, or'ing the temps to see
+      // if any were out of range, but that was slower
+      o[0] = stbi__clamp((x0+t3) >> 17);
+      o[7] = stbi__clamp((x0-t3) >> 17);
+      o[1] = stbi__clamp((x1+t2) >> 17);
+      o[6] = stbi__clamp((x1-t2) >> 17);
+      o[2] = stbi__clamp((x2+t1) >> 17);
+      o[5] = stbi__clamp((x2-t1) >> 17);
+      o[3] = stbi__clamp((x3+t0) >> 17);
+      o[4] = stbi__clamp((x3-t0) >> 17);
+   }
+}
+
+#ifdef STBI_SSE2
+// sse2 integer IDCT. not the fastest possible implementation but it
+// produces bit-identical results to the generic C version so it's
+// fully "transparent".
+static void stbi__idct_simd(stbi_uc *out, int out_stride, short data[64])
+{
+   // This is constructed to match our regular (generic) integer IDCT exactly.
+   __m128i row0, row1, row2, row3, row4, row5, row6, row7;
+   __m128i tmp;
+
+   // dot product constant: even elems=x, odd elems=y
+   #define dct_const(x,y)  _mm_setr_epi16((x),(y),(x),(y),(x),(y),(x),(y))
+
+   // out(0) = c0[even]*x + c0[odd]*y   (c0, x, y 16-bit, out 32-bit)
+   // out(1) = c1[even]*x + c1[odd]*y
+   #define dct_rot(out0,out1, x,y,c0,c1) \
+      __m128i c0##lo = _mm_unpacklo_epi16((x),(y)); \
+      __m128i c0##hi = _mm_unpackhi_epi16((x),(y)); \
+      __m128i out0##_l = _mm_madd_epi16(c0##lo, c0); \
+      __m128i out0##_h = _mm_madd_epi16(c0##hi, c0); \
+      __m128i out1##_l = _mm_madd_epi16(c0##lo, c1); \
+      __m128i out1##_h = _mm_madd_epi16(c0##hi, c1)
+
+   // out = in << 12  (in 16-bit, out 32-bit)
+   #define dct_widen(out, in) \
+      __m128i out##_l = _mm_srai_epi32(_mm_unpacklo_epi16(_mm_setzero_si128(), (in)), 4); \
+      __m128i out##_h = _mm_srai_epi32(_mm_unpackhi_epi16(_mm_setzero_si128(), (in)), 4)
+
+   // wide add
+   #define dct_wadd(out, a, b) \
+      __m128i out##_l = _mm_add_epi32(a##_l, b##_l); \
+      __m128i out##_h = _mm_add_epi32(a##_h, b##_h)
+
+   // wide sub
+   #define dct_wsub(out, a, b) \
+      __m128i out##_l = _mm_sub_epi32(a##_l, b##_l); \
+      __m128i out##_h = _mm_sub_epi32(a##_h, b##_h)
+
+   // butterfly a/b, add bias, then shift by "s" and pack
+   #define dct_bfly32o(out0, out1, a,b,bias,s) \
+      { \
+         __m128i abiased_l = _mm_add_epi32(a##_l, bias); \
+         __m128i abiased_h = _mm_add_epi32(a##_h, bias); \
+         dct_wadd(sum, abiased, b); \
+         dct_wsub(dif, abiased, b); \
+         out0 = _mm_packs_epi32(_mm_srai_epi32(sum_l, s), _mm_srai_epi32(sum_h, s)); \
+         out1 = _mm_packs_epi32(_mm_srai_epi32(dif_l, s), _mm_srai_epi32(dif_h, s)); \
+      }
+
+   // 8-bit interleave step (for transposes)
+   #define dct_interleave8(a, b) \
+      tmp = a; \
+      a = _mm_unpacklo_epi8(a, b); \
+      b = _mm_unpackhi_epi8(tmp, b)
+
+   // 16-bit interleave step (for transposes)
+   #define dct_interleave16(a, b) \
+      tmp = a; \
+      a = _mm_unpacklo_epi16(a, b); \
+      b = _mm_unpackhi_epi16(tmp, b)
+
+   #define dct_pass(bias,shift) \
+      { \
+         /* even part */ \
+         dct_rot(t2e,t3e, row2,row6, rot0_0,rot0_1); \
+         __m128i sum04 = _mm_add_epi16(row0, row4); \
+         __m128i dif04 = _mm_sub_epi16(row0, row4); \
+         dct_widen(t0e, sum04); \
+         dct_widen(t1e, dif04); \
+         dct_wadd(x0, t0e, t3e); \
+         dct_wsub(x3, t0e, t3e); \
+         dct_wadd(x1, t1e, t2e); \
+         dct_wsub(x2, t1e, t2e); \
+         /* odd part */ \
+         dct_rot(y0o,y2o, row7,row3, rot2_0,rot2_1); \
+         dct_rot(y1o,y3o, row5,row1, rot3_0,rot3_1); \
+         __m128i sum17 = _mm_add_epi16(row1, row7); \
+         __m128i sum35 = _mm_add_epi16(row3, row5); \
+         dct_rot(y4o,y5o, sum17,sum35, rot1_0,rot1_1); \
+         dct_wadd(x4, y0o, y4o); \
+         dct_wadd(x5, y1o, y5o); \
+         dct_wadd(x6, y2o, y5o); \
+         dct_wadd(x7, y3o, y4o); \
+         dct_bfly32o(row0,row7, x0,x7,bias,shift); \
+         dct_bfly32o(row1,row6, x1,x6,bias,shift); \
+         dct_bfly32o(row2,row5, x2,x5,bias,shift); \
+         dct_bfly32o(row3,row4, x3,x4,bias,shift); \
+      }
+
+   __m128i rot0_0 = dct_const(stbi__f2f(0.5411961f), stbi__f2f(0.5411961f) + stbi__f2f(-1.847759065f));
+   __m128i rot0_1 = dct_const(stbi__f2f(0.5411961f) + stbi__f2f( 0.765366865f), stbi__f2f(0.5411961f));
+   __m128i rot1_0 = dct_const(stbi__f2f(1.175875602f) + stbi__f2f(-0.899976223f), stbi__f2f(1.175875602f));
+   __m128i rot1_1 = dct_const(stbi__f2f(1.175875602f), stbi__f2f(1.175875602f) + stbi__f2f(-2.562915447f));
+   __m128i rot2_0 = dct_const(stbi__f2f(-1.961570560f) + stbi__f2f( 0.298631336f), stbi__f2f(-1.961570560f));
+   __m128i rot2_1 = dct_const(stbi__f2f(-1.961570560f), stbi__f2f(-1.961570560f) + stbi__f2f( 3.072711026f));
+   __m128i rot3_0 = dct_const(stbi__f2f(-0.390180644f) + stbi__f2f( 2.053119869f), stbi__f2f(-0.390180644f));
+   __m128i rot3_1 = dct_const(stbi__f2f(-0.390180644f), stbi__f2f(-0.390180644f) + stbi__f2f( 1.501321110f));
+
+   // rounding biases in column/row passes, see stbi__idct_block for explanation.
+   __m128i bias_0 = _mm_set1_epi32(512);
+   __m128i bias_1 = _mm_set1_epi32(65536 + (128<<17));
+
+   // load
+   row0 = _mm_load_si128((const __m128i *) (data + 0*8));
+   row1 = _mm_load_si128((const __m128i *) (data + 1*8));
+   row2 = _mm_load_si128((const __m128i *) (data + 2*8));
+   row3 = _mm_load_si128((const __m128i *) (data + 3*8));
+   row4 = _mm_load_si128((const __m128i *) (data + 4*8));
+   row5 = _mm_load_si128((const __m128i *) (data + 5*8));
+   row6 = _mm_load_si128((const __m128i *) (data + 6*8));
+   row7 = _mm_load_si128((const __m128i *) (data + 7*8));
+
+   // column pass
+   dct_pass(bias_0, 10);
+
+   {
+      // 16bit 8x8 transpose pass 1
+      dct_interleave16(row0, row4);
+      dct_interleave16(row1, row5);
+      dct_interleave16(row2, row6);
+      dct_interleave16(row3, row7);
+
+      // transpose pass 2
+      dct_interleave16(row0, row2);
+      dct_interleave16(row1, row3);
+      dct_interleave16(row4, row6);
+      dct_interleave16(row5, row7);
+
+      // transpose pass 3
+      dct_interleave16(row0, row1);
+      dct_interleave16(row2, row3);
+      dct_interleave16(row4, row5);
+      dct_interleave16(row6, row7);
+   }
+
+   // row pass
+   dct_pass(bias_1, 17);
+
+   {
+      // pack
+      __m128i p0 = _mm_packus_epi16(row0, row1); // a0a1a2a3...a7b0b1b2b3...b7
+      __m128i p1 = _mm_packus_epi16(row2, row3);
+      __m128i p2 = _mm_packus_epi16(row4, row5);
+      __m128i p3 = _mm_packus_epi16(row6, row7);
+
+      // 8bit 8x8 transpose pass 1
+      dct_interleave8(p0, p2); // a0e0a1e1...
+      dct_interleave8(p1, p3); // c0g0c1g1...
+
+      // transpose pass 2
+      dct_interleave8(p0, p1); // a0c0e0g0...
+      dct_interleave8(p2, p3); // b0d0f0h0...
+
+      // transpose pass 3
+      dct_interleave8(p0, p2); // a0b0c0d0...
+      dct_interleave8(p1, p3); // a4b4c4d4...
+
+      // store
+      _mm_storel_epi64((__m128i *) out, p0); out += out_stride;
+      _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p0, 0x4e)); out += out_stride;
+      _mm_storel_epi64((__m128i *) out, p2); out += out_stride;
+      _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p2, 0x4e)); out += out_stride;
+      _mm_storel_epi64((__m128i *) out, p1); out += out_stride;
+      _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p1, 0x4e)); out += out_stride;
+      _mm_storel_epi64((__m128i *) out, p3); out += out_stride;
+      _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p3, 0x4e));
+   }
+
+#undef dct_const
+#undef dct_rot
+#undef dct_widen
+#undef dct_wadd
+#undef dct_wsub
+#undef dct_bfly32o
+#undef dct_interleave8
+#undef dct_interleave16
+#undef dct_pass
+}
+
+#endif // STBI_SSE2
+
+#ifdef STBI_NEON
+
+// NEON integer IDCT. should produce bit-identical
+// results to the generic C version.
+static void stbi__idct_simd(stbi_uc *out, int out_stride, short data[64])
+{
+   int16x8_t row0, row1, row2, row3, row4, row5, row6, row7;
+
+   int16x4_t rot0_0 = vdup_n_s16(stbi__f2f(0.5411961f));
+   int16x4_t rot0_1 = vdup_n_s16(stbi__f2f(-1.847759065f));
+   int16x4_t rot0_2 = vdup_n_s16(stbi__f2f( 0.765366865f));
+   int16x4_t rot1_0 = vdup_n_s16(stbi__f2f( 1.175875602f));
+   int16x4_t rot1_1 = vdup_n_s16(stbi__f2f(-0.899976223f));
+   int16x4_t rot1_2 = vdup_n_s16(stbi__f2f(-2.562915447f));
+   int16x4_t rot2_0 = vdup_n_s16(stbi__f2f(-1.961570560f));
+   int16x4_t rot2_1 = vdup_n_s16(stbi__f2f(-0.390180644f));
+   int16x4_t rot3_0 = vdup_n_s16(stbi__f2f( 0.298631336f));
+   int16x4_t rot3_1 = vdup_n_s16(stbi__f2f( 2.053119869f));
+   int16x4_t rot3_2 = vdup_n_s16(stbi__f2f( 3.072711026f));
+   int16x4_t rot3_3 = vdup_n_s16(stbi__f2f( 1.501321110f));
+
+#define dct_long_mul(out, inq, coeff) \
+   int32x4_t out##_l = vmull_s16(vget_low_s16(inq), coeff); \
+   int32x4_t out##_h = vmull_s16(vget_high_s16(inq), coeff)
+
+#define dct_long_mac(out, acc, inq, coeff) \
+   int32x4_t out##_l = vmlal_s16(acc##_l, vget_low_s16(inq), coeff); \
+   int32x4_t out##_h = vmlal_s16(acc##_h, vget_high_s16(inq), coeff)
+
+#define dct_widen(out, inq) \
+   int32x4_t out##_l = vshll_n_s16(vget_low_s16(inq), 12); \
+   int32x4_t out##_h = vshll_n_s16(vget_high_s16(inq), 12)
+
+// wide add
+#define dct_wadd(out, a, b) \
+   int32x4_t out##_l = vaddq_s32(a##_l, b##_l); \
+   int32x4_t out##_h = vaddq_s32(a##_h, b##_h)
+
+// wide sub
+#define dct_wsub(out, a, b) \
+   int32x4_t out##_l = vsubq_s32(a##_l, b##_l); \
+   int32x4_t out##_h = vsubq_s32(a##_h, b##_h)
+
+// butterfly a/b, then shift using "shiftop" by "s" and pack
+#define dct_bfly32o(out0,out1, a,b,shiftop,s) \
+   { \
+      dct_wadd(sum, a, b); \
+      dct_wsub(dif, a, b); \
+      out0 = vcombine_s16(shiftop(sum_l, s), shiftop(sum_h, s)); \
+      out1 = vcombine_s16(shiftop(dif_l, s), shiftop(dif_h, s)); \
+   }
+
+#define dct_pass(shiftop, shift) \
+   { \
+      /* even part */ \
+      int16x8_t sum26 = vaddq_s16(row2, row6); \
+      dct_long_mul(p1e, sum26, rot0_0); \
+      dct_long_mac(t2e, p1e, row6, rot0_1); \
+      dct_long_mac(t3e, p1e, row2, rot0_2); \
+      int16x8_t sum04 = vaddq_s16(row0, row4); \
+      int16x8_t dif04 = vsubq_s16(row0, row4); \
+      dct_widen(t0e, sum04); \
+      dct_widen(t1e, dif04); \
+      dct_wadd(x0, t0e, t3e); \
+      dct_wsub(x3, t0e, t3e); \
+      dct_wadd(x1, t1e, t2e); \
+      dct_wsub(x2, t1e, t2e); \
+      /* odd part */ \
+      int16x8_t sum15 = vaddq_s16(row1, row5); \
+      int16x8_t sum17 = vaddq_s16(row1, row7); \
+      int16x8_t sum35 = vaddq_s16(row3, row5); \
+      int16x8_t sum37 = vaddq_s16(row3, row7); \
+      int16x8_t sumodd = vaddq_s16(sum17, sum35); \
+      dct_long_mul(p5o, sumodd, rot1_0); \
+      dct_long_mac(p1o, p5o, sum17, rot1_1); \
+      dct_long_mac(p2o, p5o, sum35, rot1_2); \
+      dct_long_mul(p3o, sum37, rot2_0); \
+      dct_long_mul(p4o, sum15, rot2_1); \
+      dct_wadd(sump13o, p1o, p3o); \
+      dct_wadd(sump24o, p2o, p4o); \
+      dct_wadd(sump23o, p2o, p3o); \
+      dct_wadd(sump14o, p1o, p4o); \
+      dct_long_mac(x4, sump13o, row7, rot3_0); \
+      dct_long_mac(x5, sump24o, row5, rot3_1); \
+      dct_long_mac(x6, sump23o, row3, rot3_2); \
+      dct_long_mac(x7, sump14o, row1, rot3_3); \
+      dct_bfly32o(row0,row7, x0,x7,shiftop,shift); \
+      dct_bfly32o(row1,row6, x1,x6,shiftop,shift); \
+      dct_bfly32o(row2,row5, x2,x5,shiftop,shift); \
+      dct_bfly32o(row3,row4, x3,x4,shiftop,shift); \
+   }
+
+   // load
+   row0 = vld1q_s16(data + 0*8);
+   row1 = vld1q_s16(data + 1*8);
+   row2 = vld1q_s16(data + 2*8);
+   row3 = vld1q_s16(data + 3*8);
+   row4 = vld1q_s16(data + 4*8);
+   row5 = vld1q_s16(data + 5*8);
+   row6 = vld1q_s16(data + 6*8);
+   row7 = vld1q_s16(data + 7*8);
+
+   // add DC bias
+   row0 = vaddq_s16(row0, vsetq_lane_s16(1024, vdupq_n_s16(0), 0));
+
+   // column pass
+   dct_pass(vrshrn_n_s32, 10);
+
+   // 16bit 8x8 transpose
+   {
+// these three map to a single VTRN.16, VTRN.32, and VSWP, respectively.
+// whether compilers actually get this is another story, sadly.
+#define dct_trn16(x, y) { int16x8x2_t t = vtrnq_s16(x, y); x = t.val[0]; y = t.val[1]; }
+#define dct_trn32(x, y) { int32x4x2_t t = vtrnq_s32(vreinterpretq_s32_s16(x), vreinterpretq_s32_s16(y)); x = vreinterpretq_s16_s32(t.val[0]); y = vreinterpretq_s16_s32(t.val[1]); }
+#define dct_trn64(x, y) { int16x8_t x0 = x; int16x8_t y0 = y; x = vcombine_s16(vget_low_s16(x0), vget_low_s16(y0)); y = vcombine_s16(vget_high_s16(x0), vget_high_s16(y0)); }
+
+      // pass 1
+      dct_trn16(row0, row1); // a0b0a2b2a4b4a6b6
+      dct_trn16(row2, row3);
+      dct_trn16(row4, row5);
+      dct_trn16(row6, row7);
+
+      // pass 2
+      dct_trn32(row0, row2); // a0b0c0d0a4b4c4d4
+      dct_trn32(row1, row3);
+      dct_trn32(row4, row6);
+      dct_trn32(row5, row7);
+
+      // pass 3
+      dct_trn64(row0, row4); // a0b0c0d0e0f0g0h0
+      dct_trn64(row1, row5);
+      dct_trn64(row2, row6);
+      dct_trn64(row3, row7);
+
+#undef dct_trn16
+#undef dct_trn32
+#undef dct_trn64
+   }
+
+   // row pass
+   // vrshrn_n_s32 only supports shifts up to 16, we need
+   // 17. so do a non-rounding shift of 16 first then follow
+   // up with a rounding shift by 1.
+   dct_pass(vshrn_n_s32, 16);
+
+   {
+      // pack and round
+      uint8x8_t p0 = vqrshrun_n_s16(row0, 1);
+      uint8x8_t p1 = vqrshrun_n_s16(row1, 1);
+      uint8x8_t p2 = vqrshrun_n_s16(row2, 1);
+      uint8x8_t p3 = vqrshrun_n_s16(row3, 1);
+      uint8x8_t p4 = vqrshrun_n_s16(row4, 1);
+      uint8x8_t p5 = vqrshrun_n_s16(row5, 1);
+      uint8x8_t p6 = vqrshrun_n_s16(row6, 1);
+      uint8x8_t p7 = vqrshrun_n_s16(row7, 1);
+
+      // again, these can translate into one instruction, but often don't.
+#define dct_trn8_8(x, y) { uint8x8x2_t t = vtrn_u8(x, y); x = t.val[0]; y = t.val[1]; }
+#define dct_trn8_16(x, y) { uint16x4x2_t t = vtrn_u16(vreinterpret_u16_u8(x), vreinterpret_u16_u8(y)); x = vreinterpret_u8_u16(t.val[0]); y = vreinterpret_u8_u16(t.val[1]); }
+#define dct_trn8_32(x, y) { uint32x2x2_t t = vtrn_u32(vreinterpret_u32_u8(x), vreinterpret_u32_u8(y)); x = vreinterpret_u8_u32(t.val[0]); y = vreinterpret_u8_u32(t.val[1]); }
+
+      // sadly can't use interleaved stores here since we only write
+      // 8 bytes to each scan line!
+
+      // 8x8 8-bit transpose pass 1
+      dct_trn8_8(p0, p1);
+      dct_trn8_8(p2, p3);
+      dct_trn8_8(p4, p5);
+      dct_trn8_8(p6, p7);
+
+      // pass 2
+      dct_trn8_16(p0, p2);
+      dct_trn8_16(p1, p3);
+      dct_trn8_16(p4, p6);
+      dct_trn8_16(p5, p7);
+
+      // pass 3
+      dct_trn8_32(p0, p4);
+      dct_trn8_32(p1, p5);
+      dct_trn8_32(p2, p6);
+      dct_trn8_32(p3, p7);
+
+      // store
+      vst1_u8(out, p0); out += out_stride;
+      vst1_u8(out, p1); out += out_stride;
+      vst1_u8(out, p2); out += out_stride;
+      vst1_u8(out, p3); out += out_stride;
+      vst1_u8(out, p4); out += out_stride;
+      vst1_u8(out, p5); out += out_stride;
+      vst1_u8(out, p6); out += out_stride;
+      vst1_u8(out, p7);
+
+#undef dct_trn8_8
+#undef dct_trn8_16
+#undef dct_trn8_32
+   }
+
+#undef dct_long_mul
+#undef dct_long_mac
+#undef dct_widen
+#undef dct_wadd
+#undef dct_wsub
+#undef dct_bfly32o
+#undef dct_pass
+}
+
+#endif // STBI_NEON
+
+#define STBI__MARKER_none  0xff
+// if there's a pending marker from the entropy stream, return that
+// otherwise, fetch from the stream and get a marker. if there's no
+// marker, return 0xff, which is never a valid marker value
+static stbi_uc stbi__get_marker(stbi__jpeg *j)
+{
+   stbi_uc x;
+   if (j->marker != STBI__MARKER_none) { x = j->marker; j->marker = STBI__MARKER_none; return x; }
+   x = stbi__get8(j->s);
+   if (x != 0xff) return STBI__MARKER_none;
+   while (x == 0xff)
+      x = stbi__get8(j->s); // consume repeated 0xff fill bytes
+   return x;
+}
+
+// in each scan, we'll have scan_n components, and the order
+// of the components is specified by order[]
+#define STBI__RESTART(x)     ((x) >= 0xd0 && (x) <= 0xd7)
+
+// after a restart interval, stbi__jpeg_reset the entropy decoder and
+// the dc prediction
+static void stbi__jpeg_reset(stbi__jpeg *j)
+{
+   j->code_bits = 0;
+   j->code_buffer = 0;
+   j->nomore = 0;
+   j->img_comp[0].dc_pred = j->img_comp[1].dc_pred = j->img_comp[2].dc_pred = j->img_comp[3].dc_pred = 0;
+   j->marker = STBI__MARKER_none;
+   j->todo = j->restart_interval ? j->restart_interval : 0x7fffffff;
+   j->eob_run = 0;
+   // no more than 1<<31 MCUs if no restart_interal? that's plenty safe,
+   // since we don't even allow 1<<30 pixels
+}
+
+static int stbi__parse_entropy_coded_data(stbi__jpeg *z)
+{
+   stbi__jpeg_reset(z);
+   if (!z->progressive) {
+      if (z->scan_n == 1) {
+         int i,j;
+         STBI_SIMD_ALIGN(short, data[64]);
+         int n = z->order[0];
+         // non-interleaved data, we just need to process one block at a time,
+         // in trivial scanline order
+         // number of blocks to do just depends on how many actual "pixels" this
+         // component has, independent of interleaved MCU blocking and such
+         int w = (z->img_comp[n].x+7) >> 3;
+         int h = (z->img_comp[n].y+7) >> 3;
+         for (j=0; j < h; ++j) {
+            for (i=0; i < w; ++i) {
+               int ha = z->img_comp[n].ha;
+               if (!stbi__jpeg_decode_block(z, data, z->huff_dc+z->img_comp[n].hd, z->huff_ac+ha, z->fast_ac[ha], n, z->dequant[z->img_comp[n].tq])) return 0;
+               z->idct_block_kernel(z->img_comp[n].data+z->img_comp[n].w2*j*8+i*8, z->img_comp[n].w2, data);
+               // every data block is an MCU, so countdown the restart interval
+               if (--z->todo <= 0) {
+                  if (z->code_bits < 24) stbi__grow_buffer_unsafe(z);
+                  // if it's NOT a restart, then just bail, so we get corrupt data
+                  // rather than no data
+                  if (!STBI__RESTART(z->marker)) return 1;
+                  stbi__jpeg_reset(z);
+               }
+            }
+         }
+         return 1;
+      } else { // interleaved
+         int i,j,k,x,y;
+         STBI_SIMD_ALIGN(short, data[64]);
+         for (j=0; j < z->img_mcu_y; ++j) {
+            for (i=0; i < z->img_mcu_x; ++i) {
+               // scan an interleaved mcu... process scan_n components in order
+               for (k=0; k < z->scan_n; ++k) {
+                  int n = z->order[k];
+                  // scan out an mcu's worth of this component; that's just determined
+                  // by the basic H and V specified for the component
+                  for (y=0; y < z->img_comp[n].v; ++y) {
+                     for (x=0; x < z->img_comp[n].h; ++x) {
+                        int x2 = (i*z->img_comp[n].h + x)*8;
+                        int y2 = (j*z->img_comp[n].v + y)*8;
+                        int ha = z->img_comp[n].ha;
+                        if (!stbi__jpeg_decode_block(z, data, z->huff_dc+z->img_comp[n].hd, z->huff_ac+ha, z->fast_ac[ha], n, z->dequant[z->img_comp[n].tq])) return 0;
+                        z->idct_block_kernel(z->img_comp[n].data+z->img_comp[n].w2*y2+x2, z->img_comp[n].w2, data);
+                     }
+                  }
+               }
+               // after all interleaved components, that's an interleaved MCU,
+               // so now count down the restart interval
+               if (--z->todo <= 0) {
+                  if (z->code_bits < 24) stbi__grow_buffer_unsafe(z);
+                  if (!STBI__RESTART(z->marker)) return 1;
+                  stbi__jpeg_reset(z);
+               }
+            }
+         }
+         return 1;
+      }
+   } else {
+      if (z->scan_n == 1) {
+         int i,j;
+         int n = z->order[0];
+         // non-interleaved data, we just need to process one block at a time,
+         // in trivial scanline order
+         // number of blocks to do just depends on how many actual "pixels" this
+         // component has, independent of interleaved MCU blocking and such
+         int w = (z->img_comp[n].x+7) >> 3;
+         int h = (z->img_comp[n].y+7) >> 3;
+         for (j=0; j < h; ++j) {
+            for (i=0; i < w; ++i) {
+               short *data = z->img_comp[n].coeff + 64 * (i + j * z->img_comp[n].coeff_w);
+               if (z->spec_start == 0) {
+                  if (!stbi__jpeg_decode_block_prog_dc(z, data, &z->huff_dc[z->img_comp[n].hd], n))
+                     return 0;
+               } else {
+                  int ha = z->img_comp[n].ha;
+                  if (!stbi__jpeg_decode_block_prog_ac(z, data, &z->huff_ac[ha], z->fast_ac[ha]))
+                     return 0;
+               }
+               // every data block is an MCU, so countdown the restart interval
+               if (--z->todo <= 0) {
+                  if (z->code_bits < 24) stbi__grow_buffer_unsafe(z);
+                  if (!STBI__RESTART(z->marker)) return 1;
+                  stbi__jpeg_reset(z);
+               }
+            }
+         }
+         return 1;
+      } else { // interleaved
+         int i,j,k,x,y;
+         for (j=0; j < z->img_mcu_y; ++j) {
+            for (i=0; i < z->img_mcu_x; ++i) {
+               // scan an interleaved mcu... process scan_n components in order
+               for (k=0; k < z->scan_n; ++k) {
+                  int n = z->order[k];
+                  // scan out an mcu's worth of this component; that's just determined
+                  // by the basic H and V specified for the component
+                  for (y=0; y < z->img_comp[n].v; ++y) {
+                     for (x=0; x < z->img_comp[n].h; ++x) {
+                        int x2 = (i*z->img_comp[n].h + x);
+                        int y2 = (j*z->img_comp[n].v + y);
+                        short *data = z->img_comp[n].coeff + 64 * (x2 + y2 * z->img_comp[n].coeff_w);
+                        if (!stbi__jpeg_decode_block_prog_dc(z, data, &z->huff_dc[z->img_comp[n].hd], n))
+                           return 0;
+                     }
+                  }
+               }
+               // after all interleaved components, that's an interleaved MCU,
+               // so now count down the restart interval
+               if (--z->todo <= 0) {
+                  if (z->code_bits < 24) stbi__grow_buffer_unsafe(z);
+                  if (!STBI__RESTART(z->marker)) return 1;
+                  stbi__jpeg_reset(z);
+               }
+            }
+         }
+         return 1;
+      }
+   }
+}
+
+static void stbi__jpeg_dequantize(short *data, stbi__uint16 *dequant)
+{
+   int i;
+   for (i=0; i < 64; ++i)
+      data[i] *= dequant[i];
+}
+
+static void stbi__jpeg_finish(stbi__jpeg *z)
+{
+   if (z->progressive) {
+      // dequantize and idct the data
+      int i,j,n;
+      for (n=0; n < z->s->img_n; ++n) {
+         int w = (z->img_comp[n].x+7) >> 3;
+         int h = (z->img_comp[n].y+7) >> 3;
+         for (j=0; j < h; ++j) {
+            for (i=0; i < w; ++i) {
+               short *data = z->img_comp[n].coeff + 64 * (i + j * z->img_comp[n].coeff_w);
+               stbi__jpeg_dequantize(data, z->dequant[z->img_comp[n].tq]);
+               z->idct_block_kernel(z->img_comp[n].data+z->img_comp[n].w2*j*8+i*8, z->img_comp[n].w2, data);
+            }
+         }
+      }
+   }
+}
+
+static int stbi__process_marker(stbi__jpeg *z, int m)
+{
+   int L;
+   switch (m) {
+      case STBI__MARKER_none: // no marker found
+         return stbi__err("expected marker","Corrupt JPEG");
+
+      case 0xDD: // DRI - specify restart interval
+         if (stbi__get16be(z->s) != 4) return stbi__err("bad DRI len","Corrupt JPEG");
+         z->restart_interval = stbi__get16be(z->s);
+         return 1;
+
+      case 0xDB: // DQT - define quantization table
+         L = stbi__get16be(z->s)-2;
+         while (L > 0) {
+            int q = stbi__get8(z->s);
+            int p = q >> 4, sixteen = (p != 0);
+            int t = q & 15,i;
+            if (p != 0 && p != 1) return stbi__err("bad DQT type","Corrupt JPEG");
+            if (t > 3) return stbi__err("bad DQT table","Corrupt JPEG");
+
+            for (i=0; i < 64; ++i)
+               z->dequant[t][stbi__jpeg_dezigzag[i]] = sixteen ? stbi__get16be(z->s) : stbi__get8(z->s);
+            L -= (sixteen ? 129 : 65);
+         }
+         return L==0;
+
+      case 0xC4: // DHT - define huffman table
+         L = stbi__get16be(z->s)-2;
+         while (L > 0) {
+            stbi_uc *v;
+            int sizes[16],i,n=0;
+            int q = stbi__get8(z->s);
+            int tc = q >> 4;
+            int th = q & 15;
+            if (tc > 1 || th > 3) return stbi__err("bad DHT header","Corrupt JPEG");
+            for (i=0; i < 16; ++i) {
+               sizes[i] = stbi__get8(z->s);
+               n += sizes[i];
+            }
+            L -= 17;
+            if (tc == 0) {
+               if (!stbi__build_huffman(z->huff_dc+th, sizes)) return 0;
+               v = z->huff_dc[th].values;
+            } else {
+               if (!stbi__build_huffman(z->huff_ac+th, sizes)) return 0;
+               v = z->huff_ac[th].values;
+            }
+            for (i=0; i < n; ++i)
+               v[i] = stbi__get8(z->s);
+            if (tc != 0)
+               stbi__build_fast_ac(z->fast_ac[th], z->huff_ac + th);
+            L -= n;
+         }
+         return L==0;
+   }
+
+   // check for comment block or APP blocks
+   if ((m >= 0xE0 && m <= 0xEF) || m == 0xFE) {
+      L = stbi__get16be(z->s);
+      if (L < 2) {
+         if (m == 0xFE)
+            return stbi__err("bad COM len","Corrupt JPEG");
+         else
+            return stbi__err("bad APP len","Corrupt JPEG");
+      }
+      L -= 2;
+
+      if (m == 0xE0 && L >= 5) { // JFIF APP0 segment
+         static const unsigned char tag[5] = {'J','F','I','F','\0'};
+         int ok = 1;
+         int i;
+         for (i=0; i < 5; ++i)
+            if (stbi__get8(z->s) != tag[i])
+               ok = 0;
+         L -= 5;
+         if (ok)
+            z->jfif = 1;
+      } else if (m == 0xEE && L >= 12) { // Adobe APP14 segment
+         static const unsigned char tag[6] = {'A','d','o','b','e','\0'};
+         int ok = 1;
+         int i;
+         for (i=0; i < 6; ++i)
+            if (stbi__get8(z->s) != tag[i])
+               ok = 0;
+         L -= 6;
+         if (ok) {
+            stbi__get8(z->s); // version
+            stbi__get16be(z->s); // flags0
+            stbi__get16be(z->s); // flags1
+            z->app14_color_transform = stbi__get8(z->s); // color transform
+            L -= 6;
+         }
+      }
+
+      stbi__skip(z->s, L);
+      return 1;
+   }
+
+   return stbi__err("unknown marker","Corrupt JPEG");
+}
+
+// after we see SOS
+static int stbi__process_scan_header(stbi__jpeg *z)
+{
+   int i;
+   int Ls = stbi__get16be(z->s);
+   z->scan_n = stbi__get8(z->s);
+   if (z->scan_n < 1 || z->scan_n > 4 || z->scan_n > (int) z->s->img_n) return stbi__err("bad SOS component count","Corrupt JPEG");
+   if (Ls != 6+2*z->scan_n) return stbi__err("bad SOS len","Corrupt JPEG");
+   for (i=0; i < z->scan_n; ++i) {
+      int id = stbi__get8(z->s), which;
+      int q = stbi__get8(z->s);
+      for (which = 0; which < z->s->img_n; ++which)
+         if (z->img_comp[which].id == id)
+            break;
+      if (which == z->s->img_n) return 0; // no match
+      z->img_comp[which].hd = q >> 4;   if (z->img_comp[which].hd > 3) return stbi__err("bad DC huff","Corrupt JPEG");
+      z->img_comp[which].ha = q & 15;   if (z->img_comp[which].ha > 3) return stbi__err("bad AC huff","Corrupt JPEG");
+      z->order[i] = which;
+   }
+
+   {
+      int aa;
+      z->spec_start = stbi__get8(z->s);
+      z->spec_end   = stbi__get8(z->s); // should be 63, but might be 0
+      aa = stbi__get8(z->s);
+      z->succ_high = (aa >> 4);
+      z->succ_low  = (aa & 15);
+      if (z->progressive) {
+         if (z->spec_start > 63 || z->spec_end > 63  || z->spec_start > z->spec_end || z->succ_high > 13 || z->succ_low > 13)
+            return stbi__err("bad SOS", "Corrupt JPEG");
+      } else {
+         if (z->spec_start != 0) return stbi__err("bad SOS","Corrupt JPEG");
+         if (z->succ_high != 0 || z->succ_low != 0) return stbi__err("bad SOS","Corrupt JPEG");
+         z->spec_end = 63;
+      }
+   }
+
+   return 1;
+}
+
+static int stbi__free_jpeg_components(stbi__jpeg *z, int ncomp, int why)
+{
+   int i;
+   for (i=0; i < ncomp; ++i) {
+      if (z->img_comp[i].raw_data) {
+         STBI_FREE(z->img_comp[i].raw_data);
+         z->img_comp[i].raw_data = NULL;
+         z->img_comp[i].data = NULL;
+      }
+      if (z->img_comp[i].raw_coeff) {
+         STBI_FREE(z->img_comp[i].raw_coeff);
+         z->img_comp[i].raw_coeff = 0;
+         z->img_comp[i].coeff = 0;
+      }
+      if (z->img_comp[i].linebuf) {
+         STBI_FREE(z->img_comp[i].linebuf);
+         z->img_comp[i].linebuf = NULL;
+      }
+   }
+   return why;
+}
+
+static int stbi__process_frame_header(stbi__jpeg *z, int scan)
+{
+   stbi__context *s = z->s;
+   int Lf,p,i,q, h_max=1,v_max=1,c;
+   Lf = stbi__get16be(s);         if (Lf < 11) return stbi__err("bad SOF len","Corrupt JPEG"); // JPEG
+   p  = stbi__get8(s);            if (p != 8) return stbi__err("only 8-bit","JPEG format not supported: 8-bit only"); // JPEG baseline
+   s->img_y = stbi__get16be(s);   if (s->img_y == 0) return stbi__err("no header height", "JPEG format not supported: delayed height"); // Legal, but we don't handle it--but neither does IJG
+   s->img_x = stbi__get16be(s);   if (s->img_x == 0) return stbi__err("0 width","Corrupt JPEG"); // JPEG requires
+   c = stbi__get8(s);
+   if (c != 3 && c != 1 && c != 4) return stbi__err("bad component count","Corrupt JPEG");
+   s->img_n = c;
+   for (i=0; i < c; ++i) {
+      z->img_comp[i].data = NULL;
+      z->img_comp[i].linebuf = NULL;
+   }
+
+   if (Lf != 8+3*s->img_n) return stbi__err("bad SOF len","Corrupt JPEG");
+
+   z->rgb = 0;
+   for (i=0; i < s->img_n; ++i) {
+      static unsigned char rgb[3] = { 'R', 'G', 'B' };
+      z->img_comp[i].id = stbi__get8(s);
+      if (s->img_n == 3 && z->img_comp[i].id == rgb[i])
+         ++z->rgb;
+      q = stbi__get8(s);
+      z->img_comp[i].h = (q >> 4);  if (!z->img_comp[i].h || z->img_comp[i].h > 4) return stbi__err("bad H","Corrupt JPEG");
+      z->img_comp[i].v = q & 15;    if (!z->img_comp[i].v || z->img_comp[i].v > 4) return stbi__err("bad V","Corrupt JPEG");
+      z->img_comp[i].tq = stbi__get8(s);  if (z->img_comp[i].tq > 3) return stbi__err("bad TQ","Corrupt JPEG");
+   }
+
+   if (scan != STBI__SCAN_load) return 1;
+
+   if (!stbi__mad3sizes_valid(s->img_x, s->img_y, s->img_n, 0)) return stbi__err("too large", "Image too large to decode");
+
+   for (i=0; i < s->img_n; ++i) {
+      if (z->img_comp[i].h > h_max) h_max = z->img_comp[i].h;
+      if (z->img_comp[i].v > v_max) v_max = z->img_comp[i].v;
+   }
+
+   // compute interleaved mcu info
+   z->img_h_max = h_max;
+   z->img_v_max = v_max;
+   z->img_mcu_w = h_max * 8;
+   z->img_mcu_h = v_max * 8;
+   // these sizes can't be more than 17 bits
+   z->img_mcu_x = (s->img_x + z->img_mcu_w-1) / z->img_mcu_w;
+   z->img_mcu_y = (s->img_y + z->img_mcu_h-1) / z->img_mcu_h;
+
+   for (i=0; i < s->img_n; ++i) {
+      // number of effective pixels (e.g. for non-interleaved MCU)
+      z->img_comp[i].x = (s->img_x * z->img_comp[i].h + h_max-1) / h_max;
+      z->img_comp[i].y = (s->img_y * z->img_comp[i].v + v_max-1) / v_max;
+      // to simplify generation, we'll allocate enough memory to decode
+      // the bogus oversized data from using interleaved MCUs and their
+      // big blocks (e.g. a 16x16 iMCU on an image of width 33); we won't
+      // discard the extra data until colorspace conversion
+      //
+      // img_mcu_x, img_mcu_y: <=17 bits; comp[i].h and .v are <=4 (checked earlier)
+      // so these muls can't overflow with 32-bit ints (which we require)
+      z->img_comp[i].w2 = z->img_mcu_x * z->img_comp[i].h * 8;
+      z->img_comp[i].h2 = z->img_mcu_y * z->img_comp[i].v * 8;
+      z->img_comp[i].coeff = 0;
+      z->img_comp[i].raw_coeff = 0;
+      z->img_comp[i].linebuf = NULL;
+      z->img_comp[i].raw_data = stbi__malloc_mad2(z->img_comp[i].w2, z->img_comp[i].h2, 15);
+      if (z->img_comp[i].raw_data == NULL)
+         return stbi__free_jpeg_components(z, i+1, stbi__err("outofmem", "Out of memory"));
+      // align blocks for idct using mmx/sse
+      z->img_comp[i].data = (stbi_uc*) (((size_t) z->img_comp[i].raw_data + 15) & ~15);
+      if (z->progressive) {
+         // w2, h2 are multiples of 8 (see above)
+         z->img_comp[i].coeff_w = z->img_comp[i].w2 / 8;
+         z->img_comp[i].coeff_h = z->img_comp[i].h2 / 8;
+         z->img_comp[i].raw_coeff = stbi__malloc_mad3(z->img_comp[i].w2, z->img_comp[i].h2, sizeof(short), 15);
+         if (z->img_comp[i].raw_coeff == NULL)
+            return stbi__free_jpeg_components(z, i+1, stbi__err("outofmem", "Out of memory"));
+         z->img_comp[i].coeff = (short*) (((size_t) z->img_comp[i].raw_coeff + 15) & ~15);
+      }
+   }
+
+   return 1;
+}
+
+// use comparisons since in some cases we handle more than one case (e.g. SOF)
+#define stbi__DNL(x)         ((x) == 0xdc)
+#define stbi__SOI(x)         ((x) == 0xd8)
+#define stbi__EOI(x)         ((x) == 0xd9)
+#define stbi__SOF(x)         ((x) == 0xc0 || (x) == 0xc1 || (x) == 0xc2)
+#define stbi__SOS(x)         ((x) == 0xda)
+
+#define stbi__SOF_progressive(x)   ((x) == 0xc2)
+
+static int stbi__decode_jpeg_header(stbi__jpeg *z, int scan)
+{
+   int m;
+   z->jfif = 0;
+   z->app14_color_transform = -1; // valid values are 0,1,2
+   z->marker = STBI__MARKER_none; // initialize cached marker to empty
+   m = stbi__get_marker(z);
+   if (!stbi__SOI(m)) return stbi__err("no SOI","Corrupt JPEG");
+   if (scan == STBI__SCAN_type) return 1;
+   m = stbi__get_marker(z);
+   while (!stbi__SOF(m)) {
+      if (!stbi__process_marker(z,m)) return 0;
+      m = stbi__get_marker(z);
+      while (m == STBI__MARKER_none) {
+         // some files have extra padding after their blocks, so ok, we'll scan
+         if (stbi__at_eof(z->s)) return stbi__err("no SOF", "Corrupt JPEG");
+         m = stbi__get_marker(z);
+      }
+   }
+   z->progressive = stbi__SOF_progressive(m);
+   if (!stbi__process_frame_header(z, scan)) return 0;
+   return 1;
+}
+
+// decode image to YCbCr format
+static int stbi__decode_jpeg_image(stbi__jpeg *j)
+{
+   int m;
+   for (m = 0; m < 4; m++) {
+      j->img_comp[m].raw_data = NULL;
+      j->img_comp[m].raw_coeff = NULL;
+   }
+   j->restart_interval = 0;
+   if (!stbi__decode_jpeg_header(j, STBI__SCAN_load)) return 0;
+   m = stbi__get_marker(j);
+   while (!stbi__EOI(m)) {
+      if (stbi__SOS(m)) {
+         if (!stbi__process_scan_header(j)) return 0;
+         if (!stbi__parse_entropy_coded_data(j)) return 0;
+         if (j->marker == STBI__MARKER_none ) {
+            // handle 0s at the end of image data from IP Kamera 9060
+            while (!stbi__at_eof(j->s)) {
+               int x = stbi__get8(j->s);
+               if (x == 255) {
+                  j->marker = stbi__get8(j->s);
+                  break;
+               }
+            }
+            // if we reach eof without hitting a marker, stbi__get_marker() below will fail and we'll eventually return 0
+         }
+      } else if (stbi__DNL(m)) {
+         int Ld = stbi__get16be(j->s);
+         stbi__uint32 NL = stbi__get16be(j->s);
+         if (Ld != 4) stbi__err("bad DNL len", "Corrupt JPEG");
+         if (NL != j->s->img_y) stbi__err("bad DNL height", "Corrupt JPEG");
+      } else {
+         if (!stbi__process_marker(j, m)) return 0;
+      }
+      m = stbi__get_marker(j);
+   }
+   if (j->progressive)
+      stbi__jpeg_finish(j);
+   return 1;
+}
+
+// static jfif-centered resampling (across block boundaries)
+
+typedef stbi_uc *(*resample_row_func)(stbi_uc *out, stbi_uc *in0, stbi_uc *in1,
+                                    int w, int hs);
+
+#define stbi__div4(x) ((stbi_uc) ((x) >> 2))
+
+static stbi_uc *resample_row_1(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
+{
+   STBI_NOTUSED(out);
+   STBI_NOTUSED(in_far);
+   STBI_NOTUSED(w);
+   STBI_NOTUSED(hs);
+   return in_near;
+}
+
+static stbi_uc* stbi__resample_row_v_2(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
+{
+   // need to generate two samples vertically for every one in input
+   int i;
+   STBI_NOTUSED(hs);
+   for (i=0; i < w; ++i)
+      out[i] = stbi__div4(3*in_near[i] + in_far[i] + 2);
+   return out;
+}
+
+static stbi_uc*  stbi__resample_row_h_2(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
+{
+   // need to generate two samples horizontally for every one in input
+   int i;
+   stbi_uc *input = in_near;
+
+   if (w == 1) {
+      // if only one sample, can't do any interpolation
+      out[0] = out[1] = input[0];
+      return out;
+   }
+
+   out[0] = input[0];
+   out[1] = stbi__div4(input[0]*3 + input[1] + 2);
+   for (i=1; i < w-1; ++i) {
+      int n = 3*input[i]+2;
+      out[i*2+0] = stbi__div4(n+input[i-1]);
+      out[i*2+1] = stbi__div4(n+input[i+1]);
+   }
+   out[i*2+0] = stbi__div4(input[w-2]*3 + input[w-1] + 2);
+   out[i*2+1] = input[w-1];
+
+   STBI_NOTUSED(in_far);
+   STBI_NOTUSED(hs);
+
+   return out;
+}
+
+#define stbi__div16(x) ((stbi_uc) ((x) >> 4))
+
+static stbi_uc *stbi__resample_row_hv_2(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
+{
+   // need to generate 2x2 samples for every one in input
+   int i,t0,t1;
+   if (w == 1) {
+      out[0] = out[1] = stbi__div4(3*in_near[0] + in_far[0] + 2);
+      return out;
+   }
+
+   t1 = 3*in_near[0] + in_far[0];
+   out[0] = stbi__div4(t1+2);
+   for (i=1; i < w; ++i) {
+      t0 = t1;
+      t1 = 3*in_near[i]+in_far[i];
+      out[i*2-1] = stbi__div16(3*t0 + t1 + 8);
+      out[i*2  ] = stbi__div16(3*t1 + t0 + 8);
+   }
+   out[w*2-1] = stbi__div4(t1+2);
+
+   STBI_NOTUSED(hs);
+
+   return out;
+}
+
+#if defined(STBI_SSE2) || defined(STBI_NEON)
+static stbi_uc *stbi__resample_row_hv_2_simd(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
+{
+   // need to generate 2x2 samples for every one in input
+   int i=0,t0,t1;
+
+   if (w == 1) {
+      out[0] = out[1] = stbi__div4(3*in_near[0] + in_far[0] + 2);
+      return out;
+   }
+
+   t1 = 3*in_near[0] + in_far[0];
+   // process groups of 8 pixels for as long as we can.
+   // note we can't handle the last pixel in a row in this loop
+   // because we need to handle the filter boundary conditions.
+   for (; i < ((w-1) & ~7); i += 8) {
+#if defined(STBI_SSE2)
+      // load and perform the vertical filtering pass
+      // this uses 3*x + y = 4*x + (y - x)
+      __m128i zero  = _mm_setzero_si128();
+      __m128i farb  = _mm_loadl_epi64((__m128i *) (in_far + i));
+      __m128i nearb = _mm_loadl_epi64((__m128i *) (in_near + i));
+      __m128i farw  = _mm_unpacklo_epi8(farb, zero);
+      __m128i nearw = _mm_unpacklo_epi8(nearb, zero);
+      __m128i diff  = _mm_sub_epi16(farw, nearw);
+      __m128i nears = _mm_slli_epi16(nearw, 2);
+      __m128i curr  = _mm_add_epi16(nears, diff); // current row
+
+      // horizontal filter works the same based on shifted vers of current
+      // row. "prev" is current row shifted right by 1 pixel; we need to
+      // insert the previous pixel value (from t1).
+      // "next" is current row shifted left by 1 pixel, with first pixel
+      // of next block of 8 pixels added in.
+      __m128i prv0 = _mm_slli_si128(curr, 2);
+      __m128i nxt0 = _mm_srli_si128(curr, 2);
+      __m128i prev = _mm_insert_epi16(prv0, t1, 0);
+      __m128i next = _mm_insert_epi16(nxt0, 3*in_near[i+8] + in_far[i+8], 7);
+
+      // horizontal filter, polyphase implementation since it's convenient:
+      // even pixels = 3*cur + prev = cur*4 + (prev - cur)
+      // odd  pixels = 3*cur + next = cur*4 + (next - cur)
+      // note the shared term.
+      __m128i bias  = _mm_set1_epi16(8);
+      __m128i curs = _mm_slli_epi16(curr, 2);
+      __m128i prvd = _mm_sub_epi16(prev, curr);
+      __m128i nxtd = _mm_sub_epi16(next, curr);
+      __m128i curb = _mm_add_epi16(curs, bias);
+      __m128i even = _mm_add_epi16(prvd, curb);
+      __m128i odd  = _mm_add_epi16(nxtd, curb);
+
+      // interleave even and odd pixels, then undo scaling.
+      __m128i int0 = _mm_unpacklo_epi16(even, odd);
+      __m128i int1 = _mm_unpackhi_epi16(even, odd);
+      __m128i de0  = _mm_srli_epi16(int0, 4);
+      __m128i de1  = _mm_srli_epi16(int1, 4);
+
+      // pack and write output
+      __m128i outv = _mm_packus_epi16(de0, de1);
+      _mm_storeu_si128((__m128i *) (out + i*2), outv);
+#elif defined(STBI_NEON)
+      // load and perform the vertical filtering pass
+      // this uses 3*x + y = 4*x + (y - x)
+      uint8x8_t farb  = vld1_u8(in_far + i);
+      uint8x8_t nearb = vld1_u8(in_near + i);
+      int16x8_t diff  = vreinterpretq_s16_u16(vsubl_u8(farb, nearb));
+      int16x8_t nears = vreinterpretq_s16_u16(vshll_n_u8(nearb, 2));
+      int16x8_t curr  = vaddq_s16(nears, diff); // current row
+
+      // horizontal filter works the same based on shifted vers of current
+      // row. "prev" is current row shifted right by 1 pixel; we need to
+      // insert the previous pixel value (from t1).
+      // "next" is current row shifted left by 1 pixel, with first pixel
+      // of next block of 8 pixels added in.
+      int16x8_t prv0 = vextq_s16(curr, curr, 7);
+      int16x8_t nxt0 = vextq_s16(curr, curr, 1);
+      int16x8_t prev = vsetq_lane_s16(t1, prv0, 0);
+      int16x8_t next = vsetq_lane_s16(3*in_near[i+8] + in_far[i+8], nxt0, 7);
+
+      // horizontal filter, polyphase implementation since it's convenient:
+      // even pixels = 3*cur + prev = cur*4 + (prev - cur)
+      // odd  pixels = 3*cur + next = cur*4 + (next - cur)
+      // note the shared term.
+      int16x8_t curs = vshlq_n_s16(curr, 2);
+      int16x8_t prvd = vsubq_s16(prev, curr);
+      int16x8_t nxtd = vsubq_s16(next, curr);
+      int16x8_t even = vaddq_s16(curs, prvd);
+      int16x8_t odd  = vaddq_s16(curs, nxtd);
+
+      // undo scaling and round, then store with even/odd phases interleaved
+      uint8x8x2_t o;
+      o.val[0] = vqrshrun_n_s16(even, 4);
+      o.val[1] = vqrshrun_n_s16(odd,  4);
+      vst2_u8(out + i*2, o);
+#endif
+
+      // "previous" value for next iter
+      t1 = 3*in_near[i+7] + in_far[i+7];
+   }
+
+   t0 = t1;
+   t1 = 3*in_near[i] + in_far[i];
+   out[i*2] = stbi__div16(3*t1 + t0 + 8);
+
+   for (++i; i < w; ++i) {
+      t0 = t1;
+      t1 = 3*in_near[i]+in_far[i];
+      out[i*2-1] = stbi__div16(3*t0 + t1 + 8);
+      out[i*2  ] = stbi__div16(3*t1 + t0 + 8);
+   }
+   out[w*2-1] = stbi__div4(t1+2);
+
+   STBI_NOTUSED(hs);
+
+   return out;
+}
+#endif
+
+static stbi_uc *stbi__resample_row_generic(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
+{
+   // resample with nearest-neighbor
+   int i,j;
+   STBI_NOTUSED(in_far);
+   for (i=0; i < w; ++i)
+      for (j=0; j < hs; ++j)
+         out[i*hs+j] = in_near[i];
+   return out;
+}
+
+// this is a reduced-precision calculation of YCbCr-to-RGB introduced
+// to make sure the code produces the same results in both SIMD and scalar
+#define stbi__float2fixed(x)  (((int) ((x) * 4096.0f + 0.5f)) << 8)
+static void stbi__YCbCr_to_RGB_row(stbi_uc *out, const stbi_uc *y, const stbi_uc *pcb, const stbi_uc *pcr, int count, int step)
+{
+   int i;
+   for (i=0; i < count; ++i) {
+      int y_fixed = (y[i] << 20) + (1<<19); // rounding
+      int r,g,b;
+      int cr = pcr[i] - 128;
+      int cb = pcb[i] - 128;
+      r = y_fixed +  cr* stbi__float2fixed(1.40200f);
+      g = y_fixed + (cr*-stbi__float2fixed(0.71414f)) + ((cb*-stbi__float2fixed(0.34414f)) & 0xffff0000);
+      b = y_fixed                                     +   cb* stbi__float2fixed(1.77200f);
+      r >>= 20;
+      g >>= 20;
+      b >>= 20;
+      if ((unsigned) r > 255) { if (r < 0) r = 0; else r = 255; }
+      if ((unsigned) g > 255) { if (g < 0) g = 0; else g = 255; }
+      if ((unsigned) b > 255) { if (b < 0) b = 0; else b = 255; }
+      out[0] = (stbi_uc)r;
+      out[1] = (stbi_uc)g;
+      out[2] = (stbi_uc)b;
+      out[3] = 255;
+      out += step;
+   }
+}
+
+#if defined(STBI_SSE2) || defined(STBI_NEON)
+static void stbi__YCbCr_to_RGB_simd(stbi_uc *out, stbi_uc const *y, stbi_uc const *pcb, stbi_uc const *pcr, int count, int step)
+{
+   int i = 0;
+
+#ifdef STBI_SSE2
+   // step == 3 is pretty ugly on the final interleave, and i'm not convinced
+   // it's useful in practice (you wouldn't use it for textures, for example).
+   // so just accelerate step == 4 case.
+   if (step == 4) {
+      // this is a fairly straightforward implementation and not super-optimized.
+      __m128i signflip  = _mm_set1_epi8(-0x80);
+      __m128i cr_const0 = _mm_set1_epi16(   (short) ( 1.40200f*4096.0f+0.5f));
+      __m128i cr_const1 = _mm_set1_epi16( - (short) ( 0.71414f*4096.0f+0.5f));
+      __m128i cb_const0 = _mm_set1_epi16( - (short) ( 0.34414f*4096.0f+0.5f));
+      __m128i cb_const1 = _mm_set1_epi16(   (short) ( 1.77200f*4096.0f+0.5f));
+      __m128i y_bias = _mm_set1_epi8((char) (unsigned char) 128);
+      __m128i xw = _mm_set1_epi16(255); // alpha channel
+
+      for (; i+7 < count; i += 8) {
+         // load
+         __m128i y_bytes = _mm_loadl_epi64((__m128i *) (y+i));
+         __m128i cr_bytes = _mm_loadl_epi64((__m128i *) (pcr+i));
+         __m128i cb_bytes = _mm_loadl_epi64((__m128i *) (pcb+i));
+         __m128i cr_biased = _mm_xor_si128(cr_bytes, signflip); // -128
+         __m128i cb_biased = _mm_xor_si128(cb_bytes, signflip); // -128
+
+         // unpack to short (and left-shift cr, cb by 8)
+         __m128i yw  = _mm_unpacklo_epi8(y_bias, y_bytes);
+         __m128i crw = _mm_unpacklo_epi8(_mm_setzero_si128(), cr_biased);
+         __m128i cbw = _mm_unpacklo_epi8(_mm_setzero_si128(), cb_biased);
+
+         // color transform
+         __m128i yws = _mm_srli_epi16(yw, 4);
+         __m128i cr0 = _mm_mulhi_epi16(cr_const0, crw);
+         __m128i cb0 = _mm_mulhi_epi16(cb_const0, cbw);
+         __m128i cb1 = _mm_mulhi_epi16(cbw, cb_const1);
+         __m128i cr1 = _mm_mulhi_epi16(crw, cr_const1);
+         __m128i rws = _mm_add_epi16(cr0, yws);
+         __m128i gwt = _mm_add_epi16(cb0, yws);
+         __m128i bws = _mm_add_epi16(yws, cb1);
+         __m128i gws = _mm_add_epi16(gwt, cr1);
+
+         // descale
+         __m128i rw = _mm_srai_epi16(rws, 4);
+         __m128i bw = _mm_srai_epi16(bws, 4);
+         __m128i gw = _mm_srai_epi16(gws, 4);
+
+         // back to byte, set up for transpose
+         __m128i brb = _mm_packus_epi16(rw, bw);
+         __m128i gxb = _mm_packus_epi16(gw, xw);
+
+         // transpose to interleave channels
+         __m128i t0 = _mm_unpacklo_epi8(brb, gxb);
+         __m128i t1 = _mm_unpackhi_epi8(brb, gxb);
+         __m128i o0 = _mm_unpacklo_epi16(t0, t1);
+         __m128i o1 = _mm_unpackhi_epi16(t0, t1);
+
+         // store
+         _mm_storeu_si128((__m128i *) (out + 0), o0);
+         _mm_storeu_si128((__m128i *) (out + 16), o1);
+         out += 32;
+      }
+   }
+#endif
+
+#ifdef STBI_NEON
+   // in this version, step=3 support would be easy to add. but is there demand?
+   if (step == 4) {
+      // this is a fairly straightforward implementation and not super-optimized.
+      uint8x8_t signflip = vdup_n_u8(0x80);
+      int16x8_t cr_const0 = vdupq_n_s16(   (short) ( 1.40200f*4096.0f+0.5f));
+      int16x8_t cr_const1 = vdupq_n_s16( - (short) ( 0.71414f*4096.0f+0.5f));
+      int16x8_t cb_const0 = vdupq_n_s16( - (short) ( 0.34414f*4096.0f+0.5f));
+      int16x8_t cb_const1 = vdupq_n_s16(   (short) ( 1.77200f*4096.0f+0.5f));
+
+      for (; i+7 < count; i += 8) {
+         // load
+         uint8x8_t y_bytes  = vld1_u8(y + i);
+         uint8x8_t cr_bytes = vld1_u8(pcr + i);
+         uint8x8_t cb_bytes = vld1_u8(pcb + i);
+         int8x8_t cr_biased = vreinterpret_s8_u8(vsub_u8(cr_bytes, signflip));
+         int8x8_t cb_biased = vreinterpret_s8_u8(vsub_u8(cb_bytes, signflip));
+
+         // expand to s16
+         int16x8_t yws = vreinterpretq_s16_u16(vshll_n_u8(y_bytes, 4));
+         int16x8_t crw = vshll_n_s8(cr_biased, 7);
+         int16x8_t cbw = vshll_n_s8(cb_biased, 7);
+
+         // color transform
+         int16x8_t cr0 = vqdmulhq_s16(crw, cr_const0);
+         int16x8_t cb0 = vqdmulhq_s16(cbw, cb_const0);
+         int16x8_t cr1 = vqdmulhq_s16(crw, cr_const1);
+         int16x8_t cb1 = vqdmulhq_s16(cbw, cb_const1);
+         int16x8_t rws = vaddq_s16(yws, cr0);
+         int16x8_t gws = vaddq_s16(vaddq_s16(yws, cb0), cr1);
+         int16x8_t bws = vaddq_s16(yws, cb1);
+
+         // undo scaling, round, convert to byte
+         uint8x8x4_t o;
+         o.val[0] = vqrshrun_n_s16(rws, 4);
+         o.val[1] = vqrshrun_n_s16(gws, 4);
+         o.val[2] = vqrshrun_n_s16(bws, 4);
+         o.val[3] = vdup_n_u8(255);
+
+         // store, interleaving r/g/b/a
+         vst4_u8(out, o);
+         out += 8*4;
+      }
+   }
+#endif
+
+   for (; i < count; ++i) {
+      int y_fixed = (y[i] << 20) + (1<<19); // rounding
+      int r,g,b;
+      int cr = pcr[i] - 128;
+      int cb = pcb[i] - 128;
+      r = y_fixed + cr* stbi__float2fixed(1.40200f);
+      g = y_fixed + cr*-stbi__float2fixed(0.71414f) + ((cb*-stbi__float2fixed(0.34414f)) & 0xffff0000);
+      b = y_fixed                                   +   cb* stbi__float2fixed(1.77200f);
+      r >>= 20;
+      g >>= 20;
+      b >>= 20;
+      if ((unsigned) r > 255) { if (r < 0) r = 0; else r = 255; }
+      if ((unsigned) g > 255) { if (g < 0) g = 0; else g = 255; }
+      if ((unsigned) b > 255) { if (b < 0) b = 0; else b = 255; }
+      out[0] = (stbi_uc)r;
+      out[1] = (stbi_uc)g;
+      out[2] = (stbi_uc)b;
+      out[3] = 255;
+      out += step;
+   }
+}
+#endif
+
+// set up the kernels
+static void stbi__setup_jpeg(stbi__jpeg *j)
+{
+   j->idct_block_kernel = stbi__idct_block;
+   j->YCbCr_to_RGB_kernel = stbi__YCbCr_to_RGB_row;
+   j->resample_row_hv_2_kernel = stbi__resample_row_hv_2;
+
+#ifdef STBI_SSE2
+   if (stbi__sse2_available()) {
+      j->idct_block_kernel = stbi__idct_simd;
+      j->YCbCr_to_RGB_kernel = stbi__YCbCr_to_RGB_simd;
+      j->resample_row_hv_2_kernel = stbi__resample_row_hv_2_simd;
+   }
+#endif
+
+#ifdef STBI_NEON
+   j->idct_block_kernel = stbi__idct_simd;
+   j->YCbCr_to_RGB_kernel = stbi__YCbCr_to_RGB_simd;
+   j->resample_row_hv_2_kernel = stbi__resample_row_hv_2_simd;
+#endif
+}
+
+// clean up the temporary component buffers
+static void stbi__cleanup_jpeg(stbi__jpeg *j)
+{
+   stbi__free_jpeg_components(j, j->s->img_n, 0);
+}
+
+typedef struct
+{
+   resample_row_func resample;
+   stbi_uc *line0,*line1;
+   int hs,vs;   // expansion factor in each axis
+   int w_lores; // horizontal pixels pre-expansion
+   int ystep;   // how far through vertical expansion we are
+   int ypos;    // which pre-expansion row we're on
+} stbi__resample;
+
+// fast 0..255 * 0..255 => 0..255 rounded multiplication
+static stbi_uc stbi__blinn_8x8(stbi_uc x, stbi_uc y)
+{
+   unsigned int t = x*y + 128;
+   return (stbi_uc) ((t + (t >>8)) >> 8);
+}
+
+static stbi_uc *load_jpeg_image(stbi__jpeg *z, int *out_x, int *out_y, int *comp, int req_comp)
+{
+   int n, decode_n, is_rgb;
+   z->s->img_n = 0; // make stbi__cleanup_jpeg safe
+
+   // validate req_comp
+   if (req_comp < 0 || req_comp > 4) return stbi__errpuc("bad req_comp", "Internal error");
+
+   // load a jpeg image from whichever source, but leave in YCbCr format
+   if (!stbi__decode_jpeg_image(z)) { stbi__cleanup_jpeg(z); return NULL; }
+
+   // determine actual number of components to generate
+   n = req_comp ? req_comp : z->s->img_n >= 3 ? 3 : 1;
+
+   is_rgb = z->s->img_n == 3 && (z->rgb == 3 || (z->app14_color_transform == 0 && !z->jfif));
+
+   if (z->s->img_n == 3 && n < 3 && !is_rgb)
+      decode_n = 1;
+   else
+      decode_n = z->s->img_n;
+
+   // resample and color-convert
+   {
+      int k;
+      unsigned int i,j;
+      stbi_uc *output;
+      stbi_uc *coutput[4];
+
+      stbi__resample res_comp[4];
+
+      for (k=0; k < decode_n; ++k) {
+         stbi__resample *r = &res_comp[k];
+
+         // allocate line buffer big enough for upsampling off the edges
+         // with upsample factor of 4
+         z->img_comp[k].linebuf = (stbi_uc *) stbi__malloc(z->s->img_x + 3);
+         if (!z->img_comp[k].linebuf) { stbi__cleanup_jpeg(z); return stbi__errpuc("outofmem", "Out of memory"); }
+
+         r->hs      = z->img_h_max / z->img_comp[k].h;
+         r->vs      = z->img_v_max / z->img_comp[k].v;
+         r->ystep   = r->vs >> 1;
+         r->w_lores = (z->s->img_x + r->hs-1) / r->hs;
+         r->ypos    = 0;
+         r->line0   = r->line1 = z->img_comp[k].data;
+
+         if      (r->hs == 1 && r->vs == 1) r->resample = resample_row_1;
+         else if (r->hs == 1 && r->vs == 2) r->resample = stbi__resample_row_v_2;
+         else if (r->hs == 2 && r->vs == 1) r->resample = stbi__resample_row_h_2;
+         else if (r->hs == 2 && r->vs == 2) r->resample = z->resample_row_hv_2_kernel;
+         else                               r->resample = stbi__resample_row_generic;
+      }
+
+      // can't error after this so, this is safe
+      output = (stbi_uc *) stbi__malloc_mad3(n, z->s->img_x, z->s->img_y, 1);
+      if (!output) { stbi__cleanup_jpeg(z); return stbi__errpuc("outofmem", "Out of memory"); }
+
+      // now go ahead and resample
+      for (j=0; j < z->s->img_y; ++j) {
+         stbi_uc *out = output + n * z->s->img_x * j;
+         for (k=0; k < decode_n; ++k) {
+            stbi__resample *r = &res_comp[k];
+            int y_bot = r->ystep >= (r->vs >> 1);
+            coutput[k] = r->resample(z->img_comp[k].linebuf,
+                                     y_bot ? r->line1 : r->line0,
+                                     y_bot ? r->line0 : r->line1,
+                                     r->w_lores, r->hs);
+            if (++r->ystep >= r->vs) {
+               r->ystep = 0;
+               r->line0 = r->line1;
+               if (++r->ypos < z->img_comp[k].y)
+                  r->line1 += z->img_comp[k].w2;
+            }
+         }
+         if (n >= 3) {
+            stbi_uc *y = coutput[0];
+            if (z->s->img_n == 3) {
+               if (is_rgb) {
+                  for (i=0; i < z->s->img_x; ++i) {
+                     out[0] = y[i];
+                     out[1] = coutput[1][i];
+                     out[2] = coutput[2][i];
+                     out[3] = 255;
+                     out += n;
+                  }
+               } else {
+                  z->YCbCr_to_RGB_kernel(out, y, coutput[1], coutput[2], z->s->img_x, n);
+               }
+            } else if (z->s->img_n == 4) {
+               if (z->app14_color_transform == 0) { // CMYK
+                  for (i=0; i < z->s->img_x; ++i) {
+                     stbi_uc k = coutput[3][i];
+                     out[0] = stbi__blinn_8x8(coutput[0][i], k);
+                     out[1] = stbi__blinn_8x8(coutput[1][i], k);
+                     out[2] = stbi__blinn_8x8(coutput[2][i], k);
+                     out[3] = 255;
+                     out += n;
+                  }
+               } else if (z->app14_color_transform == 2) { // YCCK
+                  z->YCbCr_to_RGB_kernel(out, y, coutput[1], coutput[2], z->s->img_x, n);
+                  for (i=0; i < z->s->img_x; ++i) {
+                     stbi_uc k = coutput[3][i];
+                     out[0] = stbi__blinn_8x8(255 - out[0], k);
+                     out[1] = stbi__blinn_8x8(255 - out[1], k);
+                     out[2] = stbi__blinn_8x8(255 - out[2], k);
+                     out += n;
+                  }
+               } else { // YCbCr + alpha?  Ignore the fourth channel for now
+                  z->YCbCr_to_RGB_kernel(out, y, coutput[1], coutput[2], z->s->img_x, n);
+               }
+            } else
+               for (i=0; i < z->s->img_x; ++i) {
+                  out[0] = out[1] = out[2] = y[i];
+                  out[3] = 255; // not used if n==3
+                  out += n;
+               }
+         } else {
+            if (is_rgb) {
+               if (n == 1)
+                  for (i=0; i < z->s->img_x; ++i)
+                     *out++ = stbi__compute_y(coutput[0][i], coutput[1][i], coutput[2][i]);
+               else {
+                  for (i=0; i < z->s->img_x; ++i, out += 2) {
+                     out[0] = stbi__compute_y(coutput[0][i], coutput[1][i], coutput[2][i]);
+                     out[1] = 255;
+                  }
+               }
+            } else if (z->s->img_n == 4 && z->app14_color_transform == 0) {
+               for (i=0; i < z->s->img_x; ++i) {
+                  stbi_uc k = coutput[3][i];
+                  stbi_uc r = stbi__blinn_8x8(coutput[0][i], k);
+                  stbi_uc g = stbi__blinn_8x8(coutput[1][i], k);
+                  stbi_uc b = stbi__blinn_8x8(coutput[2][i], k);
+                  out[0] = stbi__compute_y(r, g, b);
+                  out[1] = 255;
+                  out += n;
+               }
+            } else if (z->s->img_n == 4 && z->app14_color_transform == 2) {
+               for (i=0; i < z->s->img_x; ++i) {
+                  out[0] = stbi__blinn_8x8(255 - coutput[0][i], coutput[3][i]);
+                  out[1] = 255;
+                  out += n;
+               }
+            } else {
+               stbi_uc *y = coutput[0];
+               if (n == 1)
+                  for (i=0; i < z->s->img_x; ++i) out[i] = y[i];
+               else
+                  for (i=0; i < z->s->img_x; ++i) *out++ = y[i], *out++ = 255;
+            }
+         }
+      }
+      stbi__cleanup_jpeg(z);
+      *out_x = z->s->img_x;
+      *out_y = z->s->img_y;
+      if (comp) *comp = z->s->img_n >= 3 ? 3 : 1; // report original components, not output
+      return output;
+   }
+}
+
+static void *stbi__jpeg_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri)
+{
+   unsigned char* result;
+   stbi__jpeg* j = (stbi__jpeg*) stbi__malloc(sizeof(stbi__jpeg));
+   STBI_NOTUSED(ri);
+   j->s = s;
+   stbi__setup_jpeg(j);
+   result = load_jpeg_image(j, x,y,comp,req_comp);
+   STBI_FREE(j);
+   return result;
+}
+
+static int stbi__jpeg_test(stbi__context *s)
+{
+   int r;
+   stbi__jpeg* j = (stbi__jpeg*)stbi__malloc(sizeof(stbi__jpeg));
+   j->s = s;
+   stbi__setup_jpeg(j);
+   r = stbi__decode_jpeg_header(j, STBI__SCAN_type);
+   stbi__rewind(s);
+   STBI_FREE(j);
+   return r;
+}
+
+static int stbi__jpeg_info_raw(stbi__jpeg *j, int *x, int *y, int *comp)
+{
+   if (!stbi__decode_jpeg_header(j, STBI__SCAN_header)) {
+      stbi__rewind( j->s );
+      return 0;
+   }
+   if (x) *x = j->s->img_x;
+   if (y) *y = j->s->img_y;
+   if (comp) *comp = j->s->img_n >= 3 ? 3 : 1;
+   return 1;
+}
+
+static int stbi__jpeg_info(stbi__context *s, int *x, int *y, int *comp)
+{
+   int result;
+   stbi__jpeg* j = (stbi__jpeg*) (stbi__malloc(sizeof(stbi__jpeg)));
+   j->s = s;
+   result = stbi__jpeg_info_raw(j, x, y, comp);
+   STBI_FREE(j);
+   return result;
+}
+#endif
+
+// public domain zlib decode    v0.2  Sean Barrett 2006-11-18
+//    simple implementation
+//      - all input must be provided in an upfront buffer
+//      - all output is written to a single output buffer (can malloc/realloc)
+//    performance
+//      - fast huffman
+
+#ifndef STBI_NO_ZLIB
+
+// fast-way is faster to check than jpeg huffman, but slow way is slower
+#define STBI__ZFAST_BITS  9 // accelerate all cases in default tables
+#define STBI__ZFAST_MASK  ((1 << STBI__ZFAST_BITS) - 1)
+
+// zlib-style huffman encoding
+// (jpegs packs from left, zlib from right, so can't share code)
+typedef struct
+{
+   stbi__uint16 fast[1 << STBI__ZFAST_BITS];
+   stbi__uint16 firstcode[16];
+   int maxcode[17];
+   stbi__uint16 firstsymbol[16];
+   stbi_uc  size[288];
+   stbi__uint16 value[288];
+} stbi__zhuffman;
+
+stbi_inline static int stbi__bitreverse16(int n)
+{
+  n = ((n & 0xAAAA) >>  1) | ((n & 0x5555) << 1);
+  n = ((n & 0xCCCC) >>  2) | ((n & 0x3333) << 2);
+  n = ((n & 0xF0F0) >>  4) | ((n & 0x0F0F) << 4);
+  n = ((n & 0xFF00) >>  8) | ((n & 0x00FF) << 8);
+  return n;
+}
+
+stbi_inline static int stbi__bit_reverse(int v, int bits)
+{
+   STBI_ASSERT(bits <= 16);
+   // to bit reverse n bits, reverse 16 and shift
+   // e.g. 11 bits, bit reverse and shift away 5
+   return stbi__bitreverse16(v) >> (16-bits);
+}
+
+static int stbi__zbuild_huffman(stbi__zhuffman *z, const stbi_uc *sizelist, int num)
+{
+   int i,k=0;
+   int code, next_code[16], sizes[17];
+
+   // DEFLATE spec for generating codes
+   memset(sizes, 0, sizeof(sizes));
+   memset(z->fast, 0, sizeof(z->fast));
+   for (i=0; i < num; ++i)
+      ++sizes[sizelist[i]];
+   sizes[0] = 0;
+   for (i=1; i < 16; ++i)
+      if (sizes[i] > (1 << i))
+         return stbi__err("bad sizes", "Corrupt PNG");
+   code = 0;
+   for (i=1; i < 16; ++i) {
+      next_code[i] = code;
+      z->firstcode[i] = (stbi__uint16) code;
+      z->firstsymbol[i] = (stbi__uint16) k;
+      code = (code + sizes[i]);
+      if (sizes[i])
+         if (code-1 >= (1 << i)) return stbi__err("bad codelengths","Corrupt PNG");
+      z->maxcode[i] = code << (16-i); // preshift for inner loop
+      code <<= 1;
+      k += sizes[i];
+   }
+   z->maxcode[16] = 0x10000; // sentinel
+   for (i=0; i < num; ++i) {
+      int s = sizelist[i];
+      if (s) {
+         int c = next_code[s] - z->firstcode[s] + z->firstsymbol[s];
+         stbi__uint16 fastv = (stbi__uint16) ((s << 9) | i);
+         z->size [c] = (stbi_uc     ) s;
+         z->value[c] = (stbi__uint16) i;
+         if (s <= STBI__ZFAST_BITS) {
+            int j = stbi__bit_reverse(next_code[s],s);
+            while (j < (1 << STBI__ZFAST_BITS)) {
+               z->fast[j] = fastv;
+               j += (1 << s);
+            }
+         }
+         ++next_code[s];
+      }
+   }
+   return 1;
+}
+
+// zlib-from-memory implementation for PNG reading
+//    because PNG allows splitting the zlib stream arbitrarily,
+//    and it's annoying structurally to have PNG call ZLIB call PNG,
+//    we require PNG read all the IDATs and combine them into a single
+//    memory buffer
+
+typedef struct
+{
+   stbi_uc *zbuffer, *zbuffer_end;
+   int num_bits;
+   stbi__uint32 code_buffer;
+
+   char *zout;
+   char *zout_start;
+   char *zout_end;
+   int   z_expandable;
+
+   stbi__zhuffman z_length, z_distance;
+} stbi__zbuf;
+
+stbi_inline static stbi_uc stbi__zget8(stbi__zbuf *z)
+{
+   if (z->zbuffer >= z->zbuffer_end) return 0;
+   return *z->zbuffer++;
+}
+
+static void stbi__fill_bits(stbi__zbuf *z)
+{
+   do {
+      STBI_ASSERT(z->code_buffer < (1U << z->num_bits));
+      z->code_buffer |= (unsigned int) stbi__zget8(z) << z->num_bits;
+      z->num_bits += 8;
+   } while (z->num_bits <= 24);
+}
+
+stbi_inline static unsigned int stbi__zreceive(stbi__zbuf *z, int n)
+{
+   unsigned int k;
+   if (z->num_bits < n) stbi__fill_bits(z);
+   k = z->code_buffer & ((1 << n) - 1);
+   z->code_buffer >>= n;
+   z->num_bits -= n;
+   return k;
+}
+
+static int stbi__zhuffman_decode_slowpath(stbi__zbuf *a, stbi__zhuffman *z)
+{
+   int b,s,k;
+   // not resolved by fast table, so compute it the slow way
+   // use jpeg approach, which requires MSbits at top
+   k = stbi__bit_reverse(a->code_buffer, 16);
+   for (s=STBI__ZFAST_BITS+1; ; ++s)
+      if (k < z->maxcode[s])
+         break;
+   if (s == 16) return -1; // invalid code!
+   // code size is s, so:
+   b = (k >> (16-s)) - z->firstcode[s] + z->firstsymbol[s];
+   STBI_ASSERT(z->size[b] == s);
+   a->code_buffer >>= s;
+   a->num_bits -= s;
+   return z->value[b];
+}
+
+stbi_inline static int stbi__zhuffman_decode(stbi__zbuf *a, stbi__zhuffman *z)
+{
+   int b,s;
+   if (a->num_bits < 16) stbi__fill_bits(a);
+   b = z->fast[a->code_buffer & STBI__ZFAST_MASK];
+   if (b) {
+      s = b >> 9;
+      a->code_buffer >>= s;
+      a->num_bits -= s;
+      return b & 511;
+   }
+   return stbi__zhuffman_decode_slowpath(a, z);
+}
+
+static int stbi__zexpand(stbi__zbuf *z, char *zout, int n)  // need to make room for n bytes
+{
+   char *q;
+   int cur, limit, old_limit;
+   z->zout = zout;
+   if (!z->z_expandable) return stbi__err("output buffer limit","Corrupt PNG");
+   cur   = (int) (z->zout     - z->zout_start);
+   limit = old_limit = (int) (z->zout_end - z->zout_start);
+   while (cur + n > limit)
+      limit *= 2;
+   q = (char *) STBI_REALLOC_SIZED(z->zout_start, old_limit, limit);
+   STBI_NOTUSED(old_limit);
+   if (q == NULL) return stbi__err("outofmem", "Out of memory");
+   z->zout_start = q;
+   z->zout       = q + cur;
+   z->zout_end   = q + limit;
+   return 1;
+}
+
+static int stbi__zlength_base[31] = {
+   3,4,5,6,7,8,9,10,11,13,
+   15,17,19,23,27,31,35,43,51,59,
+   67,83,99,115,131,163,195,227,258,0,0 };
+
+static int stbi__zlength_extra[31]=
+{ 0,0,0,0,0,0,0,0,1,1,1,1,2,2,2,2,3,3,3,3,4,4,4,4,5,5,5,5,0,0,0 };
+
+static int stbi__zdist_base[32] = { 1,2,3,4,5,7,9,13,17,25,33,49,65,97,129,193,
+257,385,513,769,1025,1537,2049,3073,4097,6145,8193,12289,16385,24577,0,0};
+
+static int stbi__zdist_extra[32] =
+{ 0,0,0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,8,8,9,9,10,10,11,11,12,12,13,13};
+
+static int stbi__parse_huffman_block(stbi__zbuf *a)
+{
+   char *zout = a->zout;
+   for(;;) {
+      int z = stbi__zhuffman_decode(a, &a->z_length);
+      if (z < 256) {
+         if (z < 0) return stbi__err("bad huffman code","Corrupt PNG"); // error in huffman codes
+         if (zout >= a->zout_end) {
+            if (!stbi__zexpand(a, zout, 1)) return 0;
+            zout = a->zout;
+         }
+         *zout++ = (char) z;
+      } else {
+         stbi_uc *p;
+         int len,dist;
+         if (z == 256) {
+            a->zout = zout;
+            return 1;
+         }
+         z -= 257;
+         len = stbi__zlength_base[z];
+         if (stbi__zlength_extra[z]) len += stbi__zreceive(a, stbi__zlength_extra[z]);
+         z = stbi__zhuffman_decode(a, &a->z_distance);
+         if (z < 0) return stbi__err("bad huffman code","Corrupt PNG");
+         dist = stbi__zdist_base[z];
+         if (stbi__zdist_extra[z]) dist += stbi__zreceive(a, stbi__zdist_extra[z]);
+         if (zout - a->zout_start < dist) return stbi__err("bad dist","Corrupt PNG");
+         if (zout + len > a->zout_end) {
+            if (!stbi__zexpand(a, zout, len)) return 0;
+            zout = a->zout;
+         }
+         p = (stbi_uc *) (zout - dist);
+         if (dist == 1) { // run of one byte; common in images.
+            stbi_uc v = *p;
+            if (len) { do *zout++ = v; while (--len); }
+         } else {
+            if (len) { do *zout++ = *p++; while (--len); }
+         }
+      }
+   }
+}
+
+static int stbi__compute_huffman_codes(stbi__zbuf *a)
+{
+   static stbi_uc length_dezigzag[19] = { 16,17,18,0,8,7,9,6,10,5,11,4,12,3,13,2,14,1,15 };
+   stbi__zhuffman z_codelength;
+   stbi_uc lencodes[286+32+137];//padding for maximum single op
+   stbi_uc codelength_sizes[19];
+   int i,n;
+
+   int hlit  = stbi__zreceive(a,5) + 257;
+   int hdist = stbi__zreceive(a,5) + 1;
+   int hclen = stbi__zreceive(a,4) + 4;
+   int ntot  = hlit + hdist;
+
+   memset(codelength_sizes, 0, sizeof(codelength_sizes));
+   for (i=0; i < hclen; ++i) {
+      int s = stbi__zreceive(a,3);
+      codelength_sizes[length_dezigzag[i]] = (stbi_uc) s;
+   }
+   if (!stbi__zbuild_huffman(&z_codelength, codelength_sizes, 19)) return 0;
+
+   n = 0;
+   while (n < ntot) {
+      int c = stbi__zhuffman_decode(a, &z_codelength);
+      if (c < 0 || c >= 19) return stbi__err("bad codelengths", "Corrupt PNG");
+      if (c < 16)
+         lencodes[n++] = (stbi_uc) c;
+      else {
+         stbi_uc fill = 0;
+         if (c == 16) {
+            c = stbi__zreceive(a,2)+3;
+            if (n == 0) return stbi__err("bad codelengths", "Corrupt PNG");
+            fill = lencodes[n-1];
+         } else if (c == 17)
+            c = stbi__zreceive(a,3)+3;
+         else {
+            STBI_ASSERT(c == 18);
+            c = stbi__zreceive(a,7)+11;
+         }
+         if (ntot - n < c) return stbi__err("bad codelengths", "Corrupt PNG");
+         memset(lencodes+n, fill, c);
+         n += c;
+      }
+   }
+   if (n != ntot) return stbi__err("bad codelengths","Corrupt PNG");
+   if (!stbi__zbuild_huffman(&a->z_length, lencodes, hlit)) return 0;
+   if (!stbi__zbuild_huffman(&a->z_distance, lencodes+hlit, hdist)) return 0;
+   return 1;
+}
+
+static int stbi__parse_uncompressed_block(stbi__zbuf *a)
+{
+   stbi_uc header[4];
+   int len,nlen,k;
+   if (a->num_bits & 7)
+      stbi__zreceive(a, a->num_bits & 7); // discard
+   // drain the bit-packed data into header
+   k = 0;
+   while (a->num_bits > 0) {
+      header[k++] = (stbi_uc) (a->code_buffer & 255); // suppress MSVC run-time check
+      a->code_buffer >>= 8;
+      a->num_bits -= 8;
+   }
+   STBI_ASSERT(a->num_bits == 0);
+   // now fill header the normal way
+   while (k < 4)
+      header[k++] = stbi__zget8(a);
+   len  = header[1] * 256 + header[0];
+   nlen = header[3] * 256 + header[2];
+   if (nlen != (len ^ 0xffff)) return stbi__err("zlib corrupt","Corrupt PNG");
+   if (a->zbuffer + len > a->zbuffer_end) return stbi__err("read past buffer","Corrupt PNG");
+   if (a->zout + len > a->zout_end)
+      if (!stbi__zexpand(a, a->zout, len)) return 0;
+   memcpy(a->zout, a->zbuffer, len);
+   a->zbuffer += len;
+   a->zout += len;
+   return 1;
+}
+
+static int stbi__parse_zlib_header(stbi__zbuf *a)
+{
+   int cmf   = stbi__zget8(a);
+   int cm    = cmf & 15;
+   /* int cinfo = cmf >> 4; */
+   int flg   = stbi__zget8(a);
+   if ((cmf*256+flg) % 31 != 0) return stbi__err("bad zlib header","Corrupt PNG"); // zlib spec
+   if (flg & 32) return stbi__err("no preset dict","Corrupt PNG"); // preset dictionary not allowed in png
+   if (cm != 8) return stbi__err("bad compression","Corrupt PNG"); // DEFLATE required for png
+   // window = 1 << (8 + cinfo)... but who cares, we fully buffer output
+   return 1;
+}
+
+static const stbi_uc stbi__zdefault_length[288] =
+{
+   8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,
+   8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,
+   8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,
+   8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,
+   8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,
+   9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,
+   9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,
+   9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,
+   7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7, 7,7,7,7,7,7,7,7,8,8,8,8,8,8,8,8
+};
+static const stbi_uc stbi__zdefault_distance[32] =
+{
+   5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5
+};
+/*
+Init algorithm:
+{
+   int i;   // use <= to match clearly with spec
+   for (i=0; i <= 143; ++i)     stbi__zdefault_length[i]   = 8;
+   for (   ; i <= 255; ++i)     stbi__zdefault_length[i]   = 9;
+   for (   ; i <= 279; ++i)     stbi__zdefault_length[i]   = 7;
+   for (   ; i <= 287; ++i)     stbi__zdefault_length[i]   = 8;
+
+   for (i=0; i <=  31; ++i)     stbi__zdefault_distance[i] = 5;
+}
+*/
+
+static int stbi__parse_zlib(stbi__zbuf *a, int parse_header)
+{
+   int final, type;
+   if (parse_header)
+      if (!stbi__parse_zlib_header(a)) return 0;
+   a->num_bits = 0;
+   a->code_buffer = 0;
+   do {
+      final = stbi__zreceive(a,1);
+      type = stbi__zreceive(a,2);
+      if (type == 0) {
+         if (!stbi__parse_uncompressed_block(a)) return 0;
+      } else if (type == 3) {
+         return 0;
+      } else {
+         if (type == 1) {
+            // use fixed code lengths
+            if (!stbi__zbuild_huffman(&a->z_length  , stbi__zdefault_length  , 288)) return 0;
+            if (!stbi__zbuild_huffman(&a->z_distance, stbi__zdefault_distance,  32)) return 0;
+         } else {
+            if (!stbi__compute_huffman_codes(a)) return 0;
+         }
+         if (!stbi__parse_huffman_block(a)) return 0;
+      }
+   } while (!final);
+   return 1;
+}
+
+static int stbi__do_zlib(stbi__zbuf *a, char *obuf, int olen, int exp, int parse_header)
+{
+   a->zout_start = obuf;
+   a->zout       = obuf;
+   a->zout_end   = obuf + olen;
+   a->z_expandable = exp;
+
+   return stbi__parse_zlib(a, parse_header);
+}
+
+STBIDEF char *stbi_zlib_decode_malloc_guesssize(const char *buffer, int len, int initial_size, int *outlen)
+{
+   stbi__zbuf a;
+   char *p = (char *) stbi__malloc(initial_size);
+   if (p == NULL) return NULL;
+   a.zbuffer = (stbi_uc *) buffer;
+   a.zbuffer_end = (stbi_uc *) buffer + len;
+   if (stbi__do_zlib(&a, p, initial_size, 1, 1)) {
+      if (outlen) *outlen = (int) (a.zout - a.zout_start);
+      return a.zout_start;
+   } else {
+      STBI_FREE(a.zout_start);
+      return NULL;
+   }
+}
+
+STBIDEF char *stbi_zlib_decode_malloc(char const *buffer, int len, int *outlen)
+{
+   return stbi_zlib_decode_malloc_guesssize(buffer, len, 16384, outlen);
+}
+
+STBIDEF char *stbi_zlib_decode_malloc_guesssize_headerflag(const char *buffer, int len, int initial_size, int *outlen, int parse_header)
+{
+   stbi__zbuf a;
+   char *p = (char *) stbi__malloc(initial_size);
+   if (p == NULL) return NULL;
+   a.zbuffer = (stbi_uc *) buffer;
+   a.zbuffer_end = (stbi_uc *) buffer + len;
+   if (stbi__do_zlib(&a, p, initial_size, 1, parse_header)) {
+      if (outlen) *outlen = (int) (a.zout - a.zout_start);
+      return a.zout_start;
+   } else {
+      STBI_FREE(a.zout_start);
+      return NULL;
+   }
+}
+
+STBIDEF int stbi_zlib_decode_buffer(char *obuffer, int olen, char const *ibuffer, int ilen)
+{
+   stbi__zbuf a;
+   a.zbuffer = (stbi_uc *) ibuffer;
+   a.zbuffer_end = (stbi_uc *) ibuffer + ilen;
+   if (stbi__do_zlib(&a, obuffer, olen, 0, 1))
+      return (int) (a.zout - a.zout_start);
+   else
+      return -1;
+}
+
+STBIDEF char *stbi_zlib_decode_noheader_malloc(char const *buffer, int len, int *outlen)
+{
+   stbi__zbuf a;
+   char *p = (char *) stbi__malloc(16384);
+   if (p == NULL) return NULL;
+   a.zbuffer = (stbi_uc *) buffer;
+   a.zbuffer_end = (stbi_uc *) buffer+len;
+   if (stbi__do_zlib(&a, p, 16384, 1, 0)) {
+      if (outlen) *outlen = (int) (a.zout - a.zout_start);
+      return a.zout_start;
+   } else {
+      STBI_FREE(a.zout_start);
+      return NULL;
+   }
+}
+
+STBIDEF int stbi_zlib_decode_noheader_buffer(char *obuffer, int olen, const char *ibuffer, int ilen)
+{
+   stbi__zbuf a;
+   a.zbuffer = (stbi_uc *) ibuffer;
+   a.zbuffer_end = (stbi_uc *) ibuffer + ilen;
+   if (stbi__do_zlib(&a, obuffer, olen, 0, 0))
+      return (int) (a.zout - a.zout_start);
+   else
+      return -1;
+}
+#endif
+
+// public domain "baseline" PNG decoder   v0.10  Sean Barrett 2006-11-18
+//    simple implementation
+//      - only 8-bit samples
+//      - no CRC checking
+//      - allocates lots of intermediate memory
+//        - avoids problem of streaming data between subsystems
+//        - avoids explicit window management
+//    performance
+//      - uses stb_zlib, a PD zlib implementation with fast huffman decoding
+
+#ifndef STBI_NO_PNG
+typedef struct
+{
+   stbi__uint32 length;
+   stbi__uint32 type;
+} stbi__pngchunk;
+
+static stbi__pngchunk stbi__get_chunk_header(stbi__context *s)
+{
+   stbi__pngchunk c;
+   c.length = stbi__get32be(s);
+   c.type   = stbi__get32be(s);
+   return c;
+}
+
+static int stbi__check_png_header(stbi__context *s)
+{
+   static stbi_uc png_sig[8] = { 137,80,78,71,13,10,26,10 };
+   int i;
+   for (i=0; i < 8; ++i)
+      if (stbi__get8(s) != png_sig[i]) return stbi__err("bad png sig","Not a PNG");
+   return 1;
+}
+
+typedef struct
+{
+   stbi__context *s;
+   stbi_uc *idata, *expanded, *out;
+   int depth;
+} stbi__png;
+
+
+enum {
+   STBI__F_none=0,
+   STBI__F_sub=1,
+   STBI__F_up=2,
+   STBI__F_avg=3,
+   STBI__F_paeth=4,
+   // synthetic filters used for first scanline to avoid needing a dummy row of 0s
+   STBI__F_avg_first,
+   STBI__F_paeth_first
+};
+
+static stbi_uc first_row_filter[5] =
+{
+   STBI__F_none,
+   STBI__F_sub,
+   STBI__F_none,
+   STBI__F_avg_first,
+   STBI__F_paeth_first
+};
+
+static int stbi__paeth(int a, int b, int c)
+{
+   int p = a + b - c;
+   int pa = abs(p-a);
+   int pb = abs(p-b);
+   int pc = abs(p-c);
+   if (pa <= pb && pa <= pc) return a;
+   if (pb <= pc) return b;
+   return c;
+}
+
+static stbi_uc stbi__depth_scale_table[9] = { 0, 0xff, 0x55, 0, 0x11, 0,0,0, 0x01 };
+
+// create the png data from post-deflated data
+static int stbi__create_png_image_raw(stbi__png *a, stbi_uc *raw, stbi__uint32 raw_len, int out_n, stbi__uint32 x, stbi__uint32 y, int depth, int color)
+{
+   int bytes = (depth == 16? 2 : 1);
+   stbi__context *s = a->s;
+   stbi__uint32 i,j,stride = x*out_n*bytes;
+   stbi__uint32 img_len, img_width_bytes;
+   int k;
+   int img_n = s->img_n; // copy it into a local for later
+
+   int output_bytes = out_n*bytes;
+   int filter_bytes = img_n*bytes;
+   int width = x;
+
+   STBI_ASSERT(out_n == s->img_n || out_n == s->img_n+1);
+   a->out = (stbi_uc *) stbi__malloc_mad3(x, y, output_bytes, 0); // extra bytes to write off the end into
+   if (!a->out) return stbi__err("outofmem", "Out of memory");
+
+   img_width_bytes = (((img_n * x * depth) + 7) >> 3);
+   img_len = (img_width_bytes + 1) * y;
+   if (s->img_x == x && s->img_y == y) {
+      if (raw_len != img_len) return stbi__err("not enough pixels","Corrupt PNG");
+   } else { // interlaced:
+      if (raw_len < img_len) return stbi__err("not enough pixels","Corrupt PNG");
+   }
+
+   for (j=0; j < y; ++j) {
+      stbi_uc *cur = a->out + stride*j;
+      stbi_uc *prior;
+      int filter = *raw++;
+
+      if (filter > 4)
+         return stbi__err("invalid filter","Corrupt PNG");
+
+      if (depth < 8) {
+         STBI_ASSERT(img_width_bytes <= x);
+         cur += x*out_n - img_width_bytes; // store output to the rightmost img_len bytes, so we can decode in place
+         filter_bytes = 1;
+         width = img_width_bytes;
+      }
+      prior = cur - stride; // bugfix: need to compute this after 'cur +=' computation above
+
+      // if first row, use special filter that doesn't sample previous row
+      if (j == 0) filter = first_row_filter[filter];
+
+      // handle first byte explicitly
+      for (k=0; k < filter_bytes; ++k) {
+         switch (filter) {
+            case STBI__F_none       : cur[k] = raw[k]; break;
+            case STBI__F_sub        : cur[k] = raw[k]; break;
+            case STBI__F_up         : cur[k] = STBI__BYTECAST(raw[k] + prior[k]); break;
+            case STBI__F_avg        : cur[k] = STBI__BYTECAST(raw[k] + (prior[k]>>1)); break;
+            case STBI__F_paeth      : cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(0,prior[k],0)); break;
+            case STBI__F_avg_first  : cur[k] = raw[k]; break;
+            case STBI__F_paeth_first: cur[k] = raw[k]; break;
+         }
+      }
+
+      if (depth == 8) {
+         if (img_n != out_n)
+            cur[img_n] = 255; // first pixel
+         raw += img_n;
+         cur += out_n;
+         prior += out_n;
+      } else if (depth == 16) {
+         if (img_n != out_n) {
+            cur[filter_bytes]   = 255; // first pixel top byte
+            cur[filter_bytes+1] = 255; // first pixel bottom byte
+         }
+         raw += filter_bytes;
+         cur += output_bytes;
+         prior += output_bytes;
+      } else {
+         raw += 1;
+         cur += 1;
+         prior += 1;
+      }
+
+      // this is a little gross, so that we don't switch per-pixel or per-component
+      if (depth < 8 || img_n == out_n) {
+         int nk = (width - 1)*filter_bytes;
+         #define STBI__CASE(f) \
+             case f:     \
+                for (k=0; k < nk; ++k)
+         switch (filter) {
+            // "none" filter turns into a memcpy here; make that explicit.
+            case STBI__F_none:         memcpy(cur, raw, nk); break;
+            STBI__CASE(STBI__F_sub)          { cur[k] = STBI__BYTECAST(raw[k] + cur[k-filter_bytes]); } break;
+            STBI__CASE(STBI__F_up)           { cur[k] = STBI__BYTECAST(raw[k] + prior[k]); } break;
+            STBI__CASE(STBI__F_avg)          { cur[k] = STBI__BYTECAST(raw[k] + ((prior[k] + cur[k-filter_bytes])>>1)); } break;
+            STBI__CASE(STBI__F_paeth)        { cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k-filter_bytes],prior[k],prior[k-filter_bytes])); } break;
+            STBI__CASE(STBI__F_avg_first)    { cur[k] = STBI__BYTECAST(raw[k] + (cur[k-filter_bytes] >> 1)); } break;
+            STBI__CASE(STBI__F_paeth_first)  { cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k-filter_bytes],0,0)); } break;
+         }
+         #undef STBI__CASE
+         raw += nk;
+      } else {
+         STBI_ASSERT(img_n+1 == out_n);
+         #define STBI__CASE(f) \
+             case f:     \
+                for (i=x-1; i >= 1; --i, cur[filter_bytes]=255,raw+=filter_bytes,cur+=output_bytes,prior+=output_bytes) \
+                   for (k=0; k < filter_bytes; ++k)
+         switch (filter) {
+            STBI__CASE(STBI__F_none)         { cur[k] = raw[k]; } break;
+            STBI__CASE(STBI__F_sub)          { cur[k] = STBI__BYTECAST(raw[k] + cur[k- output_bytes]); } break;
+            STBI__CASE(STBI__F_up)           { cur[k] = STBI__BYTECAST(raw[k] + prior[k]); } break;
+            STBI__CASE(STBI__F_avg)          { cur[k] = STBI__BYTECAST(raw[k] + ((prior[k] + cur[k- output_bytes])>>1)); } break;
+            STBI__CASE(STBI__F_paeth)        { cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k- output_bytes],prior[k],prior[k- output_bytes])); } break;
+            STBI__CASE(STBI__F_avg_first)    { cur[k] = STBI__BYTECAST(raw[k] + (cur[k- output_bytes] >> 1)); } break;
+            STBI__CASE(STBI__F_paeth_first)  { cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k- output_bytes],0,0)); } break;
+         }
+         #undef STBI__CASE
+
+         // the loop above sets the high byte of the pixels' alpha, but for
+         // 16 bit png files we also need the low byte set. we'll do that here.
+         if (depth == 16) {
+            cur = a->out + stride*j; // start at the beginning of the row again
+            for (i=0; i < x; ++i,cur+=output_bytes) {
+               cur[filter_bytes+1] = 255;
+            }
+         }
+      }
+   }
+
+   // we make a separate pass to expand bits to pixels; for performance,
+   // this could run two scanlines behind the above code, so it won't
+   // intefere with filtering but will still be in the cache.
+   if (depth < 8) {
+      for (j=0; j < y; ++j) {
+         stbi_uc *cur = a->out + stride*j;
+         stbi_uc *in  = a->out + stride*j + x*out_n - img_width_bytes;
+         // unpack 1/2/4-bit into a 8-bit buffer. allows us to keep the common 8-bit path optimal at minimal cost for 1/2/4-bit
+         // png guarante byte alignment, if width is not multiple of 8/4/2 we'll decode dummy trailing data that will be skipped in the later loop
+         stbi_uc scale = (color == 0) ? stbi__depth_scale_table[depth] : 1; // scale grayscale values to 0..255 range
+
+         // note that the final byte might overshoot and write more data than desired.
+         // we can allocate enough data that this never writes out of memory, but it
+         // could also overwrite the next scanline. can it overwrite non-empty data
+         // on the next scanline? yes, consider 1-pixel-wide scanlines with 1-bit-per-pixel.
+         // so we need to explicitly clamp the final ones
+
+         if (depth == 4) {
+            for (k=x*img_n; k >= 2; k-=2, ++in) {
+               *cur++ = scale * ((*in >> 4)       );
+               *cur++ = scale * ((*in     ) & 0x0f);
+            }
+            if (k > 0) *cur++ = scale * ((*in >> 4)       );
+         } else if (depth == 2) {
+            for (k=x*img_n; k >= 4; k-=4, ++in) {
+               *cur++ = scale * ((*in >> 6)       );
+               *cur++ = scale * ((*in >> 4) & 0x03);
+               *cur++ = scale * ((*in >> 2) & 0x03);
+               *cur++ = scale * ((*in     ) & 0x03);
+            }
+            if (k > 0) *cur++ = scale * ((*in >> 6)       );
+            if (k > 1) *cur++ = scale * ((*in >> 4) & 0x03);
+            if (k > 2) *cur++ = scale * ((*in >> 2) & 0x03);
+         } else if (depth == 1) {
+            for (k=x*img_n; k >= 8; k-=8, ++in) {
+               *cur++ = scale * ((*in >> 7)       );
+               *cur++ = scale * ((*in >> 6) & 0x01);
+               *cur++ = scale * ((*in >> 5) & 0x01);
+               *cur++ = scale * ((*in >> 4) & 0x01);
+               *cur++ = scale * ((*in >> 3) & 0x01);
+               *cur++ = scale * ((*in >> 2) & 0x01);
+               *cur++ = scale * ((*in >> 1) & 0x01);
+               *cur++ = scale * ((*in     ) & 0x01);
+            }
+            if (k > 0) *cur++ = scale * ((*in >> 7)       );
+            if (k > 1) *cur++ = scale * ((*in >> 6) & 0x01);
+            if (k > 2) *cur++ = scale * ((*in >> 5) & 0x01);
+            if (k > 3) *cur++ = scale * ((*in >> 4) & 0x01);
+            if (k > 4) *cur++ = scale * ((*in >> 3) & 0x01);
+            if (k > 5) *cur++ = scale * ((*in >> 2) & 0x01);
+            if (k > 6) *cur++ = scale * ((*in >> 1) & 0x01);
+         }
+         if (img_n != out_n) {
+            int q;
+            // insert alpha = 255
+            cur = a->out + stride*j;
+            if (img_n == 1) {
+               for (q=x-1; q >= 0; --q) {
+                  cur[q*2+1] = 255;
+                  cur[q*2+0] = cur[q];
+               }
+            } else {
+               STBI_ASSERT(img_n == 3);
+               for (q=x-1; q >= 0; --q) {
+                  cur[q*4+3] = 255;
+                  cur[q*4+2] = cur[q*3+2];
+                  cur[q*4+1] = cur[q*3+1];
+                  cur[q*4+0] = cur[q*3+0];
+               }
+            }
+         }
+      }
+   } else if (depth == 16) {
+      // force the image data from big-endian to platform-native.
+      // this is done in a separate pass due to the decoding relying
+      // on the data being untouched, but could probably be done
+      // per-line during decode if care is taken.
+      stbi_uc *cur = a->out;
+      stbi__uint16 *cur16 = (stbi__uint16*)cur;
+
+      for(i=0; i < x*y*out_n; ++i,cur16++,cur+=2) {
+         *cur16 = (cur[0] << 8) | cur[1];
+      }
+   }
+
+   return 1;
+}
+
+static int stbi__create_png_image(stbi__png *a, stbi_uc *image_data, stbi__uint32 image_data_len, int out_n, int depth, int color, int interlaced)
+{
+   int bytes = (depth == 16 ? 2 : 1);
+   int out_bytes = out_n * bytes;
+   stbi_uc *final;
+   int p;
+   if (!interlaced)
+      return stbi__create_png_image_raw(a, image_data, image_data_len, out_n, a->s->img_x, a->s->img_y, depth, color);
+
+   // de-interlacing
+   final = (stbi_uc *) stbi__malloc_mad3(a->s->img_x, a->s->img_y, out_bytes, 0);
+   for (p=0; p < 7; ++p) {
+      int xorig[] = { 0,4,0,2,0,1,0 };
+      int yorig[] = { 0,0,4,0,2,0,1 };
+      int xspc[]  = { 8,8,4,4,2,2,1 };
+      int yspc[]  = { 8,8,8,4,4,2,2 };
+      int i,j,x,y;
+      // pass1_x[4] = 0, pass1_x[5] = 1, pass1_x[12] = 1
+      x = (a->s->img_x - xorig[p] + xspc[p]-1) / xspc[p];
+      y = (a->s->img_y - yorig[p] + yspc[p]-1) / yspc[p];
+      if (x && y) {
+         stbi__uint32 img_len = ((((a->s->img_n * x * depth) + 7) >> 3) + 1) * y;
+         if (!stbi__create_png_image_raw(a, image_data, image_data_len, out_n, x, y, depth, color)) {
+            STBI_FREE(final);
+            return 0;
+         }
+         for (j=0; j < y; ++j) {
+            for (i=0; i < x; ++i) {
+               int out_y = j*yspc[p]+yorig[p];
+               int out_x = i*xspc[p]+xorig[p];
+               memcpy(final + out_y*a->s->img_x*out_bytes + out_x*out_bytes,
+                      a->out + (j*x+i)*out_bytes, out_bytes);
+            }
+         }
+         STBI_FREE(a->out);
+         image_data += img_len;
+         image_data_len -= img_len;
+      }
+   }
+   a->out = final;
+
+   return 1;
+}
+
+static int stbi__compute_transparency(stbi__png *z, stbi_uc tc[3], int out_n)
+{
+   stbi__context *s = z->s;
+   stbi__uint32 i, pixel_count = s->img_x * s->img_y;
+   stbi_uc *p = z->out;
+
+   // compute color-based transparency, assuming we've
+   // already got 255 as the alpha value in the output
+   STBI_ASSERT(out_n == 2 || out_n == 4);
+
+   if (out_n == 2) {
+      for (i=0; i < pixel_count; ++i) {
+         p[1] = (p[0] == tc[0] ? 0 : 255);
+         p += 2;
+      }
+   } else {
+      for (i=0; i < pixel_count; ++i) {
+         if (p[0] == tc[0] && p[1] == tc[1] && p[2] == tc[2])
+            p[3] = 0;
+         p += 4;
+      }
+   }
+   return 1;
+}
+
+static int stbi__compute_transparency16(stbi__png *z, stbi__uint16 tc[3], int out_n)
+{
+   stbi__context *s = z->s;
+   stbi__uint32 i, pixel_count = s->img_x * s->img_y;
+   stbi__uint16 *p = (stbi__uint16*) z->out;
+
+   // compute color-based transparency, assuming we've
+   // already got 65535 as the alpha value in the output
+   STBI_ASSERT(out_n == 2 || out_n == 4);
+
+   if (out_n == 2) {
+      for (i = 0; i < pixel_count; ++i) {
+         p[1] = (p[0] == tc[0] ? 0 : 65535);
+         p += 2;
+      }
+   } else {
+      for (i = 0; i < pixel_count; ++i) {
+         if (p[0] == tc[0] && p[1] == tc[1] && p[2] == tc[2])
+            p[3] = 0;
+         p += 4;
+      }
+   }
+   return 1;
+}
+
+static int stbi__expand_png_palette(stbi__png *a, stbi_uc *palette, int len, int pal_img_n)
+{
+   stbi__uint32 i, pixel_count = a->s->img_x * a->s->img_y;
+   stbi_uc *p, *temp_out, *orig = a->out;
+
+   p = (stbi_uc *) stbi__malloc_mad2(pixel_count, pal_img_n, 0);
+   if (p == NULL) return stbi__err("outofmem", "Out of memory");
+
+   // between here and free(out) below, exitting would leak
+   temp_out = p;
+
+   if (pal_img_n == 3) {
+      for (i=0; i < pixel_count; ++i) {
+         int n = orig[i]*4;
+         p[0] = palette[n  ];
+         p[1] = palette[n+1];
+         p[2] = palette[n+2];
+         p += 3;
+      }
+   } else {
+      for (i=0; i < pixel_count; ++i) {
+         int n = orig[i]*4;
+         p[0] = palette[n  ];
+         p[1] = palette[n+1];
+         p[2] = palette[n+2];
+         p[3] = palette[n+3];
+         p += 4;
+      }
+   }
+   STBI_FREE(a->out);
+   a->out = temp_out;
+
+   STBI_NOTUSED(len);
+
+   return 1;
+}
+
+static int stbi__unpremultiply_on_load = 0;
+static int stbi__de_iphone_flag = 0;
+
+STBIDEF void stbi_set_unpremultiply_on_load(int flag_true_if_should_unpremultiply)
+{
+   stbi__unpremultiply_on_load = flag_true_if_should_unpremultiply;
+}
+
+STBIDEF void stbi_convert_iphone_png_to_rgb(int flag_true_if_should_convert)
+{
+   stbi__de_iphone_flag = flag_true_if_should_convert;
+}
+
+static void stbi__de_iphone(stbi__png *z)
+{
+   stbi__context *s = z->s;
+   stbi__uint32 i, pixel_count = s->img_x * s->img_y;
+   stbi_uc *p = z->out;
+
+   if (s->img_out_n == 3) {  // convert bgr to rgb
+      for (i=0; i < pixel_count; ++i) {
+         stbi_uc t = p[0];
+         p[0] = p[2];
+         p[2] = t;
+         p += 3;
+      }
+   } else {
+      STBI_ASSERT(s->img_out_n == 4);
+      if (stbi__unpremultiply_on_load) {
+         // convert bgr to rgb and unpremultiply
+         for (i=0; i < pixel_count; ++i) {
+            stbi_uc a = p[3];
+            stbi_uc t = p[0];
+            if (a) {
+               p[0] = p[2] * 255 / a;
+               p[1] = p[1] * 255 / a;
+               p[2] =  t   * 255 / a;
+            } else {
+               p[0] = p[2];
+               p[2] = t;
+            }
+            p += 4;
+         }
+      } else {
+         // convert bgr to rgb
+         for (i=0; i < pixel_count; ++i) {
+            stbi_uc t = p[0];
+            p[0] = p[2];
+            p[2] = t;
+            p += 4;
+         }
+      }
+   }
+}
+
+#define STBI__PNG_TYPE(a,b,c,d)  (((a) << 24) + ((b) << 16) + ((c) << 8) + (d))
+
+static int stbi__parse_png_file(stbi__png *z, int scan, int req_comp)
+{
+   stbi_uc palette[1024], pal_img_n=0;
+   stbi_uc has_trans=0, tc[3];
+   stbi__uint16 tc16[3];
+   stbi__uint32 ioff=0, idata_limit=0, i, pal_len=0;
+   int first=1,k,interlace=0, color=0, is_iphone=0;
+   stbi__context *s = z->s;
+
+   z->expanded = NULL;
+   z->idata = NULL;
+   z->out = NULL;
+
+   if (!stbi__check_png_header(s)) return 0;
+
+   if (scan == STBI__SCAN_type) return 1;
+
+   for (;;) {
+      stbi__pngchunk c = stbi__get_chunk_header(s);
+      switch (c.type) {
+         case STBI__PNG_TYPE('C','g','B','I'):
+            is_iphone = 1;
+            stbi__skip(s, c.length);
+            break;
+         case STBI__PNG_TYPE('I','H','D','R'): {
+            int comp,filter;
+            if (!first) return stbi__err("multiple IHDR","Corrupt PNG");
+            first = 0;
+            if (c.length != 13) return stbi__err("bad IHDR len","Corrupt PNG");
+            s->img_x = stbi__get32be(s); if (s->img_x > (1 << 24)) return stbi__err("too large","Very large image (corrupt?)");
+            s->img_y = stbi__get32be(s); if (s->img_y > (1 << 24)) return stbi__err("too large","Very large image (corrupt?)");
+            z->depth = stbi__get8(s);  if (z->depth != 1 && z->depth != 2 && z->depth != 4 && z->depth != 8 && z->depth != 16)  return stbi__err("1/2/4/8/16-bit only","PNG not supported: 1/2/4/8/16-bit only");
+            color = stbi__get8(s);  if (color > 6)         return stbi__err("bad ctype","Corrupt PNG");
+            if (color == 3 && z->depth == 16)                  return stbi__err("bad ctype","Corrupt PNG");
+            if (color == 3) pal_img_n = 3; else if (color & 1) return stbi__err("bad ctype","Corrupt PNG");
+            comp  = stbi__get8(s);  if (comp) return stbi__err("bad comp method","Corrupt PNG");
+            filter= stbi__get8(s);  if (filter) return stbi__err("bad filter method","Corrupt PNG");
+            interlace = stbi__get8(s); if (interlace>1) return stbi__err("bad interlace method","Corrupt PNG");
+            if (!s->img_x || !s->img_y) return stbi__err("0-pixel image","Corrupt PNG");
+            if (!pal_img_n) {
+               s->img_n = (color & 2 ? 3 : 1) + (color & 4 ? 1 : 0);
+               if ((1 << 30) / s->img_x / s->img_n < s->img_y) return stbi__err("too large", "Image too large to decode");
+               if (scan == STBI__SCAN_header) return 1;
+            } else {
+               // if paletted, then pal_n is our final components, and
+               // img_n is # components to decompress/filter.
+               s->img_n = 1;
+               if ((1 << 30) / s->img_x / 4 < s->img_y) return stbi__err("too large","Corrupt PNG");
+               // if SCAN_header, have to scan to see if we have a tRNS
+            }
+            break;
+         }
+
+         case STBI__PNG_TYPE('P','L','T','E'):  {
+            if (first) return stbi__err("first not IHDR", "Corrupt PNG");
+            if (c.length > 256*3) return stbi__err("invalid PLTE","Corrupt PNG");
+            pal_len = c.length / 3;
+            if (pal_len * 3 != c.length) return stbi__err("invalid PLTE","Corrupt PNG");
+            for (i=0; i < pal_len; ++i) {
+               palette[i*4+0] = stbi__get8(s);
+               palette[i*4+1] = stbi__get8(s);
+               palette[i*4+2] = stbi__get8(s);
+               palette[i*4+3] = 255;
+            }
+            break;
+         }
+
+         case STBI__PNG_TYPE('t','R','N','S'): {
+            if (first) return stbi__err("first not IHDR", "Corrupt PNG");
+            if (z->idata) return stbi__err("tRNS after IDAT","Corrupt PNG");
+            if (pal_img_n) {
+               if (scan == STBI__SCAN_header) { s->img_n = 4; return 1; }
+               if (pal_len == 0) return stbi__err("tRNS before PLTE","Corrupt PNG");
+               if (c.length > pal_len) return stbi__err("bad tRNS len","Corrupt PNG");
+               pal_img_n = 4;
+               for (i=0; i < c.length; ++i)
+                  palette[i*4+3] = stbi__get8(s);
+            } else {
+               if (!(s->img_n & 1)) return stbi__err("tRNS with alpha","Corrupt PNG");
+               if (c.length != (stbi__uint32) s->img_n*2) return stbi__err("bad tRNS len","Corrupt PNG");
+               has_trans = 1;
+               if (z->depth == 16) {
+                  for (k = 0; k < s->img_n; ++k) tc16[k] = (stbi__uint16)stbi__get16be(s); // copy the values as-is
+               } else {
+                  for (k = 0; k < s->img_n; ++k) tc[k] = (stbi_uc)(stbi__get16be(s) & 255) * stbi__depth_scale_table[z->depth]; // non 8-bit images will be larger
+               }
+            }
+            break;
+         }
+
+         case STBI__PNG_TYPE('I','D','A','T'): {
+            if (first) return stbi__err("first not IHDR", "Corrupt PNG");
+            if (pal_img_n && !pal_len) return stbi__err("no PLTE","Corrupt PNG");
+            if (scan == STBI__SCAN_header) { s->img_n = pal_img_n; return 1; }
+            if ((int)(ioff + c.length) < (int)ioff) return 0;
+            if (ioff + c.length > idata_limit) {
+               stbi__uint32 idata_limit_old = idata_limit;
+               stbi_uc *p;
+               if (idata_limit == 0) idata_limit = c.length > 4096 ? c.length : 4096;
+               while (ioff + c.length > idata_limit)
+                  idata_limit *= 2;
+               STBI_NOTUSED(idata_limit_old);
+               p = (stbi_uc *) STBI_REALLOC_SIZED(z->idata, idata_limit_old, idata_limit); if (p == NULL) return stbi__err("outofmem", "Out of memory");
+               z->idata = p;
+            }
+            if (!stbi__getn(s, z->idata+ioff,c.length)) return stbi__err("outofdata","Corrupt PNG");
+            ioff += c.length;
+            break;
+         }
+
+         case STBI__PNG_TYPE('I','E','N','D'): {
+            stbi__uint32 raw_len, bpl;
+            if (first) return stbi__err("first not IHDR", "Corrupt PNG");
+            if (scan != STBI__SCAN_load) return 1;
+            if (z->idata == NULL) return stbi__err("no IDAT","Corrupt PNG");
+            // initial guess for decoded data size to avoid unnecessary reallocs
+            bpl = (s->img_x * z->depth + 7) / 8; // bytes per line, per component
+            raw_len = bpl * s->img_y * s->img_n /* pixels */ + s->img_y /* filter mode per row */;
+            z->expanded = (stbi_uc *) stbi_zlib_decode_malloc_guesssize_headerflag((char *) z->idata, ioff, raw_len, (int *) &raw_len, !is_iphone);
+            if (z->expanded == NULL) return 0; // zlib should set error
+            STBI_FREE(z->idata); z->idata = NULL;
+            if ((req_comp == s->img_n+1 && req_comp != 3 && !pal_img_n) || has_trans)
+               s->img_out_n = s->img_n+1;
+            else
+               s->img_out_n = s->img_n;
+            if (!stbi__create_png_image(z, z->expanded, raw_len, s->img_out_n, z->depth, color, interlace)) return 0;
+            if (has_trans) {
+               if (z->depth == 16) {
+                  if (!stbi__compute_transparency16(z, tc16, s->img_out_n)) return 0;
+               } else {
+                  if (!stbi__compute_transparency(z, tc, s->img_out_n)) return 0;
+               }
+            }
+            if (is_iphone && stbi__de_iphone_flag && s->img_out_n > 2)
+               stbi__de_iphone(z);
+            if (pal_img_n) {
+               // pal_img_n == 3 or 4
+               s->img_n = pal_img_n; // record the actual colors we had
+               s->img_out_n = pal_img_n;
+               if (req_comp >= 3) s->img_out_n = req_comp;
+               if (!stbi__expand_png_palette(z, palette, pal_len, s->img_out_n))
+                  return 0;
+            }
+            STBI_FREE(z->expanded); z->expanded = NULL;
+            return 1;
+         }
+
+         default:
+            // if critical, fail
+            if (first) return stbi__err("first not IHDR", "Corrupt PNG");
+            if ((c.type & (1 << 29)) == 0) {
+               #ifndef STBI_NO_FAILURE_STRINGS
+               // not threadsafe
+               static char invalid_chunk[] = "XXXX PNG chunk not known";
+               invalid_chunk[0] = STBI__BYTECAST(c.type >> 24);
+               invalid_chunk[1] = STBI__BYTECAST(c.type >> 16);
+               invalid_chunk[2] = STBI__BYTECAST(c.type >>  8);
+               invalid_chunk[3] = STBI__BYTECAST(c.type >>  0);
+               #endif
+               return stbi__err(invalid_chunk, "PNG not supported: unknown PNG chunk type");
+            }
+            stbi__skip(s, c.length);
+            break;
+      }
+      // end of PNG chunk, read and skip CRC
+      stbi__get32be(s);
+   }
+}
+
+static void *stbi__do_png(stbi__png *p, int *x, int *y, int *n, int req_comp, stbi__result_info *ri)
+{
+   void *result=NULL;
+   if (req_comp < 0 || req_comp > 4) return stbi__errpuc("bad req_comp", "Internal error");
+   if (stbi__parse_png_file(p, STBI__SCAN_load, req_comp)) {
+      if (p->depth < 8)
+         ri->bits_per_channel = 8;
+      else
+         ri->bits_per_channel = p->depth;
+      result = p->out;
+      p->out = NULL;
+      if (req_comp && req_comp != p->s->img_out_n) {
+         if (ri->bits_per_channel == 8)
+            result = stbi__convert_format((unsigned char *) result, p->s->img_out_n, req_comp, p->s->img_x, p->s->img_y);
+         else
+            result = stbi__convert_format16((stbi__uint16 *) result, p->s->img_out_n, req_comp, p->s->img_x, p->s->img_y);
+         p->s->img_out_n = req_comp;
+         if (result == NULL) return result;
+      }
+      *x = p->s->img_x;
+      *y = p->s->img_y;
+      if (n) *n = p->s->img_n;
+   }
+   STBI_FREE(p->out);      p->out      = NULL;
+   STBI_FREE(p->expanded); p->expanded = NULL;
+   STBI_FREE(p->idata);    p->idata    = NULL;
+
+   return result;
+}
+
+static void *stbi__png_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri)
+{
+   stbi__png p;
+   p.s = s;
+   return stbi__do_png(&p, x,y,comp,req_comp, ri);
+}
+
+static int stbi__png_test(stbi__context *s)
+{
+   int r;
+   r = stbi__check_png_header(s);
+   stbi__rewind(s);
+   return r;
+}
+
+static int stbi__png_info_raw(stbi__png *p, int *x, int *y, int *comp)
+{
+   if (!stbi__parse_png_file(p, STBI__SCAN_header, 0)) {
+      stbi__rewind( p->s );
+      return 0;
+   }
+   if (x) *x = p->s->img_x;
+   if (y) *y = p->s->img_y;
+   if (comp) *comp = p->s->img_n;
+   return 1;
+}
+
+static int stbi__png_info(stbi__context *s, int *x, int *y, int *comp)
+{
+   stbi__png p;
+   p.s = s;
+   return stbi__png_info_raw(&p, x, y, comp);
+}
+#endif
+
+// Microsoft/Windows BMP image
+
+#ifndef STBI_NO_BMP
+static int stbi__bmp_test_raw(stbi__context *s)
+{
+   int r;
+   int sz;
+   if (stbi__get8(s) != 'B') return 0;
+   if (stbi__get8(s) != 'M') return 0;
+   stbi__get32le(s); // discard filesize
+   stbi__get16le(s); // discard reserved
+   stbi__get16le(s); // discard reserved
+   stbi__get32le(s); // discard data offset
+   sz = stbi__get32le(s);
+   r = (sz == 12 || sz == 40 || sz == 56 || sz == 108 || sz == 124);
+   return r;
+}
+
+static int stbi__bmp_test(stbi__context *s)
+{
+   int r = stbi__bmp_test_raw(s);
+   stbi__rewind(s);
+   return r;
+}
+
+
+// returns 0..31 for the highest set bit
+static int stbi__high_bit(unsigned int z)
+{
+   int n=0;
+   if (z == 0) return -1;
+   if (z >= 0x10000) n += 16, z >>= 16;
+   if (z >= 0x00100) n +=  8, z >>=  8;
+   if (z >= 0x00010) n +=  4, z >>=  4;
+   if (z >= 0x00004) n +=  2, z >>=  2;
+   if (z >= 0x00002) n +=  1, z >>=  1;
+   return n;
+}
+
+static int stbi__bitcount(unsigned int a)
+{
+   a = (a & 0x55555555) + ((a >>  1) & 0x55555555); // max 2
+   a = (a & 0x33333333) + ((a >>  2) & 0x33333333); // max 4
+   a = (a + (a >> 4)) & 0x0f0f0f0f; // max 8 per 4, now 8 bits
+   a = (a + (a >> 8)); // max 16 per 8 bits
+   a = (a + (a >> 16)); // max 32 per 8 bits
+   return a & 0xff;
+}
+
+static int stbi__shiftsigned(int v, int shift, int bits)
+{
+   int result;
+   int z=0;
+
+   if (shift < 0) v <<= -shift;
+   else v >>= shift;
+   result = v;
+
+   z = bits;
+   while (z < 8) {
+      result += v >> z;
+      z += bits;
+   }
+   return result;
+}
+
+typedef struct
+{
+   int bpp, offset, hsz;
+   unsigned int mr,mg,mb,ma, all_a;
+} stbi__bmp_data;
+
+static void *stbi__bmp_parse_header(stbi__context *s, stbi__bmp_data *info)
+{
+   int hsz;
+   if (stbi__get8(s) != 'B' || stbi__get8(s) != 'M') return stbi__errpuc("not BMP", "Corrupt BMP");
+   stbi__get32le(s); // discard filesize
+   stbi__get16le(s); // discard reserved
+   stbi__get16le(s); // discard reserved
+   info->offset = stbi__get32le(s);
+   info->hsz = hsz = stbi__get32le(s);
+   info->mr = info->mg = info->mb = info->ma = 0;
+
+   if (hsz != 12 && hsz != 40 && hsz != 56 && hsz != 108 && hsz != 124) return stbi__errpuc("unknown BMP", "BMP type not supported: unknown");
+   if (hsz == 12) {
+      s->img_x = stbi__get16le(s);
+      s->img_y = stbi__get16le(s);
+   } else {
+      s->img_x = stbi__get32le(s);
+      s->img_y = stbi__get32le(s);
+   }
+   if (stbi__get16le(s) != 1) return stbi__errpuc("bad BMP", "bad BMP");
+   info->bpp = stbi__get16le(s);
+   if (info->bpp == 1) return stbi__errpuc("monochrome", "BMP type not supported: 1-bit");
+   if (hsz != 12) {
+      int compress = stbi__get32le(s);
+      if (compress == 1 || compress == 2) return stbi__errpuc("BMP RLE", "BMP type not supported: RLE");
+      stbi__get32le(s); // discard sizeof
+      stbi__get32le(s); // discard hres
+      stbi__get32le(s); // discard vres
+      stbi__get32le(s); // discard colorsused
+      stbi__get32le(s); // discard max important
+      if (hsz == 40 || hsz == 56) {
+         if (hsz == 56) {
+            stbi__get32le(s);
+            stbi__get32le(s);
+            stbi__get32le(s);
+            stbi__get32le(s);
+         }
+         if (info->bpp == 16 || info->bpp == 32) {
+            if (compress == 0) {
+               if (info->bpp == 32) {
+                  info->mr = 0xffu << 16;
+                  info->mg = 0xffu <<  8;
+                  info->mb = 0xffu <<  0;
+                  info->ma = 0xffu << 24;
+                  info->all_a = 0; // if all_a is 0 at end, then we loaded alpha channel but it was all 0
+               } else {
+                  info->mr = 31u << 10;
+                  info->mg = 31u <<  5;
+                  info->mb = 31u <<  0;
+               }
+            } else if (compress == 3) {
+               info->mr = stbi__get32le(s);
+               info->mg = stbi__get32le(s);
+               info->mb = stbi__get32le(s);
+               // not documented, but generated by photoshop and handled by mspaint
+               if (info->mr == info->mg && info->mg == info->mb) {
+                  // ?!?!?
+                  return stbi__errpuc("bad BMP", "bad BMP");
+               }
+            } else
+               return stbi__errpuc("bad BMP", "bad BMP");
+         }
+      } else {
+         int i;
+         if (hsz != 108 && hsz != 124)
+            return stbi__errpuc("bad BMP", "bad BMP");
+         info->mr = stbi__get32le(s);
+         info->mg = stbi__get32le(s);
+         info->mb = stbi__get32le(s);
+         info->ma = stbi__get32le(s);
+         stbi__get32le(s); // discard color space
+         for (i=0; i < 12; ++i)
+            stbi__get32le(s); // discard color space parameters
+         if (hsz == 124) {
+            stbi__get32le(s); // discard rendering intent
+            stbi__get32le(s); // discard offset of profile data
+            stbi__get32le(s); // discard size of profile data
+            stbi__get32le(s); // discard reserved
+         }
+      }
+   }
+   return (void *) 1;
+}
+
+
+static void *stbi__bmp_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri)
+{
+   stbi_uc *out;
+   unsigned int mr=0,mg=0,mb=0,ma=0, all_a;
+   stbi_uc pal[256][4];
+   int psize=0,i,j,width;
+   int flip_vertically, pad, target;
+   stbi__bmp_data info;
+   STBI_NOTUSED(ri);
+
+   info.all_a = 255;
+   if (stbi__bmp_parse_header(s, &info) == NULL)
+      return NULL; // error code already set
+
+   flip_vertically = ((int) s->img_y) > 0;
+   s->img_y = abs((int) s->img_y);
+
+   mr = info.mr;
+   mg = info.mg;
+   mb = info.mb;
+   ma = info.ma;
+   all_a = info.all_a;
+
+   if (info.hsz == 12) {
+      if (info.bpp < 24)
+         psize = (info.offset - 14 - 24) / 3;
+   } else {
+      if (info.bpp < 16)
+         psize = (info.offset - 14 - info.hsz) >> 2;
+   }
+
+   s->img_n = ma ? 4 : 3;
+   if (req_comp && req_comp >= 3) // we can directly decode 3 or 4
+      target = req_comp;
+   else
+      target = s->img_n; // if they want monochrome, we'll post-convert
+
+   // sanity-check size
+   if (!stbi__mad3sizes_valid(target, s->img_x, s->img_y, 0))
+      return stbi__errpuc("too large", "Corrupt BMP");
+
+   out = (stbi_uc *) stbi__malloc_mad3(target, s->img_x, s->img_y, 0);
+   if (!out) return stbi__errpuc("outofmem", "Out of memory");
+   if (info.bpp < 16) {
+      int z=0;
+      if (psize == 0 || psize > 256) { STBI_FREE(out); return stbi__errpuc("invalid", "Corrupt BMP"); }
+      for (i=0; i < psize; ++i) {
+         pal[i][2] = stbi__get8(s);
+         pal[i][1] = stbi__get8(s);
+         pal[i][0] = stbi__get8(s);
+         if (info.hsz != 12) stbi__get8(s);
+         pal[i][3] = 255;
+      }
+      stbi__skip(s, info.offset - 14 - info.hsz - psize * (info.hsz == 12 ? 3 : 4));
+      if (info.bpp == 4) width = (s->img_x + 1) >> 1;
+      else if (info.bpp == 8) width = s->img_x;
+      else { STBI_FREE(out); return stbi__errpuc("bad bpp", "Corrupt BMP"); }
+      pad = (-width)&3;
+      for (j=0; j < (int) s->img_y; ++j) {
+         for (i=0; i < (int) s->img_x; i += 2) {
+            int v=stbi__get8(s),v2=0;
+            if (info.bpp == 4) {
+               v2 = v & 15;
+               v >>= 4;
+            }
+            out[z++] = pal[v][0];
+            out[z++] = pal[v][1];
+            out[z++] = pal[v][2];
+            if (target == 4) out[z++] = 255;
+            if (i+1 == (int) s->img_x) break;
+            v = (info.bpp == 8) ? stbi__get8(s) : v2;
+            out[z++] = pal[v][0];
+            out[z++] = pal[v][1];
+            out[z++] = pal[v][2];
+            if (target == 4) out[z++] = 255;
+         }
+         stbi__skip(s, pad);
+      }
+   } else {
+      int rshift=0,gshift=0,bshift=0,ashift=0,rcount=0,gcount=0,bcount=0,acount=0;
+      int z = 0;
+      int easy=0;
+      stbi__skip(s, info.offset - 14 - info.hsz);
+      if (info.bpp == 24) width = 3 * s->img_x;
+      else if (info.bpp == 16) width = 2*s->img_x;
+      else /* bpp = 32 and pad = 0 */ width=0;
+      pad = (-width) & 3;
+      if (info.bpp == 24) {
+         easy = 1;
+      } else if (info.bpp == 32) {
+         if (mb == 0xff && mg == 0xff00 && mr == 0x00ff0000 && ma == 0xff000000)
+            easy = 2;
+      }
+      if (!easy) {
+         if (!mr || !mg || !mb) { STBI_FREE(out); return stbi__errpuc("bad masks", "Corrupt BMP"); }
+         // right shift amt to put high bit in position #7
+         rshift = stbi__high_bit(mr)-7; rcount = stbi__bitcount(mr);
+         gshift = stbi__high_bit(mg)-7; gcount = stbi__bitcount(mg);
+         bshift = stbi__high_bit(mb)-7; bcount = stbi__bitcount(mb);
+         ashift = stbi__high_bit(ma)-7; acount = stbi__bitcount(ma);
+      }
+      for (j=0; j < (int) s->img_y; ++j) {
+         if (easy) {
+            for (i=0; i < (int) s->img_x; ++i) {
+               unsigned char a;
+               out[z+2] = stbi__get8(s);
+               out[z+1] = stbi__get8(s);
+               out[z+0] = stbi__get8(s);
+               z += 3;
+               a = (easy == 2 ? stbi__get8(s) : 255);
+               all_a |= a;
+               if (target == 4) out[z++] = a;
+            }
+         } else {
+            int bpp = info.bpp;
+            for (i=0; i < (int) s->img_x; ++i) {
+               stbi__uint32 v = (bpp == 16 ? (stbi__uint32) stbi__get16le(s) : stbi__get32le(s));
+               int a;
+               out[z++] = STBI__BYTECAST(stbi__shiftsigned(v & mr, rshift, rcount));
+               out[z++] = STBI__BYTECAST(stbi__shiftsigned(v & mg, gshift, gcount));
+               out[z++] = STBI__BYTECAST(stbi__shiftsigned(v & mb, bshift, bcount));
+               a = (ma ? stbi__shiftsigned(v & ma, ashift, acount) : 255);
+               all_a |= a;
+               if (target == 4) out[z++] = STBI__BYTECAST(a);
+            }
+         }
+         stbi__skip(s, pad);
+      }
+   }
+
+   // if alpha channel is all 0s, replace with all 255s
+   if (target == 4 && all_a == 0)
+      for (i=4*s->img_x*s->img_y-1; i >= 0; i -= 4)
+         out[i] = 255;
+
+   if (flip_vertically) {
+      stbi_uc t;
+      for (j=0; j < (int) s->img_y>>1; ++j) {
+         stbi_uc *p1 = out +      j     *s->img_x*target;
+         stbi_uc *p2 = out + (s->img_y-1-j)*s->img_x*target;
+         for (i=0; i < (int) s->img_x*target; ++i) {
+            t = p1[i], p1[i] = p2[i], p2[i] = t;
+         }
+      }
+   }
+
+   if (req_comp && req_comp != target) {
+      out = stbi__convert_format(out, target, req_comp, s->img_x, s->img_y);
+      if (out == NULL) return out; // stbi__convert_format frees input on failure
+   }
+
+   *x = s->img_x;
+   *y = s->img_y;
+   if (comp) *comp = s->img_n;
+   return out;
+}
+#endif
+
+// Targa Truevision - TGA
+// by Jonathan Dummer
+#ifndef STBI_NO_TGA
+// returns STBI_rgb or whatever, 0 on error
+static int stbi__tga_get_comp(int bits_per_pixel, int is_grey, int* is_rgb16)
+{
+   // only RGB or RGBA (incl. 16bit) or grey allowed
+   if(is_rgb16) *is_rgb16 = 0;
+   switch(bits_per_pixel) {
+      case 8:  return STBI_grey;
+      case 16: if(is_grey) return STBI_grey_alpha;
+            // else: fall-through
+      case 15: if(is_rgb16) *is_rgb16 = 1;
+            return STBI_rgb;
+      case 24: // fall-through
+      case 32: return bits_per_pixel/8;
+      default: return 0;
+   }
+}
+
+static int stbi__tga_info(stbi__context *s, int *x, int *y, int *comp)
+{
+    int tga_w, tga_h, tga_comp, tga_image_type, tga_bits_per_pixel, tga_colormap_bpp;
+    int sz, tga_colormap_type;
+    stbi__get8(s);                   // discard Offset
+    tga_colormap_type = stbi__get8(s); // colormap type
+    if( tga_colormap_type > 1 ) {
+        stbi__rewind(s);
+        return 0;      // only RGB or indexed allowed
+    }
+    tga_image_type = stbi__get8(s); // image type
+    if ( tga_colormap_type == 1 ) { // colormapped (paletted) image
+        if (tga_image_type != 1 && tga_image_type != 9) {
+            stbi__rewind(s);
+            return 0;
+        }
+        stbi__skip(s,4);       // skip index of first colormap entry and number of entries
+        sz = stbi__get8(s);    //   check bits per palette color entry
+        if ( (sz != 8) && (sz != 15) && (sz != 16) && (sz != 24) && (sz != 32) ) {
+            stbi__rewind(s);
+            return 0;
+        }
+        stbi__skip(s,4);       // skip image x and y origin
+        tga_colormap_bpp = sz;
+    } else { // "normal" image w/o colormap - only RGB or grey allowed, +/- RLE
+        if ( (tga_image_type != 2) && (tga_image_type != 3) && (tga_image_type != 10) && (tga_image_type != 11) ) {
+            stbi__rewind(s);
+            return 0; // only RGB or grey allowed, +/- RLE
+        }
+        stbi__skip(s,9); // skip colormap specification and image x/y origin
+        tga_colormap_bpp = 0;
+    }
+    tga_w = stbi__get16le(s);
+    if( tga_w < 1 ) {
+        stbi__rewind(s);
+        return 0;   // test width
+    }
+    tga_h = stbi__get16le(s);
+    if( tga_h < 1 ) {
+        stbi__rewind(s);
+        return 0;   // test height
+    }
+    tga_bits_per_pixel = stbi__get8(s); // bits per pixel
+    stbi__get8(s); // ignore alpha bits
+    if (tga_colormap_bpp != 0) {
+        if((tga_bits_per_pixel != 8) && (tga_bits_per_pixel != 16)) {
+            // when using a colormap, tga_bits_per_pixel is the size of the indexes
+            // I don't think anything but 8 or 16bit indexes makes sense
+            stbi__rewind(s);
+            return 0;
+        }
+        tga_comp = stbi__tga_get_comp(tga_colormap_bpp, 0, NULL);
+    } else {
+        tga_comp = stbi__tga_get_comp(tga_bits_per_pixel, (tga_image_type == 3) || (tga_image_type == 11), NULL);
+    }
+    if(!tga_comp) {
+      stbi__rewind(s);
+      return 0;
+    }
+    if (x) *x = tga_w;
+    if (y) *y = tga_h;
+    if (comp) *comp = tga_comp;
+    return 1;                   // seems to have passed everything
+}
+
+static int stbi__tga_test(stbi__context *s)
+{
+   int res = 0;
+   int sz, tga_color_type;
+   stbi__get8(s);      //   discard Offset
+   tga_color_type = stbi__get8(s);   //   color type
+   if ( tga_color_type > 1 ) goto errorEnd;   //   only RGB or indexed allowed
+   sz = stbi__get8(s);   //   image type
+   if ( tga_color_type == 1 ) { // colormapped (paletted) image
+      if (sz != 1 && sz != 9) goto errorEnd; // colortype 1 demands image type 1 or 9
+      stbi__skip(s,4);       // skip index of first colormap entry and number of entries
+      sz = stbi__get8(s);    //   check bits per palette color entry
+      if ( (sz != 8) && (sz != 15) && (sz != 16) && (sz != 24) && (sz != 32) ) goto errorEnd;
+      stbi__skip(s,4);       // skip image x and y origin
+   } else { // "normal" image w/o colormap
+      if ( (sz != 2) && (sz != 3) && (sz != 10) && (sz != 11) ) goto errorEnd; // only RGB or grey allowed, +/- RLE
+      stbi__skip(s,9); // skip colormap specification and image x/y origin
+   }
+   if ( stbi__get16le(s) < 1 ) goto errorEnd;      //   test width
+   if ( stbi__get16le(s) < 1 ) goto errorEnd;      //   test height
+   sz = stbi__get8(s);   //   bits per pixel
+   if ( (tga_color_type == 1) && (sz != 8) && (sz != 16) ) goto errorEnd; // for colormapped images, bpp is size of an index
+   if ( (sz != 8) && (sz != 15) && (sz != 16) && (sz != 24) && (sz != 32) ) goto errorEnd;
+
+   res = 1; // if we got this far, everything's good and we can return 1 instead of 0
+
+errorEnd:
+   stbi__rewind(s);
+   return res;
+}
+
+// read 16bit value and convert to 24bit RGB
+static void stbi__tga_read_rgb16(stbi__context *s, stbi_uc* out)
+{
+   stbi__uint16 px = (stbi__uint16)stbi__get16le(s);
+   stbi__uint16 fiveBitMask = 31;
+   // we have 3 channels with 5bits each
+   int r = (px >> 10) & fiveBitMask;
+   int g = (px >> 5) & fiveBitMask;
+   int b = px & fiveBitMask;
+   // Note that this saves the data in RGB(A) order, so it doesn't need to be swapped later
+   out[0] = (stbi_uc)((r * 255)/31);
+   out[1] = (stbi_uc)((g * 255)/31);
+   out[2] = (stbi_uc)((b * 255)/31);
+
+   // some people claim that the most significant bit might be used for alpha
+   // (possibly if an alpha-bit is set in the "image descriptor byte")
+   // but that only made 16bit test images completely translucent..
+   // so let's treat all 15 and 16bit TGAs as RGB with no alpha.
+}
+
+static void *stbi__tga_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri)
+{
+   //   read in the TGA header stuff
+   int tga_offset = stbi__get8(s);
+   int tga_indexed = stbi__get8(s);
+   int tga_image_type = stbi__get8(s);
+   int tga_is_RLE = 0;
+   int tga_palette_start = stbi__get16le(s);
+   int tga_palette_len = stbi__get16le(s);
+   int tga_palette_bits = stbi__get8(s);
+   int tga_x_origin = stbi__get16le(s);
+   int tga_y_origin = stbi__get16le(s);
+   int tga_width = stbi__get16le(s);
+   int tga_height = stbi__get16le(s);
+   int tga_bits_per_pixel = stbi__get8(s);
+   int tga_comp, tga_rgb16=0;
+   int tga_inverted = stbi__get8(s);
+   // int tga_alpha_bits = tga_inverted & 15; // the 4 lowest bits - unused (useless?)
+   //   image data
+   unsigned char *tga_data;
+   unsigned char *tga_palette = NULL;
+   int i, j;
+   unsigned char raw_data[4] = {0};
+   int RLE_count = 0;
+   int RLE_repeating = 0;
+   int read_next_pixel = 1;
+   STBI_NOTUSED(ri);
+
+   //   do a tiny bit of precessing
+   if ( tga_image_type >= 8 )
+   {
+      tga_image_type -= 8;
+      tga_is_RLE = 1;
+   }
+   tga_inverted = 1 - ((tga_inverted >> 5) & 1);
+
+   //   If I'm paletted, then I'll use the number of bits from the palette
+   if ( tga_indexed ) tga_comp = stbi__tga_get_comp(tga_palette_bits, 0, &tga_rgb16);
+   else tga_comp = stbi__tga_get_comp(tga_bits_per_pixel, (tga_image_type == 3), &tga_rgb16);
+
+   if(!tga_comp) // shouldn't really happen, stbi__tga_test() should have ensured basic consistency
+      return stbi__errpuc("bad format", "Can't find out TGA pixelformat");
+
+   //   tga info
+   *x = tga_width;
+   *y = tga_height;
+   if (comp) *comp = tga_comp;
+
+   if (!stbi__mad3sizes_valid(tga_width, tga_height, tga_comp, 0))
+      return stbi__errpuc("too large", "Corrupt TGA");
+
+   tga_data = (unsigned char*)stbi__malloc_mad3(tga_width, tga_height, tga_comp, 0);
+   if (!tga_data) return stbi__errpuc("outofmem", "Out of memory");
+
+   // skip to the data's starting position (offset usually = 0)
+   stbi__skip(s, tga_offset );
+
+   if ( !tga_indexed && !tga_is_RLE && !tga_rgb16 ) {
+      for (i=0; i < tga_height; ++i) {
+         int row = tga_inverted ? tga_height -i - 1 : i;
+         stbi_uc *tga_row = tga_data + row*tga_width*tga_comp;
+         stbi__getn(s, tga_row, tga_width * tga_comp);
+      }
+   } else  {
+      //   do I need to load a palette?
+      if ( tga_indexed)
+      {
+         //   any data to skip? (offset usually = 0)
+         stbi__skip(s, tga_palette_start );
+         //   load the palette
+         tga_palette = (unsigned char*)stbi__malloc_mad2(tga_palette_len, tga_comp, 0);
+         if (!tga_palette) {
+            STBI_FREE(tga_data);
+            return stbi__errpuc("outofmem", "Out of memory");
+         }
+         if (tga_rgb16) {
+            stbi_uc *pal_entry = tga_palette;
+            STBI_ASSERT(tga_comp == STBI_rgb);
+            for (i=0; i < tga_palette_len; ++i) {
+               stbi__tga_read_rgb16(s, pal_entry);
+               pal_entry += tga_comp;
+            }
+         } else if (!stbi__getn(s, tga_palette, tga_palette_len * tga_comp)) {
+               STBI_FREE(tga_data);
+               STBI_FREE(tga_palette);
+               return stbi__errpuc("bad palette", "Corrupt TGA");
+         }
+      }
+      //   load the data
+      for (i=0; i < tga_width * tga_height; ++i)
+      {
+         //   if I'm in RLE mode, do I need to get a RLE stbi__pngchunk?
+         if ( tga_is_RLE )
+         {
+            if ( RLE_count == 0 )
+            {
+               //   yep, get the next byte as a RLE command
+               int RLE_cmd = stbi__get8(s);
+               RLE_count = 1 + (RLE_cmd & 127);
+               RLE_repeating = RLE_cmd >> 7;
+               read_next_pixel = 1;
+            } else if ( !RLE_repeating )
+            {
+               read_next_pixel = 1;
+            }
+         } else
+         {
+            read_next_pixel = 1;
+         }
+         //   OK, if I need to read a pixel, do it now
+         if ( read_next_pixel )
+         {
+            //   load however much data we did have
+            if ( tga_indexed )
+            {
+               // read in index, then perform the lookup
+               int pal_idx = (tga_bits_per_pixel == 8) ? stbi__get8(s) : stbi__get16le(s);
+               if ( pal_idx >= tga_palette_len ) {
+                  // invalid index
+                  pal_idx = 0;
+               }
+               pal_idx *= tga_comp;
+               for (j = 0; j < tga_comp; ++j) {
+                  raw_data[j] = tga_palette[pal_idx+j];
+               }
+            } else if(tga_rgb16) {
+               STBI_ASSERT(tga_comp == STBI_rgb);
+               stbi__tga_read_rgb16(s, raw_data);
+            } else {
+               //   read in the data raw
+               for (j = 0; j < tga_comp; ++j) {
+                  raw_data[j] = stbi__get8(s);
+               }
+            }
+            //   clear the reading flag for the next pixel
+            read_next_pixel = 0;
+         } // end of reading a pixel
+
+         // copy data
+         for (j = 0; j < tga_comp; ++j)
+           tga_data[i*tga_comp+j] = raw_data[j];
+
+         //   in case we're in RLE mode, keep counting down
+         --RLE_count;
+      }
+      //   do I need to invert the image?
+      if ( tga_inverted )
+      {
+         for (j = 0; j*2 < tga_height; ++j)
+         {
+            int index1 = j * tga_width * tga_comp;
+            int index2 = (tga_height - 1 - j) * tga_width * tga_comp;
+            for (i = tga_width * tga_comp; i > 0; --i)
+            {
+               unsigned char temp = tga_data[index1];
+               tga_data[index1] = tga_data[index2];
+               tga_data[index2] = temp;
+               ++index1;
+               ++index2;
+            }
+         }
+      }
+      //   clear my palette, if I had one
+      if ( tga_palette != NULL )
+      {
+         STBI_FREE( tga_palette );
+      }
+   }
+
+   // swap RGB - if the source data was RGB16, it already is in the right order
+   if (tga_comp >= 3 && !tga_rgb16)
+   {
+      unsigned char* tga_pixel = tga_data;
+      for (i=0; i < tga_width * tga_height; ++i)
+      {
+         unsigned char temp = tga_pixel[0];
+         tga_pixel[0] = tga_pixel[2];
+         tga_pixel[2] = temp;
+         tga_pixel += tga_comp;
+      }
+   }
+
+   // convert to target component count
+   if (req_comp && req_comp != tga_comp)
+      tga_data = stbi__convert_format(tga_data, tga_comp, req_comp, tga_width, tga_height);
+
+   //   the things I do to get rid of an error message, and yet keep
+   //   Microsoft's C compilers happy... [8^(
+   tga_palette_start = tga_palette_len = tga_palette_bits =
+         tga_x_origin = tga_y_origin = 0;
+   //   OK, done
+   return tga_data;
+}
+#endif
+
+// *************************************************************************************************
+// Photoshop PSD loader -- PD by Thatcher Ulrich, integration by Nicolas Schulz, tweaked by STB
+
+#ifndef STBI_NO_PSD
+static int stbi__psd_test(stbi__context *s)
+{
+   int r = (stbi__get32be(s) == 0x38425053);
+   stbi__rewind(s);
+   return r;
+}
+
+static int stbi__psd_decode_rle(stbi__context *s, stbi_uc *p, int pixelCount)
+{
+   int count, nleft, len;
+
+   count = 0;
+   while ((nleft = pixelCount - count) > 0) {
+      len = stbi__get8(s);
+      if (len == 128) {
+         // No-op.
+      } else if (len < 128) {
+         // Copy next len+1 bytes literally.
+         len++;
+         if (len > nleft) return 0; // corrupt data
+         count += len;
+         while (len) {
+            *p = stbi__get8(s);
+            p += 4;
+            len--;
+         }
+      } else if (len > 128) {
+         stbi_uc   val;
+         // Next -len+1 bytes in the dest are replicated from next source byte.
+         // (Interpret len as a negative 8-bit int.)
+         len = 257 - len;
+         if (len > nleft) return 0; // corrupt data
+         val = stbi__get8(s);
+         count += len;
+         while (len) {
+            *p = val;
+            p += 4;
+            len--;
+         }
+      }
+   }
+
+   return 1;
+}
+
+static void *stbi__psd_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri, int bpc)
+{
+   int pixelCount;
+   int channelCount, compression;
+   int channel, i;
+   int bitdepth;
+   int w,h;
+   stbi_uc *out;
+   STBI_NOTUSED(ri);
+
+   // Check identifier
+   if (stbi__get32be(s) != 0x38425053)   // "8BPS"
+      return stbi__errpuc("not PSD", "Corrupt PSD image");
+
+   // Check file type version.
+   if (stbi__get16be(s) != 1)
+      return stbi__errpuc("wrong version", "Unsupported version of PSD image");
+
+   // Skip 6 reserved bytes.
+   stbi__skip(s, 6 );
+
+   // Read the number of channels (R, G, B, A, etc).
+   channelCount = stbi__get16be(s);
+   if (channelCount < 0 || channelCount > 16)
+      return stbi__errpuc("wrong channel count", "Unsupported number of channels in PSD image");
+
+   // Read the rows and columns of the image.
+   h = stbi__get32be(s);
+   w = stbi__get32be(s);
+
+   // Make sure the depth is 8 bits.
+   bitdepth = stbi__get16be(s);
+   if (bitdepth != 8 && bitdepth != 16)
+      return stbi__errpuc("unsupported bit depth", "PSD bit depth is not 8 or 16 bit");
+
+   // Make sure the color mode is RGB.
+   // Valid options are:
+   //   0: Bitmap
+   //   1: Grayscale
+   //   2: Indexed color
+   //   3: RGB color
+   //   4: CMYK color
+   //   7: Multichannel
+   //   8: Duotone
+   //   9: Lab color
+   if (stbi__get16be(s) != 3)
+      return stbi__errpuc("wrong color format", "PSD is not in RGB color format");
+
+   // Skip the Mode Data.  (It's the palette for indexed color; other info for other modes.)
+   stbi__skip(s,stbi__get32be(s) );
+
+   // Skip the image resources.  (resolution, pen tool paths, etc)
+   stbi__skip(s, stbi__get32be(s) );
+
+   // Skip the reserved data.
+   stbi__skip(s, stbi__get32be(s) );
+
+   // Find out if the data is compressed.
+   // Known values:
+   //   0: no compression
+   //   1: RLE compressed
+   compression = stbi__get16be(s);
+   if (compression > 1)
+      return stbi__errpuc("bad compression", "PSD has an unknown compression format");
+
+   // Check size
+   if (!stbi__mad3sizes_valid(4, w, h, 0))
+      return stbi__errpuc("too large", "Corrupt PSD");
+
+   // Create the destination image.
+
+   if (!compression && bitdepth == 16 && bpc == 16) {
+      out = (stbi_uc *) stbi__malloc_mad3(8, w, h, 0);
+      ri->bits_per_channel = 16;
+   } else
+      out = (stbi_uc *) stbi__malloc(4 * w*h);
+
+   if (!out) return stbi__errpuc("outofmem", "Out of memory");
+   pixelCount = w*h;
+
+   // Initialize the data to zero.
+   //memset( out, 0, pixelCount * 4 );
+
+   // Finally, the image data.
+   if (compression) {
+      // RLE as used by .PSD and .TIFF
+      // Loop until you get the number of unpacked bytes you are expecting:
+      //     Read the next source byte into n.
+      //     If n is between 0 and 127 inclusive, copy the next n+1 bytes literally.
+      //     Else if n is between -127 and -1 inclusive, copy the next byte -n+1 times.
+      //     Else if n is 128, noop.
+      // Endloop
+
+      // The RLE-compressed data is preceeded by a 2-byte data count for each row in the data,
+      // which we're going to just skip.
+      stbi__skip(s, h * channelCount * 2 );
+
+      // Read the RLE data by channel.
+      for (channel = 0; channel < 4; channel++) {
+         stbi_uc *p;
+
+         p = out+channel;
+         if (channel >= channelCount) {
+            // Fill this channel with default data.
+            for (i = 0; i < pixelCount; i++, p += 4)
+               *p = (channel == 3 ? 255 : 0);
+         } else {
+            // Read the RLE data.
+            if (!stbi__psd_decode_rle(s, p, pixelCount)) {
+               STBI_FREE(out);
+               return stbi__errpuc("corrupt", "bad RLE data");
+            }
+         }
+      }
+
+   } else {
+      // We're at the raw image data.  It's each channel in order (Red, Green, Blue, Alpha, ...)
+      // where each channel consists of an 8-bit (or 16-bit) value for each pixel in the image.
+
+      // Read the data by channel.
+      for (channel = 0; channel < 4; channel++) {
+         if (channel >= channelCount) {
+            // Fill this channel with default data.
+            if (bitdepth == 16 && bpc == 16) {
+               stbi__uint16 *q = ((stbi__uint16 *) out) + channel;
+               stbi__uint16 val = channel == 3 ? 65535 : 0;
+               for (i = 0; i < pixelCount; i++, q += 4)
+                  *q = val;
+            } else {
+               stbi_uc *p = out+channel;
+               stbi_uc val = channel == 3 ? 255 : 0;
+               for (i = 0; i < pixelCount; i++, p += 4)
+                  *p = val;
+            }
+         } else {
+            if (ri->bits_per_channel == 16) {    // output bpc
+               stbi__uint16 *q = ((stbi__uint16 *) out) + channel;
+               for (i = 0; i < pixelCount; i++, q += 4)
+                  *q = (stbi__uint16) stbi__get16be(s);
+            } else {
+               stbi_uc *p = out+channel;
+               if (bitdepth == 16) {  // input bpc
+                  for (i = 0; i < pixelCount; i++, p += 4)
+                     *p = (stbi_uc) (stbi__get16be(s) >> 8);
+               } else {
+                  for (i = 0; i < pixelCount; i++, p += 4)
+                     *p = stbi__get8(s);
+               }
+            }
+         }
+      }
+   }
+
+   // remove weird white matte from PSD
+   if (channelCount >= 4) {
+      if (ri->bits_per_channel == 16) {
+         for (i=0; i < w*h; ++i) {
+            stbi__uint16 *pixel = (stbi__uint16 *) out + 4*i;
+            if (pixel[3] != 0 && pixel[3] != 65535) {
+               float a = pixel[3] / 65535.0f;
+               float ra = 1.0f / a;
+               float inv_a = 65535.0f * (1 - ra);
+               pixel[0] = (stbi__uint16) (pixel[0]*ra + inv_a);
+               pixel[1] = (stbi__uint16) (pixel[1]*ra + inv_a);
+               pixel[2] = (stbi__uint16) (pixel[2]*ra + inv_a);
+            }
+         }
+      } else {
+         for (i=0; i < w*h; ++i) {
+            unsigned char *pixel = out + 4*i;
+            if (pixel[3] != 0 && pixel[3] != 255) {
+               float a = pixel[3] / 255.0f;
+               float ra = 1.0f / a;
+               float inv_a = 255.0f * (1 - ra);
+               pixel[0] = (unsigned char) (pixel[0]*ra + inv_a);
+               pixel[1] = (unsigned char) (pixel[1]*ra + inv_a);
+               pixel[2] = (unsigned char) (pixel[2]*ra + inv_a);
+            }
+         }
+      }
+   }
+
+   // convert to desired output format
+   if (req_comp && req_comp != 4) {
+      if (ri->bits_per_channel == 16)
+         out = (stbi_uc *) stbi__convert_format16((stbi__uint16 *) out, 4, req_comp, w, h);
+      else
+         out = stbi__convert_format(out, 4, req_comp, w, h);
+      if (out == NULL) return out; // stbi__convert_format frees input on failure
+   }
+
+   if (comp) *comp = 4;
+   *y = h;
+   *x = w;
+
+   return out;
+}
+#endif
+
+// *************************************************************************************************
+// Softimage PIC loader
+// by Tom Seddon
+//
+// See http://softimage.wiki.softimage.com/index.php/INFO:_PIC_file_format
+// See http://ozviz.wasp.uwa.edu.au/~pbourke/dataformats/softimagepic/
+
+#ifndef STBI_NO_PIC
+static int stbi__pic_is4(stbi__context *s,const char *str)
+{
+   int i;
+   for (i=0; i<4; ++i)
+      if (stbi__get8(s) != (stbi_uc)str[i])
+         return 0;
+
+   return 1;
+}
+
+static int stbi__pic_test_core(stbi__context *s)
+{
+   int i;
+
+   if (!stbi__pic_is4(s,"\x53\x80\xF6\x34"))
+      return 0;
+
+   for(i=0;i<84;++i)
+      stbi__get8(s);
+
+   if (!stbi__pic_is4(s,"PICT"))
+      return 0;
+
+   return 1;
+}
+
+typedef struct
+{
+   stbi_uc size,type,channel;
+} stbi__pic_packet;
+
+static stbi_uc *stbi__readval(stbi__context *s, int channel, stbi_uc *dest)
+{
+   int mask=0x80, i;
+
+   for (i=0; i<4; ++i, mask>>=1) {
+      if (channel & mask) {
+         if (stbi__at_eof(s)) return stbi__errpuc("bad file","PIC file too short");
+         dest[i]=stbi__get8(s);
+      }
+   }
+
+   return dest;
+}
+
+static void stbi__copyval(int channel,stbi_uc *dest,const stbi_uc *src)
+{
+   int mask=0x80,i;
+
+   for (i=0;i<4; ++i, mask>>=1)
+      if (channel&mask)
+         dest[i]=src[i];
+}
+
+static stbi_uc *stbi__pic_load_core(stbi__context *s,int width,int height,int *comp, stbi_uc *result)
+{
+   int act_comp=0,num_packets=0,y,chained;
+   stbi__pic_packet packets[10];
+
+   // this will (should...) cater for even some bizarre stuff like having data
+    // for the same channel in multiple packets.
+   do {
+      stbi__pic_packet *packet;
+
+      if (num_packets==sizeof(packets)/sizeof(packets[0]))
+         return stbi__errpuc("bad format","too many packets");
+
+      packet = &packets[num_packets++];
+
+      chained = stbi__get8(s);
+      packet->size    = stbi__get8(s);
+      packet->type    = stbi__get8(s);
+      packet->channel = stbi__get8(s);
+
+      act_comp |= packet->channel;
+
+      if (stbi__at_eof(s))          return stbi__errpuc("bad file","file too short (reading packets)");
+      if (packet->size != 8)  return stbi__errpuc("bad format","packet isn't 8bpp");
+   } while (chained);
+
+   *comp = (act_comp & 0x10 ? 4 : 3); // has alpha channel?
+
+   for(y=0; y<height; ++y) {
+      int packet_idx;
+
+      for(packet_idx=0; packet_idx < num_packets; ++packet_idx) {
+         stbi__pic_packet *packet = &packets[packet_idx];
+         stbi_uc *dest = result+y*width*4;
+
+         switch (packet->type) {
+            default:
+               return stbi__errpuc("bad format","packet has bad compression type");
+
+            case 0: {//uncompressed
+               int x;
+
+               for(x=0;x<width;++x, dest+=4)
+                  if (!stbi__readval(s,packet->channel,dest))
+                     return 0;
+               break;
+            }
+
+            case 1://Pure RLE
+               {
+                  int left=width, i;
+
+                  while (left>0) {
+                     stbi_uc count,value[4];
+
+                     count=stbi__get8(s);
+                     if (stbi__at_eof(s))   return stbi__errpuc("bad file","file too short (pure read count)");
+
+                     if (count > left)
+                        count = (stbi_uc) left;
+
+                     if (!stbi__readval(s,packet->channel,value))  return 0;
+
+                     for(i=0; i<count; ++i,dest+=4)
+                        stbi__copyval(packet->channel,dest,value);
+                     left -= count;
+                  }
+               }
+               break;
+
+            case 2: {//Mixed RLE
+               int left=width;
+               while (left>0) {
+                  int count = stbi__get8(s), i;
+                  if (stbi__at_eof(s))  return stbi__errpuc("bad file","file too short (mixed read count)");
+
+                  if (count >= 128) { // Repeated
+                     stbi_uc value[4];
+
+                     if (count==128)
+                        count = stbi__get16be(s);
+                     else
+                        count -= 127;
+                     if (count > left)
+                        return stbi__errpuc("bad file","scanline overrun");
+
+                     if (!stbi__readval(s,packet->channel,value))
+                        return 0;
+
+                     for(i=0;i<count;++i, dest += 4)
+                        stbi__copyval(packet->channel,dest,value);
+                  } else { // Raw
+                     ++count;
+                     if (count>left) return stbi__errpuc("bad file","scanline overrun");
+
+                     for(i=0;i<count;++i, dest+=4)
+                        if (!stbi__readval(s,packet->channel,dest))
+                           return 0;
+                  }
+                  left-=count;
+               }
+               break;
+            }
+         }
+      }
+   }
+
+   return result;
+}
+
+static void *stbi__pic_load(stbi__context *s,int *px,int *py,int *comp,int req_comp, stbi__result_info *ri)
+{
+   stbi_uc *result;
+   int i, x,y, internal_comp;
+   STBI_NOTUSED(ri);
+
+   if (!comp) comp = &internal_comp;
+
+   for (i=0; i<92; ++i)
+      stbi__get8(s);
+
+   x = stbi__get16be(s);
+   y = stbi__get16be(s);
+   if (stbi__at_eof(s))  return stbi__errpuc("bad file","file too short (pic header)");
+   if (!stbi__mad3sizes_valid(x, y, 4, 0)) return stbi__errpuc("too large", "PIC image too large to decode");
+
+   stbi__get32be(s); //skip `ratio'
+   stbi__get16be(s); //skip `fields'
+   stbi__get16be(s); //skip `pad'
+
+   // intermediate buffer is RGBA
+   result = (stbi_uc *) stbi__malloc_mad3(x, y, 4, 0);
+   memset(result, 0xff, x*y*4);
+
+   if (!stbi__pic_load_core(s,x,y,comp, result)) {
+      STBI_FREE(result);
+      result=0;
+   }
+   *px = x;
+   *py = y;
+   if (req_comp == 0) req_comp = *comp;
+   result=stbi__convert_format(result,4,req_comp,x,y);
+
+   return result;
+}
+
+static int stbi__pic_test(stbi__context *s)
+{
+   int r = stbi__pic_test_core(s);
+   stbi__rewind(s);
+   return r;
+}
+#endif
+
+// *************************************************************************************************
+// GIF loader -- public domain by Jean-Marc Lienher -- simplified/shrunk by stb
+
+#ifndef STBI_NO_GIF
+typedef struct
+{
+   stbi__int16 prefix;
+   stbi_uc first;
+   stbi_uc suffix;
+} stbi__gif_lzw;
+
+typedef struct
+{
+   int w,h;
+   stbi_uc *out, *old_out;             // output buffer (always 4 components)
+   int flags, bgindex, ratio, transparent, eflags, delay;
+   stbi_uc  pal[256][4];
+   stbi_uc lpal[256][4];
+   stbi__gif_lzw codes[4096];
+   stbi_uc *color_table;
+   int parse, step;
+   int lflags;
+   int start_x, start_y;
+   int max_x, max_y;
+   int cur_x, cur_y;
+   int line_size;
+} stbi__gif;
+
+static int stbi__gif_test_raw(stbi__context *s)
+{
+   int sz;
+   if (stbi__get8(s) != 'G' || stbi__get8(s) != 'I' || stbi__get8(s) != 'F' || stbi__get8(s) != '8') return 0;
+   sz = stbi__get8(s);
+   if (sz != '9' && sz != '7') return 0;
+   if (stbi__get8(s) != 'a') return 0;
+   return 1;
+}
+
+static int stbi__gif_test(stbi__context *s)
+{
+   int r = stbi__gif_test_raw(s);
+   stbi__rewind(s);
+   return r;
+}
+
+static void stbi__gif_parse_colortable(stbi__context *s, stbi_uc pal[256][4], int num_entries, int transp)
+{
+   int i;
+   for (i=0; i < num_entries; ++i) {
+      pal[i][2] = stbi__get8(s);
+      pal[i][1] = stbi__get8(s);
+      pal[i][0] = stbi__get8(s);
+      pal[i][3] = transp == i ? 0 : 255;
+   }
+}
+
+static int stbi__gif_header(stbi__context *s, stbi__gif *g, int *comp, int is_info)
+{
+   stbi_uc version;
+   if (stbi__get8(s) != 'G' || stbi__get8(s) != 'I' || stbi__get8(s) != 'F' || stbi__get8(s) != '8')
+      return stbi__err("not GIF", "Corrupt GIF");
+
+   version = stbi__get8(s);
+   if (version != '7' && version != '9')    return stbi__err("not GIF", "Corrupt GIF");
+   if (stbi__get8(s) != 'a')                return stbi__err("not GIF", "Corrupt GIF");
+
+   stbi__g_failure_reason = "";
+   g->w = stbi__get16le(s);
+   g->h = stbi__get16le(s);
+   g->flags = stbi__get8(s);
+   g->bgindex = stbi__get8(s);
+   g->ratio = stbi__get8(s);
+   g->transparent = -1;
+
+   if (comp != 0) *comp = 4;  // can't actually tell whether it's 3 or 4 until we parse the comments
+
+   if (is_info) return 1;
+
+   if (g->flags & 0x80)
+      stbi__gif_parse_colortable(s,g->pal, 2 << (g->flags & 7), -1);
+
+   return 1;
+}
+
+static int stbi__gif_info_raw(stbi__context *s, int *x, int *y, int *comp)
+{
+   stbi__gif* g = (stbi__gif*) stbi__malloc(sizeof(stbi__gif));
+   if (!stbi__gif_header(s, g, comp, 1)) {
+      STBI_FREE(g);
+      stbi__rewind( s );
+      return 0;
+   }
+   if (x) *x = g->w;
+   if (y) *y = g->h;
+   STBI_FREE(g);
+   return 1;
+}
+
+static void stbi__out_gif_code(stbi__gif *g, stbi__uint16 code)
+{
+   stbi_uc *p, *c;
+
+   // recurse to decode the prefixes, since the linked-list is backwards,
+   // and working backwards through an interleaved image would be nasty
+   if (g->codes[code].prefix >= 0)
+      stbi__out_gif_code(g, g->codes[code].prefix);
+
+   if (g->cur_y >= g->max_y) return;
+
+   p = &g->out[g->cur_x + g->cur_y];
+   c = &g->color_table[g->codes[code].suffix * 4];
+
+   if (c[3] >= 128) {
+      p[0] = c[2];
+      p[1] = c[1];
+      p[2] = c[0];
+      p[3] = c[3];
+   }
+   g->cur_x += 4;
+
+   if (g->cur_x >= g->max_x) {
+      g->cur_x = g->start_x;
+      g->cur_y += g->step;
+
+      while (g->cur_y >= g->max_y && g->parse > 0) {
+         g->step = (1 << g->parse) * g->line_size;
+         g->cur_y = g->start_y + (g->step >> 1);
+         --g->parse;
+      }
+   }
+}
+
+static stbi_uc *stbi__process_gif_raster(stbi__context *s, stbi__gif *g)
+{
+   stbi_uc lzw_cs;
+   stbi__int32 len, init_code;
+   stbi__uint32 first;
+   stbi__int32 codesize, codemask, avail, oldcode, bits, valid_bits, clear;
+   stbi__gif_lzw *p;
+
+   lzw_cs = stbi__get8(s);
+   if (lzw_cs > 12) return NULL;
+   clear = 1 << lzw_cs;
+   first = 1;
+   codesize = lzw_cs + 1;
+   codemask = (1 << codesize) - 1;
+   bits = 0;
+   valid_bits = 0;
+   for (init_code = 0; init_code < clear; init_code++) {
+      g->codes[init_code].prefix = -1;
+      g->codes[init_code].first = (stbi_uc) init_code;
+      g->codes[init_code].suffix = (stbi_uc) init_code;
+   }
+
+   // support no starting clear code
+   avail = clear+2;
+   oldcode = -1;
+
+   len = 0;
+   for(;;) {
+      if (valid_bits < codesize) {
+         if (len == 0) {
+            len = stbi__get8(s); // start new block
+            if (len == 0)
+               return g->out;
+         }
+         --len;
+         bits |= (stbi__int32) stbi__get8(s) << valid_bits;
+         valid_bits += 8;
+      } else {
+         stbi__int32 code = bits & codemask;
+         bits >>= codesize;
+         valid_bits -= codesize;
+         // @OPTIMIZE: is there some way we can accelerate the non-clear path?
+         if (code == clear) {  // clear code
+            codesize = lzw_cs + 1;
+            codemask = (1 << codesize) - 1;
+            avail = clear + 2;
+            oldcode = -1;
+            first = 0;
+         } else if (code == clear + 1) { // end of stream code
+            stbi__skip(s, len);
+            while ((len = stbi__get8(s)) > 0)
+               stbi__skip(s,len);
+            return g->out;
+         } else if (code <= avail) {
+            if (first) return stbi__errpuc("no clear code", "Corrupt GIF");
+
+            if (oldcode >= 0) {
+               p = &g->codes[avail++];
+               if (avail > 4096)        return stbi__errpuc("too many codes", "Corrupt GIF");
+               p->prefix = (stbi__int16) oldcode;
+               p->first = g->codes[oldcode].first;
+               p->suffix = (code == avail) ? p->first : g->codes[code].first;
+            } else if (code == avail)
+               return stbi__errpuc("illegal code in raster", "Corrupt GIF");
+
+            stbi__out_gif_code(g, (stbi__uint16) code);
+
+            if ((avail & codemask) == 0 && avail <= 0x0FFF) {
+               codesize++;
+               codemask = (1 << codesize) - 1;
+            }
+
+            oldcode = code;
+         } else {
+            return stbi__errpuc("illegal code in raster", "Corrupt GIF");
+         }
+      }
+   }
+}
+
+static void stbi__fill_gif_background(stbi__gif *g, int x0, int y0, int x1, int y1)
+{
+   int x, y;
+   stbi_uc *c = g->pal[g->bgindex];
+   for (y = y0; y < y1; y += 4 * g->w) {
+      for (x = x0; x < x1; x += 4) {
+         stbi_uc *p  = &g->out[y + x];
+         p[0] = c[2];
+         p[1] = c[1];
+         p[2] = c[0];
+         p[3] = 0;
+      }
+   }
+}
+
+// this function is designed to support animated gifs, although stb_image doesn't support it
+static stbi_uc *stbi__gif_load_next(stbi__context *s, stbi__gif *g, int *comp, int req_comp)
+{
+   int i;
+   stbi_uc *prev_out = 0;
+
+   if (g->out == 0 && !stbi__gif_header(s, g, comp,0))
+      return 0; // stbi__g_failure_reason set by stbi__gif_header
+
+   if (!stbi__mad3sizes_valid(g->w, g->h, 4, 0))
+      return stbi__errpuc("too large", "GIF too large");
+
+   prev_out = g->out;
+   g->out = (stbi_uc *) stbi__malloc_mad3(4, g->w, g->h, 0);
+   if (g->out == 0) return stbi__errpuc("outofmem", "Out of memory");
+
+   switch ((g->eflags & 0x1C) >> 2) {
+      case 0: // unspecified (also always used on 1st frame)
+         stbi__fill_gif_background(g, 0, 0, 4 * g->w, 4 * g->w * g->h);
+         break;
+      case 1: // do not dispose
+         if (prev_out) memcpy(g->out, prev_out, 4 * g->w * g->h);
+         g->old_out = prev_out;
+         break;
+      case 2: // dispose to background
+         if (prev_out) memcpy(g->out, prev_out, 4 * g->w * g->h);
+         stbi__fill_gif_background(g, g->start_x, g->start_y, g->max_x, g->max_y);
+         break;
+      case 3: // dispose to previous
+         if (g->old_out) {
+            for (i = g->start_y; i < g->max_y; i += 4 * g->w)
+               memcpy(&g->out[i + g->start_x], &g->old_out[i + g->start_x], g->max_x - g->start_x);
+         }
+         break;
+   }
+
+   for (;;) {
+      switch (stbi__get8(s)) {
+         case 0x2C: /* Image Descriptor */
+         {
+            int prev_trans = -1;
+            stbi__int32 x, y, w, h;
+            stbi_uc *o;
+
+            x = stbi__get16le(s);
+            y = stbi__get16le(s);
+            w = stbi__get16le(s);
+            h = stbi__get16le(s);
+            if (((x + w) > (g->w)) || ((y + h) > (g->h)))
+               return stbi__errpuc("bad Image Descriptor", "Corrupt GIF");
+
+            g->line_size = g->w * 4;
+            g->start_x = x * 4;
+            g->start_y = y * g->line_size;
+            g->max_x   = g->start_x + w * 4;
+            g->max_y   = g->start_y + h * g->line_size;
+            g->cur_x   = g->start_x;
+            g->cur_y   = g->start_y;
+
+            g->lflags = stbi__get8(s);
+
+            if (g->lflags & 0x40) {
+               g->step = 8 * g->line_size; // first interlaced spacing
+               g->parse = 3;
+            } else {
+               g->step = g->line_size;
+               g->parse = 0;
+            }
+
+            if (g->lflags & 0x80) {
+               stbi__gif_parse_colortable(s,g->lpal, 2 << (g->lflags & 7), g->eflags & 0x01 ? g->transparent : -1);
+               g->color_table = (stbi_uc *) g->lpal;
+            } else if (g->flags & 0x80) {
+               if (g->transparent >= 0 && (g->eflags & 0x01)) {
+                  prev_trans = g->pal[g->transparent][3];
+                  g->pal[g->transparent][3] = 0;
+               }
+               g->color_table = (stbi_uc *) g->pal;
+            } else
+               return stbi__errpuc("missing color table", "Corrupt GIF");
+
+            o = stbi__process_gif_raster(s, g);
+            if (o == NULL) return NULL;
+
+            if (prev_trans != -1)
+               g->pal[g->transparent][3] = (stbi_uc) prev_trans;
+
+            return o;
+         }
+
+         case 0x21: // Comment Extension.
+         {
+            int len;
+            if (stbi__get8(s) == 0xF9) { // Graphic Control Extension.
+               len = stbi__get8(s);
+               if (len == 4) {
+                  g->eflags = stbi__get8(s);
+                  g->delay = stbi__get16le(s);
+                  g->transparent = stbi__get8(s);
+               } else {
+                  stbi__skip(s, len);
+                  break;
+               }
+            }
+            while ((len = stbi__get8(s)) != 0)
+               stbi__skip(s, len);
+            break;
+         }
+
+         case 0x3B: // gif stream termination code
+            return (stbi_uc *) s; // using '1' causes warning on some compilers
+
+         default:
+            return stbi__errpuc("unknown code", "Corrupt GIF");
+      }
+   }
+
+   STBI_NOTUSED(req_comp);
+}
+
+static void *stbi__gif_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri)
+{
+   stbi_uc *u = 0;
+   stbi__gif* g = (stbi__gif*) stbi__malloc(sizeof(stbi__gif));
+   memset(g, 0, sizeof(*g));
+   STBI_NOTUSED(ri);
+
+   u = stbi__gif_load_next(s, g, comp, req_comp);
+   if (u == (stbi_uc *) s) u = 0;  // end of animated gif marker
+   if (u) {
+      *x = g->w;
+      *y = g->h;
+      if (req_comp && req_comp != 4)
+         u = stbi__convert_format(u, 4, req_comp, g->w, g->h);
+   }
+   else if (g->out)
+      STBI_FREE(g->out);
+   STBI_FREE(g);
+   return u;
+}
+
+static int stbi__gif_info(stbi__context *s, int *x, int *y, int *comp)
+{
+   return stbi__gif_info_raw(s,x,y,comp);
+}
+#endif
+
+// *************************************************************************************************
+// Radiance RGBE HDR loader
+// originally by Nicolas Schulz
+#ifndef STBI_NO_HDR
+static int stbi__hdr_test_core(stbi__context *s, const char *signature)
+{
+   int i;
+   for (i=0; signature[i]; ++i)
+      if (stbi__get8(s) != signature[i])
+          return 0;
+   stbi__rewind(s);
+   return 1;
+}
+
+static int stbi__hdr_test(stbi__context* s)
+{
+   int r = stbi__hdr_test_core(s, "#?RADIANCE\n");
+   stbi__rewind(s);
+   if(!r) {
+       r = stbi__hdr_test_core(s, "#?RGBE\n");
+       stbi__rewind(s);
+   }
+   return r;
+}
+
+#define STBI__HDR_BUFLEN  1024
+static char *stbi__hdr_gettoken(stbi__context *z, char *buffer)
+{
+   int len=0;
+   char c = '\0';
+
+   c = (char) stbi__get8(z);
+
+   while (!stbi__at_eof(z) && c != '\n') {
+      buffer[len++] = c;
+      if (len == STBI__HDR_BUFLEN-1) {
+         // flush to end of line
+         while (!stbi__at_eof(z) && stbi__get8(z) != '\n')
+            ;
+         break;
+      }
+      c = (char) stbi__get8(z);
+   }
+
+   buffer[len] = 0;
+   return buffer;
+}
+
+static void stbi__hdr_convert(float *output, stbi_uc *input, int req_comp)
+{
+   if ( input[3] != 0 ) {
+      float f1;
+      // Exponent
+      f1 = (float) ldexp(1.0f, input[3] - (int)(128 + 8));
+      if (req_comp <= 2)
+         output[0] = (input[0] + input[1] + input[2]) * f1 / 3;
+      else {
+         output[0] = input[0] * f1;
+         output[1] = input[1] * f1;
+         output[2] = input[2] * f1;
+      }
+      if (req_comp == 2) output[1] = 1;
+      if (req_comp == 4) output[3] = 1;
+   } else {
+      switch (req_comp) {
+         case 4: output[3] = 1; /* fallthrough */
+         case 3: output[0] = output[1] = output[2] = 0;
+                 break;
+         case 2: output[1] = 1; /* fallthrough */
+         case 1: output[0] = 0;
+                 break;
+      }
+   }
+}
+
+static float *stbi__hdr_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri)
+{
+   char buffer[STBI__HDR_BUFLEN];
+   char *token;
+   int valid = 0;
+   int width, height;
+   stbi_uc *scanline;
+   float *hdr_data;
+   int len;
+   unsigned char count, value;
+   int i, j, k, c1,c2, z;
+   const char *headerToken;
+   STBI_NOTUSED(ri);
+
+   // Check identifier
+   headerToken = stbi__hdr_gettoken(s,buffer);
+   if (strcmp(headerToken, "#?RADIANCE") != 0 && strcmp(headerToken, "#?RGBE") != 0)
+      return stbi__errpf("not HDR", "Corrupt HDR image");
+
+   // Parse header
+   for(;;) {
+      token = stbi__hdr_gettoken(s,buffer);
+      if (token[0] == 0) break;
+      if (strcmp(token, "FORMAT=32-bit_rle_rgbe") == 0) valid = 1;
+   }
+
+   if (!valid)    return stbi__errpf("unsupported format", "Unsupported HDR format");
+
+   // Parse width and height
+   // can't use sscanf() if we're not using stdio!
+   token = stbi__hdr_gettoken(s,buffer);
+   if (strncmp(token, "-Y ", 3))  return stbi__errpf("unsupported data layout", "Unsupported HDR format");
+   token += 3;
+   height = (int) strtol(token, &token, 10);
+   while (*token == ' ') ++token;
+   if (strncmp(token, "+X ", 3))  return stbi__errpf("unsupported data layout", "Unsupported HDR format");
+   token += 3;
+   width = (int) strtol(token, NULL, 10);
+
+   *x = width;
+   *y = height;
+
+   if (comp) *comp = 3;
+   if (req_comp == 0) req_comp = 3;
+
+   if (!stbi__mad4sizes_valid(width, height, req_comp, sizeof(float), 0))
+      return stbi__errpf("too large", "HDR image is too large");
+
+   // Read data
+   hdr_data = (float *) stbi__malloc_mad4(width, height, req_comp, sizeof(float), 0);
+   if (!hdr_data)
+      return stbi__errpf("outofmem", "Out of memory");
+
+   // Load image data
+   // image data is stored as some number of sca
+   if ( width < 8 || width >= 32768) {
+      // Read flat data
+      for (j=0; j < height; ++j) {
+         for (i=0; i < width; ++i) {
+            stbi_uc rgbe[4];
+           main_decode_loop:
+            stbi__getn(s, rgbe, 4);
+            stbi__hdr_convert(hdr_data + j * width * req_comp + i * req_comp, rgbe, req_comp);
+         }
+      }
+   } else {
+      // Read RLE-encoded data
+      scanline = NULL;
+
+      for (j = 0; j < height; ++j) {
+         c1 = stbi__get8(s);
+         c2 = stbi__get8(s);
+         len = stbi__get8(s);
+         if (c1 != 2 || c2 != 2 || (len & 0x80)) {
+            // not run-length encoded, so we have to actually use THIS data as a decoded
+            // pixel (note this can't be a valid pixel--one of RGB must be >= 128)
+            stbi_uc rgbe[4];
+            rgbe[0] = (stbi_uc) c1;
+            rgbe[1] = (stbi_uc) c2;
+            rgbe[2] = (stbi_uc) len;
+            rgbe[3] = (stbi_uc) stbi__get8(s);
+            stbi__hdr_convert(hdr_data, rgbe, req_comp);
+            i = 1;
+            j = 0;
+            STBI_FREE(scanline);
+            goto main_decode_loop; // yes, this makes no sense
+         }
+         len <<= 8;
+         len |= stbi__get8(s);
+         if (len != width) { STBI_FREE(hdr_data); STBI_FREE(scanline); return stbi__errpf("invalid decoded scanline length", "corrupt HDR"); }
+         if (scanline == NULL) {
+            scanline = (stbi_uc *) stbi__malloc_mad2(width, 4, 0);
+            if (!scanline) {
+               STBI_FREE(hdr_data);
+               return stbi__errpf("outofmem", "Out of memory");
+            }
+         }
+
+         for (k = 0; k < 4; ++k) {
+            int nleft;
+            i = 0;
+            while ((nleft = width - i) > 0) {
+               count = stbi__get8(s);
+               if (count > 128) {
+                  // Run
+                  value = stbi__get8(s);
+                  count -= 128;
+                  if (count > nleft) { STBI_FREE(hdr_data); STBI_FREE(scanline); return stbi__errpf("corrupt", "bad RLE data in HDR"); }
+                  for (z = 0; z < count; ++z)
+                     scanline[i++ * 4 + k] = value;
+               } else {
+                  // Dump
+                  if (count > nleft) { STBI_FREE(hdr_data); STBI_FREE(scanline); return stbi__errpf("corrupt", "bad RLE data in HDR"); }
+                  for (z = 0; z < count; ++z)
+                     scanline[i++ * 4 + k] = stbi__get8(s);
+               }
+            }
+         }
+         for (i=0; i < width; ++i)
+            stbi__hdr_convert(hdr_data+(j*width + i)*req_comp, scanline + i*4, req_comp);
+      }
+      if (scanline)
+         STBI_FREE(scanline);
+   }
+
+   return hdr_data;
+}
+
+static int stbi__hdr_info(stbi__context *s, int *x, int *y, int *comp)
+{
+   char buffer[STBI__HDR_BUFLEN];
+   char *token;
+   int valid = 0;
+   int dummy;
+
+   if (!x) x = &dummy;
+   if (!y) y = &dummy;
+   if (!comp) comp = &dummy;
+
+   if (stbi__hdr_test(s) == 0) {
+       stbi__rewind( s );
+       return 0;
+   }
+
+   for(;;) {
+      token = stbi__hdr_gettoken(s,buffer);
+      if (token[0] == 0) break;
+      if (strcmp(token, "FORMAT=32-bit_rle_rgbe") == 0) valid = 1;
+   }
+
+   if (!valid) {
+       stbi__rewind( s );
+       return 0;
+   }
+   token = stbi__hdr_gettoken(s,buffer);
+   if (strncmp(token, "-Y ", 3)) {
+       stbi__rewind( s );
+       return 0;
+   }
+   token += 3;
+   *y = (int) strtol(token, &token, 10);
+   while (*token == ' ') ++token;
+   if (strncmp(token, "+X ", 3)) {
+       stbi__rewind( s );
+       return 0;
+   }
+   token += 3;
+   *x = (int) strtol(token, NULL, 10);
+   *comp = 3;
+   return 1;
+}
+#endif // STBI_NO_HDR
+
+#ifndef STBI_NO_BMP
+static int stbi__bmp_info(stbi__context *s, int *x, int *y, int *comp)
+{
+   void *p;
+   stbi__bmp_data info;
+
+   info.all_a = 255;
+   p = stbi__bmp_parse_header(s, &info);
+   stbi__rewind( s );
+   if (p == NULL)
+      return 0;
+   if (x) *x = s->img_x;
+   if (y) *y = s->img_y;
+   if (comp) *comp = info.ma ? 4 : 3;
+   return 1;
+}
+#endif
+
+#ifndef STBI_NO_PSD
+static int stbi__psd_info(stbi__context *s, int *x, int *y, int *comp)
+{
+   int channelCount, dummy;
+   if (!x) x = &dummy;
+   if (!y) y = &dummy;
+   if (!comp) comp = &dummy;
+   if (stbi__get32be(s) != 0x38425053) {
+       stbi__rewind( s );
+       return 0;
+   }
+   if (stbi__get16be(s) != 1) {
+       stbi__rewind( s );
+       return 0;
+   }
+   stbi__skip(s, 6);
+   channelCount = stbi__get16be(s);
+   if (channelCount < 0 || channelCount > 16) {
+       stbi__rewind( s );
+       return 0;
+   }
+   *y = stbi__get32be(s);
+   *x = stbi__get32be(s);
+   if (stbi__get16be(s) != 8) {
+       stbi__rewind( s );
+       return 0;
+   }
+   if (stbi__get16be(s) != 3) {
+       stbi__rewind( s );
+       return 0;
+   }
+   *comp = 4;
+   return 1;
+}
+#endif
+
+#ifndef STBI_NO_PIC
+static int stbi__pic_info(stbi__context *s, int *x, int *y, int *comp)
+{
+   int act_comp=0,num_packets=0,chained,dummy;
+   stbi__pic_packet packets[10];
+
+   if (!x) x = &dummy;
+   if (!y) y = &dummy;
+   if (!comp) comp = &dummy;
+
+   if (!stbi__pic_is4(s,"\x53\x80\xF6\x34")) {
+      stbi__rewind(s);
+      return 0;
+   }
+
+   stbi__skip(s, 88);
+
+   *x = stbi__get16be(s);
+   *y = stbi__get16be(s);
+   if (stbi__at_eof(s)) {
+      stbi__rewind( s);
+      return 0;
+   }
+   if ( (*x) != 0 && (1 << 28) / (*x) < (*y)) {
+      stbi__rewind( s );
+      return 0;
+   }
+
+   stbi__skip(s, 8);
+
+   do {
+      stbi__pic_packet *packet;
+
+      if (num_packets==sizeof(packets)/sizeof(packets[0]))
+         return 0;
+
+      packet = &packets[num_packets++];
+      chained = stbi__get8(s);
+      packet->size    = stbi__get8(s);
+      packet->type    = stbi__get8(s);
+      packet->channel = stbi__get8(s);
+      act_comp |= packet->channel;
+
+      if (stbi__at_eof(s)) {
+          stbi__rewind( s );
+          return 0;
+      }
+      if (packet->size != 8) {
+          stbi__rewind( s );
+          return 0;
+      }
+   } while (chained);
+
+   *comp = (act_comp & 0x10 ? 4 : 3);
+
+   return 1;
+}
+#endif
+
+// *************************************************************************************************
+// Portable Gray Map and Portable Pixel Map loader
+// by Ken Miller
+//
+// PGM: http://netpbm.sourceforge.net/doc/pgm.html
+// PPM: http://netpbm.sourceforge.net/doc/ppm.html
+//
+// Known limitations:
+//    Does not support comments in the header section
+//    Does not support ASCII image data (formats P2 and P3)
+//    Does not support 16-bit-per-channel
+
+#ifndef STBI_NO_PNM
+
+static int      stbi__pnm_test(stbi__context *s)
+{
+   char p, t;
+   p = (char) stbi__get8(s);
+   t = (char) stbi__get8(s);
+   if (p != 'P' || (t != '5' && t != '6')) {
+       stbi__rewind( s );
+       return 0;
+   }
+   return 1;
+}
+
+static void *stbi__pnm_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri)
+{
+   stbi_uc *out;
+   STBI_NOTUSED(ri);
+
+   if (!stbi__pnm_info(s, (int *)&s->img_x, (int *)&s->img_y, (int *)&s->img_n))
+      return 0;
+
+   *x = s->img_x;
+   *y = s->img_y;
+   if (comp) *comp = s->img_n;
+
+   if (!stbi__mad3sizes_valid(s->img_n, s->img_x, s->img_y, 0))
+      return stbi__errpuc("too large", "PNM too large");
+
+   out = (stbi_uc *) stbi__malloc_mad3(s->img_n, s->img_x, s->img_y, 0);
+   if (!out) return stbi__errpuc("outofmem", "Out of memory");
+   stbi__getn(s, out, s->img_n * s->img_x * s->img_y);
+
+   if (req_comp && req_comp != s->img_n) {
+      out = stbi__convert_format(out, s->img_n, req_comp, s->img_x, s->img_y);
+      if (out == NULL) return out; // stbi__convert_format frees input on failure
+   }
+   return out;
+}
+
+static int      stbi__pnm_isspace(char c)
+{
+   return c == ' ' || c == '\t' || c == '\n' || c == '\v' || c == '\f' || c == '\r';
+}
+
+static void     stbi__pnm_skip_whitespace(stbi__context *s, char *c)
+{
+   for (;;) {
+      while (!stbi__at_eof(s) && stbi__pnm_isspace(*c))
+         *c = (char) stbi__get8(s);
+
+      if (stbi__at_eof(s) || *c != '#')
+         break;
+
+      while (!stbi__at_eof(s) && *c != '\n' && *c != '\r' )
+         *c = (char) stbi__get8(s);
+   }
+}
+
+static int      stbi__pnm_isdigit(char c)
+{
+   return c >= '0' && c <= '9';
+}
+
+static int      stbi__pnm_getinteger(stbi__context *s, char *c)
+{
+   int value = 0;
+
+   while (!stbi__at_eof(s) && stbi__pnm_isdigit(*c)) {
+      value = value*10 + (*c - '0');
+      *c = (char) stbi__get8(s);
+   }
+
+   return value;
+}
+
+static int      stbi__pnm_info(stbi__context *s, int *x, int *y, int *comp)
+{
+   int maxv, dummy;
+   char c, p, t;
+
+   if (!x) x = &dummy;
+   if (!y) y = &dummy;
+   if (!comp) comp = &dummy;
+
+   stbi__rewind(s);
+
+   // Get identifier
+   p = (char) stbi__get8(s);
+   t = (char) stbi__get8(s);
+   if (p != 'P' || (t != '5' && t != '6')) {
+       stbi__rewind(s);
+       return 0;
+   }
+
+   *comp = (t == '6') ? 3 : 1;  // '5' is 1-component .pgm; '6' is 3-component .ppm
+
+   c = (char) stbi__get8(s);
+   stbi__pnm_skip_whitespace(s, &c);
+
+   *x = stbi__pnm_getinteger(s, &c); // read width
+   stbi__pnm_skip_whitespace(s, &c);
+
+   *y = stbi__pnm_getinteger(s, &c); // read height
+   stbi__pnm_skip_whitespace(s, &c);
+
+   maxv = stbi__pnm_getinteger(s, &c);  // read max value
+
+   if (maxv > 255)
+      return stbi__err("max value > 255", "PPM image not 8-bit");
+   else
+      return 1;
+}
+#endif
+
+static int stbi__info_main(stbi__context *s, int *x, int *y, int *comp)
+{
+   #ifndef STBI_NO_JPEG
+   if (stbi__jpeg_info(s, x, y, comp)) return 1;
+   #endif
+
+   #ifndef STBI_NO_PNG
+   if (stbi__png_info(s, x, y, comp))  return 1;
+   #endif
+
+   #ifndef STBI_NO_GIF
+   if (stbi__gif_info(s, x, y, comp))  return 1;
+   #endif
+
+   #ifndef STBI_NO_BMP
+   if (stbi__bmp_info(s, x, y, comp))  return 1;
+   #endif
+
+   #ifndef STBI_NO_PSD
+   if (stbi__psd_info(s, x, y, comp))  return 1;
+   #endif
+
+   #ifndef STBI_NO_PIC
+   if (stbi__pic_info(s, x, y, comp))  return 1;
+   #endif
+
+   #ifndef STBI_NO_PNM
+   if (stbi__pnm_info(s, x, y, comp))  return 1;
+   #endif
+
+   #ifndef STBI_NO_HDR
+   if (stbi__hdr_info(s, x, y, comp))  return 1;
+   #endif
+
+   // test tga last because it's a crappy test!
+   #ifndef STBI_NO_TGA
+   if (stbi__tga_info(s, x, y, comp))
+       return 1;
+   #endif
+   return stbi__err("unknown image type", "Image not of any known type, or corrupt");
+}
+
+#ifndef STBI_NO_STDIO
+STBIDEF int stbi_info(char const *filename, int *x, int *y, int *comp)
+{
+    FILE *f = stbi__fopen(filename, "rb");
+    int result;
+    if (!f) return stbi__err("can't fopen", "Unable to open file");
+    result = stbi_info_from_file(f, x, y, comp);
+    fclose(f);
+    return result;
+}
+
+STBIDEF int stbi_info_from_file(FILE *f, int *x, int *y, int *comp)
+{
+   int r;
+   stbi__context s;
+   long pos = ftell(f);
+   stbi__start_file(&s, f);
+   r = stbi__info_main(&s,x,y,comp);
+   fseek(f,pos,SEEK_SET);
+   return r;
+}
+#endif // !STBI_NO_STDIO
+
+STBIDEF int stbi_info_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp)
+{
+   stbi__context s;
+   stbi__start_mem(&s,buffer,len);
+   return stbi__info_main(&s,x,y,comp);
+}
+
+STBIDEF int stbi_info_from_callbacks(stbi_io_callbacks const *c, void *user, int *x, int *y, int *comp)
+{
+   stbi__context s;
+   stbi__start_callbacks(&s, (stbi_io_callbacks *) c, user);
+   return stbi__info_main(&s,x,y,comp);
+}
+
+#endif // STB_IMAGE_IMPLEMENTATION
+
+/*
+   revision history:
+      2.15  (2017-03-18) fix png-1,2,4 bug; now all Imagenet JPGs decode;
+                         warning fixes; disable run-time SSE detection on gcc;
+                         uniform handling of optional "return" values;
+                         thread-safe initialization of zlib tables
+      2.14  (2017-03-03) remove deprecated STBI_JPEG_OLD; fixes for Imagenet JPGs
+      2.13  (2016-11-29) add 16-bit API, only supported for PNG right now
+      2.12  (2016-04-02) fix typo in 2.11 PSD fix that caused crashes
+      2.11  (2016-04-02) allocate large structures on the stack
+                         remove white matting for transparent PSD
+                         fix reported channel count for PNG & BMP
+                         re-enable SSE2 in non-gcc 64-bit
+                         support RGB-formatted JPEG
+                         read 16-bit PNGs (only as 8-bit)
+      2.10  (2016-01-22) avoid warning introduced in 2.09 by STBI_REALLOC_SIZED
+      2.09  (2016-01-16) allow comments in PNM files
+                         16-bit-per-pixel TGA (not bit-per-component)
+                         info() for TGA could break due to .hdr handling
+                         info() for BMP to shares code instead of sloppy parse
+                         can use STBI_REALLOC_SIZED if allocator doesn't support realloc
+                         code cleanup
+      2.08  (2015-09-13) fix to 2.07 cleanup, reading RGB PSD as RGBA
+      2.07  (2015-09-13) fix compiler warnings
+                         partial animated GIF support
+                         limited 16-bpc PSD support
+                         #ifdef unused functions
+                         bug with < 92 byte PIC,PNM,HDR,TGA
+      2.06  (2015-04-19) fix bug where PSD returns wrong '*comp' value
+      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) extra corruption checking (mmozeiko)
+                         stbi_set_flip_vertically_on_load (nguillemot)
+                         fix NEON support; fix mingw support
+      2.02  (2015-01-19) fix incorrect assert, fix warning
+      2.01  (2015-01-17) fix various warnings; suppress SIMD on gcc 32-bit without -msse2
+      2.00b (2014-12-25) fix STBI_MALLOC in progressive JPEG
+      2.00  (2014-12-25) optimize JPG, including x86 SSE2 & NEON SIMD (ryg)
+                         progressive JPEG (stb)
+                         PGM/PPM support (Ken Miller)
+                         STBI_MALLOC,STBI_REALLOC,STBI_FREE
+                         GIF bugfix -- seemingly never worked
+                         STBI_NO_*, STBI_ONLY_*
+      1.48  (2014-12-14) fix incorrectly-named assert()
+      1.47  (2014-12-14) 1/2/4-bit PNG support, both direct and paletted (Omar Cornut & stb)
+                         optimize PNG (ryg)
+                         fix bug in interlaced PNG with user-specified channel count (stb)
+      1.46  (2014-08-26)
+              fix broken tRNS chunk (colorkey-style transparency) in non-paletted PNG
+      1.45  (2014-08-16)
+              fix MSVC-ARM internal compiler error by wrapping malloc
+      1.44  (2014-08-07)
+              various warning fixes from Ronny Chevalier
+      1.43  (2014-07-15)
+              fix MSVC-only compiler problem in code changed in 1.42
+      1.42  (2014-07-09)
+              don't define _CRT_SECURE_NO_WARNINGS (affects user code)
+              fixes to stbi__cleanup_jpeg path
+              added STBI_ASSERT to avoid requiring assert.h
+      1.41  (2014-06-25)
+              fix search&replace from 1.36 that messed up comments/error messages
+      1.40  (2014-06-22)
+              fix gcc struct-initialization warning
+      1.39  (2014-06-15)
+              fix to TGA optimization when req_comp != number of components in TGA;
+              fix to GIF loading because BMP wasn't rewinding (whoops, no GIFs in my test suite)
+              add support for BMP version 5 (more ignored fields)
+      1.38  (2014-06-06)
+              suppress MSVC warnings on integer casts truncating values
+              fix accidental rename of 'skip' field of I/O
+      1.37  (2014-06-04)
+              remove duplicate typedef
+      1.36  (2014-06-03)
+              convert to header file single-file library
+              if de-iphone isn't set, load iphone images color-swapped instead of returning NULL
+      1.35  (2014-05-27)
+              various warnings
+              fix broken STBI_SIMD path
+              fix bug where stbi_load_from_file no longer left file pointer in correct place
+              fix broken non-easy path for 32-bit BMP (possibly never used)
+              TGA optimization by Arseny Kapoulkine
+      1.34  (unknown)
+              use STBI_NOTUSED in stbi__resample_row_generic(), fix one more leak in tga failure case
+      1.33  (2011-07-14)
+              make stbi_is_hdr work in STBI_NO_HDR (as specified), minor compiler-friendly improvements
+      1.32  (2011-07-13)
+              support for "info" function for all supported filetypes (SpartanJ)
+      1.31  (2011-06-20)
+              a few more leak fixes, bug in PNG handling (SpartanJ)
+      1.30  (2011-06-11)
+              added ability to load files via callbacks to accomidate custom input streams (Ben Wenger)
+              removed deprecated format-specific test/load functions
+              removed support for installable file formats (stbi_loader) -- would have been broken for IO callbacks anyway
+              error cases in bmp and tga give messages and don't leak (Raymond Barbiero, grisha)
+              fix inefficiency in decoding 32-bit BMP (David Woo)
+      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-stbi_uc to fix warnings
+      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 on platforms with lock-heavy fgetc()
+              minor perf improvements for jpeg
+              deprecated type-specific functions so we'll get feedback if they're needed
+              attempt to fix trans_data warning (Won Chun)
+      1.23    fixed bug in iPhone support
+      1.22  (2010-07-10)
+              removed image *writing* support
+              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 (i.stbi__err. Janez (U+017D)emva)
+      1.21    fix use of 'stbi_uc' in header (reported by jon blow)
+      1.20    added support for Softimage PIC, by Tom Seddon
+      1.19    bug in interlaced PNG corruption check (found by ryg)
+      1.18  (2008-08-02)
+              fix a threading bug (local mutable static)
+      1.17    support interlaced PNG
+      1.16    major bugfix - stbi__convert_format converted one too many pixels
+      1.15    initialize some fields for thread safety
+      1.14    fix threadsafe conversion bug
+              header-file-only version (#define STBI_HEADER_FILE_ONLY before including)
+      1.13    threadsafe
+      1.12    const qualifiers in the API
+      1.11    Support installable IDCT, colorspace conversion routines
+      1.10    Fixes for 64-bit (don't use "unsigned long")
+              optimized upsampling by Fabian "ryg" Giesen
+      1.09    Fix format-conversion for PSD code (bad global variables!)
+      1.08    Thatcher Ulrich's PSD code integrated by Nicolas Schulz
+      1.07    attempt to fix C++ warning/errors again
+      1.06    attempt to fix C++ warning/errors again
+      1.05    fix TGA loading to return correct *comp and use good luminance calc
+      1.04    default float alpha is 1, not 255; use 'void *' for stbi_image_free
+      1.03    bugfixes to STBI_NO_STDIO, STBI_NO_HDR
+      1.02    support for (subset of) HDR files, float interface for preferred access to them
+      1.01    fix bug: possible bug in handling right-side up bmps... not sure
+              fix bug: the stbi__bmp_load() and stbi__tga_load() functions didn't work at all
+      1.00    interface to zlib that skips zlib header
+      0.99    correct handling of alpha in palette
+      0.98    TGA loader by lonesock; dynamically add loaders (untested)
+      0.97    jpeg errors on too large a file; also catch another malloc failure
+      0.96    fix detection of invalid v value - particleman@mollyrocket forum
+      0.95    during header scan, seek to markers in case of padding
+      0.94    STBI_NO_STDIO to disable stdio usage; rename all #defines the same
+      0.93    handle jpegtran output; verbose errors
+      0.92    read 4,8,16,24,32-bit BMP files of several formats
+      0.91    output 24-bit Windows 3.0 BMP files
+      0.90    fix a few more warnings; bump version number to approach 1.0
+      0.61    bugfixes due to Marc LeBlanc, Christopher Lloyd
+      0.60    fix compiling as c++
+      0.59    fix warnings: merge Dave Moore's -Wall fixes
+      0.58    fix bug: zlib uncompressed mode len/nlen was wrong endian
+      0.57    fix bug: jpg last huffman symbol before marker was >9 bits but less than 16 available
+      0.56    fix bug: zlib uncompressed mode len vs. nlen
+      0.55    fix bug: restart_interval not initialized to 0
+      0.54    allow NULL for 'int *comp'
+      0.53    fix bug in png 3->4; speedup png decoding
+      0.52    png handles req_comp=3,4 directly; minor cleanup; jpeg comments
+      0.51    obey req_comp requests, 1-component jpegs return as 1-component,
+              on 'test' only check type, not whether we support this variant
+      0.50  (2006-11-19)
+              first released version
+*/
+
+
+/*
+------------------------------------------------------------------------------
+This software is available under 2 licenses -- choose whichever you prefer.
+------------------------------------------------------------------------------
+ALTERNATIVE A - MIT License
+Copyright (c) 2017 Sean Barrett
+Permission is hereby granted, free of charge, to any person obtaining a copy of
+this software and associated documentation files (the "Software"), to deal in
+the Software without restriction, including without limitation the rights to
+use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies
+of the Software, and to permit persons to whom the Software is furnished to do
+so, subject to the following conditions:
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
+------------------------------------------------------------------------------
+ALTERNATIVE B - Public Domain (www.unlicense.org)
+This is free and unencumbered software released into the public domain.
+Anyone is free to copy, modify, publish, use, compile, sell, or distribute this
+software, either in source code form or as a compiled binary, for any purpose,
+commercial or non-commercial, and by any means.
+In jurisdictions that recognize copyright laws, the author or authors of this
+software dedicate any and all copyright interest in the software to the public
+domain. We make this dedication for the benefit of the public at large and to
+the detriment of our heirs and successors. We intend this dedication to be an
+overt act of relinquishment in perpetuity of all present and future rights to
+this software under copyright law.
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+------------------------------------------------------------------------------
 */

Source/ThirdParty/STB/stb_image_write.cpp

@@ -1,879 +0,0 @@
-#include "stb_image_write.h"
-#include "../../Atomic/Container/Str.h"
-#define STB_IMAGE_WRITE_IMPLEMENTATION
-
-#ifdef STB_IMAGE_WRITE_IMPLEMENTATION
-
-#ifdef _WIN32
-   #ifndef _CRT_SECURE_NO_WARNINGS
-   #define _CRT_SECURE_NO_WARNINGS
-   #endif
-   #ifndef _CRT_NONSTDC_NO_DEPRECATE
-   #define _CRT_NONSTDC_NO_DEPRECATE
-   #endif
-#endif
-
-#ifndef STBI_WRITE_NO_STDIO
-#include <stdio.h>
-#endif // STBI_WRITE_NO_STDIO
-
-#include <stdarg.h>
-#include <stdlib.h>
-#include <string.h>
-#include <math.h>
-
-#if defined(STBIW_MALLOC) && defined(STBIW_FREE) && (defined(STBIW_REALLOC) || defined(STBIW_REALLOC_SIZED))
-// ok
-#elif !defined(STBIW_MALLOC) && !defined(STBIW_FREE) && !defined(STBIW_REALLOC) && !defined(STBIW_REALLOC_SIZED)
-// ok
-#else
-#error "Must define all or none of STBIW_MALLOC, STBIW_FREE, and STBIW_REALLOC (or STBIW_REALLOC_SIZED)."
-#endif
-
-#ifndef STBIW_MALLOC
-#define STBIW_MALLOC(sz)        malloc(sz)
-#define STBIW_REALLOC(p,newsz)  realloc(p,newsz)
-#define STBIW_FREE(p)           free(p)
-#endif
-
-#ifndef STBIW_REALLOC_SIZED
-#define STBIW_REALLOC_SIZED(p,oldsz,newsz) STBIW_REALLOC(p,newsz)
-#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
-
-#define STBIW_UCHAR(x) (unsigned char) ((x) & 0xff)
-
-typedef struct
-{
-   stbi_write_func *func;
-   void *context;
-} stbi__write_context;
-
-// initialize a callback-based context
-static void stbi__start_write_callbacks(stbi__write_context *s, stbi_write_func *c, void *context)
-{
-   s->func    = c;
-   s->context = context;
-}
-
-#ifndef STBI_WRITE_NO_STDIO
-
-static void stbi__stdio_write(void *context, void *data, int size)
-{
-   fwrite(data,1,size,(FILE*) context);
-}
-
-static int stbi__start_write_file(stbi__write_context *s, const char *filename)
-{
-   // Urho3D: proper UTF8 handling for Windows, requires Urho3D WString class
-#ifndef _WIN32
-   FILE *f = fopen(filename, "wb");
-#else
-    Atomic::WString wstr(filename);
-    FILE *f = _wfopen(wstr.CString(), L"wb");
-#endif
-   stbi__start_write_callbacks(s, stbi__stdio_write, (void *) f);
-   return f != NULL;
-}
-
-static void stbi__end_write_file(stbi__write_context *s)
-{
-   fclose((FILE *)s->context);
-}
-
-#endif // !STBI_WRITE_NO_STDIO
-
-typedef unsigned int stbiw_uint32;
-typedef int stb_image_write_test[sizeof(stbiw_uint32)==4 ? 1 : -1];
-
-#ifdef STB_IMAGE_WRITE_STATIC
-static int stbi_write_tga_with_rle = 1;
-#else
-int stbi_write_tga_with_rle = 1;
-#endif
-
-static void stbiw__writefv(stbi__write_context *s, const char *fmt, va_list v)
-{
-   while (*fmt) {
-      switch (*fmt++) {
-         case ' ': break;
-         case '1': { unsigned char x = STBIW_UCHAR(va_arg(v, int));
-                     s->func(s->context,&x,1);
-                     break; }
-         case '2': { int x = va_arg(v,int);
-                     unsigned char b[2];
-                     b[0] = STBIW_UCHAR(x);
-                     b[1] = STBIW_UCHAR(x>>8);
-                     s->func(s->context,b,2);
-                     break; }
-         case '4': { stbiw_uint32 x = va_arg(v,int);
-                     unsigned char b[4];
-                     b[0]=STBIW_UCHAR(x);
-                     b[1]=STBIW_UCHAR(x>>8);
-                     b[2]=STBIW_UCHAR(x>>16);
-                     b[3]=STBIW_UCHAR(x>>24);
-                     s->func(s->context,b,4);
-                     break; }
-         default:
-            STBIW_ASSERT(0);
-            return;
-      }
-   }
-}
-
-static void stbiw__writef(stbi__write_context *s, const char *fmt, ...)
-{
-   va_list v;
-   va_start(v, fmt);
-   stbiw__writefv(s, fmt, v);
-   va_end(v);
-}
-
-static void stbiw__write3(stbi__write_context *s, unsigned char a, unsigned char b, unsigned char c)
-{
-   unsigned char arr[3];
-   arr[0] = a, arr[1] = b, arr[2] = c;
-   s->func(s->context, arr, 3);
-}
-
-static void stbiw__write_pixel(stbi__write_context *s, int rgb_dir, int comp, int write_alpha, int expand_mono, unsigned char *d)
-{
-   unsigned char bg[3] = { 255, 0, 255}, px[3];
-   int k;
-
-   if (write_alpha < 0)
-      s->func(s->context, &d[comp - 1], 1);
-
-   switch (comp) {
-      case 1:
-         s->func(s->context,d,1);
-         break;
-      case 2:
-         if (expand_mono)
-            stbiw__write3(s, d[0], d[0], d[0]); // monochrome bmp
-         else
-            s->func(s->context, d, 1);  // monochrome TGA
-         break;
-      case 4:
-         if (!write_alpha) {
-            // composite against pink background
-            for (k = 0; k < 3; ++k)
-               px[k] = bg[k] + ((d[k] - bg[k]) * d[3]) / 255;
-            stbiw__write3(s, px[1 - rgb_dir], px[1], px[1 + rgb_dir]);
-            break;
-         }
-         /* FALLTHROUGH */
-      case 3:
-         stbiw__write3(s, d[1 - rgb_dir], d[1], d[1 + rgb_dir]);
-         break;
-   }
-   if (write_alpha > 0)
-      s->func(s->context, &d[comp - 1], 1);
-}
-
-static void stbiw__write_pixels(stbi__write_context *s, int rgb_dir, int vdir, int x, int y, int comp, void *data, int write_alpha, int scanline_pad, int expand_mono)
-{
-   stbiw_uint32 zero = 0;
-   int i,j, j_end;
-
-   if (y <= 0)
-      return;
-
-   if (vdir < 0)
-      j_end = -1, j = y-1;
-   else
-      j_end =  y, j = 0;
-
-   for (; j != j_end; j += vdir) {
-      for (i=0; i < x; ++i) {
-         unsigned char *d = (unsigned char *) data + (j*x+i)*comp;
-         stbiw__write_pixel(s, rgb_dir, comp, write_alpha, expand_mono, d);
-      }
-      s->func(s->context, &zero, scanline_pad);
-   }
-}
-
-static int stbiw__outfile(stbi__write_context *s, int rgb_dir, int vdir, int x, int y, int comp, int expand_mono, void *data, int alpha, int pad, const char *fmt, ...)
-{
-   if (y < 0 || x < 0) {
-      return 0;
-   } else {
-      va_list v;
-      va_start(v, fmt);
-      stbiw__writefv(s, fmt, v);
-      va_end(v);
-      stbiw__write_pixels(s,rgb_dir,vdir,x,y,comp,data,alpha,pad, expand_mono);
-      return 1;
-   }
-}
-
-static int stbi_write_bmp_core(stbi__write_context *s, int x, int y, int comp, const void *data)
-{
-   int pad = (-x*3) & 3;
-   return stbiw__outfile(s,-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
-}
-
-STBIWDEF int stbi_write_bmp_to_func(stbi_write_func *func, void *context, int x, int y, int comp, const void *data)
-{
-   stbi__write_context s;
-   stbi__start_write_callbacks(&s, func, context);
-   return stbi_write_bmp_core(&s, x, y, comp, data);
-}
-
-#ifndef STBI_WRITE_NO_STDIO
-STBIWDEF int stbi_write_bmp(char const *filename, int x, int y, int comp, const void *data)
-{
-   stbi__write_context s;
-   if (stbi__start_write_file(&s,filename)) {
-      int r = stbi_write_bmp_core(&s, x, y, comp, data);
-      stbi__end_write_file(&s);
-      return r;
-   } else
-      return 0;
-}
-#endif //!STBI_WRITE_NO_STDIO
-
-static int stbi_write_tga_core(stbi__write_context *s, int x, int y, int comp, void *data)
-{
-   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
-
-   if (y < 0 || x < 0)
-      return 0;
-
-   if (!stbi_write_tga_with_rle) {
-      return stbiw__outfile(s, -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);
-   } else {
-      int i,j,k;
-
-      stbiw__writef(s, "111 221 2222 11", 0,0,format+8, 0,0,0, 0,0,x,y, (colorbytes + has_alpha) * 8, has_alpha * 8);
-
-      for (j = y - 1; j >= 0; --j) {
-          unsigned char *row = (unsigned char *) data + j * x * comp;
-         int len;
-
-         for (i = 0; i < x; i += len) {
-            unsigned char *begin = row + i * comp;
-            int diff = 1;
-            len = 1;
-
-            if (i < x - 1) {
-               ++len;
-               diff = memcmp(begin, row + (i + 1) * comp, comp);
-               if (diff) {
-                  const unsigned char *prev = begin;
-                  for (k = i + 2; k < x && len < 128; ++k) {
-                     if (memcmp(prev, row + k * comp, comp)) {
-                        prev += comp;
-                        ++len;
-                     } else {
-                        --len;
-                        break;
-                     }
-                  }
-               } else {
-                  for (k = i + 2; k < x && len < 128; ++k) {
-                     if (!memcmp(begin, row + k * comp, comp)) {
-                        ++len;
-                     } else {
-                        break;
-                     }
-                  }
-               }
-            }
-
-            if (diff) {
-               unsigned char header = STBIW_UCHAR(len - 1);
-               s->func(s->context, &header, 1);
-               for (k = 0; k < len; ++k) {
-                  stbiw__write_pixel(s, -1, comp, has_alpha, 0, begin + k * comp);
-               }
-            } else {
-               unsigned char header = STBIW_UCHAR(len - 129);
-               s->func(s->context, &header, 1);
-               stbiw__write_pixel(s, -1, comp, has_alpha, 0, begin);
-            }
-         }
-      }
-   }
-   return 1;
-}
-
-int stbi_write_tga_to_func(stbi_write_func *func, void *context, int x, int y, int comp, const void *data)
-{
-   stbi__write_context s;
-   stbi__start_write_callbacks(&s, func, context);
-   return stbi_write_tga_core(&s, x, y, comp, (void *) data);
-}
-
-#ifndef STBI_WRITE_NO_STDIO
-int stbi_write_tga(char const *filename, int x, int y, int comp, const void *data)
-{
-   stbi__write_context s;
-   if (stbi__start_write_file(&s,filename)) {
-      int r = stbi_write_tga_core(&s, x, y, comp, (void *) data);
-      stbi__end_write_file(&s);
-      return r;
-   } else
-      return 0;
-}
-#endif
-
-// *************************************************************************************************
-// Radiance RGBE HDR writer
-// by Baldur Karlsson
-#ifndef STBI_WRITE_NO_STDIO
-
-#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-32f) {
-      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(stbi__write_context *s, int length, unsigned char databyte)
-{
-   unsigned char lengthbyte = STBIW_UCHAR(length+128);
-   STBIW_ASSERT(length+128 <= 255);
-   s->func(s->context, &lengthbyte, 1);
-   s->func(s->context, &databyte, 1);
-}
-
-void stbiw__write_dump_data(stbi__write_context *s, int length, unsigned char *data)
-{
-   unsigned char lengthbyte = STBIW_UCHAR(length);
-   STBIW_ASSERT(length <= 128); // inconsistent with spec but consistent with official code
-   s->func(s->context, &lengthbyte, 1);
-   s->func(s->context, data, length);
-}
-
-void stbiw__write_hdr_scanline(stbi__write_context *s, int width, int ncomp, unsigned char *scratch, 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 (ncomp) {
-            case 4: /* fallthrough */
-            case 3: linear[2] = scanline[x*ncomp + 2];
-                    linear[1] = scanline[x*ncomp + 1];
-                    linear[0] = scanline[x*ncomp + 0];
-                    break;
-            default:
-                    linear[0] = linear[1] = linear[2] = scanline[x*ncomp + 0];
-                    break;
-         }
-         stbiw__linear_to_rgbe(rgbe, linear);
-         s->func(s->context, rgbe, 4);
-      }
-   } else {
-      int c,r;
-      /* encode into scratch buffer */
-      for (x=0; x < width; x++) {
-         switch(ncomp) {
-            case 4: /* fallthrough */
-            case 3: linear[2] = scanline[x*ncomp + 2];
-                    linear[1] = scanline[x*ncomp + 1];
-                    linear[0] = scanline[x*ncomp + 0];
-                    break;
-            default:
-                    linear[0] = linear[1] = linear[2] = scanline[x*ncomp + 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];
-      }
-
-      s->func(s->context, scanlineheader, 4);
-
-      /* 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(s, 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(s, len, comp[x]);
-                  x += len;
-               }
-            }
-         }
-      }
-   }
-}
-
-static int stbi_write_hdr_core(stbi__write_context *s, int x, int y, int comp, float *data)
-{
-   if (y <= 0 || x <= 0 || data == NULL)
-      return 0;
-   else {
-      // Each component is stored separately. Allocate scratch space for full output scanline.
-      unsigned char *scratch = (unsigned char *) STBIW_MALLOC(x*4);
-      int i, len;
-      char buffer[128];
-      char header[] = "#?RADIANCE\n# Written by stb_image_write.h\nFORMAT=32-bit_rle_rgbe\n";
-      s->func(s->context, header, sizeof(header)-1);
-
-      len = sprintf(buffer, "EXPOSURE=          1.0000000000000\n\n-Y %d +X %d\n", y, x);
-      s->func(s->context, buffer, len);
-
-      for(i=0; i < y; i++)
-         stbiw__write_hdr_scanline(s, x, comp, scratch, data + comp*i*x);
-      STBIW_FREE(scratch);
-      return 1;
-   }
-}
-
-int stbi_write_hdr_to_func(stbi_write_func *func, void *context, int x, int y, int comp, const float *data)
-{
-   stbi__write_context s;
-   stbi__start_write_callbacks(&s, func, context);
-   return stbi_write_hdr_core(&s, x, y, comp, (float *) data);
-}
-
-int stbi_write_hdr(char const *filename, int x, int y, int comp, const float *data)
-{
-   stbi__write_context s;
-   if (stbi__start_write_file(&s,filename)) {
-      int r = stbi_write_hdr_core(&s, x, y, comp, (float *) data);
-      stbi__end_write_file(&s);
-      return r;
-   } else
-      return 0;
-}
-#endif // STBI_WRITE_NO_STDIO
-
-
-//////////////////////////////////////////////////////////////////////////////
-//
-// PNG writer
-//
-
-// 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 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 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 *stbiw__sbgrowf(void **arr, int increment, int itemsize)
-{
-   int m = *arr ? 2*stbiw__sbm(*arr)+increment : increment+1;
-   void *p = STBIW_REALLOC_SIZED(*arr ? stbiw__sbraw(*arr) : 0, *arr ? (stbiw__sbm(*arr)*itemsize + sizeof(int)*2) : 0, itemsize * m + sizeof(int)*2);
-   STBIW_ASSERT(p);
-   if (p) {
-      if (!*arr) ((int *) p)[1] = 0;
-      *arr = (void *) ((int *) p + 2);
-      stbiw__sbm(*arr) = m;
-   }
-   return *arr;
-}
-
-static unsigned char *stbiw__zlib_flushf(unsigned char *data, unsigned int *bitbuffer, int *bitcount)
-{
-   while (*bitcount >= 8) {
-      stbiw__sbpush(data, STBIW_UCHAR(*bitbuffer));
-      *bitbuffer >>= 8;
-      *bitcount -= 8;
-   }
-   return data;
-}
-
-static int stbiw__zlib_bitrev(int code, int codebits)
-{
-   int res=0;
-   while (codebits--) {
-      res = (res << 1) | (code & 1);
-      code >>= 1;
-   }
-   return res;
-}
-
-static unsigned int stbiw__zlib_countm(unsigned char *a, unsigned char *b, int limit)
-{
-   int i;
-   for (i=0; i < limit && i < 258; ++i)
-      if (a[i] != b[i]) break;
-   return i;
-}
-
-static unsigned int stbiw__zhash(unsigned char *data)
-{
-   stbiw_uint32 hash = data[0] + (data[1] << 8) + (data[2] << 16);
-   hash ^= hash << 3;
-   hash += hash >> 5;
-   hash ^= hash << 4;
-   hash += hash >> 17;
-   hash ^= hash << 25;
-   hash += hash >> 6;
-   return hash;
-}
-
-#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 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 stbiw__ZHASH   16384
-
-unsigned char * stbi_zlib_compress(unsigned char *data, int data_len, int *out_len, int quality)
-{
-   static unsigned short lengthc[] = { 3,4,5,6,7,8,9,10,11,13,15,17,19,23,27,31,35,43,51,59,67,83,99,115,131,163,195,227,258, 259 };
-   static unsigned char  lengtheb[]= { 0,0,0,0,0,0,0, 0, 1, 1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 4, 4, 4,  4,  5,  5,  5,  5,  0 };
-   static unsigned short distc[]   = { 1,2,3,4,5,7,9,13,17,25,33,49,65,97,129,193,257,385,513,769,1025,1537,2049,3073,4097,6145,8193,12289,16385,24577, 32768 };
-   static unsigned char  disteb[]  = { 0,0,0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,8,8,9,9,10,10,11,11,12,12,13,13 };
-   unsigned int bitbuf=0;
-   int i,j, bitcount=0;
-   unsigned char *out = NULL;
-   unsigned char ***hash_table = (unsigned char***) STBIW_MALLOC(stbiw__ZHASH * sizeof(char**));
-   if (quality < 5) quality = 5;
-
-   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 < 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 = stbiw__zhash(data+i)&(stbiw__ZHASH-1), best=3;
-      unsigned char *bestloc = 0;
-      unsigned char **hlist = hash_table[h];
-      int n = stbiw__sbcount(hlist);
-      for (j=0; j < n; ++j) {
-         if (hlist[j]-data > i-32768) { // if entry lies within window
-            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] && 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;
-      }
-      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 = stbiw__zhash(data+i+1)&(stbiw__ZHASH-1);
-         hlist = hash_table[h];
-         n = stbiw__sbcount(hlist);
-         for (j=0; j < n; ++j) {
-            if (hlist[j]-data > i-32767) {
-               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;
-               }
-            }
-         }
-      }
-
-      if (bestloc) {
-         int d = (int) (data+i - bestloc); // distance back
-         STBIW_ASSERT(d <= 32767 && best <= 258);
-         for (j=0; best > lengthc[j+1]-1; ++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);
-         stbiw__zlib_add(stbiw__zlib_bitrev(j,5),5);
-         if (disteb[j]) stbiw__zlib_add(d - distc[j], disteb[j]);
-         i += best;
-      } else {
-         stbiw__zlib_huffb(data[i]);
-         ++i;
-      }
-   }
-   // write out final bytes
-   for (;i < data_len; ++i)
-      stbiw__zlib_huffb(data[i]);
-   stbiw__zlib_huff(256); // end of block
-   // pad with 0 bits to byte boundary
-   while (bitcount)
-      stbiw__zlib_add(0,1);
-
-   for (i=0; i < stbiw__ZHASH; ++i)
-      (void) stbiw__sbfree(hash_table[i]);
-   STBIW_FREE(hash_table);
-
-   {
-      // compute adler32 on input
-      unsigned int s1=1, s2=0;
-      int blocklen = (int) (data_len % 5552);
-      j=0;
-      while (j < data_len) {
-         for (i=0; i < blocklen; ++i) s1 += data[j+i], s2 += s1;
-         s1 %= 65521, s2 %= 65521;
-         j += blocklen;
-         blocklen = 5552;
-      }
-      stbiw__sbpush(out, STBIW_UCHAR(s2 >> 8));
-      stbiw__sbpush(out, STBIW_UCHAR(s2));
-      stbiw__sbpush(out, STBIW_UCHAR(s1 >> 8));
-      stbiw__sbpush(out, STBIW_UCHAR(s1));
-   }
-   *out_len = stbiw__sbn(out);
-   // make returned pointer freeable
-   STBIW_MEMMOVE(stbiw__sbraw(out), out, *out_len);
-   return (unsigned char *) stbiw__sbraw(out);
-}
-
-static unsigned int stbiw__crc32(unsigned char *buffer, int len)
-{
-   static unsigned int crc_table[256] =
-   {
-      0x00000000, 0x77073096, 0xEE0E612C, 0x990951BA, 0x076DC419, 0x706AF48F, 0xE963A535, 0x9E6495A3,
-      0x0eDB8832, 0x79DCB8A4, 0xE0D5E91E, 0x97D2D988, 0x09B64C2B, 0x7EB17CBD, 0xE7B82D07, 0x90BF1D91,
-      0x1DB71064, 0x6AB020F2, 0xF3B97148, 0x84BE41DE, 0x1ADAD47D, 0x6DDDE4EB, 0xF4D4B551, 0x83D385C7,
-      0x136C9856, 0x646BA8C0, 0xFD62F97A, 0x8A65C9EC, 0x14015C4F, 0x63066CD9, 0xFA0F3D63, 0x8D080DF5,
-      0x3B6E20C8, 0x4C69105E, 0xD56041E4, 0xA2677172, 0x3C03E4D1, 0x4B04D447, 0xD20D85FD, 0xA50AB56B,
-      0x35B5A8FA, 0x42B2986C, 0xDBBBC9D6, 0xACBCF940, 0x32D86CE3, 0x45DF5C75, 0xDCD60DCF, 0xABD13D59,
-      0x26D930AC, 0x51DE003A, 0xC8D75180, 0xBFD06116, 0x21B4F4B5, 0x56B3C423, 0xCFBA9599, 0xB8BDA50F,
-      0x2802B89E, 0x5F058808, 0xC60CD9B2, 0xB10BE924, 0x2F6F7C87, 0x58684C11, 0xC1611DAB, 0xB6662D3D,
-      0x76DC4190, 0x01DB7106, 0x98D220BC, 0xEFD5102A, 0x71B18589, 0x06B6B51F, 0x9FBFE4A5, 0xE8B8D433,
-      0x7807C9A2, 0x0F00F934, 0x9609A88E, 0xE10E9818, 0x7F6A0DBB, 0x086D3D2D, 0x91646C97, 0xE6635C01,
-      0x6B6B51F4, 0x1C6C6162, 0x856530D8, 0xF262004E, 0x6C0695ED, 0x1B01A57B, 0x8208F4C1, 0xF50FC457,
-      0x65B0D9C6, 0x12B7E950, 0x8BBEB8EA, 0xFCB9887C, 0x62DD1DDF, 0x15DA2D49, 0x8CD37CF3, 0xFBD44C65,
-      0x4DB26158, 0x3AB551CE, 0xA3BC0074, 0xD4BB30E2, 0x4ADFA541, 0x3DD895D7, 0xA4D1C46D, 0xD3D6F4FB,
-      0x4369E96A, 0x346ED9FC, 0xAD678846, 0xDA60B8D0, 0x44042D73, 0x33031DE5, 0xAA0A4C5F, 0xDD0D7CC9,
-      0x5005713C, 0x270241AA, 0xBE0B1010, 0xC90C2086, 0x5768B525, 0x206F85B3, 0xB966D409, 0xCE61E49F,
-      0x5EDEF90E, 0x29D9C998, 0xB0D09822, 0xC7D7A8B4, 0x59B33D17, 0x2EB40D81, 0xB7BD5C3B, 0xC0BA6CAD,
-      0xEDB88320, 0x9ABFB3B6, 0x03B6E20C, 0x74B1D29A, 0xEAD54739, 0x9DD277AF, 0x04DB2615, 0x73DC1683,
-      0xE3630B12, 0x94643B84, 0x0D6D6A3E, 0x7A6A5AA8, 0xE40ECF0B, 0x9309FF9D, 0x0A00AE27, 0x7D079EB1,
-      0xF00F9344, 0x8708A3D2, 0x1E01F268, 0x6906C2FE, 0xF762575D, 0x806567CB, 0x196C3671, 0x6E6B06E7,
-      0xFED41B76, 0x89D32BE0, 0x10DA7A5A, 0x67DD4ACC, 0xF9B9DF6F, 0x8EBEEFF9, 0x17B7BE43, 0x60B08ED5,
-      0xD6D6A3E8, 0xA1D1937E, 0x38D8C2C4, 0x4FDFF252, 0xD1BB67F1, 0xA6BC5767, 0x3FB506DD, 0x48B2364B,
-      0xD80D2BDA, 0xAF0A1B4C, 0x36034AF6, 0x41047A60, 0xDF60EFC3, 0xA867DF55, 0x316E8EEF, 0x4669BE79,
-      0xCB61B38C, 0xBC66831A, 0x256FD2A0, 0x5268E236, 0xCC0C7795, 0xBB0B4703, 0x220216B9, 0x5505262F,
-      0xC5BA3BBE, 0xB2BD0B28, 0x2BB45A92, 0x5CB36A04, 0xC2D7FFA7, 0xB5D0CF31, 0x2CD99E8B, 0x5BDEAE1D,
-      0x9B64C2B0, 0xEC63F226, 0x756AA39C, 0x026D930A, 0x9C0906A9, 0xEB0E363F, 0x72076785, 0x05005713,
-      0x95BF4A82, 0xE2B87A14, 0x7BB12BAE, 0x0CB61B38, 0x92D28E9B, 0xE5D5BE0D, 0x7CDCEFB7, 0x0BDBDF21,
-      0x86D3D2D4, 0xF1D4E242, 0x68DDB3F8, 0x1FDA836E, 0x81BE16CD, 0xF6B9265B, 0x6FB077E1, 0x18B74777,
-      0x88085AE6, 0xFF0F6A70, 0x66063BCA, 0x11010B5C, 0x8F659EFF, 0xF862AE69, 0x616BFFD3, 0x166CCF45,
-      0xA00AE278, 0xD70DD2EE, 0x4E048354, 0x3903B3C2, 0xA7672661, 0xD06016F7, 0x4969474D, 0x3E6E77DB,
-      0xAED16A4A, 0xD9D65ADC, 0x40DF0B66, 0x37D83BF0, 0xA9BCAE53, 0xDEBB9EC5, 0x47B2CF7F, 0x30B5FFE9,
-      0xBDBDF21C, 0xCABAC28A, 0x53B39330, 0x24B4A3A6, 0xBAD03605, 0xCDD70693, 0x54DE5729, 0x23D967BF,
-      0xB3667A2E, 0xC4614AB8, 0x5D681B02, 0x2A6F2B94, 0xB40BBE37, 0xC30C8EA1, 0x5A05DF1B, 0x2D02EF8D
-   };
-
-   unsigned int crc = ~0u;
-   int i;
-   for (i=0; i < len; ++i)
-      crc = (crc >> 8) ^ crc_table[buffer[i] ^ (crc & 0xff)];
-   return ~crc;
-}
-
-#define stbiw__wpng4(o,a,b,c,d) ((o)[0]=STBIW_UCHAR(a),(o)[1]=STBIW_UCHAR(b),(o)[2]=STBIW_UCHAR(c),(o)[3]=STBIW_UCHAR(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 stbiw__wpcrc(unsigned char **data, int len)
-{
-   unsigned int crc = stbiw__crc32(*data - len - 4, len+4);
-   stbiw__wp32(*data, crc);
-}
-
-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 STBIW_UCHAR(a);
-   if (pb <= pc) return STBIW_UCHAR(b);
-   return STBIW_UCHAR(c);
-}
-
-unsigned char *stbi_write_png_to_mem(unsigned char *pixels, int stride_bytes, int x, int y, int n, int *out_len)
-{
-   int ctype[5] = { -1, 0, 4, 2, 6 };
-   unsigned char sig[8] = { 137,80,78,71,13,10,26,10 };
-   unsigned char *out,*o, *filt, *zlib;
-   signed char *line_buffer;
-   int i,j,k,p,zlen;
-
-   if (stride_bytes == 0)
-      stride_bytes = x * n;
-
-   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 };
-      int *mymap = j ? mapping : firstmap;
-      int best = 0, bestval = 0x7fffffff;
-      for (p=0; p < 2; ++p) {
-         for (k= p?best:0; k < 5; ++k) {
-            int type = mymap[k],est=0;
-            unsigned char *z = pixels + stride_bytes*j;
-            for (i=0; i < n; ++i)
-               switch (type) {
-                  case 0: line_buffer[i] = z[i]; break;
-                  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] - 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;
-               }
-            for (i=n; i < x*n; ++i) {
-               switch (type) {
-                  case 0: line_buffer[i] = z[i]; break;
-                  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] - 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] - stbiw__paeth(z[i-n], 0,0); break;
-               }
-            }
-            if (p) break;
-            for (i=0; i < x*n; ++i)
-               est += abs((signed char) line_buffer[i]);
-            if (est < bestval) { bestval = est; best = k; }
-         }
-      }
-      // when we get here, best contains the filter type, and line_buffer contains the data
-      filt[j*(x*n+1)] = (unsigned char) best;
-      STBIW_MEMMOVE(filt+j*(x*n+1)+1, line_buffer, x*n);
-   }
-   STBIW_FREE(line_buffer);
-   zlib = stbi_zlib_compress(filt, y*( x*n+1), &zlen, 8); // increase 8 to get smaller but use more memory
-   STBIW_FREE(filt);
-   if (!zlib) return 0;
-
-   // each tag requires 12 bytes of overhead
-   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;
-   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++ = STBIW_UCHAR(ctype[n]);
-   *o++ = 0;
-   *o++ = 0;
-   *o++ = 0;
-   stbiw__wpcrc(&o,13);
-
-   stbiw__wp32(o, zlen);
-   stbiw__wptag(o, "IDAT");
-   STBIW_MEMMOVE(o, zlib, zlen);
-   o += zlen;
-   STBIW_FREE(zlib);
-   stbiw__wpcrc(&o, zlen);
-
-   stbiw__wp32(o,0);
-   stbiw__wptag(o, "IEND");
-   stbiw__wpcrc(&o,0);
-
-   STBIW_ASSERT(o == out + *out_len);
-
-   return out;
-}
-
-#ifndef STBI_WRITE_NO_STDIO
-STBIWDEF int stbi_write_png(char const *filename, int x, int y, int comp, const void *data, int stride_bytes)
-{
-   FILE *f;
-   int len;
-   unsigned char *png = stbi_write_png_to_mem((unsigned char *) data, stride_bytes, x, y, comp, &len);
-   if (png == NULL) return 0;
-   f = fopen(filename, "wb");
-   if (!f) { STBIW_FREE(png); return 0; }
-   fwrite(png, 1, len, f);
-   fclose(f);
-   STBIW_FREE(png);
-   return 1;
-}
-#endif
-
-STBIWDEF int stbi_write_png_to_func(stbi_write_func *func, void *context, int x, int y, int comp, const void *data, int stride_bytes)
-{
-   int len;
-   unsigned char *png = stbi_write_png_to_mem((unsigned char *) data, stride_bytes, x, y, comp, &len);
-   if (png == NULL) return 0;
-   func(context, png, len);
-   STBIW_FREE(png);
-   return 1;
-}
-
-#endif // STB_IMAGE_WRITE_IMPLEMENTATION

Source/ThirdParty/STB/stb_image_write.h

@@ -1,4 +1,4 @@
-/* stb_image_write - v1.02 - public domain - http://nothings.org/stb/stb_image_write.h
+/* stb_image_write - v1.05 - public domain - http://nothings.org/stb/stb_image_write.h
    writes out PNG/BMP/TGA images to C stdio - Sean Barrett 2010-2015
                                      no warranty implied; use at your own risk
 
@@ -103,20 +103,22 @@ CREDITS:
       Jonas Karlsson
       Filip Wasil
       Thatcher Ulrich
-
+      github:poppolopoppo
+      Patrick Boettcher
+      
 LICENSE
 
-This software is dual-licensed to the public domain and under the following
-license: you are granted a perpetual, irrevocable license to copy, modify,
-publish, and distribute this file as you see fit.
+  See end of file for license information.
 
 */
 
-// Modified by Lasse Oorni for Urho3D
-
 #ifndef INCLUDE_STB_IMAGE_WRITE_H
 #define INCLUDE_STB_IMAGE_WRITE_H
 
+// ATOMIC BEGIN
+#include "../../Atomic/Container/Str.h"
+// ATOMIC END
+
 #ifdef __cplusplus
 extern "C" {
 #endif
@@ -152,7 +154,885 @@ unsigned char *stbi_write_png_to_mem(unsigned char *pixels, int stride_bytes, in
 
 #endif//INCLUDE_STB_IMAGE_WRITE_H
 
+#ifdef STB_IMAGE_WRITE_IMPLEMENTATION
+
+#ifdef _WIN32
+   #ifndef _CRT_SECURE_NO_WARNINGS
+   #define _CRT_SECURE_NO_WARNINGS
+   #endif
+   #ifndef _CRT_NONSTDC_NO_DEPRECATE
+   #define _CRT_NONSTDC_NO_DEPRECATE
+   #endif
+#endif
+
+#ifndef STBI_WRITE_NO_STDIO
+#include <stdio.h>
+#endif // STBI_WRITE_NO_STDIO
+
+#include <stdarg.h>
+#include <stdlib.h>
+#include <string.h>
+#include <math.h>
+
+#if defined(STBIW_MALLOC) && defined(STBIW_FREE) && (defined(STBIW_REALLOC) || defined(STBIW_REALLOC_SIZED))
+// ok
+#elif !defined(STBIW_MALLOC) && !defined(STBIW_FREE) && !defined(STBIW_REALLOC) && !defined(STBIW_REALLOC_SIZED)
+// ok
+#else
+#error "Must define all or none of STBIW_MALLOC, STBIW_FREE, and STBIW_REALLOC (or STBIW_REALLOC_SIZED)."
+#endif
+
+#ifndef STBIW_MALLOC
+#define STBIW_MALLOC(sz)        malloc(sz)
+#define STBIW_REALLOC(p,newsz)  realloc(p,newsz)
+#define STBIW_FREE(p)           free(p)
+#endif
+
+#ifndef STBIW_REALLOC_SIZED
+#define STBIW_REALLOC_SIZED(p,oldsz,newsz) STBIW_REALLOC(p,newsz)
+#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
+
+#define STBIW_UCHAR(x) (unsigned char) ((x) & 0xff)
+
+typedef struct
+{
+   stbi_write_func *func;
+   void *context;
+} stbi__write_context;
+
+// initialize a callback-based context
+static void stbi__start_write_callbacks(stbi__write_context *s, stbi_write_func *c, void *context)
+{
+   s->func    = c;
+   s->context = context;
+}
+
+#ifndef STBI_WRITE_NO_STDIO
+
+static void stbi__stdio_write(void *context, void *data, int size)
+{
+   fwrite(data,1,size,(FILE*) context);
+}
+
+static int stbi__start_write_file(stbi__write_context *s, const char *filename)
+{
+   // Urho3D: proper UTF8 handling for Windows, requires Urho3D WString class
+#ifndef _WIN32
+   FILE *f = fopen(filename, "wb");
+#else
+    Atomic::WString wstr(filename);
+    FILE *f = _wfopen(wstr.CString(), L"wb");
+#endif
+   stbi__start_write_callbacks(s, stbi__stdio_write, (void *) f);
+   return f != NULL;
+}
+
+static void stbi__end_write_file(stbi__write_context *s)
+{
+   fclose((FILE *)s->context);
+}
+
+#endif // !STBI_WRITE_NO_STDIO
+
+typedef unsigned int stbiw_uint32;
+typedef int stb_image_write_test[sizeof(stbiw_uint32)==4 ? 1 : -1];
+
+#ifdef STB_IMAGE_WRITE_STATIC
+static int stbi_write_tga_with_rle = 1;
+#else
+int stbi_write_tga_with_rle = 1;
+#endif
+
+static void stbiw__writefv(stbi__write_context *s, const char *fmt, va_list v)
+{
+   while (*fmt) {
+      switch (*fmt++) {
+         case ' ': break;
+         case '1': { unsigned char x = STBIW_UCHAR(va_arg(v, int));
+                     s->func(s->context,&x,1);
+                     break; }
+         case '2': { int x = va_arg(v,int);
+                     unsigned char b[2];
+                     b[0] = STBIW_UCHAR(x);
+                     b[1] = STBIW_UCHAR(x>>8);
+                     s->func(s->context,b,2);
+                     break; }
+         case '4': { stbiw_uint32 x = va_arg(v,int);
+                     unsigned char b[4];
+                     b[0]=STBIW_UCHAR(x);
+                     b[1]=STBIW_UCHAR(x>>8);
+                     b[2]=STBIW_UCHAR(x>>16);
+                     b[3]=STBIW_UCHAR(x>>24);
+                     s->func(s->context,b,4);
+                     break; }
+         default:
+            STBIW_ASSERT(0);
+            return;
+      }
+   }
+}
+
+static void stbiw__writef(stbi__write_context *s, const char *fmt, ...)
+{
+   va_list v;
+   va_start(v, fmt);
+   stbiw__writefv(s, fmt, v);
+   va_end(v);
+}
+
+static void stbiw__write3(stbi__write_context *s, unsigned char a, unsigned char b, unsigned char c)
+{
+   unsigned char arr[3];
+   arr[0] = a, arr[1] = b, arr[2] = c;
+   s->func(s->context, arr, 3);
+}
+
+static void stbiw__write_pixel(stbi__write_context *s, int rgb_dir, int comp, int write_alpha, int expand_mono, unsigned char *d)
+{
+   unsigned char bg[3] = { 255, 0, 255}, px[3];
+   int k;
+
+   if (write_alpha < 0)
+      s->func(s->context, &d[comp - 1], 1);
+
+   switch (comp) {
+      case 2: // 2 pixels = mono + alpha, alpha is written separately, so same as 1-channel case
+      case 1:
+         if (expand_mono)
+            stbiw__write3(s, d[0], d[0], d[0]); // monochrome bmp
+         else
+            s->func(s->context, d, 1);  // monochrome TGA
+         break;
+      case 4:
+         if (!write_alpha) {
+            // composite against pink background
+            for (k = 0; k < 3; ++k)
+               px[k] = bg[k] + ((d[k] - bg[k]) * d[3]) / 255;
+            stbiw__write3(s, px[1 - rgb_dir], px[1], px[1 + rgb_dir]);
+            break;
+         }
+         /* FALLTHROUGH */
+      case 3:
+         stbiw__write3(s, d[1 - rgb_dir], d[1], d[1 + rgb_dir]);
+         break;
+   }
+   if (write_alpha > 0)
+      s->func(s->context, &d[comp - 1], 1);
+}
+
+static void stbiw__write_pixels(stbi__write_context *s, int rgb_dir, int vdir, int x, int y, int comp, void *data, int write_alpha, int scanline_pad, int expand_mono)
+{
+   stbiw_uint32 zero = 0;
+   int i,j, j_end;
+
+   if (y <= 0)
+      return;
+
+   if (vdir < 0)
+      j_end = -1, j = y-1;
+   else
+      j_end =  y, j = 0;
+
+   for (; j != j_end; j += vdir) {
+      for (i=0; i < x; ++i) {
+         unsigned char *d = (unsigned char *) data + (j*x+i)*comp;
+         stbiw__write_pixel(s, rgb_dir, comp, write_alpha, expand_mono, d);
+      }
+      s->func(s->context, &zero, scanline_pad);
+   }
+}
+
+static int stbiw__outfile(stbi__write_context *s, int rgb_dir, int vdir, int x, int y, int comp, int expand_mono, void *data, int alpha, int pad, const char *fmt, ...)
+{
+   if (y < 0 || x < 0) {
+      return 0;
+   } else {
+      va_list v;
+      va_start(v, fmt);
+      stbiw__writefv(s, fmt, v);
+      va_end(v);
+      stbiw__write_pixels(s,rgb_dir,vdir,x,y,comp,data,alpha,pad, expand_mono);
+      return 1;
+   }
+}
+
+static int stbi_write_bmp_core(stbi__write_context *s, int x, int y, int comp, const void *data)
+{
+   int pad = (-x*3) & 3;
+   return stbiw__outfile(s,-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
+}
+
+STBIWDEF int stbi_write_bmp_to_func(stbi_write_func *func, void *context, int x, int y, int comp, const void *data)
+{
+   stbi__write_context s;
+   stbi__start_write_callbacks(&s, func, context);
+   return stbi_write_bmp_core(&s, x, y, comp, data);
+}
+
+#ifndef STBI_WRITE_NO_STDIO
+STBIWDEF int stbi_write_bmp(char const *filename, int x, int y, int comp, const void *data)
+{
+   stbi__write_context s;
+   if (stbi__start_write_file(&s,filename)) {
+      int r = stbi_write_bmp_core(&s, x, y, comp, data);
+      stbi__end_write_file(&s);
+      return r;
+   } else
+      return 0;
+}
+#endif //!STBI_WRITE_NO_STDIO
+
+static int stbi_write_tga_core(stbi__write_context *s, int x, int y, int comp, void *data)
+{
+   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
+
+   if (y < 0 || x < 0)
+      return 0;
+
+   if (!stbi_write_tga_with_rle) {
+      return stbiw__outfile(s, -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);
+   } else {
+      int i,j,k;
+
+      stbiw__writef(s, "111 221 2222 11", 0,0,format+8, 0,0,0, 0,0,x,y, (colorbytes + has_alpha) * 8, has_alpha * 8);
+
+      for (j = y - 1; j >= 0; --j) {
+          unsigned char *row = (unsigned char *) data + j * x * comp;
+         int len;
+
+         for (i = 0; i < x; i += len) {
+            unsigned char *begin = row + i * comp;
+            int diff = 1;
+            len = 1;
+
+            if (i < x - 1) {
+               ++len;
+               diff = memcmp(begin, row + (i + 1) * comp, comp);
+               if (diff) {
+                  const unsigned char *prev = begin;
+                  for (k = i + 2; k < x && len < 128; ++k) {
+                     if (memcmp(prev, row + k * comp, comp)) {
+                        prev += comp;
+                        ++len;
+                     } else {
+                        --len;
+                        break;
+                     }
+                  }
+               } else {
+                  for (k = i + 2; k < x && len < 128; ++k) {
+                     if (!memcmp(begin, row + k * comp, comp)) {
+                        ++len;
+                     } else {
+                        break;
+                     }
+                  }
+               }
+            }
+
+            if (diff) {
+               unsigned char header = STBIW_UCHAR(len - 1);
+               s->func(s->context, &header, 1);
+               for (k = 0; k < len; ++k) {
+                  stbiw__write_pixel(s, -1, comp, has_alpha, 0, begin + k * comp);
+               }
+            } else {
+               unsigned char header = STBIW_UCHAR(len - 129);
+               s->func(s->context, &header, 1);
+               stbiw__write_pixel(s, -1, comp, has_alpha, 0, begin);
+            }
+         }
+      }
+   }
+   return 1;
+}
+
+int stbi_write_tga_to_func(stbi_write_func *func, void *context, int x, int y, int comp, const void *data)
+{
+   stbi__write_context s;
+   stbi__start_write_callbacks(&s, func, context);
+   return stbi_write_tga_core(&s, x, y, comp, (void *) data);
+}
+
+#ifndef STBI_WRITE_NO_STDIO
+int stbi_write_tga(char const *filename, int x, int y, int comp, const void *data)
+{
+   stbi__write_context s;
+   if (stbi__start_write_file(&s,filename)) {
+      int r = stbi_write_tga_core(&s, x, y, comp, (void *) data);
+      stbi__end_write_file(&s);
+      return r;
+   } else
+      return 0;
+}
+#endif
+
+// *************************************************************************************************
+// 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-32f) {
+      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(stbi__write_context *s, int length, unsigned char databyte)
+{
+   unsigned char lengthbyte = STBIW_UCHAR(length+128);
+   STBIW_ASSERT(length+128 <= 255);
+   s->func(s->context, &lengthbyte, 1);
+   s->func(s->context, &databyte, 1);
+}
+
+void stbiw__write_dump_data(stbi__write_context *s, int length, unsigned char *data)
+{
+   unsigned char lengthbyte = STBIW_UCHAR(length);
+   STBIW_ASSERT(length <= 128); // inconsistent with spec but consistent with official code
+   s->func(s->context, &lengthbyte, 1);
+   s->func(s->context, data, length);
+}
+
+void stbiw__write_hdr_scanline(stbi__write_context *s, int width, int ncomp, unsigned char *scratch, 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 (ncomp) {
+            case 4: /* fallthrough */
+            case 3: linear[2] = scanline[x*ncomp + 2];
+                    linear[1] = scanline[x*ncomp + 1];
+                    linear[0] = scanline[x*ncomp + 0];
+                    break;
+            default:
+                    linear[0] = linear[1] = linear[2] = scanline[x*ncomp + 0];
+                    break;
+         }
+         stbiw__linear_to_rgbe(rgbe, linear);
+         s->func(s->context, rgbe, 4);
+      }
+   } else {
+      int c,r;
+      /* encode into scratch buffer */
+      for (x=0; x < width; x++) {
+         switch(ncomp) {
+            case 4: /* fallthrough */
+            case 3: linear[2] = scanline[x*ncomp + 2];
+                    linear[1] = scanline[x*ncomp + 1];
+                    linear[0] = scanline[x*ncomp + 0];
+                    break;
+            default:
+                    linear[0] = linear[1] = linear[2] = scanline[x*ncomp + 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];
+      }
+
+      s->func(s->context, scanlineheader, 4);
+
+      /* 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(s, 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(s, len, comp[x]);
+                  x += len;
+               }
+            }
+         }
+      }
+   }
+}
+
+static int stbi_write_hdr_core(stbi__write_context *s, int x, int y, int comp, float *data)
+{
+   if (y <= 0 || x <= 0 || data == NULL)
+      return 0;
+   else {
+      // Each component is stored separately. Allocate scratch space for full output scanline.
+      unsigned char *scratch = (unsigned char *) STBIW_MALLOC(x*4);
+      int i, len;
+      char buffer[128];
+      char header[] = "#?RADIANCE\n# Written by stb_image_write.h\nFORMAT=32-bit_rle_rgbe\n";
+      s->func(s->context, header, sizeof(header)-1);
+
+      len = sprintf(buffer, "EXPOSURE=          1.0000000000000\n\n-Y %d +X %d\n", y, x);
+      s->func(s->context, buffer, len);
+
+      for(i=0; i < y; i++)
+         stbiw__write_hdr_scanline(s, x, comp, scratch, data + comp*i*x);
+      STBIW_FREE(scratch);
+      return 1;
+   }
+}
+
+int stbi_write_hdr_to_func(stbi_write_func *func, void *context, int x, int y, int comp, const float *data)
+{
+   stbi__write_context s;
+   stbi__start_write_callbacks(&s, func, context);
+   return stbi_write_hdr_core(&s, x, y, comp, (float *) data);
+}
+
+#ifndef STBI_WRITE_NO_STDIO
+int stbi_write_hdr(char const *filename, int x, int y, int comp, const float *data)
+{
+   stbi__write_context s;
+   if (stbi__start_write_file(&s,filename)) {
+      int r = stbi_write_hdr_core(&s, x, y, comp, (float *) data);
+      stbi__end_write_file(&s);
+      return r;
+   } else
+      return 0;
+}
+#endif // STBI_WRITE_NO_STDIO
+
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// PNG writer
+//
+
+// 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 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 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 *stbiw__sbgrowf(void **arr, int increment, int itemsize)
+{
+   int m = *arr ? 2*stbiw__sbm(*arr)+increment : increment+1;
+   void *p = STBIW_REALLOC_SIZED(*arr ? stbiw__sbraw(*arr) : 0, *arr ? (stbiw__sbm(*arr)*itemsize + sizeof(int)*2) : 0, itemsize * m + sizeof(int)*2);
+   STBIW_ASSERT(p);
+   if (p) {
+      if (!*arr) ((int *) p)[1] = 0;
+      *arr = (void *) ((int *) p + 2);
+      stbiw__sbm(*arr) = m;
+   }
+   return *arr;
+}
+
+static unsigned char *stbiw__zlib_flushf(unsigned char *data, unsigned int *bitbuffer, int *bitcount)
+{
+   while (*bitcount >= 8) {
+      stbiw__sbpush(data, STBIW_UCHAR(*bitbuffer));
+      *bitbuffer >>= 8;
+      *bitcount -= 8;
+   }
+   return data;
+}
+
+static int stbiw__zlib_bitrev(int code, int codebits)
+{
+   int res=0;
+   while (codebits--) {
+      res = (res << 1) | (code & 1);
+      code >>= 1;
+   }
+   return res;
+}
+
+static unsigned int stbiw__zlib_countm(unsigned char *a, unsigned char *b, int limit)
+{
+   int i;
+   for (i=0; i < limit && i < 258; ++i)
+      if (a[i] != b[i]) break;
+   return i;
+}
+
+static unsigned int stbiw__zhash(unsigned char *data)
+{
+   stbiw_uint32 hash = data[0] + (data[1] << 8) + (data[2] << 16);
+   hash ^= hash << 3;
+   hash += hash >> 5;
+   hash ^= hash << 4;
+   hash += hash >> 17;
+   hash ^= hash << 25;
+   hash += hash >> 6;
+   return hash;
+}
+
+#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 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 stbiw__ZHASH   16384
+
+unsigned char * stbi_zlib_compress(unsigned char *data, int data_len, int *out_len, int quality)
+{
+   static unsigned short lengthc[] = { 3,4,5,6,7,8,9,10,11,13,15,17,19,23,27,31,35,43,51,59,67,83,99,115,131,163,195,227,258, 259 };
+   static unsigned char  lengtheb[]= { 0,0,0,0,0,0,0, 0, 1, 1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 4, 4, 4,  4,  5,  5,  5,  5,  0 };
+   static unsigned short distc[]   = { 1,2,3,4,5,7,9,13,17,25,33,49,65,97,129,193,257,385,513,769,1025,1537,2049,3073,4097,6145,8193,12289,16385,24577, 32768 };
+   static unsigned char  disteb[]  = { 0,0,0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,8,8,9,9,10,10,11,11,12,12,13,13 };
+   unsigned int bitbuf=0;
+   int i,j, bitcount=0;
+   unsigned char *out = NULL;
+   unsigned char ***hash_table = (unsigned char***) STBIW_MALLOC(stbiw__ZHASH * sizeof(char**));
+   if (quality < 5) quality = 5;
+
+   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 < 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 = stbiw__zhash(data+i)&(stbiw__ZHASH-1), best=3;
+      unsigned char *bestloc = 0;
+      unsigned char **hlist = hash_table[h];
+      int n = stbiw__sbcount(hlist);
+      for (j=0; j < n; ++j) {
+         if (hlist[j]-data > i-32768) { // if entry lies within window
+            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] && 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;
+      }
+      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 = stbiw__zhash(data+i+1)&(stbiw__ZHASH-1);
+         hlist = hash_table[h];
+         n = stbiw__sbcount(hlist);
+         for (j=0; j < n; ++j) {
+            if (hlist[j]-data > i-32767) {
+               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;
+               }
+            }
+         }
+      }
+
+      if (bestloc) {
+         int d = (int) (data+i - bestloc); // distance back
+         STBIW_ASSERT(d <= 32767 && best <= 258);
+         for (j=0; best > lengthc[j+1]-1; ++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);
+         stbiw__zlib_add(stbiw__zlib_bitrev(j,5),5);
+         if (disteb[j]) stbiw__zlib_add(d - distc[j], disteb[j]);
+         i += best;
+      } else {
+         stbiw__zlib_huffb(data[i]);
+         ++i;
+      }
+   }
+   // write out final bytes
+   for (;i < data_len; ++i)
+      stbiw__zlib_huffb(data[i]);
+   stbiw__zlib_huff(256); // end of block
+   // pad with 0 bits to byte boundary
+   while (bitcount)
+      stbiw__zlib_add(0,1);
+
+   for (i=0; i < stbiw__ZHASH; ++i)
+      (void) stbiw__sbfree(hash_table[i]);
+   STBIW_FREE(hash_table);
+
+   {
+      // compute adler32 on input
+      unsigned int s1=1, s2=0;
+      int blocklen = (int) (data_len % 5552);
+      j=0;
+      while (j < data_len) {
+         for (i=0; i < blocklen; ++i) s1 += data[j+i], s2 += s1;
+         s1 %= 65521, s2 %= 65521;
+         j += blocklen;
+         blocklen = 5552;
+      }
+      stbiw__sbpush(out, STBIW_UCHAR(s2 >> 8));
+      stbiw__sbpush(out, STBIW_UCHAR(s2));
+      stbiw__sbpush(out, STBIW_UCHAR(s1 >> 8));
+      stbiw__sbpush(out, STBIW_UCHAR(s1));
+   }
+   *out_len = stbiw__sbn(out);
+   // make returned pointer freeable
+   STBIW_MEMMOVE(stbiw__sbraw(out), out, *out_len);
+   return (unsigned char *) stbiw__sbraw(out);
+}
+
+static unsigned int stbiw__crc32(unsigned char *buffer, int len)
+{
+   static unsigned int crc_table[256] =
+   {
+      0x00000000, 0x77073096, 0xEE0E612C, 0x990951BA, 0x076DC419, 0x706AF48F, 0xE963A535, 0x9E6495A3,
+      0x0eDB8832, 0x79DCB8A4, 0xE0D5E91E, 0x97D2D988, 0x09B64C2B, 0x7EB17CBD, 0xE7B82D07, 0x90BF1D91,
+      0x1DB71064, 0x6AB020F2, 0xF3B97148, 0x84BE41DE, 0x1ADAD47D, 0x6DDDE4EB, 0xF4D4B551, 0x83D385C7,
+      0x136C9856, 0x646BA8C0, 0xFD62F97A, 0x8A65C9EC, 0x14015C4F, 0x63066CD9, 0xFA0F3D63, 0x8D080DF5,
+      0x3B6E20C8, 0x4C69105E, 0xD56041E4, 0xA2677172, 0x3C03E4D1, 0x4B04D447, 0xD20D85FD, 0xA50AB56B,
+      0x35B5A8FA, 0x42B2986C, 0xDBBBC9D6, 0xACBCF940, 0x32D86CE3, 0x45DF5C75, 0xDCD60DCF, 0xABD13D59,
+      0x26D930AC, 0x51DE003A, 0xC8D75180, 0xBFD06116, 0x21B4F4B5, 0x56B3C423, 0xCFBA9599, 0xB8BDA50F,
+      0x2802B89E, 0x5F058808, 0xC60CD9B2, 0xB10BE924, 0x2F6F7C87, 0x58684C11, 0xC1611DAB, 0xB6662D3D,
+      0x76DC4190, 0x01DB7106, 0x98D220BC, 0xEFD5102A, 0x71B18589, 0x06B6B51F, 0x9FBFE4A5, 0xE8B8D433,
+      0x7807C9A2, 0x0F00F934, 0x9609A88E, 0xE10E9818, 0x7F6A0DBB, 0x086D3D2D, 0x91646C97, 0xE6635C01,
+      0x6B6B51F4, 0x1C6C6162, 0x856530D8, 0xF262004E, 0x6C0695ED, 0x1B01A57B, 0x8208F4C1, 0xF50FC457,
+      0x65B0D9C6, 0x12B7E950, 0x8BBEB8EA, 0xFCB9887C, 0x62DD1DDF, 0x15DA2D49, 0x8CD37CF3, 0xFBD44C65,
+      0x4DB26158, 0x3AB551CE, 0xA3BC0074, 0xD4BB30E2, 0x4ADFA541, 0x3DD895D7, 0xA4D1C46D, 0xD3D6F4FB,
+      0x4369E96A, 0x346ED9FC, 0xAD678846, 0xDA60B8D0, 0x44042D73, 0x33031DE5, 0xAA0A4C5F, 0xDD0D7CC9,
+      0x5005713C, 0x270241AA, 0xBE0B1010, 0xC90C2086, 0x5768B525, 0x206F85B3, 0xB966D409, 0xCE61E49F,
+      0x5EDEF90E, 0x29D9C998, 0xB0D09822, 0xC7D7A8B4, 0x59B33D17, 0x2EB40D81, 0xB7BD5C3B, 0xC0BA6CAD,
+      0xEDB88320, 0x9ABFB3B6, 0x03B6E20C, 0x74B1D29A, 0xEAD54739, 0x9DD277AF, 0x04DB2615, 0x73DC1683,
+      0xE3630B12, 0x94643B84, 0x0D6D6A3E, 0x7A6A5AA8, 0xE40ECF0B, 0x9309FF9D, 0x0A00AE27, 0x7D079EB1,
+      0xF00F9344, 0x8708A3D2, 0x1E01F268, 0x6906C2FE, 0xF762575D, 0x806567CB, 0x196C3671, 0x6E6B06E7,
+      0xFED41B76, 0x89D32BE0, 0x10DA7A5A, 0x67DD4ACC, 0xF9B9DF6F, 0x8EBEEFF9, 0x17B7BE43, 0x60B08ED5,
+      0xD6D6A3E8, 0xA1D1937E, 0x38D8C2C4, 0x4FDFF252, 0xD1BB67F1, 0xA6BC5767, 0x3FB506DD, 0x48B2364B,
+      0xD80D2BDA, 0xAF0A1B4C, 0x36034AF6, 0x41047A60, 0xDF60EFC3, 0xA867DF55, 0x316E8EEF, 0x4669BE79,
+      0xCB61B38C, 0xBC66831A, 0x256FD2A0, 0x5268E236, 0xCC0C7795, 0xBB0B4703, 0x220216B9, 0x5505262F,
+      0xC5BA3BBE, 0xB2BD0B28, 0x2BB45A92, 0x5CB36A04, 0xC2D7FFA7, 0xB5D0CF31, 0x2CD99E8B, 0x5BDEAE1D,
+      0x9B64C2B0, 0xEC63F226, 0x756AA39C, 0x026D930A, 0x9C0906A9, 0xEB0E363F, 0x72076785, 0x05005713,
+      0x95BF4A82, 0xE2B87A14, 0x7BB12BAE, 0x0CB61B38, 0x92D28E9B, 0xE5D5BE0D, 0x7CDCEFB7, 0x0BDBDF21,
+      0x86D3D2D4, 0xF1D4E242, 0x68DDB3F8, 0x1FDA836E, 0x81BE16CD, 0xF6B9265B, 0x6FB077E1, 0x18B74777,
+      0x88085AE6, 0xFF0F6A70, 0x66063BCA, 0x11010B5C, 0x8F659EFF, 0xF862AE69, 0x616BFFD3, 0x166CCF45,
+      0xA00AE278, 0xD70DD2EE, 0x4E048354, 0x3903B3C2, 0xA7672661, 0xD06016F7, 0x4969474D, 0x3E6E77DB,
+      0xAED16A4A, 0xD9D65ADC, 0x40DF0B66, 0x37D83BF0, 0xA9BCAE53, 0xDEBB9EC5, 0x47B2CF7F, 0x30B5FFE9,
+      0xBDBDF21C, 0xCABAC28A, 0x53B39330, 0x24B4A3A6, 0xBAD03605, 0xCDD70693, 0x54DE5729, 0x23D967BF,
+      0xB3667A2E, 0xC4614AB8, 0x5D681B02, 0x2A6F2B94, 0xB40BBE37, 0xC30C8EA1, 0x5A05DF1B, 0x2D02EF8D
+   };
+
+   unsigned int crc = ~0u;
+   int i;
+   for (i=0; i < len; ++i)
+      crc = (crc >> 8) ^ crc_table[buffer[i] ^ (crc & 0xff)];
+   return ~crc;
+}
+
+#define stbiw__wpng4(o,a,b,c,d) ((o)[0]=STBIW_UCHAR(a),(o)[1]=STBIW_UCHAR(b),(o)[2]=STBIW_UCHAR(c),(o)[3]=STBIW_UCHAR(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 stbiw__wpcrc(unsigned char **data, int len)
+{
+   unsigned int crc = stbiw__crc32(*data - len - 4, len+4);
+   stbiw__wp32(*data, crc);
+}
+
+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 STBIW_UCHAR(a);
+   if (pb <= pc) return STBIW_UCHAR(b);
+   return STBIW_UCHAR(c);
+}
+
+// @OPTIMIZE: provide an option that always forces left-predict or paeth predict
+unsigned char *stbi_write_png_to_mem(unsigned char *pixels, int stride_bytes, int x, int y, int n, int *out_len)
+{
+   int ctype[5] = { -1, 0, 4, 2, 6 };
+   unsigned char sig[8] = { 137,80,78,71,13,10,26,10 };
+   unsigned char *out,*o, *filt, *zlib;
+   signed char *line_buffer;
+   int i,j,k,p,zlen;
+
+   if (stride_bytes == 0)
+      stride_bytes = x * n;
+
+   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 };
+      int *mymap = (j != 0) ? mapping : firstmap;
+      int best = 0, bestval = 0x7fffffff;
+      for (p=0; p < 2; ++p) {
+         for (k= p?best:0; k < 5; ++k) { // @TODO: clarity: rewrite this to go 0..5, and 'continue' the unwanted ones during 2nd pass
+            int type = mymap[k],est=0;
+            unsigned char *z = pixels + stride_bytes*j;
+            for (i=0; i < n; ++i)
+               switch (type) {
+                  case 0: line_buffer[i] = z[i]; break;
+                  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] - 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;
+               }
+            for (i=n; i < x*n; ++i) {
+               switch (type) {
+                  case 0: line_buffer[i] = z[i]; break;
+                  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] - 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] - stbiw__paeth(z[i-n], 0,0); break;
+               }
+            }
+            if (p) break;
+            for (i=0; i < x*n; ++i)
+               est += abs((signed char) line_buffer[i]);
+            if (est < bestval) { bestval = est; best = k; }
+         }
+      }
+      // when we get here, best contains the filter type, and line_buffer contains the data
+      filt[j*(x*n+1)] = (unsigned char) best;
+      STBIW_MEMMOVE(filt+j*(x*n+1)+1, line_buffer, x*n);
+   }
+   STBIW_FREE(line_buffer);
+   zlib = stbi_zlib_compress(filt, y*( x*n+1), &zlen, 8); // increase 8 to get smaller but use more memory
+   STBIW_FREE(filt);
+   if (!zlib) return 0;
+
+   // each tag requires 12 bytes of overhead
+   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;
+   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++ = STBIW_UCHAR(ctype[n]);
+   *o++ = 0;
+   *o++ = 0;
+   *o++ = 0;
+   stbiw__wpcrc(&o,13);
+
+   stbiw__wp32(o, zlen);
+   stbiw__wptag(o, "IDAT");
+   STBIW_MEMMOVE(o, zlib, zlen);
+   o += zlen;
+   STBIW_FREE(zlib);
+   stbiw__wpcrc(&o, zlen);
+
+   stbiw__wp32(o,0);
+   stbiw__wptag(o, "IEND");
+   stbiw__wpcrc(&o,0);
+
+   STBIW_ASSERT(o == out + *out_len);
+
+   return out;
+}
+
+#ifndef STBI_WRITE_NO_STDIO
+STBIWDEF int stbi_write_png(char const *filename, int x, int y, int comp, const void *data, int stride_bytes)
+{
+   FILE *f;
+   int len;
+   unsigned char *png = stbi_write_png_to_mem((unsigned char *) data, stride_bytes, x, y, comp, &len);
+   if (png == NULL) return 0;
+   f = fopen(filename, "wb");
+   if (!f) { STBIW_FREE(png); return 0; }
+   fwrite(png, 1, len, f);
+   fclose(f);
+   STBIW_FREE(png);
+   return 1;
+}
+#endif
+
+STBIWDEF int stbi_write_png_to_func(stbi_write_func *func, void *context, int x, int y, int comp, const void *data, int stride_bytes)
+{
+   int len;
+   unsigned char *png = stbi_write_png_to_mem((unsigned char *) data, stride_bytes, x, y, comp, &len);
+   if (png == NULL) return 0;
+   func(context, png, len);
+   STBIW_FREE(png);
+   return 1;
+}
+
+#endif // STB_IMAGE_WRITE_IMPLEMENTATION
+
 /* Revision history
+      1.04 (2017-03-03)
+             monochrome BMP expansion
+      1.03   ???
       1.02 (2016-04-02)
              avoid allocating large structures on the stack
       1.01 (2016-01-16)
@@ -182,3 +1062,45 @@ unsigned char *stbi_write_png_to_mem(unsigned char *pixels, int stride_bytes, in
              first public release
       0.90   first internal release
 */
+
+/*
+------------------------------------------------------------------------------
+This software is available under 2 licenses -- choose whichever you prefer.
+------------------------------------------------------------------------------
+ALTERNATIVE A - MIT License
+Copyright (c) 2017 Sean Barrett
+Permission is hereby granted, free of charge, to any person obtaining a copy of 
+this software and associated documentation files (the "Software"), to deal in 
+the Software without restriction, including without limitation the rights to 
+use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies 
+of the Software, and to permit persons to whom the Software is furnished to do 
+so, subject to the following conditions:
+The above copyright notice and this permission notice shall be included in all 
+copies or substantial portions of the Software.
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 
+SOFTWARE.
+------------------------------------------------------------------------------
+ALTERNATIVE B - Public Domain (www.unlicense.org)
+This is free and unencumbered software released into the public domain.
+Anyone is free to copy, modify, publish, use, compile, sell, or distribute this 
+software, either in source code form or as a compiled binary, for any purpose, 
+commercial or non-commercial, and by any means.
+In jurisdictions that recognize copyright laws, the author or authors of this 
+software dedicate any and all copyright interest in the software to the public 
+domain. We make this dedication for the benefit of the public at large and to 
+the detriment of our heirs and successors. We intend this dedication to be an 
+overt act of relinquishment in perpetuity of all present and future rights to 
+this software under copyright law.
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 
+AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 
+ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 
+WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+------------------------------------------------------------------------------
+*/

Source/ThirdParty/STB/stb_impl.cpp

@@ -0,0 +1,41 @@
+//
+// Copyright (c) 2017 the Atomic project.
+//
+// Permission is hereby granted, free of charge, to any person obtaining a copy
+// of this software and associated documentation files (the "Software"), to deal
+// in the Software without restriction, including without limitation the rights
+// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+// copies of the Software, and to permit persons to whom the Software is
+// furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in
+// all copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+// THE SOFTWARE.
+//
+#define STB_RECT_PACK_IMPLEMENTATION 1
+#include "stb_rect_pack.h"
+
+#define STB_IMAGE_IMPLEMENTATION 1
+#include "stb_image.h"
+
+#define STB_IMAGE_WRITE_IMPLEMENTATION 1
+#include "stb_image_write.h"
+
+#undef STB_VORBIS_HEADER_ONLY
+#include "stb_vorbis.h"
+
+#define STB_TRUETYPE_IMPLEMENTATION 1
+#include "stb_truetype.h"
+
+// So far only imgui is using textedit. Also use of this header requires extensive configuration. Due to these reasons
+// header implementation is not enabled here until other parts of project require it.
+//#define STB_TEXTEDIT_IMPLEMENTATION 1
+//#include "stb_textedit.h"
+

Source/ThirdParty/STB/stb_rect_pack.c

@@ -1,392 +0,0 @@
-#include "stb_rect_pack.h"
-
-#define STB_RECT_PACK_IMPLEMENTATION
-
-//////////////////////////////////////////////////////////////////////////////
-//
-//     IMPLEMENTATION SECTION
-//
-
-#ifdef STB_RECT_PACK_IMPLEMENTATION
-#ifndef STBRP_SORT
-#include <stdlib.h>
-#define STBRP_SORT qsort
-#endif
-
-#ifndef STBRP_ASSERT
-#include <assert.h>
-#define STBRP_ASSERT assert
-#endif
-
-enum
-{
-   STBRP__INIT_skyline = 1,
-};
-
-STBRP_DEF void stbrp_setup_heuristic(stbrp_context *context, int heuristic)
-{
-   switch (context->init_mode) {
-      case STBRP__INIT_skyline:
-         STBRP_ASSERT(heuristic == STBRP_HEURISTIC_Skyline_BL_sortHeight || heuristic == STBRP_HEURISTIC_Skyline_BF_sortHeight);
-         context->heuristic = heuristic;
-         break;
-      default:
-         STBRP_ASSERT(0);
-   }
-}
-
-STBRP_DEF void stbrp_setup_allow_out_of_mem(stbrp_context *context, int allow_out_of_mem)
-{
-   if (allow_out_of_mem)
-      // if it's ok to run out of memory, then don't bother aligning them;
-      // this gives better packing, but may fail due to OOM (even though
-      // the rectangles easily fit). @TODO a smarter approach would be to only
-      // quantize once we've hit OOM, then we could get rid of this parameter.
-      context->align = 1;
-   else {
-      // if it's not ok to run out of memory, then quantize the widths
-      // so that num_nodes is always enough nodes.
-      //
-      // I.e. num_nodes * align >= width
-      //                  align >= width / num_nodes
-      //                  align = ceil(width/num_nodes)
-
-      context->align = (context->width + context->num_nodes-1) / context->num_nodes;
-   }
-}
-
-STBRP_DEF void stbrp_init_target(stbrp_context *context, int width, int height, stbrp_node *nodes, int num_nodes)
-{
-   int i;
-#ifndef STBRP_LARGE_RECTS
-   STBRP_ASSERT(width <= 0xffff && height <= 0xffff);
-#endif
-
-   for (i=0; i < num_nodes-1; ++i)
-      nodes[i].next = &nodes[i+1];
-   nodes[i].next = NULL;
-   context->init_mode = STBRP__INIT_skyline;
-   context->heuristic = STBRP_HEURISTIC_Skyline_default;
-   context->free_head = &nodes[0];
-   context->active_head = &context->extra[0];
-   context->width = width;
-   context->height = height;
-   context->num_nodes = num_nodes;
-   stbrp_setup_allow_out_of_mem(context, 0);
-
-   // node 0 is the full width, node 1 is the sentinel (lets us not store width explicitly)
-   context->extra[0].x = 0;
-   context->extra[0].y = 0;
-   context->extra[0].next = &context->extra[1];
-   context->extra[1].x = (stbrp_coord) width;
-#ifdef STBRP_LARGE_RECTS
-   context->extra[1].y = (1<<30);
-#else
-   context->extra[1].y = 65535;
-#endif
-   context->extra[1].next = NULL;
-}
-
-// find minimum y position if it starts at x1
-static int stbrp__skyline_find_min_y(stbrp_context *c, stbrp_node *first, int x0, int width, int *pwaste)
-{
-   stbrp_node *node = first;
-   int x1 = x0 + width;
-   int min_y, visited_width, waste_area;
-   STBRP_ASSERT(first->x <= x0);
-
-   #if 0
-   // skip in case we're past the node
-   while (node->next->x <= x0)
-      ++node;
-   #else
-   STBRP_ASSERT(node->next->x > x0); // we ended up handling this in the caller for efficiency
-   #endif
-
-   STBRP_ASSERT(node->x <= x0);
-
-   min_y = 0;
-   waste_area = 0;
-   visited_width = 0;
-   while (node->x < x1) {
-      if (node->y > min_y) {
-         // raise min_y higher.
-         // we've accounted for all waste up to min_y,
-         // but we'll now add more waste for everything we've visted
-         waste_area += visited_width * (node->y - min_y);
-         min_y = node->y;
-         // the first time through, visited_width might be reduced
-         if (node->x < x0)
-            visited_width += node->next->x - x0;
-         else
-            visited_width += node->next->x - node->x;
-      } else {
-         // add waste area
-         int under_width = node->next->x - node->x;
-         if (under_width + visited_width > width)
-            under_width = width - visited_width;
-         waste_area += under_width * (min_y - node->y);
-         visited_width += under_width;
-      }
-      node = node->next;
-   }
-
-   *pwaste = waste_area;
-   return min_y;
-}
-
-typedef struct
-{
-   int x,y;
-   stbrp_node **prev_link;
-} stbrp__findresult;
-
-static stbrp__findresult stbrp__skyline_find_best_pos(stbrp_context *c, int width, int height)
-{
-   int best_waste = (1<<30), best_x, best_y = (1 << 30);
-   stbrp__findresult fr;
-   stbrp_node **prev, *node, *tail, **best = NULL;
-
-   // align to multiple of c->align
-   width = (width + c->align - 1);
-   width -= width % c->align;
-   STBRP_ASSERT(width % c->align == 0);
-
-   node = c->active_head;
-   prev = &c->active_head;
-   while (node->x + width <= c->width) {
-      int y,waste;
-      y = stbrp__skyline_find_min_y(c, node, node->x, width, &waste);
-      if (c->heuristic == STBRP_HEURISTIC_Skyline_BL_sortHeight) { // actually just want to test BL
-         // bottom left
-         if (y < best_y) {
-            best_y = y;
-            best = prev;
-         }
-      } else {
-         // best-fit
-         if (y + height <= c->height) {
-            // can only use it if it first vertically
-            if (y < best_y || (y == best_y && waste < best_waste)) {
-               best_y = y;
-               best_waste = waste;
-               best = prev;
-            }
-         }
-      }
-      prev = &node->next;
-      node = node->next;
-   }
-
-   best_x = (best == NULL) ? 0 : (*best)->x;
-
-   // if doing best-fit (BF), we also have to try aligning right edge to each node position
-   //
-   // e.g, if fitting
-   //
-   //     ____________________
-   //    |____________________|
-   //
-   //            into
-   //
-   //   |                         |
-   //   |             ____________|
-   //   |____________|
-   //
-   // then right-aligned reduces waste, but bottom-left BL is always chooses left-aligned
-   //
-   // This makes BF take about 2x the time
-
-   if (c->heuristic == STBRP_HEURISTIC_Skyline_BF_sortHeight) {
-      tail = c->active_head;
-      node = c->active_head;
-      prev = &c->active_head;
-      // find first node that's admissible
-      while (tail->x < width)
-         tail = tail->next;
-      while (tail) {
-         int xpos = tail->x - width;
-         int y,waste;
-         STBRP_ASSERT(xpos >= 0);
-         // find the left position that matches this
-         while (node->next->x <= xpos) {
-            prev = &node->next;
-            node = node->next;
-         }
-         STBRP_ASSERT(node->next->x > xpos && node->x <= xpos);
-         y = stbrp__skyline_find_min_y(c, node, xpos, width, &waste);
-         if (y + height < c->height) {
-            if (y <= best_y) {
-               if (y < best_y || waste < best_waste || (waste==best_waste && xpos < best_x)) {
-                  best_x = xpos;
-                  STBRP_ASSERT(y <= best_y);
-                  best_y = y;
-                  best_waste = waste;
-                  best = prev;
-               }
-            }
-         }
-         tail = tail->next;
-      }
-   }
-
-   fr.prev_link = best;
-   fr.x = best_x;
-   fr.y = best_y;
-   return fr;
-}
-
-static stbrp__findresult stbrp__skyline_pack_rectangle(stbrp_context *context, int width, int height)
-{
-   // find best position according to heuristic
-   stbrp__findresult res = stbrp__skyline_find_best_pos(context, width, height);
-   stbrp_node *node, *cur;
-
-   // bail if:
-   //    1. it failed
-   //    2. the best node doesn't fit (we don't always check this)
-   //    3. we're out of memory
-   if (res.prev_link == NULL || res.y + height > context->height || context->free_head == NULL) {
-      res.prev_link = NULL;
-      return res;
-   }
-
-   // on success, create new node
-   node = context->free_head;
-   node->x = (stbrp_coord) res.x;
-   node->y = (stbrp_coord) (res.y + height);
-
-   context->free_head = node->next;
-
-   // insert the new node into the right starting point, and
-   // let 'cur' point to the remaining nodes needing to be
-   // stiched back in
-
-   cur = *res.prev_link;
-   if (cur->x < res.x) {
-      // preserve the existing one, so start testing with the next one
-      stbrp_node *next = cur->next;
-      cur->next = node;
-      cur = next;
-   } else {
-      *res.prev_link = node;
-   }
-
-   // from here, traverse cur and free the nodes, until we get to one
-   // that shouldn't be freed
-   while (cur->next && cur->next->x <= res.x + width) {
-      stbrp_node *next = cur->next;
-      // move the current node to the free list
-      cur->next = context->free_head;
-      context->free_head = cur;
-      cur = next;
-   }
-
-   // stitch the list back in
-   node->next = cur;
-
-   if (cur->x < res.x + width)
-      cur->x = (stbrp_coord) (res.x + width);
-
-#ifdef _DEBUG
-   cur = context->active_head;
-   while (cur->x < context->width) {
-      STBRP_ASSERT(cur->x < cur->next->x);
-      cur = cur->next;
-   }
-   STBRP_ASSERT(cur->next == NULL);
-
-   {
-      stbrp_node *L1 = NULL, *L2 = NULL;
-      int count=0;
-      cur = context->active_head;
-      while (cur) {
-         L1 = cur;
-         cur = cur->next;
-         ++count;
-      }
-      cur = context->free_head;
-      while (cur) {
-         L2 = cur;
-         cur = cur->next;
-         ++count;
-      }
-      STBRP_ASSERT(count == context->num_nodes+2);
-   }
-#endif
-
-   return res;
-}
-
-static int rect_height_compare(const void *a, const void *b)
-{
-   stbrp_rect *p = (stbrp_rect *) a;
-   stbrp_rect *q = (stbrp_rect *) b;
-   if (p->h > q->h)
-      return -1;
-   if (p->h < q->h)
-      return  1;
-   return (p->w > q->w) ? -1 : (p->w < q->w);
-}
-
-static int rect_width_compare(const void *a, const void *b)
-{
-   stbrp_rect *p = (stbrp_rect *) a;
-   stbrp_rect *q = (stbrp_rect *) b;
-   if (p->w > q->w)
-      return -1;
-   if (p->w < q->w)
-      return  1;
-   return (p->h > q->h) ? -1 : (p->h < q->h);
-}
-
-static int rect_original_order(const void *a, const void *b)
-{
-   stbrp_rect *p = (stbrp_rect *) a;
-   stbrp_rect *q = (stbrp_rect *) b;
-   return (p->was_packed < q->was_packed) ? -1 : (p->was_packed > q->was_packed);
-}
-
-#ifdef STBRP_LARGE_RECTS
-#define STBRP__MAXVAL  0xffffffff
-#else
-#define STBRP__MAXVAL  0xffff
-#endif
-
-STBRP_DEF void stbrp_pack_rects(stbrp_context *context, stbrp_rect *rects, int num_rects)
-{
-   int i;
-
-   // we use the 'was_packed' field internally to allow sorting/unsorting
-   for (i=0; i < num_rects; ++i) {
-      rects[i].was_packed = i;
-      #ifndef STBRP_LARGE_RECTS
-      STBRP_ASSERT(rects[i].w <= 0xffff && rects[i].h <= 0xffff);
-      #endif
-   }
-
-   // sort according to heuristic
-   STBRP_SORT(rects, num_rects, sizeof(rects[0]), rect_height_compare);
-
-   for (i=0; i < num_rects; ++i) {
-      if (rects[i].w == 0 || rects[i].h == 0) {
-         rects[i].x = rects[i].y = 0;  // empty rect needs no space
-      } else {
-         stbrp__findresult fr = stbrp__skyline_pack_rectangle(context, rects[i].w, rects[i].h);
-         if (fr.prev_link) {
-            rects[i].x = (stbrp_coord) fr.x;
-            rects[i].y = (stbrp_coord) fr.y;
-         } else {
-            rects[i].x = rects[i].y = STBRP__MAXVAL;
-         }
-      }
-   }
-
-   // unsort
-   STBRP_SORT(rects, num_rects, sizeof(rects[0]), rect_original_order);
-
-   // set was_packed flags
-   for (i=0; i < num_rects; ++i)
-      rects[i].was_packed = !(rects[i].x == STBRP__MAXVAL && rects[i].y == STBRP__MAXVAL);
-}
-#endif

Source/ThirdParty/STB/stb_rect_pack.h

@@ -1,4 +1,4 @@
-// stb_rect_pack.h - v0.08 - public domain - rectangle packing
+// stb_rect_pack.h - v0.11 - public domain - rectangle packing
 // Sean Barrett 2014
 //
 // Useful for e.g. packing rectangular textures into an atlas.
@@ -27,11 +27,16 @@
 //    Sean Barrett
 //  Minor features
 //    Martins Mozeiko
+//    github:IntellectualKitty
+//    
 //  Bugfixes / warning fixes
 //    Jeremy Jaussaud
 //
 // Version history:
 //
+//     0.11  (2017-03-03)  return packing success/fail result
+//     0.10  (2016-10-25)  remove cast-away-const to avoid warnings
+//     0.09  (2016-08-27)  fix compiler warnings
 //     0.08  (2015-09-13)  really fix bug with empty rects (w=0 or h=0)
 //     0.07  (2015-09-13)  fix bug with empty rects (w=0 or h=0)
 //     0.06  (2015-04-15)  added STBRP_SORT to allow replacing qsort
@@ -41,9 +46,7 @@
 //
 // LICENSE
 //
-//   This software is dual-licensed to the public domain and under the following
-//   license: you are granted a perpetual, irrevocable license to copy, modify,
-//   publish, and distribute this file as you see fit.
+//   See end of file for license information.
 
 //////////////////////////////////////////////////////////////////////////////
 //
@@ -75,7 +78,7 @@ typedef int            stbrp_coord;
 typedef unsigned short stbrp_coord;
 #endif
 
-STBRP_DEF void stbrp_pack_rects (stbrp_context *context, stbrp_rect *rects, int num_rects);
+STBRP_DEF int stbrp_pack_rects (stbrp_context *context, stbrp_rect *rects, int num_rects);
 // Assign packed locations to rectangles. The rectangles are of type
 // 'stbrp_rect' defined below, stored in the array 'rects', and there
 // are 'num_rects' many of them.
@@ -96,6 +99,9 @@ STBRP_DEF void stbrp_pack_rects (stbrp_context *context, stbrp_rect *rects, int
 // arrays will probably produce worse packing results than calling it
 // a single time with the full rectangle array, but the option is
 // available.
+//
+// The function returns 1 if all of the rectangles were successfully
+// packed and 0 otherwise.
 
 struct stbrp_rect
 {
@@ -148,7 +154,7 @@ enum
 {
    STBRP_HEURISTIC_Skyline_default=0,
    STBRP_HEURISTIC_Skyline_BL_sortHeight = STBRP_HEURISTIC_Skyline_default,
-   STBRP_HEURISTIC_Skyline_BF_sortHeight,
+   STBRP_HEURISTIC_Skyline_BF_sortHeight
 };
 
 
@@ -181,3 +187,449 @@ struct stbrp_context
 #endif
 
 #endif
+
+//////////////////////////////////////////////////////////////////////////////
+//
+//     IMPLEMENTATION SECTION
+//
+
+#ifdef STB_RECT_PACK_IMPLEMENTATION
+#ifndef STBRP_SORT
+#include <stdlib.h>
+#define STBRP_SORT qsort
+#endif
+
+#ifndef STBRP_ASSERT
+#include <assert.h>
+#define STBRP_ASSERT assert
+#endif
+
+#ifdef _MSC_VER
+#define STBRP__NOTUSED(v)  (void)(v)
+#else
+#define STBRP__NOTUSED(v)  (void)sizeof(v)
+#endif
+
+enum
+{
+   STBRP__INIT_skyline = 1
+};
+
+STBRP_DEF void stbrp_setup_heuristic(stbrp_context *context, int heuristic)
+{
+   switch (context->init_mode) {
+      case STBRP__INIT_skyline:
+         STBRP_ASSERT(heuristic == STBRP_HEURISTIC_Skyline_BL_sortHeight || heuristic == STBRP_HEURISTIC_Skyline_BF_sortHeight);
+         context->heuristic = heuristic;
+         break;
+      default:
+         STBRP_ASSERT(0);
+   }
+}
+
+STBRP_DEF void stbrp_setup_allow_out_of_mem(stbrp_context *context, int allow_out_of_mem)
+{
+   if (allow_out_of_mem)
+      // if it's ok to run out of memory, then don't bother aligning them;
+      // this gives better packing, but may fail due to OOM (even though
+      // the rectangles easily fit). @TODO a smarter approach would be to only
+      // quantize once we've hit OOM, then we could get rid of this parameter.
+      context->align = 1;
+   else {
+      // if it's not ok to run out of memory, then quantize the widths
+      // so that num_nodes is always enough nodes.
+      //
+      // I.e. num_nodes * align >= width
+      //                  align >= width / num_nodes
+      //                  align = ceil(width/num_nodes)
+
+      context->align = (context->width + context->num_nodes-1) / context->num_nodes;
+   }
+}
+
+STBRP_DEF void stbrp_init_target(stbrp_context *context, int width, int height, stbrp_node *nodes, int num_nodes)
+{
+   int i;
+#ifndef STBRP_LARGE_RECTS
+   STBRP_ASSERT(width <= 0xffff && height <= 0xffff);
+#endif
+
+   for (i=0; i < num_nodes-1; ++i)
+      nodes[i].next = &nodes[i+1];
+   nodes[i].next = NULL;
+   context->init_mode = STBRP__INIT_skyline;
+   context->heuristic = STBRP_HEURISTIC_Skyline_default;
+   context->free_head = &nodes[0];
+   context->active_head = &context->extra[0];
+   context->width = width;
+   context->height = height;
+   context->num_nodes = num_nodes;
+   stbrp_setup_allow_out_of_mem(context, 0);
+
+   // node 0 is the full width, node 1 is the sentinel (lets us not store width explicitly)
+   context->extra[0].x = 0;
+   context->extra[0].y = 0;
+   context->extra[0].next = &context->extra[1];
+   context->extra[1].x = (stbrp_coord) width;
+#ifdef STBRP_LARGE_RECTS
+   context->extra[1].y = (1<<30);
+#else
+   context->extra[1].y = 65535;
+#endif
+   context->extra[1].next = NULL;
+}
+
+// find minimum y position if it starts at x1
+static int stbrp__skyline_find_min_y(stbrp_context *c, stbrp_node *first, int x0, int width, int *pwaste)
+{
+   stbrp_node *node = first;
+   int x1 = x0 + width;
+   int min_y, visited_width, waste_area;
+
+   STBRP__NOTUSED(c);
+
+   STBRP_ASSERT(first->x <= x0);
+
+   #if 0
+   // skip in case we're past the node
+   while (node->next->x <= x0)
+      ++node;
+   #else
+   STBRP_ASSERT(node->next->x > x0); // we ended up handling this in the caller for efficiency
+   #endif
+
+   STBRP_ASSERT(node->x <= x0);
+
+   min_y = 0;
+   waste_area = 0;
+   visited_width = 0;
+   while (node->x < x1) {
+      if (node->y > min_y) {
+         // raise min_y higher.
+         // we've accounted for all waste up to min_y,
+         // but we'll now add more waste for everything we've visted
+         waste_area += visited_width * (node->y - min_y);
+         min_y = node->y;
+         // the first time through, visited_width might be reduced
+         if (node->x < x0)
+            visited_width += node->next->x - x0;
+         else
+            visited_width += node->next->x - node->x;
+      } else {
+         // add waste area
+         int under_width = node->next->x - node->x;
+         if (under_width + visited_width > width)
+            under_width = width - visited_width;
+         waste_area += under_width * (min_y - node->y);
+         visited_width += under_width;
+      }
+      node = node->next;
+   }
+
+   *pwaste = waste_area;
+   return min_y;
+}
+
+typedef struct
+{
+   int x,y;
+   stbrp_node **prev_link;
+} stbrp__findresult;
+
+static stbrp__findresult stbrp__skyline_find_best_pos(stbrp_context *c, int width, int height)
+{
+   int best_waste = (1<<30), best_x, best_y = (1 << 30);
+   stbrp__findresult fr;
+   stbrp_node **prev, *node, *tail, **best = NULL;
+
+   // align to multiple of c->align
+   width = (width + c->align - 1);
+   width -= width % c->align;
+   STBRP_ASSERT(width % c->align == 0);
+
+   node = c->active_head;
+   prev = &c->active_head;
+   while (node->x + width <= c->width) {
+      int y,waste;
+      y = stbrp__skyline_find_min_y(c, node, node->x, width, &waste);
+      if (c->heuristic == STBRP_HEURISTIC_Skyline_BL_sortHeight) { // actually just want to test BL
+         // bottom left
+         if (y < best_y) {
+            best_y = y;
+            best = prev;
+         }
+      } else {
+         // best-fit
+         if (y + height <= c->height) {
+            // can only use it if it first vertically
+            if (y < best_y || (y == best_y && waste < best_waste)) {
+               best_y = y;
+               best_waste = waste;
+               best = prev;
+            }
+         }
+      }
+      prev = &node->next;
+      node = node->next;
+   }
+
+   best_x = (best == NULL) ? 0 : (*best)->x;
+
+   // if doing best-fit (BF), we also have to try aligning right edge to each node position
+   //
+   // e.g, if fitting
+   //
+   //     ____________________
+   //    |____________________|
+   //
+   //            into
+   //
+   //   |                         |
+   //   |             ____________|
+   //   |____________|
+   //
+   // then right-aligned reduces waste, but bottom-left BL is always chooses left-aligned
+   //
+   // This makes BF take about 2x the time
+
+   if (c->heuristic == STBRP_HEURISTIC_Skyline_BF_sortHeight) {
+      tail = c->active_head;
+      node = c->active_head;
+      prev = &c->active_head;
+      // find first node that's admissible
+      while (tail->x < width)
+         tail = tail->next;
+      while (tail) {
+         int xpos = tail->x - width;
+         int y,waste;
+         STBRP_ASSERT(xpos >= 0);
+         // find the left position that matches this
+         while (node->next->x <= xpos) {
+            prev = &node->next;
+            node = node->next;
+         }
+         STBRP_ASSERT(node->next->x > xpos && node->x <= xpos);
+         y = stbrp__skyline_find_min_y(c, node, xpos, width, &waste);
+         if (y + height < c->height) {
+            if (y <= best_y) {
+               if (y < best_y || waste < best_waste || (waste==best_waste && xpos < best_x)) {
+                  best_x = xpos;
+                  STBRP_ASSERT(y <= best_y);
+                  best_y = y;
+                  best_waste = waste;
+                  best = prev;
+               }
+            }
+         }
+         tail = tail->next;
+      }         
+   }
+
+   fr.prev_link = best;
+   fr.x = best_x;
+   fr.y = best_y;
+   return fr;
+}
+
+static stbrp__findresult stbrp__skyline_pack_rectangle(stbrp_context *context, int width, int height)
+{
+   // find best position according to heuristic
+   stbrp__findresult res = stbrp__skyline_find_best_pos(context, width, height);
+   stbrp_node *node, *cur;
+
+   // bail if:
+   //    1. it failed
+   //    2. the best node doesn't fit (we don't always check this)
+   //    3. we're out of memory
+   if (res.prev_link == NULL || res.y + height > context->height || context->free_head == NULL) {
+      res.prev_link = NULL;
+      return res;
+   }
+
+   // on success, create new node
+   node = context->free_head;
+   node->x = (stbrp_coord) res.x;
+   node->y = (stbrp_coord) (res.y + height);
+
+   context->free_head = node->next;
+
+   // insert the new node into the right starting point, and
+   // let 'cur' point to the remaining nodes needing to be
+   // stiched back in
+
+   cur = *res.prev_link;
+   if (cur->x < res.x) {
+      // preserve the existing one, so start testing with the next one
+      stbrp_node *next = cur->next;
+      cur->next = node;
+      cur = next;
+   } else {
+      *res.prev_link = node;
+   }
+
+   // from here, traverse cur and free the nodes, until we get to one
+   // that shouldn't be freed
+   while (cur->next && cur->next->x <= res.x + width) {
+      stbrp_node *next = cur->next;
+      // move the current node to the free list
+      cur->next = context->free_head;
+      context->free_head = cur;
+      cur = next;
+   }
+
+   // stitch the list back in
+   node->next = cur;
+
+   if (cur->x < res.x + width)
+      cur->x = (stbrp_coord) (res.x + width);
+
+#ifdef _DEBUG
+   cur = context->active_head;
+   while (cur->x < context->width) {
+      STBRP_ASSERT(cur->x < cur->next->x);
+      cur = cur->next;
+   }
+   STBRP_ASSERT(cur->next == NULL);
+
+   {
+      stbrp_node *L1 = NULL, *L2 = NULL;
+      int count=0;
+      cur = context->active_head;
+      while (cur) {
+         L1 = cur;
+         cur = cur->next;
+         ++count;
+      }
+      cur = context->free_head;
+      while (cur) {
+         L2 = cur;
+         cur = cur->next;
+         ++count;
+      }
+      STBRP_ASSERT(count == context->num_nodes+2);
+   }
+#endif
+
+   return res;
+}
+
+static int rect_height_compare(const void *a, const void *b)
+{
+   const stbrp_rect *p = (const stbrp_rect *) a;
+   const stbrp_rect *q = (const stbrp_rect *) b;
+   if (p->h > q->h)
+      return -1;
+   if (p->h < q->h)
+      return  1;
+   return (p->w > q->w) ? -1 : (p->w < q->w);
+}
+
+static int rect_width_compare(const void *a, const void *b)
+{
+   const stbrp_rect *p = (const stbrp_rect *) a;
+   const stbrp_rect *q = (const stbrp_rect *) b;
+   if (p->w > q->w)
+      return -1;
+   if (p->w < q->w)
+      return  1;
+   return (p->h > q->h) ? -1 : (p->h < q->h);
+}
+
+static int rect_original_order(const void *a, const void *b)
+{
+   const stbrp_rect *p = (const stbrp_rect *) a;
+   const stbrp_rect *q = (const stbrp_rect *) b;
+   return (p->was_packed < q->was_packed) ? -1 : (p->was_packed > q->was_packed);
+}
+
+#ifdef STBRP_LARGE_RECTS
+#define STBRP__MAXVAL  0xffffffff
+#else
+#define STBRP__MAXVAL  0xffff
+#endif
+
+STBRP_DEF int stbrp_pack_rects(stbrp_context *context, stbrp_rect *rects, int num_rects)
+{
+   int i, all_rects_packed = 1;
+
+   // we use the 'was_packed' field internally to allow sorting/unsorting
+   for (i=0; i < num_rects; ++i) {
+      rects[i].was_packed = i;
+      #ifndef STBRP_LARGE_RECTS
+      STBRP_ASSERT(rects[i].w <= 0xffff && rects[i].h <= 0xffff);
+      #endif
+   }
+
+   // sort according to heuristic
+   STBRP_SORT(rects, num_rects, sizeof(rects[0]), rect_height_compare);
+
+   for (i=0; i < num_rects; ++i) {
+      if (rects[i].w == 0 || rects[i].h == 0) {
+         rects[i].x = rects[i].y = 0;  // empty rect needs no space
+      } else {
+         stbrp__findresult fr = stbrp__skyline_pack_rectangle(context, rects[i].w, rects[i].h);
+         if (fr.prev_link) {
+            rects[i].x = (stbrp_coord) fr.x;
+            rects[i].y = (stbrp_coord) fr.y;
+         } else {
+            rects[i].x = rects[i].y = STBRP__MAXVAL;
+         }
+      }
+   }
+
+   // unsort
+   STBRP_SORT(rects, num_rects, sizeof(rects[0]), rect_original_order);
+
+   // set was_packed flags and all_rects_packed status
+   for (i=0; i < num_rects; ++i) {
+      rects[i].was_packed = !(rects[i].x == STBRP__MAXVAL && rects[i].y == STBRP__MAXVAL);
+      if (!rects[i].was_packed)
+         all_rects_packed = 0;
+   }
+
+   // return the all_rects_packed status
+   return all_rects_packed;
+}
+#endif
+
+/*
+------------------------------------------------------------------------------
+This software is available under 2 licenses -- choose whichever you prefer.
+------------------------------------------------------------------------------
+ALTERNATIVE A - MIT License
+Copyright (c) 2017 Sean Barrett
+Permission is hereby granted, free of charge, to any person obtaining a copy of 
+this software and associated documentation files (the "Software"), to deal in 
+the Software without restriction, including without limitation the rights to 
+use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies 
+of the Software, and to permit persons to whom the Software is furnished to do 
+so, subject to the following conditions:
+The above copyright notice and this permission notice shall be included in all 
+copies or substantial portions of the Software.
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 
+SOFTWARE.
+------------------------------------------------------------------------------
+ALTERNATIVE B - Public Domain (www.unlicense.org)
+This is free and unencumbered software released into the public domain.
+Anyone is free to copy, modify, publish, use, compile, sell, or distribute this 
+software, either in source code form or as a compiled binary, for any purpose, 
+commercial or non-commercial, and by any means.
+In jurisdictions that recognize copyright laws, the author or authors of this 
+software dedicate any and all copyright interest in the software to the public 
+domain. We make this dedication for the benefit of the public at large and to 
+the detriment of our heirs and successors. We intend this dedication to be an 
+overt act of relinquishment in perpetuity of all present and future rights to 
+this software under copyright law.
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 
+AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 
+ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 
+WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+------------------------------------------------------------------------------
+*/

Source/ThirdParty/imgui/stb_truetype.h → Source/ThirdParty/STB/stb_truetype.h

@@ -1,4 +1,4 @@
-// stb_truetype.h - v1.14 - public domain
+// stb_truetype.h - v1.15 - public domain
 // authored from 2009-2016 by Sean Barrett / RAD Game Tools
 //
 //   This library processes TrueType files:
@@ -50,10 +50,13 @@
 //       Higor Euripedes
 //       Thomas Fields
 //       Derek Vinyard
+//       Cort Stratton
 //
 // VERSION HISTORY
 //
-//   1.13 (2017-01-02) support OpenType fonts, certain Apple fonts, num-fonts-in-TTC function
+//   1.15 (2017-03-03) make more arguments const
+//   1.14 (2017-01-16) num-fonts-in-TTC function
+//   1.13 (2017-01-02) support OpenType fonts, certain Apple fonts
 //   1.12 (2016-10-25) suppress warnings about casting away const with -Wcast-qual
 //   1.11 (2016-04-02) fix unused-variable warning
 //   1.10 (2016-04-02) user-defined fabs(); rare memory leak; remove duplicate typedef
@@ -69,9 +72,7 @@
 //
 // LICENSE
 //
-//   This software is dual-licensed to the public domain and under the following
-//   license: you are granted a perpetual, irrevocable license to copy, modify,
-//   publish, and distribute this file as you see fit.
+//   See end of file for license information.
 //
 // USAGE
 //
@@ -494,7 +495,7 @@ typedef struct
    float x1,y1,s1,t1; // bottom-right
 } stbtt_aligned_quad;
 
-STBTT_DEF void stbtt_GetBakedQuad(stbtt_bakedchar *chardata, int pw, int ph,  // same data as above
+STBTT_DEF void stbtt_GetBakedQuad(const stbtt_bakedchar *chardata, int pw, int ph,  // same data as above
                                int char_index,             // character to display
                                float *xpos, float *ypos,   // pointers to current position in screen pixel space
                                stbtt_aligned_quad *q,      // output: quad to draw
@@ -594,7 +595,7 @@ STBTT_DEF void stbtt_PackSetOversampling(stbtt_pack_context *spc, unsigned int h
 // To use with PackFontRangesGather etc., you must set it before calls
 // call to PackFontRangesGatherRects.
 
-STBTT_DEF void stbtt_GetPackedQuad(stbtt_packedchar *chardata, int pw, int ph,  // same data as above
+STBTT_DEF void stbtt_GetPackedQuad(const stbtt_packedchar *chardata, int pw, int ph,  // same data as above
                                int char_index,             // character to display
                                float *xpos, float *ypos,   // pointers to current position in screen pixel space
                                stbtt_aligned_quad *q,      // output: quad to draw
@@ -3287,11 +3288,11 @@ static int stbtt_BakeFontBitmap_internal(unsigned char *data, int offset,  // fo
    return bottom_y;
 }
 
-STBTT_DEF void stbtt_GetBakedQuad(stbtt_bakedchar *chardata, int pw, int ph, int char_index, float *xpos, float *ypos, stbtt_aligned_quad *q, int opengl_fillrule)
+STBTT_DEF void stbtt_GetBakedQuad(const stbtt_bakedchar *chardata, int pw, int ph, int char_index, float *xpos, float *ypos, stbtt_aligned_quad *q, int opengl_fillrule)
 {
    float d3d_bias = opengl_fillrule ? 0 : -0.5f;
    float ipw = 1.0f / pw, iph = 1.0f / ph;
-   stbtt_bakedchar *b = chardata + char_index;
+   const stbtt_bakedchar *b = chardata + char_index;
    int round_x = STBTT_ifloor((*xpos + b->xoff) + 0.5f);
    int round_y = STBTT_ifloor((*ypos + b->yoff) + 0.5f);
 
@@ -3735,10 +3736,10 @@ STBTT_DEF int stbtt_PackFontRange(stbtt_pack_context *spc, unsigned char *fontda
    return stbtt_PackFontRanges(spc, fontdata, font_index, &range, 1);
 }
 
-STBTT_DEF void stbtt_GetPackedQuad(stbtt_packedchar *chardata, int pw, int ph, int char_index, float *xpos, float *ypos, stbtt_aligned_quad *q, int align_to_integer)
+STBTT_DEF void stbtt_GetPackedQuad(const stbtt_packedchar *chardata, int pw, int ph, int char_index, float *xpos, float *ypos, stbtt_aligned_quad *q, int align_to_integer)
 {
    float ipw = 1.0f / pw, iph = 1.0f / ph;
-   stbtt_packedchar *b = chardata + char_index;
+   const stbtt_packedchar *b = chardata + char_index;
 
    if (align_to_integer) {
       float x = (float) STBTT_ifloor((*xpos + b->xoff) + 0.5f);
@@ -4016,3 +4017,45 @@ STBTT_DEF int stbtt_CompareUTF8toUTF16_bigendian(const char *s1, int len1, const
 //   0.2  (2009-03-11) Fix unsigned/signed char warnings
 //   0.1  (2009-03-09) First public release
 //
+
+/*
+------------------------------------------------------------------------------
+This software is available under 2 licenses -- choose whichever you prefer.
+------------------------------------------------------------------------------
+ALTERNATIVE A - MIT License
+Copyright (c) 2017 Sean Barrett
+Permission is hereby granted, free of charge, to any person obtaining a copy of 
+this software and associated documentation files (the "Software"), to deal in 
+the Software without restriction, including without limitation the rights to 
+use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies 
+of the Software, and to permit persons to whom the Software is furnished to do 
+so, subject to the following conditions:
+The above copyright notice and this permission notice shall be included in all 
+copies or substantial portions of the Software.
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 
+SOFTWARE.
+------------------------------------------------------------------------------
+ALTERNATIVE B - Public Domain (www.unlicense.org)
+This is free and unencumbered software released into the public domain.
+Anyone is free to copy, modify, publish, use, compile, sell, or distribute this 
+software, either in source code form or as a compiled binary, for any purpose, 
+commercial or non-commercial, and by any means.
+In jurisdictions that recognize copyright laws, the author or authors of this 
+software dedicate any and all copyright interest in the software to the public 
+domain. We make this dedication for the benefit of the public at large and to 
+the detriment of our heirs and successors. We intend this dedication to be an 
+overt act of relinquishment in perpetuity of all present and future rights to 
+this software under copyright law.
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 
+AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 
+ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 
+WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+------------------------------------------------------------------------------
+*/

Source/ThirdParty/STB/stb_vorbis.c

@@ -1,5011 +0,0 @@
-
-#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
-
-#ifdef STB_VORBIS_CODEBOOK_SHORTS
-#error "STB_VORBIS_CODEBOOK_SHORTS is no longer supported as it produced incorrect results for some input formats"
-#endif
-
-// STB_VORBIS_DIVIDE_TABLE
-//     this replaces small integer divides in the floor decode loop with
-//     table lookups. made less than 1% difference, so disabled by default.
-
-// STB_VORBIS_NO_INLINE_DECODE
-//     disables the inlining of the scalar codebook fast-huffman decode.
-//     might save a little codespace; useful for debugging
-// #define STB_VORBIS_NO_INLINE_DECODE
-
-// STB_VORBIS_NO_DEFER_FLOOR
-//     Normally we only decode the floor without synthesizing the actual
-//     full curve. We can instead synthesize the curve immediately. This
-//     requires more memory and is very likely slower, so I don't think
-//     you'd ever want to do it except for debugging.
-// #define STB_VORBIS_NO_DEFER_FLOOR
-
-
-
-
-//////////////////////////////////////////////////////////////////////////////
-
-#ifdef STB_VORBIS_NO_PULLDATA_API
-   #define STB_VORBIS_NO_INTEGER_CONVERSION
-   #define STB_VORBIS_NO_STDIO
-#endif
-
-#if defined(STB_VORBIS_NO_CRT) && !defined(STB_VORBIS_NO_STDIO)
-   #define STB_VORBIS_NO_STDIO 1
-#endif
-
-#ifndef STB_VORBIS_NO_INTEGER_CONVERSION
-#ifndef STB_VORBIS_NO_FAST_SCALED_FLOAT
-
-   // only need endianness for fast-float-to-int, which we don't
-   // use for pushdata
-
-   #ifndef STB_VORBIS_BIG_ENDIAN
-     #define STB_VORBIS_ENDIAN  0
-   #else
-     #define STB_VORBIS_ENDIAN  1
-   #endif
-
-#endif
-#endif
-
-
-#ifndef STB_VORBIS_NO_STDIO
-#include <stdio.h>
-#endif
-
-#ifndef STB_VORBIS_NO_CRT
-#include <stdlib.h>
-#include <string.h>
-#include <assert.h>
-#include <math.h>
-#if !(defined(__APPLE__) || defined(MACOSX) || defined(macintosh) || defined(Macintosh))
-#include <malloc.h>
-#if defined(__linux__) || defined(__linux) || defined(__EMSCRIPTEN__)
-#include <alloca.h>
-#endif
-#endif
-#else // STB_VORBIS_NO_CRT
-#define NULL 0
-#define malloc(s)   0
-#define free(s)     ((void) 0)
-#define realloc(s)  0
-#endif // STB_VORBIS_NO_CRT
-
-#include <limits.h>
-
-#ifdef __MINGW32__
-   // eff you mingw:
-   //     "fixed":
-   //         http://sourceforge.net/p/mingw-w64/mailman/message/32882927/
-   //     "no that broke the build, reverted, who cares about C":
-   //         http://sourceforge.net/p/mingw-w64/mailman/message/32890381/
-   #ifdef __forceinline
-   #undef __forceinline
-   #endif
-   #define __forceinline
-#elif !defined(_MSC_VER)
-   #if __GNUC__
-      #define __forceinline inline
-   #else
-      #define __forceinline
-   #endif
-#endif
-
-#if STB_VORBIS_MAX_CHANNELS > 256
-#error "Value of STB_VORBIS_MAX_CHANNELS outside of allowed range"
-#endif
-
-#if STB_VORBIS_FAST_HUFFMAN_LENGTH > 24
-#error "Value of STB_VORBIS_FAST_HUFFMAN_LENGTH outside of allowed range"
-#endif
-
-
-#if 0
-#include <crtdbg.h>
-#define CHECK(f)   _CrtIsValidHeapPointer(f->channel_buffers[1])
-#else
-#define CHECK(f)   ((void) 0)
-#endif
-
-#define MAX_BLOCKSIZE_LOG  13   // from specification
-#define MAX_BLOCKSIZE      (1 << MAX_BLOCKSIZE_LOG)
-
-
-typedef unsigned char  uint8;
-typedef   signed char   int8;
-typedef unsigned short uint16;
-typedef   signed short  int16;
-typedef unsigned int   uint32;
-typedef   signed int    int32;
-
-#ifndef TRUE
-#define TRUE 1
-#define FALSE 0
-#endif
-
-typedef float codetype;
-
-// @NOTE
-//
-// Some arrays below are tagged "//varies", which means it's actually
-// a variable-sized piece of data, but rather than malloc I assume it's
-// small enough it's better to just allocate it all together with the
-// main thing
-//
-// Most of the variables are specified with the smallest size I could pack
-// them into. It might give better performance to make them all full-sized
-// integers. It should be safe to freely rearrange the structures or change
-// the sizes larger--nothing relies on silently truncating etc., nor the
-// order of variables.
-
-#define FAST_HUFFMAN_TABLE_SIZE   (1 << STB_VORBIS_FAST_HUFFMAN_LENGTH)
-#define FAST_HUFFMAN_TABLE_MASK   (FAST_HUFFMAN_TABLE_SIZE - 1)
-
-typedef struct
-{
-   int dimensions, entries;
-   uint8 *codeword_lengths;
-   float  minimum_value;
-   float  delta_value;
-   uint8  value_bits;
-   uint8  lookup_type;
-   uint8  sequence_p;
-   uint8  sparse;
-   uint32 lookup_values;
-   codetype *multiplicands;
-   uint32 *codewords;
-   #ifdef STB_VORBIS_FAST_HUFFMAN_SHORT
-    int16  fast_huffman[FAST_HUFFMAN_TABLE_SIZE];
-   #else
-    int32  fast_huffman[FAST_HUFFMAN_TABLE_SIZE];
-   #endif
-   uint32 *sorted_codewords;
-   int    *sorted_values;
-   int     sorted_entries;
-} Codebook;
-
-typedef struct
-{
-   uint8 order;
-   uint16 rate;
-   uint16 bark_map_size;
-   uint8 amplitude_bits;
-   uint8 amplitude_offset;
-   uint8 number_of_books;
-   uint8 book_list[16]; // varies
-} Floor0;
-
-typedef struct
-{
-   uint8 partitions;
-   uint8 partition_class_list[32]; // varies
-   uint8 class_dimensions[16]; // varies
-   uint8 class_subclasses[16]; // varies
-   uint8 class_masterbooks[16]; // varies
-   int16 subclass_books[16][8]; // varies
-   uint16 Xlist[31*8+2]; // varies
-   uint8 sorted_order[31*8+2];
-   uint8 neighbors[31*8+2][2];
-   uint8 floor1_multiplier;
-   uint8 rangebits;
-   int values;
-} Floor1;
-
-typedef union
-{
-   Floor0 floor0;
-   Floor1 floor1;
-} Floor;
-
-typedef struct
-{
-   uint32 begin, end;
-   uint32 part_size;
-   uint8 classifications;
-   uint8 classbook;
-   uint8 **classdata;
-   int16 (*residue_books)[8];
-} Residue;
-
-typedef struct
-{
-   uint8 magnitude;
-   uint8 angle;
-   uint8 mux;
-} MappingChannel;
-
-typedef struct
-{
-   uint16 coupling_steps;
-   MappingChannel *chan;
-   uint8  submaps;
-   uint8  submap_floor[15]; // varies
-   uint8  submap_residue[15]; // varies
-} Mapping;
-
-typedef struct
-{
-   uint8 blockflag;
-   uint8 mapping;
-   uint16 windowtype;
-   uint16 transformtype;
-} Mode;
-
-typedef struct
-{
-   uint32  goal_crc;    // expected crc if match
-   int     bytes_left;  // bytes left in packet
-   uint32  crc_so_far;  // running crc
-   int     bytes_done;  // bytes processed in _current_ chunk
-   uint32  sample_loc;  // granule pos encoded in page
-} CRCscan;
-
-typedef struct
-{
-   uint32 page_start, page_end;
-   uint32 last_decoded_sample;
-} ProbedPage;
-
-struct stb_vorbis
-{
-  // user-accessible info
-   unsigned int sample_rate;
-   int channels;
-
-   unsigned int setup_memory_required;
-   unsigned int temp_memory_required;
-   unsigned int setup_temp_memory_required;
-
-  // 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;
-};
-
-#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 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=(uint32) i << 24, j=0; j < 8; ++j)
-         s = (s << 1) ^ (s >= (1U<<31) ? CRC32_POLY : 0);
-      crc_table[i] = s;
-   }
-}
-
-static __forceinline uint32 crc32_update(uint32 crc, uint8 byte)
-{
-   return (crc << 8) ^ crc_table[byte ^ (crc >> 24)];
-}
-
-
-// used in setup, and for huffman that doesn't go fast path
-static unsigned int bit_reverse(unsigned int n)
-{
-  n = ((n & 0xAAAAAAAA) >>  1) | ((n & 0x55555555) << 1);
-  n = ((n & 0xCCCCCCCC) >>  2) | ((n & 0x33333333) << 2);
-  n = ((n & 0xF0F0F0F0) >>  4) | ((n & 0x0F0F0F0F) << 4);
-  n = ((n & 0xFF00FF00) >>  8) | ((n & 0x00FF00FF) << 8);
-  return (n >> 16) | (n << 16);
-}
-
-static float square(float x)
-{
-   return x*x;
-}
-
-// this is a weird definition of log2() for which log2(1) = 1, log2(2) = 2, log2(4) = 3
-// as required by the specification. fast(?) implementation from stb.h
-// @OPTIMIZE: called multiple times per-packet with "constants"; move to setup
-static int ilog(int32 n)
-{
-   static signed char log2_4[16] = { 0,1,2,2,3,3,3,3,4,4,4,4,4,4,4,4 };
-
-   // 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] = 1U << (32-i);
-   // note that the above code treats the first case specially,
-   // but it's really the same as the following code, so they
-   // could probably be combined (except the initial code is 0,
-   // and I use 0 in available[] to mean 'empty')
-   for (i=k+1; i < n; ++i) {
-      uint32 res;
-      int z = len[i], y;
-      if (z == NO_CODE) continue;
-      // find lowest available leaf (should always be earliest,
-      // which is what the specification calls for)
-      // note that this property, and the fact we can never have
-      // more than one free leaf at a given level, isn't totally
-      // trivial to prove, but it seems true and the assert never
-      // fires, so!
-      while (z > 0 && !available[z]) --z;
-      if (z == 0) { return FALSE; }
-      res = available[z];
-      assert(z >= 0 && z < 32);
-      available[z] = 0;
-      add_entry(c, bit_reverse(res), i, m++, len[i], values);
-      // propogate availability up the tree
-      if (z != len[i]) {
-         assert(len[i] >= 0 && len[i] < 32);
-         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 += (uint32) get8(f) << 24;
-   return x;
-}
-
-static int getn(vorb *z, uint8 *data, int n)
-{
-   if (USE_MEMORY(z)) {
-      if (z->stream+n > z->stream_end) { z->eof = 1; return 0; }
-      memcpy(data, z->stream, n);
-      z->stream += n;
-      return 1;
-   }
-
-   #ifndef STB_VORBIS_NO_STDIO
-   if (fread(data, n, 1, z->f) == 1)
-      return 1;
-   else {
-      z->eof = 1;
-      return 0;
-   }
-   #endif
-}
-
-static void skip(vorb *z, int n)
-{
-   if (USE_MEMORY(z)) {
-      z->stream += n;
-      if (z->stream >= z->stream_end) z->eof = 1;
-      return;
-   }
-   #ifndef STB_VORBIS_NO_STDIO
-   {
-      long x = ftell(z->f);
-      fseek(z->f, x+n, SEEK_SET);
-   }
-   #endif
-}
-
-static int set_file_offset(stb_vorbis *f, unsigned int loc)
-{
-   #ifndef STB_VORBIS_NO_PUSHDATA_API
-   if (f->push_mode) return 0;
-   #endif
-   f->eof = 0;
-   if (USE_MEMORY(f)) {
-      if (f->stream_start + loc >= f->stream_end || f->stream_start + loc < f->stream_start) {
-         f->stream = f->stream_end;
-         f->eof = 1;
-         return 0;
-      } else {
-         f->stream = f->stream_start + loc;
-         return 1;
-      }
-   }
-   #ifndef STB_VORBIS_NO_STDIO
-   if (loc + f->f_start < loc || loc >= 0x80000000) {
-      loc = 0x7fffffff;
-      f->eof = 1;
-   } else {
-      loc += f->f_start;
-   }
-   if (!fseek(f->f, loc, SEEK_SET))
-      return 1;
-   f->eof = 1;
-   fseek(f->f, f->f_start, SEEK_END);
-   return 0;
-   #endif
-}
-
-
-static uint8 ogg_page_header[4] = { 0x4f, 0x67, 0x67, 0x53 };
-
-static int capture_pattern(vorb *f)
-{
-   if (0x4f != get8(f)) return FALSE;
-   if (0x67 != get8(f)) return FALSE;
-   if (0x67 != get8(f)) return FALSE;
-   if (0x53 != get8(f)) return FALSE;
-   return TRUE;
-}
-
-#define PAGEFLAG_continued_packet   1
-#define PAGEFLAG_first_page         2
-#define PAGEFLAG_last_page          4
-
-static int start_page_no_capturepattern(vorb *f)
-{
-   uint32 loc0,loc1,n;
-   // 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.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 += (unsigned) z << f->valid_bits;
-         f->valid_bits += 8;
-      } while (f->valid_bits <= 24);
-   }
-}
-
-enum
-{
-   VORBIS_packet_id = 1,
-   VORBIS_packet_comment = 3,
-   VORBIS_packet_setup = 5
-};
-
-static int codebook_decode_scalar_raw(vorb *f, Codebook *c)
-{
-   int i;
-   prep_huffman(f);
-
-   if (c->codewords == NULL && c->sorted_codewords == NULL)
-      return -1;
-
-   // cases to use binary search: sorted_codewords && !c->codewords
-   //                             sorted_codewords && c->entries > 8
-   if (c->entries > 8 ? c->sorted_codewords!=NULL : !c->codewords) {
-      // binary search
-      uint32 code = bit_reverse(f->acc);
-      int x=0, n=c->sorted_entries, len;
-
-      while (n > 1) {
-         // invariant: sc[x] <= code < sc[x+n]
-         int m = x + (n >> 1);
-         if (c->sorted_codewords[m] <= code) {
-            x = m;
-            n -= (n>>1);
-         } else {
-            n >>= 1;
-         }
-      }
-      // x is now the sorted index
-      if (!c->sparse) x = c->sorted_values[x];
-      // x is now sorted index if sparse, or symbol otherwise
-      len = c->codeword_lengths[x];
-      if (f->valid_bits >= len) {
-         f->acc >>= len;
-         f->valid_bits -= len;
-         return x;
-      }
-
-      f->valid_bits = 0;
-      return -1;
-   }
-
-   // if small, linear search
-   assert(!c->sparse);
-   for (i=0; i < c->entries; ++i) {
-      if (c->codeword_lengths[i] == NO_CODE) continue;
-      if (c->codewords[i] == (f->acc & ((1 << c->codeword_lengths[i])-1))) {
-         if (f->valid_bits >= c->codeword_lengths[i]) {
-            f->acc >>= c->codeword_lengths[i];
-            f->valid_bits -= c->codeword_lengths[i];
-            return i;
-         }
-         f->valid_bits = 0;
-         return -1;
-      }
-   }
-
-   error(f, VORBIS_invalid_stream);
-   f->valid_bits = 0;
-   return -1;
-}
-
-#ifndef STB_VORBIS_NO_INLINE_DECODE
-
-#define DECODE_RAW(var, f,c)                                  \
-   if (f->valid_bits < STB_VORBIS_FAST_HUFFMAN_LENGTH)        \
-      prep_huffman(f);                                        \
-   var = f->acc & FAST_HUFFMAN_TABLE_MASK;                    \
-   var = c->fast_huffman[var];                                \
-   if (var >= 0) {                                            \
-      int n = c->codeword_lengths[var];                       \
-      f->acc >>= n;                                           \
-      f->valid_bits -= n;                                     \
-      if (f->valid_bits < 0) { f->valid_bits = 0; var = -1; } \
-   } else {                                                   \
-      var = codebook_decode_scalar_raw(f,c);                  \
-   }
-
-#else
-
-static int codebook_decode_scalar(vorb *f, Codebook *c)
-{
-   int i;
-   if (f->valid_bits < STB_VORBIS_FAST_HUFFMAN_LENGTH)
-      prep_huffman(f);
-   // fast huffman table lookup
-   i = f->acc & FAST_HUFFMAN_TABLE_MASK;
-   i = c->fast_huffman[i];
-   if (i >= 0) {
-      f->acc >>= c->codeword_lengths[i];
-      f->valid_bits -= c->codeword_lengths[i];
-      if (f->valid_bits < 0) { f->valid_bits = 0; return -1; }
-      return i;
-   }
-   return codebook_decode_scalar_raw(f,c);
-}
-
-#define DECODE_RAW(var,f,c)    var = codebook_decode_scalar(f,c);
-
-#endif
-
-#define DECODE(var,f,c)                                       \
-   DECODE_RAW(var,f,c)                                        \
-   if (c->sparse) var = c->sorted_values[var];
-
-#ifndef STB_VORBIS_DIVIDES_IN_CODEBOOK
-  #define DECODE_VQ(var,f,c)   DECODE_RAW(var,f,c)
-#else
-  #define DECODE_VQ(var,f,c)   DECODE(var,f,c)
-#endif
-
-
-
-
-
-
-// CODEBOOK_ELEMENT_FAST is an optimization for the CODEBOOK_FLOATS case
-// where we avoid one addition
-#define CODEBOOK_ELEMENT(c,off)          (c->multiplicands[off])
-#define CODEBOOK_ELEMENT_FAST(c,off)     (c->multiplicands[off])
-#define CODEBOOK_ELEMENT_BASE(c)         (0)
-
-static int codebook_decode_start(vorb *f, Codebook *c)
-{
-   int z = -1;
-
-   // type 0 is only legal in a scalar context
-   if (c->lookup_type == 0)
-      error(f, VORBIS_invalid_stream);
-   else {
-      DECODE_VQ(z,f,c);
-      if (c->sparse) assert(z < c->sorted_entries);
-      if (z < 0) {  // check for EOP
-         if (!f->bytes_in_seg)
-            if (f->last_seg)
-               return z;
-         error(f, VORBIS_invalid_stream);
-      }
-   }
-   return z;
-}
-
-static int codebook_decode(vorb *f, Codebook *c, float *output, int len)
-{
-   int i,z = codebook_decode_start(f,c);
-   if (z < 0) return FALSE;
-   if (len > c->dimensions) len = c->dimensions;
-
-#ifdef STB_VORBIS_DIVIDES_IN_CODEBOOK
-   if (c->lookup_type == 1) {
-      float last = CODEBOOK_ELEMENT_BASE(c);
-      int div = 1;
-      for (i=0; i < len; ++i) {
-         int off = (z / div) % c->lookup_values;
-         float val = CODEBOOK_ELEMENT_FAST(c,off) + last;
-         output[i] += val;
-         if (c->sequence_p) last = val + c->minimum_value;
-         div *= c->lookup_values;
-      }
-      return TRUE;
-   }
-#endif
-
-   z *= c->dimensions;
-   if (c->sequence_p) {
-      float last = CODEBOOK_ELEMENT_BASE(c);
-      for (i=0; i < len; ++i) {
-         float val = CODEBOOK_ELEMENT_FAST(c,z+i) + last;
-         output[i] += val;
-         last = val + c->minimum_value;
-      }
-   } else {
-      float last = CODEBOOK_ELEMENT_BASE(c);
-      for (i=0; i < len; ++i) {
-         output[i] += CODEBOOK_ELEMENT_FAST(c,z+i) + last;
-      }
-   }
-
-   return TRUE;
-}
-
-static int codebook_decode_step(vorb *f, Codebook *c, float *output, int len, int step)
-{
-   int i,z = codebook_decode_start(f,c);
-   float last = CODEBOOK_ELEMENT_BASE(c);
-   if (z < 0) return FALSE;
-   if (len > c->dimensions) len = c->dimensions;
-
-#ifdef STB_VORBIS_DIVIDES_IN_CODEBOOK
-   if (c->lookup_type == 1) {
-      int div = 1;
-      for (i=0; i < len; ++i) {
-         int off = (z / div) % c->lookup_values;
-         float val = CODEBOOK_ELEMENT_FAST(c,off) + last;
-         output[i*step] += val;
-         if (c->sequence_p) last = val;
-         div *= c->lookup_values;
-      }
-      return TRUE;
-   }
-#endif
-
-   z *= c->dimensions;
-   for (i=0; i < len; ++i) {
-      float val = CODEBOOK_ELEMENT_FAST(c,z+i) + last;
-      output[i*step] += val;
-      if (c->sequence_p) last = val;
-   }
-
-   return TRUE;
-}
-
-static int codebook_decode_deinterleave_repeat(vorb *f, Codebook *c, float **outputs, int ch, int *c_inter_p, int *p_inter_p, int len, int total_decode)
-{
-   int c_inter = *c_inter_p;
-   int p_inter = *p_inter_p;
-   int i,z, effective = c->dimensions;
-
-   // type 0 is only legal in a scalar context
-   if (c->lookup_type == 0)   return error(f, VORBIS_invalid_stream);
-
-   while (total_decode > 0) {
-      float last = CODEBOOK_ELEMENT_BASE(c);
-      DECODE_VQ(z,f,c);
-      #ifndef STB_VORBIS_DIVIDES_IN_CODEBOOK
-      assert(!c->sparse || z < c->sorted_entries);
-      #endif
-      if (z < 0) {
-         if (!f->bytes_in_seg)
-            if (f->last_seg) return FALSE;
-         return error(f, VORBIS_invalid_stream);
-      }
-
-      // if this will take us off the end of the buffers, stop short!
-      // we check by computing the length of the virtual interleaved
-      // buffer (len*ch), our current offset within it (p_inter*ch)+(c_inter),
-      // and the length we'll be using (effective)
-      if (c_inter + p_inter*ch + effective > len * ch) {
-         effective = len*ch - (p_inter*ch - c_inter);
-      }
-
-   #ifdef STB_VORBIS_DIVIDES_IN_CODEBOOK
-      if (c->lookup_type == 1) {
-         int div = 1;
-         for (i=0; i < effective; ++i) {
-            int off = (z / div) % c->lookup_values;
-            float val = CODEBOOK_ELEMENT_FAST(c,off) + last;
-            if (outputs[c_inter])
-               outputs[c_inter][p_inter] += val;
-            if (++c_inter == ch) { c_inter = 0; ++p_inter; }
-            if (c->sequence_p) last = val;
-            div *= c->lookup_values;
-         }
-      } else
-   #endif
-      {
-         z *= c->dimensions;
-         if (c->sequence_p) {
-            for (i=0; i < effective; ++i) {
-               float val = CODEBOOK_ELEMENT_FAST(c,z+i) + last;
-               if (outputs[c_inter])
-                  outputs[c_inter][p_inter] += val;
-               if (++c_inter == ch) { c_inter = 0; ++p_inter; }
-               last = val;
-            }
-         } else {
-            for (i=0; i < effective; ++i) {
-               float val = CODEBOOK_ELEMENT_FAST(c,z+i) + last;
-               if (outputs[c_inter])
-                  outputs[c_inter][p_inter] += val;
-               if (++c_inter == ch) { c_inter = 0; ++p_inter; }
-            }
-         }
-      }
-
-      total_decode -= effective;
-   }
-   *c_inter_p = c_inter;
-   *p_inter_p = p_inter;
-   return TRUE;
-}
-
-static int predict_point(int x, int x0, int x1, int y0, int y1)
-{
-   int dy = y1 - y0;
-   int adx = x1 - x0;
-   // @OPTIMIZE: force int division to round in the right direction... is this necessary on x86?
-   int err = abs(dy) * (x - x0);
-   int off = err / adx;
-   return dy < 0 ? y0 - off : y0 + off;
-}
-
-// the following table is block-copied from the specification
-static float inverse_db_table[256] =
-{
-  1.0649863e-07f, 1.1341951e-07f, 1.2079015e-07f, 1.2863978e-07f,
-  1.3699951e-07f, 1.4590251e-07f, 1.5538408e-07f, 1.6548181e-07f,
-  1.7623575e-07f, 1.8768855e-07f, 1.9988561e-07f, 2.1287530e-07f,
-  2.2670913e-07f, 2.4144197e-07f, 2.5713223e-07f, 2.7384213e-07f,
-  2.9163793e-07f, 3.1059021e-07f, 3.3077411e-07f, 3.5226968e-07f,
-  3.7516214e-07f, 3.9954229e-07f, 4.2550680e-07f, 4.5315863e-07f,
-  4.8260743e-07f, 5.1396998e-07f, 5.4737065e-07f, 5.8294187e-07f,
-  6.2082472e-07f, 6.6116941e-07f, 7.0413592e-07f, 7.4989464e-07f,
-  7.9862701e-07f, 8.5052630e-07f, 9.0579828e-07f, 9.6466216e-07f,
-  1.0273513e-06f, 1.0941144e-06f, 1.1652161e-06f, 1.2409384e-06f,
-  1.3215816e-06f, 1.4074654e-06f, 1.4989305e-06f, 1.5963394e-06f,
-  1.7000785e-06f, 1.8105592e-06f, 1.9282195e-06f, 2.0535261e-06f,
-  2.1869758e-06f, 2.3290978e-06f, 2.4804557e-06f, 2.6416497e-06f,
-  2.8133190e-06f, 2.9961443e-06f, 3.1908506e-06f, 3.3982101e-06f,
-  3.6190449e-06f, 3.8542308e-06f, 4.1047004e-06f, 4.3714470e-06f,
-  4.6555282e-06f, 4.9580707e-06f, 5.2802740e-06f, 5.6234160e-06f,
-  5.9888572e-06f, 6.3780469e-06f, 6.7925283e-06f, 7.2339451e-06f,
-  7.7040476e-06f, 8.2047000e-06f, 8.7378876e-06f, 9.3057248e-06f,
-  9.9104632e-06f, 1.0554501e-05f, 1.1240392e-05f, 1.1970856e-05f,
-  1.2748789e-05f, 1.3577278e-05f, 1.4459606e-05f, 1.5399272e-05f,
-  1.6400004e-05f, 1.7465768e-05f, 1.8600792e-05f, 1.9809576e-05f,
-  2.1096914e-05f, 2.2467911e-05f, 2.3928002e-05f, 2.5482978e-05f,
-  2.7139006e-05f, 2.8902651e-05f, 3.0780908e-05f, 3.2781225e-05f,
-  3.4911534e-05f, 3.7180282e-05f, 3.9596466e-05f, 4.2169667e-05f,
-  4.4910090e-05f, 4.7828601e-05f, 5.0936773e-05f, 5.4246931e-05f,
-  5.7772202e-05f, 6.1526565e-05f, 6.5524908e-05f, 6.9783085e-05f,
-  7.4317983e-05f, 7.9147585e-05f, 8.4291040e-05f, 8.9768747e-05f,
-  9.5602426e-05f, 0.00010181521f, 0.00010843174f, 0.00011547824f,
-  0.00012298267f, 0.00013097477f, 0.00013948625f, 0.00014855085f,
-  0.00015820453f, 0.00016848555f, 0.00017943469f, 0.00019109536f,
-  0.00020351382f, 0.00021673929f, 0.00023082423f, 0.00024582449f,
-  0.00026179955f, 0.00027881276f, 0.00029693158f, 0.00031622787f,
-  0.00033677814f, 0.00035866388f, 0.00038197188f, 0.00040679456f,
-  0.00043323036f, 0.00046138411f, 0.00049136745f, 0.00052329927f,
-  0.00055730621f, 0.00059352311f, 0.00063209358f, 0.00067317058f,
-  0.00071691700f, 0.00076350630f, 0.00081312324f, 0.00086596457f,
-  0.00092223983f, 0.00098217216f, 0.0010459992f,  0.0011139742f,
-  0.0011863665f,  0.0012634633f,  0.0013455702f,  0.0014330129f,
-  0.0015261382f,  0.0016253153f,  0.0017309374f,  0.0018434235f,
-  0.0019632195f,  0.0020908006f,  0.0022266726f,  0.0023713743f,
-  0.0025254795f,  0.0026895994f,  0.0028643847f,  0.0030505286f,
-  0.0032487691f,  0.0034598925f,  0.0036847358f,  0.0039241906f,
-  0.0041792066f,  0.0044507950f,  0.0047400328f,  0.0050480668f,
-  0.0053761186f,  0.0057254891f,  0.0060975636f,  0.0064938176f,
-  0.0069158225f,  0.0073652516f,  0.0078438871f,  0.0083536271f,
-  0.0088964928f,  0.009474637f,   0.010090352f,   0.010746080f,
-  0.011444421f,   0.012188144f,   0.012980198f,   0.013823725f,
-  0.014722068f,   0.015678791f,   0.016697687f,   0.017782797f,
-  0.018938423f,   0.020169149f,   0.021479854f,   0.022875735f,
-  0.024362330f,   0.025945531f,   0.027631618f,   0.029427276f,
-  0.031339626f,   0.033376252f,   0.035545228f,   0.037855157f,
-  0.040315199f,   0.042935108f,   0.045725273f,   0.048696758f,
-  0.051861348f,   0.055231591f,   0.058820850f,   0.062643361f,
-  0.066714279f,   0.071049749f,   0.075666962f,   0.080584227f,
-  0.085821044f,   0.091398179f,   0.097337747f,   0.10366330f,
-  0.11039993f,    0.11757434f,    0.12521498f,    0.13335215f,
-  0.14201813f,    0.15124727f,    0.16107617f,    0.17154380f,
-  0.18269168f,    0.19456402f,    0.20720788f,    0.22067342f,
-  0.23501402f,    0.25028656f,    0.26655159f,    0.28387361f,
-  0.30232132f,    0.32196786f,    0.34289114f,    0.36517414f,
-  0.38890521f,    0.41417847f,    0.44109412f,    0.46975890f,
-  0.50028648f,    0.53279791f,    0.56742212f,    0.60429640f,
-  0.64356699f,    0.68538959f,    0.72993007f,    0.77736504f,
-  0.82788260f,    0.88168307f,    0.9389798f,     1.0f
-};
-
-
-// @OPTIMIZE: if you want to replace this bresenham line-drawing routine,
-// note that you must produce bit-identical output to decode correctly;
-// this specific sequence of operations is specified in the spec (it's
-// drawing integer-quantized frequency-space lines that the encoder
-// expects to be exactly the same)
-//     ... also, isn't the whole point of Bresenham's algorithm to NOT
-// have to divide in the setup? sigh.
-#ifndef STB_VORBIS_NO_DEFER_FLOOR
-#define LINE_OP(a,b)   a *= b
-#else
-#define LINE_OP(a,b)   a = b
-#endif
-
-#ifdef STB_VORBIS_DIVIDE_TABLE
-#define DIVTAB_NUMER   32
-#define DIVTAB_DENOM   64
-int8 integer_divide_table[DIVTAB_NUMER][DIVTAB_DENOM]; // 2KB
-#endif
-
-static __forceinline void draw_line(float *output, int x0, int y0, int x1, int y1, int n)
-{
-   int dy = y1 - y0;
-   int adx = x1 - x0;
-   int ady = abs(dy);
-   int base;
-   int x=x0,y=y0;
-   int err = 0;
-   int sy;
-
-#ifdef STB_VORBIS_DIVIDE_TABLE
-   if (adx < DIVTAB_DENOM && ady < DIVTAB_NUMER) {
-      if (dy < 0) {
-         base = -integer_divide_table[ady][adx];
-         sy = base-1;
-      } else {
-         base =  integer_divide_table[ady][adx];
-         sy = base+1;
-      }
-   } else {
-      base = dy / adx;
-      if (dy < 0)
-         sy = base - 1;
-      else
-         sy = base+1;
-   }
-#else
-   base = dy / adx;
-   if (dy < 0)
-      sy = base - 1;
-   else
-      sy = base+1;
-#endif
-   ady -= abs(base) * adx;
-   if (x1 > n) x1 = n;
-   if (x < x1) {
-      LINE_OP(output[x], inverse_db_table[y]);
-      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
-
-   CHECK(f);
-
-   for (i=0; i < ch; ++i)
-      if (!do_not_decode[i])
-         memset(residue_buffers[i], 0, sizeof(float) * n);
-
-   if (rtype == 2 && ch != 1) {
-      for (j=0; j < ch; ++j)
-         if (!do_not_decode[j])
-            break;
-      if (j == ch)
-         goto done;
-
-      for (pass=0; pass < 8; ++pass) {
-         int pcount = 0, class_set = 0;
-         if (ch == 2) {
-            while (pcount < part_read) {
-               int z = r->begin + pcount*r->part_size;
-               int c_inter = (z & 1), p_inter = z>>1;
-               if (pass == 0) {
-                  Codebook *c = f->codebooks+r->classbook;
-                  int q;
-                  DECODE(q,f,c);
-                  if (q == EOP) goto done;
-                  #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE
-                  part_classdata[0][class_set] = r->classdata[q];
-                  #else
-                  for (i=classwords-1; i >= 0; --i) {
-                     classifications[0][i+pcount] = q % r->classifications;
-                     q /= r->classifications;
-                  }
-                  #endif
-               }
-               for (i=0; i < classwords && pcount < part_read; ++i, ++pcount) {
-                  int z = r->begin + pcount*r->part_size;
-                  #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE
-                  int c = part_classdata[0][class_set][i];
-                  #else
-                  int c = classifications[0][pcount];
-                  #endif
-                  int b = r->residue_books[c][pass];
-                  if (b >= 0) {
-                     Codebook *book = f->codebooks + b;
-                     #ifdef STB_VORBIS_DIVIDES_IN_CODEBOOK
-                     if (!codebook_decode_deinterleave_repeat(f, book, residue_buffers, ch, &c_inter, &p_inter, n, r->part_size))
-                        goto done;
-                     #else
-                     // saves 1%
-                     if (!codebook_decode_deinterleave_repeat(f, book, residue_buffers, ch, &c_inter, &p_inter, n, r->part_size))
-                        goto done;
-                     #endif
-                  } else {
-                     z += r->part_size;
-                     c_inter = z & 1;
-                     p_inter = z >> 1;
-                  }
-               }
-               #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE
-               ++class_set;
-               #endif
-            }
-         } else if (ch == 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;
-                     if (!codebook_decode_deinterleave_repeat(f, book, residue_buffers, ch, &c_inter, &p_inter, n, r->part_size))
-                        goto done;
-                  } else {
-                     z += r->part_size;
-                     c_inter = 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;
-                     if (!codebook_decode_deinterleave_repeat(f, book, residue_buffers, ch, &c_inter, &p_inter, n, r->part_size))
-                        goto done;
-                  } else {
-                     z += r->part_size;
-                     c_inter = z % ch;
-                     p_inter = z / ch;
-                  }
-               }
-               #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE
-               ++class_set;
-               #endif
-            }
-         }
-      }
-      goto done;
-   }
-   CHECK(f);
-
-   for (pass=0; pass < 8; ++pass) {
-      int pcount = 0, class_set=0;
-      while (pcount < part_read) {
-         if (pass == 0) {
-            for (j=0; j < ch; ++j) {
-               if (!do_not_decode[j]) {
-                  Codebook *c = f->codebooks+r->classbook;
-                  int temp;
-                  DECODE(temp,f,c);
-                  if (temp == EOP) goto done;
-                  #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE
-                  part_classdata[j][class_set] = r->classdata[temp];
-                  #else
-                  for (i=classwords-1; i >= 0; --i) {
-                     classifications[j][i+pcount] = temp % r->classifications;
-                     temp /= r->classifications;
-                  }
-                  #endif
-               }
-            }
-         }
-         for (i=0; i < classwords && pcount < part_read; ++i, ++pcount) {
-            for (j=0; j < ch; ++j) {
-               if (!do_not_decode[j]) {
-                  #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE
-                  int c = part_classdata[j][class_set][i];
-                  #else
-                  int c = classifications[j][pcount];
-                  #endif
-                  int b = r->residue_books[c][pass];
-                  if (b >= 0) {
-                     float *target = residue_buffers[j];
-                     int offset = r->begin + pcount * r->part_size;
-                     int n = r->part_size;
-                     Codebook *book = f->codebooks + b;
-                     if (!residue_decode(f, book, target, offset, n, rtype))
-                        goto done;
-                  }
-               }
-            }
-         }
-         #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE
-         ++class_set;
-         #endif
-      }
-   }
-  done:
-   CHECK(f);
-   #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE
-   temp_free(f,part_classdata);
-   #else
-   temp_free(f,classifications);
-   #endif
-   temp_alloc_restore(f,temp_alloc_point);
-}
-
-
-#if 0
-// slow way for debugging
-void inverse_mdct_slow(float *buffer, int n)
-{
-   int i,j;
-   int n2 = n >> 1;
-   float *x = (float *) malloc(sizeof(*x) * n2);
-   memcpy(x, buffer, sizeof(*x) * n2);
-   for (i=0; i < n; ++i) {
-      float acc = 0;
-      for (j=0; j < n2; ++j)
-         // formula from paper:
-         //acc += n/4.0f * x[j] * (float) cos(M_PI / 2 / n * (2 * i + 1 + n/2.0)*(2*j+1));
-         // formula from wikipedia
-         //acc += 2.0f / n2 * x[j] * (float) cos(M_PI/n2 * (i + 0.5 + n2/2)*(j + 0.5));
-         // these are equivalent, except the formula from the paper inverts the multiplier!
-         // however, what actually works is NO MULTIPLIER!?!
-         //acc += 64 * 2.0f / n2 * x[j] * (float) cos(M_PI/n2 * (i + 0.5 + n2/2)*(j + 0.5));
-         acc += x[j] * (float) cos(M_PI / 2 / n * (2 * i + 1 + n/2.0)*(2*j+1));
-      buffer[i] = acc;
-   }
-   free(x);
-}
-#elif 0
-// same as above, but just barely able to run in real time on modern machines
-void inverse_mdct_slow(float *buffer, int n, vorb *f, int blocktype)
-{
-   float mcos[16384];
-   int i,j;
-   int n2 = n >> 1, nmask = (n << 2) -1;
-   float *x = (float *) malloc(sizeof(*x) * n2);
-   memcpy(x, buffer, sizeof(*x) * n2);
-   for (i=0; i < 4*n; ++i)
-      mcos[i] = (float) cos(M_PI / 2 * i / n);
-
-   for (i=0; i < n; ++i) {
-      float acc = 0;
-      for (j=0; j < n2; ++j)
-         acc += x[j] * mcos[(2 * i + 1 + n2)*(2*j+1) & nmask];
-      buffer[i] = acc;
-   }
-   free(x);
-}
-#elif 0
-// transform to use a slow dct-iv; this is STILL basically trivial,
-// but only requires half as many ops
-void dct_iv_slow(float *buffer, int n)
-{
-   float mcos[16384];
-   float x[2048];
-   int i,j;
-   int n2 = n >> 1, nmask = (n << 3) - 1;
-   memcpy(x, buffer, sizeof(*x) * n);
-   for (i=0; i < 8*n; ++i)
-      mcos[i] = (float) cos(M_PI / 4 * i / n);
-   for (i=0; i < n; ++i) {
-      float acc = 0;
-      for (j=0; j < n; ++j)
-         acc += x[j] * mcos[((2 * i + 1)*(2*j+1)) & nmask];
-      buffer[i] = acc;
-   }
-}
-
-void inverse_mdct_slow(float *buffer, int n, vorb *f, int blocktype)
-{
-   int i, n4 = n >> 2, n2 = n >> 1, n3_4 = n - n4;
-   float temp[4096];
-
-   memcpy(temp, buffer, n2 * sizeof(float));
-   dct_iv_slow(temp, n2);  // returns -c'-d, a-b'
-
-   for (i=0; i < n4  ; ++i) buffer[i] = temp[i+n4];            // a-b'
-   for (   ; i < n3_4; ++i) buffer[i] = -temp[n3_4 - i - 1];   // b-a', c+d'
-   for (   ; i < n   ; ++i) buffer[i] = -temp[i - n3_4];       // c'+d
-}
-#endif
-
-#ifndef LIBVORBIS_MDCT
-#define LIBVORBIS_MDCT 0
-#endif
-
-#if LIBVORBIS_MDCT
-// directly call the vorbis MDCT using an interface documented
-// by Jeff Roberts... useful for performance comparison
-typedef struct
-{
-  int n;
-  int log2n;
-
-  float *trig;
-  int   *bitrev;
-
-  float scale;
-} mdct_lookup;
-
-extern void mdct_init(mdct_lookup *lookup, int n);
-extern void mdct_clear(mdct_lookup *l);
-extern void mdct_backward(mdct_lookup *init, float *in, float *out);
-
-mdct_lookup M1,M2;
-
-void inverse_mdct(float *buffer, int n, vorb *f, int blocktype)
-{
-   mdct_lookup *M;
-   if (M1.n == n) M = &M1;
-   else if (M2.n == n) M = &M2;
-   else if (M1.n == 0) { mdct_init(&M1, n); M = &M1; }
-   else {
-      if (M2.n) __asm int 3;
-      mdct_init(&M2, n);
-      M = &M2;
-   }
-
-   mdct_backward(M, buffer, buffer);
-}
-#endif
-
-
-// the following were split out into separate functions while optimizing;
-// they could be pushed back up but eh. __forceinline showed no change;
-// they're probably already being inlined.
-static void imdct_step3_iter0_loop(int n, float *e, int i_off, int k_off, float *A)
-{
-   float *ee0 = e + i_off;
-   float *ee2 = ee0 + k_off;
-   int i;
-
-   assert((n & 3) == 0);
-   for (i=(n>>2); i > 0; --i) {
-      float k00_20, k01_21;
-      k00_20  = ee0[ 0] - ee2[ 0];
-      k01_21  = ee0[-1] - ee2[-1];
-      ee0[ 0] += ee2[ 0];//ee0[ 0] = ee0[ 0] + ee2[ 0];
-      ee0[-1] += ee2[-1];//ee0[-1] = ee0[-1] + ee2[-1];
-      ee2[ 0] = k00_20 * A[0] - k01_21 * A[1];
-      ee2[-1] = k01_21 * A[0] + k00_20 * A[1];
-      A += 8;
-
-      k00_20  = ee0[-2] - ee2[-2];
-      k01_21  = ee0[-3] - ee2[-3];
-      ee0[-2] += ee2[-2];//ee0[-2] = ee0[-2] + ee2[-2];
-      ee0[-3] += ee2[-3];//ee0[-3] = ee0[-3] + ee2[-3];
-      ee2[-2] = k00_20 * A[0] - k01_21 * A[1];
-      ee2[-3] = k01_21 * A[0] + k00_20 * A[1];
-      A += 8;
-
-      k00_20  = ee0[-4] - ee2[-4];
-      k01_21  = ee0[-5] - ee2[-5];
-      ee0[-4] += ee2[-4];//ee0[-4] = ee0[-4] + ee2[-4];
-      ee0[-5] += ee2[-5];//ee0[-5] = ee0[-5] + ee2[-5];
-      ee2[-4] = k00_20 * A[0] - k01_21 * A[1];
-      ee2[-5] = k01_21 * A[0] + k00_20 * A[1];
-      A += 8;
-
-      k00_20  = ee0[-6] - ee2[-6];
-      k01_21  = ee0[-7] - ee2[-7];
-      ee0[-6] += ee2[-6];//ee0[-6] = ee0[-6] + ee2[-6];
-      ee0[-7] += ee2[-7];//ee0[-7] = ee0[-7] + ee2[-7];
-      ee2[-6] = k00_20 * A[0] - k01_21 * A[1];
-      ee2[-7] = k01_21 * A[0] + k00_20 * A[1];
-      A += 8;
-      ee0 -= 8;
-      ee2 -= 8;
-   }
-}
-
-static void imdct_step3_inner_r_loop(int lim, float *e, int d0, int k_off, float *A, int k1)
-{
-   int i;
-   float k00_20, k01_21;
-
-   float *e0 = e + d0;
-   float *e2 = e0 + k_off;
-
-   for (i=lim >> 2; i > 0; --i) {
-      k00_20 = e0[-0] - e2[-0];
-      k01_21 = e0[-1] - e2[-1];
-      e0[-0] += e2[-0];//e0[-0] = e0[-0] + e2[-0];
-      e0[-1] += e2[-1];//e0[-1] = e0[-1] + e2[-1];
-      e2[-0] = (k00_20)*A[0] - (k01_21) * A[1];
-      e2[-1] = (k01_21)*A[0] + (k00_20) * A[1];
-
-      A += k1;
-
-      k00_20 = e0[-2] - e2[-2];
-      k01_21 = e0[-3] - e2[-3];
-      e0[-2] += e2[-2];//e0[-2] = e0[-2] + e2[-2];
-      e0[-3] += e2[-3];//e0[-3] = e0[-3] + e2[-3];
-      e2[-2] = (k00_20)*A[0] - (k01_21) * A[1];
-      e2[-3] = (k01_21)*A[0] + (k00_20) * A[1];
-
-      A += k1;
-
-      k00_20 = e0[-4] - e2[-4];
-      k01_21 = e0[-5] - e2[-5];
-      e0[-4] += e2[-4];//e0[-4] = e0[-4] + e2[-4];
-      e0[-5] += e2[-5];//e0[-5] = e0[-5] + e2[-5];
-      e2[-4] = (k00_20)*A[0] - (k01_21) * A[1];
-      e2[-5] = (k01_21)*A[0] + (k00_20) * A[1];
-
-      A += k1;
-
-      k00_20 = e0[-6] - e2[-6];
-      k01_21 = e0[-7] - e2[-7];
-      e0[-6] += e2[-6];//e0[-6] = e0[-6] + e2[-6];
-      e0[-7] += e2[-7];//e0[-7] = e0[-7] + e2[-7];
-      e2[-6] = (k00_20)*A[0] - (k01_21) * A[1];
-      e2[-7] = (k01_21)*A[0] + (k00_20) * A[1];
-
-      e0 -= 8;
-      e2 -= 8;
-
-      A += k1;
-   }
-}
-
-static void imdct_step3_inner_s_loop(int n, float *e, int i_off, int k_off, float *A, int a_off, int k0)
-{
-   int i;
-   float A0 = A[0];
-   float A1 = A[0+1];
-   float A2 = A[0+a_off];
-   float A3 = A[0+a_off+1];
-   float A4 = A[0+a_off*2+0];
-   float A5 = A[0+a_off*2+1];
-   float A6 = A[0+a_off*3+0];
-   float A7 = A[0+a_off*3+1];
-
-   float k00,k11;
-
-   float *ee0 = e  +i_off;
-   float *ee2 = ee0+k_off;
-
-   for (i=n; i > 0; --i) {
-      k00     = ee0[ 0] - ee2[ 0];
-      k11     = ee0[-1] - ee2[-1];
-      ee0[ 0] =  ee0[ 0] + ee2[ 0];
-      ee0[-1] =  ee0[-1] + ee2[-1];
-      ee2[ 0] = (k00) * A0 - (k11) * A1;
-      ee2[-1] = (k11) * A0 + (k00) * A1;
-
-      k00     = ee0[-2] - ee2[-2];
-      k11     = ee0[-3] - ee2[-3];
-      ee0[-2] =  ee0[-2] + ee2[-2];
-      ee0[-3] =  ee0[-3] + ee2[-3];
-      ee2[-2] = (k00) * A2 - (k11) * A3;
-      ee2[-3] = (k11) * A2 + (k00) * A3;
-
-      k00     = ee0[-4] - ee2[-4];
-      k11     = ee0[-5] - ee2[-5];
-      ee0[-4] =  ee0[-4] + ee2[-4];
-      ee0[-5] =  ee0[-5] + ee2[-5];
-      ee2[-4] = (k00) * A4 - (k11) * A5;
-      ee2[-5] = (k11) * A4 + (k00) * A5;
-
-      k00     = ee0[-6] - ee2[-6];
-      k11     = ee0[-7] - ee2[-7];
-      ee0[-6] =  ee0[-6] + ee2[-6];
-      ee0[-7] =  ee0[-7] + ee2[-7];
-      ee2[-6] = (k00) * A6 - (k11) * A7;
-      ee2[-7] = (k11) * A6 + (k00) * A7;
-
-      ee0 -= k0;
-      ee2 -= k0;
-   }
-}
-
-static __forceinline void iter_54(float *z)
-{
-   float k00,k11,k22,k33;
-   float y0,y1,y2,y3;
-
-   k00  = z[ 0] - z[-4];
-   y0   = z[ 0] + z[-4];
-   y2   = z[-2] + z[-6];
-   k22  = z[-2] - z[-6];
-
-   z[-0] = y0 + y2;      // z0 + z4 + z2 + z6
-   z[-2] = y0 - y2;      // z0 + z4 - z2 - z6
-
-   // done with y0,y2
-
-   k33  = z[-3] - z[-7];
-
-   z[-4] = k00 + k33;    // z0 - z4 + z3 - z7
-   z[-6] = k00 - k33;    // z0 - z4 - z3 + z7
-
-   // done with k33
-
-   k11  = z[-1] - z[-5];
-   y1   = z[-1] + z[-5];
-   y3   = z[-3] + z[-7];
-
-   z[-1] = y1 + y3;      // z1 + z5 + z3 + z7
-   z[-3] = y1 - y3;      // z1 + z5 - z3 - z7
-   z[-5] = k11 - k22;    // z1 - z5 + z2 - z6
-   z[-7] = k11 + k22;    // z1 - z5 - z2 + z6
-}
-
-static void imdct_step3_inner_s_loop_ld654(int n, float *e, int i_off, float *A, int base_n)
-{
-   int a_off = base_n >> 3;
-   float A2 = A[0+a_off];
-   float *z = e + i_off;
-   float *base = z - 16 * n;
-
-   while (z > base) {
-      float k00,k11;
-
-      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_free(f,buf2);
-   temp_alloc_restore(f,save_point);
-}
-
-#if 0
-// this is the original version of the above code, if you want to optimize it from scratch
-void inverse_mdct_naive(float *buffer, int n)
-{
-   float s;
-   float A[1 << 12], B[1 << 12], C[1 << 11];
-   int i,k,k2,k4, n2 = n >> 1, n4 = n >> 2, n8 = n >> 3, l;
-   int n3_4 = n - n4, ld;
-   // how can they claim this only uses N words?!
-   // oh, because they're only used sparsely, whoops
-   float u[1 << 13], X[1 << 13], v[1 << 13], w[1 << 13];
-   // set up twiddle factors
-
-   for (k=k2=0; k < n4; ++k,k2+=2) {
-      A[k2  ] = (float)  cos(4*k*M_PI/n);
-      A[k2+1] = (float) -sin(4*k*M_PI/n);
-      B[k2  ] = (float)  cos((k2+1)*M_PI/n/2);
-      B[k2+1] = (float)  sin((k2+1)*M_PI/n/2);
-   }
-   for (k=k2=0; k < n8; ++k,k2+=2) {
-      C[k2  ] = (float)  cos(2*(k2+1)*M_PI/n);
-      C[k2+1] = (float) -sin(2*(k2+1)*M_PI/n);
-   }
-
-   // IMDCT algorithm from "The use of multirate filter banks for coding of high quality digital audio"
-   // Note there are bugs in that pseudocode, presumably due to them attempting
-   // to rename the arrays nicely rather than representing the way their actual
-   // implementation bounces buffers back and forth. As a result, even in the
-   // "some formulars corrected" version, a direct implementation fails. These
-   // are noted below as "paper bug".
-
-   // copy and reflect spectral data
-   for (k=0; k < n2; ++k) u[k] = buffer[k];
-   for (   ; k < n ; ++k) u[k] = -buffer[n - k - 1];
-   // kernel from paper
-   // step 1
-   for (k=k2=k4=0; k < n4; k+=1, k2+=2, k4+=4) {
-      v[n-k4-1] = (u[k4] - u[n-k4-1]) * A[k2]   - (u[k4+2] - u[n-k4-3])*A[k2+1];
-      v[n-k4-3] = (u[k4] - u[n-k4-1]) * A[k2+1] + (u[k4+2] - u[n-k4-3])*A[k2];
-   }
-   // step 2
-   for (k=k4=0; k < n8; k+=1, k4+=4) {
-      w[n2+3+k4] = v[n2+3+k4] + v[k4+3];
-      w[n2+1+k4] = v[n2+1+k4] + v[k4+1];
-      w[k4+3]    = (v[n2+3+k4] - v[k4+3])*A[n2-4-k4] - (v[n2+1+k4]-v[k4+1])*A[n2-3-k4];
-      w[k4+1]    = (v[n2+1+k4] - v[k4+1])*A[n2-4-k4] + (v[n2+3+k4]-v[k4+3])*A[n2-3-k4];
-   }
-   // step 3
-   ld = ilog(n) - 1; // ilog is off-by-one from normal definitions
-   for (l=0; l < ld-3; ++l) {
-      int k0 = n >> (l+2), k1 = 1 << (l+3);
-      int rlim = n >> (l+4), r4, r;
-      int s2lim = 1 << (l+2), s2;
-      for (r=r4=0; r < rlim; r4+=4,++r) {
-         for (s2=0; s2 < s2lim; s2+=2) {
-            u[n-1-k0*s2-r4] = w[n-1-k0*s2-r4] + w[n-1-k0*(s2+1)-r4];
-            u[n-3-k0*s2-r4] = w[n-3-k0*s2-r4] + w[n-3-k0*(s2+1)-r4];
-            u[n-1-k0*(s2+1)-r4] = (w[n-1-k0*s2-r4] - w[n-1-k0*(s2+1)-r4]) * A[r*k1]
-                                - (w[n-3-k0*s2-r4] - w[n-3-k0*(s2+1)-r4]) * A[r*k1+1];
-            u[n-3-k0*(s2+1)-r4] = (w[n-3-k0*s2-r4] - w[n-3-k0*(s2+1)-r4]) * A[r*k1]
-                                + (w[n-1-k0*s2-r4] - w[n-1-k0*(s2+1)-r4]) * A[r*k1+1];
-         }
-      }
-      if (l+1 < ld-3) {
-         // paper bug: ping-ponging of u&w here is omitted
-         memcpy(w, u, sizeof(u));
-      }
-   }
-
-   // step 4
-   for (i=0; i < n8; ++i) {
-      int j = bit_reverse(i) >> (32-ld+3);
-      assert(j < n8);
-      if (i == j) {
-         // paper bug: original code probably swapped in place; if copying,
-         //            need to directly copy in this case
-         int i8 = i << 3;
-         v[i8+1] = u[i8+1];
-         v[i8+3] = u[i8+3];
-         v[i8+5] = u[i8+5];
-         v[i8+7] = u[i8+7];
-      } else if (i < j) {
-         int i8 = i << 3, j8 = j << 3;
-         v[j8+1] = u[i8+1], v[i8+1] = u[j8 + 1];
-         v[j8+3] = u[i8+3], v[i8+3] = u[j8 + 3];
-         v[j8+5] = u[i8+5], v[i8+5] = u[j8 + 5];
-         v[j8+7] = u[i8+7], v[i8+7] = u[j8 + 7];
-      }
-   }
-   // step 5
-   for (k=0; k < n2; ++k) {
-      w[k] = v[k*2+1];
-   }
-   // step 6
-   for (k=k2=k4=0; k < n8; ++k, k2 += 2, k4 += 4) {
-      u[n-1-k2] = w[k4];
-      u[n-2-k2] = w[k4+1];
-      u[n3_4 - 1 - k2] = w[k4+2];
-      u[n3_4 - 2 - k2] = w[k4+3];
-   }
-   // step 7
-   for (k=k2=0; k < n8; ++k, k2 += 2) {
-      v[n2 + k2 ] = ( u[n2 + k2] + u[n-2-k2] + C[k2+1]*(u[n2+k2]-u[n-2-k2]) + C[k2]*(u[n2+k2+1]+u[n-2-k2+1]))/2;
-      v[n-2 - k2] = ( u[n2 + k2] + u[n-2-k2] - C[k2+1]*(u[n2+k2]-u[n-2-k2]) - C[k2]*(u[n2+k2+1]+u[n-2-k2+1]))/2;
-      v[n2+1+ k2] = ( u[n2+1+k2] - u[n-1-k2] + C[k2+1]*(u[n2+1+k2]+u[n-1-k2]) - C[k2]*(u[n2+k2]-u[n-2-k2]))/2;
-      v[n-1 - k2] = (-u[n2+1+k2] + u[n-1-k2] + C[k2+1]*(u[n2+1+k2]+u[n-1-k2]) - C[k2]*(u[n2+k2]-u[n-2-k2]))/2;
-   }
-   // step 8
-   for (k=k2=0; k < n4; ++k,k2 += 2) {
-      X[k]      = v[k2+n2]*B[k2  ] + v[k2+1+n2]*B[k2+1];
-      X[n2-1-k] = v[k2+n2]*B[k2+1] - v[k2+1+n2]*B[k2  ];
-   }
-
-   // decode kernel to output
-   // determined the following value experimentally
-   // (by first figuring out what made inverse_mdct_slow work); then matching that here
-   // (probably vorbis encoder premultiplies by n or n/2, to save it on the decoder?)
-   s = 0.5; // theoretically would be n4
-
-   // [[[ note! the s value of 0.5 is compensated for by the B[] in the current code,
-   //     so it needs to use the "old" B values to behave correctly, or else
-   //     set s to 1.0 ]]]
-   for (i=0; i < n4  ; ++i) buffer[i] = s * X[i+n4];
-   for (   ; i < n3_4; ++i) buffer[i] = -s * X[n3_4 - i - 1];
-   for (   ; i < n   ; ++i) buffer[i] = -s * X[i - n3_4];
-}
-#endif
-
-static float *get_window(vorb *f, int len)
-{
-   len <<= 1;
-   if (len == f->blocksize_0) return f->window[0];
-   if (len == f->blocksize_1) return f->window[1];
-   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];
-            if (lx != hx)
-               draw_line(target, lx,ly, hx,hy, n2);
-            CHECK(f);
-            lx = hx, ly = hy;
-         }
-      }
-      if (lx < n2) {
-         // optimization of: draw_line(target, lx,ly, n,ly, n2);
-         for (j=lx; j < n2; ++j)
-            LINE_OP(target[j], inverse_db_table[ly]);
-         CHECK(f);
-      }
-   }
-   return TRUE;
-}
-
-// The meaning of "left" and "right"
-//
-// For a given frame:
-//     we compute samples from 0..n
-//     window_center is n/2
-//     we'll window and mix the samples from left_start to left_end with data from the previous frame
-//     all of the samples from left_end to right_start can be output without mixing; however,
-//        this interval is 0-length except when transitioning between short and long frames
-//     all of the samples from right_start to right_end need to be mixed with the next frame,
-//        which we don't have, so those get saved in a buffer
-//     frame N's right_end-right_start, the number of samples to mix with the next frame,
-//        has to be the same as frame N+1's left_end-left_start (which they are by
-//        construction)
-
-static int vorbis_decode_initial(vorb *f, int *p_left_start, int *p_left_end, int *p_right_start, int *p_right_end, int *mode)
-{
-   Mode *m;
-   int i, n, prev, next, window_center;
-   f->channel_buffer_start = f->channel_buffer_end = 0;
-
-  retry:
-   if (f->eof) return FALSE;
-   if (!maybe_start_packet(f))
-      return FALSE;
-   // check packet type
-   if (get_bits(f,1) != 0) {
-      if (IS_PUSH_MODE(f))
-         return error(f,VORBIS_bad_packet_type);
-      while (EOP != get8_packet(f));
-      goto retry;
-   }
-
-   if (f->alloc.alloc_buffer)
-      assert(f->alloc.alloc_buffer_length_in_bytes == f->temp_offset);
-
-   i = get_bits(f, ilog(f->mode_count-1));
-   if (i == EOP) return FALSE;
-   if (i >= f->mode_count) return FALSE;
-   *mode = i;
-   m = f->mode_config + i;
-   if (m->blockflag) {
-      n = f->blocksize_1;
-      prev = get_bits(f,1);
-      next = get_bits(f,1);
-   } else {
-      prev = next = 0;
-      n = f->blocksize_0;
-   }
-
-// WINDOWING
-
-   window_center = n >> 1;
-   if (m->blockflag && !prev) {
-      *p_left_start = (n - f->blocksize_0) >> 2;
-      *p_left_end   = (n + f->blocksize_0) >> 2;
-   } else {
-      *p_left_start = 0;
-      *p_left_end   = window_center;
-   }
-   if (m->blockflag && !next) {
-      *p_right_start = (n*3 - f->blocksize_0) >> 2;
-      *p_right_end   = (n*3 + f->blocksize_0) >> 2;
-   } else {
-      *p_right_start = window_center;
-      *p_right_end   = n;
-   }
-
-   return TRUE;
-}
-
-static int vorbis_decode_packet_rest(vorb *f, int *len, Mode *m, int left_start, int left_end, int right_start, int right_end, int *p_left)
-{
-   Mapping *map;
-   int i,j,k,n,n2;
-   int zero_channel[256];
-   int really_zero_channel[256];
-
-// WINDOWING
-
-   n = f->blocksize[m->blockflag];
-   map = &f->mapping[m->mapping];
-
-// FLOORS
-   n2 = n >> 1;
-
-   CHECK(f);
-
-   for (i=0; i < f->channels; ++i) {
-      int s = map->chan[i].mux, floor;
-      zero_channel[i] = FALSE;
-      floor = map->submap_floor[s];
-      if (f->floor_types[floor] == 0) {
-         return error(f, VORBIS_invalid_stream);
-      } else {
-         Floor1 *g = &f->floor_config[floor].floor1;
-         if (get_bits(f, 1)) {
-            short *finalY;
-            uint8 step2_flag[256];
-            static int range_list[4] = { 256, 128, 86, 64 };
-            int range = range_list[g->floor1_multiplier-1];
-            int offset = 2;
-            finalY = f->finalY[i];
-            finalY[0] = get_bits(f, ilog(range)-1);
-            finalY[1] = get_bits(f, ilog(range)-1);
-            for (j=0; j < g->partitions; ++j) {
-               int pclass = g->partition_class_list[j];
-               int cdim = g->class_dimensions[pclass];
-               int cbits = g->class_subclasses[pclass];
-               int csub = (1 << cbits)-1;
-               int cval = 0;
-               if (cbits) {
-                  Codebook *c = f->codebooks + g->class_masterbooks[pclass];
-                  DECODE(cval,f,c);
-               }
-               for (k=0; k < cdim; ++k) {
-                  int book = g->subclass_books[pclass][cval & csub];
-                  cval = cval >> cbits;
-                  if (book >= 0) {
-                     int temp;
-                     Codebook *c = f->codebooks + book;
-                     DECODE(temp,f,c);
-                     finalY[offset++] = temp;
-                  } else
-                     finalY[offset++] = 0;
-               }
-            }
-            if (f->valid_bits == INVALID_BITS) goto error; // behavior according to spec
-            step2_flag[0] = step2_flag[1] = 1;
-            for (j=2; j < g->values; ++j) {
-               int low, high, pred, highroom, lowroom, room, val;
-               low = g->neighbors[j][0];
-               high = g->neighbors[j][1];
-               //neighbors(g->Xlist, j, &low, &high);
-               pred = predict_point(g->Xlist[j], g->Xlist[low], g->Xlist[high], finalY[low], finalY[high]);
-               val = finalY[j];
-               highroom = range - pred;
-               lowroom = pred;
-               if (highroom < lowroom)
-                  room = highroom * 2;
-               else
-                  room = lowroom * 2;
-               if (val) {
-                  step2_flag[low] = step2_flag[high] = 1;
-                  step2_flag[j] = 1;
-                  if (val >= room)
-                     if (highroom > lowroom)
-                        finalY[j] = val - lowroom + pred;
-                     else
-                        finalY[j] = pred - val + highroom - 1;
-                  else
-                     if (val & 1)
-                        finalY[j] = pred - ((val+1)>>1);
-                     else
-                        finalY[j] = pred + (val>>1);
-               } else {
-                  step2_flag[j] = 0;
-                  finalY[j] = pred;
-               }
-            }
-
-#ifdef STB_VORBIS_NO_DEFER_FLOOR
-            do_floor(f, map, i, n, f->floor_buffers[i], finalY, step2_flag);
-#else
-            // defer final floor computation until _after_ residue
-            for (j=0; j < g->values; ++j) {
-               if (!step2_flag[j])
-                  finalY[j] = -1;
-            }
-#endif
-         } else {
-           error:
-            zero_channel[i] = TRUE;
-         }
-         // So we just defer everything else to later
-
-         // at this point we've decoded the floor into buffer
-      }
-   }
-   CHECK(f);
-   // 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;
-      }
-
-   CHECK(f);
-// 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);
-   CHECK(f);
-
-// INVERSE COUPLING
-   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;
-      }
-   }
-   CHECK(f);
-
-   // finish decoding the floors
-#ifndef STB_VORBIS_NO_DEFER_FLOOR
-   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
-   CHECK(f);
-   for (i=0; i < f->channels; ++i)
-      inverse_mdct(f->channel_buffers[i], n, f, m->blockflag);
-   CHECK(f);
-
-   // 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) {
-      if (f->discard_samples_deferred >= right_start - left_start) {
-         f->discard_samples_deferred -= (right_start - left_start);
-         left_start = right_start;
-         *p_left = left_start;
-      } else {
-         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-left_start)) {
-            if (current_end < f->current_loc) {
-               // negative truncation, that's impossible!
-               *len = 0;
-            } else {
-               *len = current_end - f->current_loc;
-            }
-            *len += left_start;
-            if (*len > right_end) *len = right_end; // this should never happen
-            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
-   CHECK(f);
-
-   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;
-      CHECK(f);
-      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->dimensions == 0 && c->entries != 0)    return error(f, VORBIS_invalid_setup);
-
-      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;
-               if (lengths[j] == 32)
-                  return error(f, VORBIS_invalid_setup);
-            } 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);
-         if (c->codeword_lengths == NULL) return error(f, VORBIS_outofmem);
-         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;
-
-      CHECK(f);
-      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));
-         if (c->sorted_codewords == NULL) return error(f, VORBIS_outofmem);
-         // 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 == NULL) return error(f, VORBIS_outofmem);
-         ++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);
-
-      CHECK(f);
-      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;
-         }
-         if (c->lookup_values == 0) return error(f, VORBIS_invalid_setup);
-         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;
-            float last=0;
-            // 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) {
-               unsigned int z = sparse ? c->sorted_values[j] : j;
-               unsigned int div=1;
-               for (k=0; k < c->dimensions; ++k) {
-                  int off = (z / div) % c->lookup_values;
-                  float val = mults[off];
-                  val = mults[off]*c->delta_value + c->minimum_value + last;
-                  c->multiplicands[j*c->dimensions + k] = val;
-                  if (c->sequence_p)
-                     last = val;
-                  if (k+1 < c->dimensions) {
-                     if (div > UINT_MAX / (unsigned int) c->lookup_values) {
-                        setup_temp_free(f, mults,sizeof(mults[0])*c->lookup_values);
-                        return error(f, VORBIS_invalid_setup);
-                     }
-                     div *= c->lookup_values;
-                  }
-               }
-            }
-            c->lookup_type = 2;
-         }
-         else
-#endif
-         {
-            float last=0;
-            CHECK(f);
-            c->multiplicands = (codetype *) setup_malloc(f, sizeof(c->multiplicands[0]) * c->lookup_values);
-            if (c->multiplicands == NULL) { setup_temp_free(f, mults,sizeof(mults[0])*c->lookup_values); return error(f, VORBIS_outofmem); }
-            for (j=0; j < (int) c->lookup_values; ++j) {
-               float val = mults[j] * c->delta_value + c->minimum_value + last;
-               c->multiplicands[j] = val;
-               if (c->sequence_p)
-                  last = val;
-            }
-         }
-#ifndef STB_VORBIS_DIVIDES_IN_CODEBOOK
-        skip:;
-#endif
-         setup_temp_free(f, mults, sizeof(mults[0])*c->lookup_values);
-
-         CHECK(f);
-      }
-      CHECK(f);
-   }
-
-   // 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));
-   if (f->floor_config == NULL) return error(f, VORBIS_outofmem);
-   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[0]));
-   if (f->residue_config == NULL) return error(f, VORBIS_outofmem);
-   memset(f->residue_config, 0, f->residue_count * sizeof(f->residue_config[0]));
-   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);
-      if (r->end < r->begin) return error(f, VORBIS_invalid_setup);
-      r->part_size = get_bits(f,24)+1;
-      r->classifications = get_bits(f,6)+1;
-      r->classbook = get_bits(f,8);
-      if (r->classbook >= f->codebook_count) return error(f, VORBIS_invalid_setup);
-      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);
-      if (r->residue_books == NULL) return error(f, VORBIS_outofmem);
-      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);
-         if (r->classdata[j] == NULL) return error(f, VORBIS_outofmem);
-         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));
-   if (f->mapping == NULL) return error(f, VORBIS_outofmem);
-   memset(f->mapping, 0, 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 (m->chan == NULL) return error(f, VORBIS_outofmem);
-      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);
-      if (f->channel_buffers[i] == NULL || f->previous_window[i] == NULL || f->finalY[i] == NULL) return error(f, VORBIS_outofmem);
-      #ifdef STB_VORBIS_NO_DEFER_FLOOR
-      f->floor_buffers[i]   = (float *) setup_malloc(f, sizeof(float) * f->blocksize_1/2);
-      if (f->floor_buffers[i] == NULL) return error(f, VORBIS_outofmem);
-      #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;
-   if (p->residue_config) {
-      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) {
-      CHECK(p);
-      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);
-   if (p->mapping) {
-      for (i=0; i < p->mapping_count; ++i)
-         setup_free(p, p->mapping[i].chan);
-      setup_free(p, p->mapping);
-   }
-   CHECK(p);
-   for (i=0; i < p->channels && i < STB_VORBIS_MAX_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, const 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
-         const 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, (uint8 *) data, data_len);
-   }
-
-   f->stream     = (uint8 *) data;
-   f->stream_end = (uint8 *) 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 (int) (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 (int) (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 (int) (f->stream - data);
-}
-
-stb_vorbis *stb_vorbis_open_pushdata(
-         const unsigned char *data, int data_len, // the memory available for decoding
-         int *data_used,              // only defined if result is not NULL
-         int *error, const stb_vorbis_alloc *alloc)
-{
-   stb_vorbis *f, p;
-   vorbis_init(&p, alloc);
-   p.stream     = (uint8 *) data;
-   p.stream_end = (uint8 *) 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 = (int) (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 (unsigned int) (f->stream - f->stream_start);
-   #ifndef STB_VORBIS_NO_STDIO
-   return (unsigned int) (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 candidate
-         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);
-      }
-   }
-}
-
-
-#define SAMPLE_unknown  0xffffffff
-
-// seeking is implemented with a binary search, which narrows down the range to
-// 64K, before using a linear search (because finding the synchronization
-// pattern can be expensive, and the chance we'd find the end page again is
-// relatively high for small ranges)
-//
-// two initial interpolation-style probes are used at the start of the search
-// to try to bound either side of the binary search sensibly, while still
-// working in O(log n) time if they fail.
-
-static int get_seek_page_info(stb_vorbis *f, ProbedPage *z)
-{
-   uint8 header[27], lacing[255];
-   int i,len;
-
-   // record where the page starts
-   z->page_start = stb_vorbis_get_file_offset(f);
-
-   // parse the header
-   getn(f, header, 27);
-   if (header[0] != 'O' || header[1] != 'g' || header[2] != 'g' || header[3] != 'S')
-      return 0;
-   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] << 24);
-
-   // restore file state to where we were
-   set_file_offset(f, z->page_start);
-   return 1;
-}
-
-// rarely used function to seek back to the preceeding page while finding the
-// start of a packet
-static int go_to_page_before(stb_vorbis *f, unsigned int limit_offset)
-{
-   unsigned int previous_safe, end;
-
-   // now we want to seek back 64K from the limit
-   if (limit_offset >= 65536 && limit_offset-65536 >= f->first_audio_page_offset)
-      previous_safe = limit_offset - 65536;
-   else
-      previous_safe = f->first_audio_page_offset;
-
-   set_file_offset(f, previous_safe);
-
-   while (vorbis_find_page(f, &end, NULL)) {
-      if (end >= limit_offset && stb_vorbis_get_file_offset(f) < limit_offset)
-         return 1;
-      set_file_offset(f, end);
-   }
-
-   return 0;
-}
-
-// implements the search logic for finding a page and starting decoding. if
-// the function succeeds, current_loc_valid will be true and current_loc will
-// be less than or equal to the provided sample number (the closer the
-// better).
-static int seek_to_sample_coarse(stb_vorbis *f, uint32 sample_number)
-{
-   ProbedPage left, right, mid;
-   int i, start_seg_with_known_loc, end_pos, page_start;
-   uint32 delta, stream_length, padding;
-   double offset, bytes_per_sample;
-   int probe = 0;
-
-   // find the last page and validate the target sample
-   stream_length = stb_vorbis_stream_length_in_samples(f);
-   if (stream_length == 0)            return error(f, VORBIS_seek_without_length);
-   if (sample_number > stream_length) return error(f, VORBIS_seek_invalid);
-
-   // this is the maximum difference between the window-center (which is the
-   // actual granule position value), and the right-start (which the spec
-   // indicates should be the granule position (give or take one)).
-   padding = ((f->blocksize_1 - f->blocksize_0) >> 2);
-   if (sample_number < padding)
-      sample_number = 0;
-   else
-      sample_number -= padding;
-
-   left = f->p_first;
-   while (left.last_decoded_sample == ~0U) {
-      // (untested) the first page does not have a 'last_decoded_sample'
-      set_file_offset(f, left.page_end);
-      if (!get_seek_page_info(f, &left)) goto error;
-   }
-
-   right = f->p_last;
-   assert(right.last_decoded_sample != ~0U);
-
-   // starting from the start is handled differently
-   if (sample_number <= left.last_decoded_sample) {
-      stb_vorbis_seek_start(f);
-      return 1;
-   }
-
-   while (left.page_end != right.page_start) {
-      assert(left.page_end < right.page_start);
-      // search range in bytes
-      delta = right.page_start - left.page_end;
-      if (delta <= 65536) {
-         // there's only 64K left to search - handle it linearly
-         set_file_offset(f, left.page_end);
-      } else {
-         if (probe < 2) {
-            if (probe == 0) {
-               // first probe (interpolate)
-               double data_bytes = right.page_end - left.page_start;
-               bytes_per_sample = data_bytes / right.last_decoded_sample;
-               offset = left.page_start + bytes_per_sample * (sample_number - left.last_decoded_sample);
-            } else {
-               // second probe (try to bound the other side)
-               double error = ((double) sample_number - mid.last_decoded_sample) * bytes_per_sample;
-               if (error >= 0 && error <  8000) error =  8000;
-               if (error <  0 && error > -8000) error = -8000;
-               offset += error * 2;
-            }
-
-            // ensure the offset is valid
-            if (offset < left.page_end)
-               offset = left.page_end;
-            if (offset > right.page_start - 65536)
-               offset = right.page_start - 65536;
-
-            set_file_offset(f, (unsigned int) offset);
-         } else {
-            // binary search for large ranges (offset by 32K to ensure
-            // we don't hit the right page)
-            set_file_offset(f, left.page_end + (delta / 2) - 32768);
-         }
-
-         if (!vorbis_find_page(f, NULL, NULL)) goto error;
-      }
-
-      for (;;) {
-         if (!get_seek_page_info(f, &mid)) goto error;
-         if (mid.last_decoded_sample != ~0U) break;
-         // (untested) no frames end on this page
-         set_file_offset(f, mid.page_end);
-         assert(mid.page_start < right.page_start);
-      }
-
-      // if we've just found the last page again then we're in a tricky file,
-      // and we're close enough.
-      if (mid.page_start == right.page_start)
-         break;
-
-      if (sample_number < mid.last_decoded_sample)
-         right = mid;
-      else
-         left = mid;
-
-      ++probe;
-   }
-
-   // seek back to start of the last packet
-   page_start = left.page_start;
-   set_file_offset(f, page_start);
-   if (!start_page(f)) return error(f, VORBIS_seek_failed);
-   end_pos = f->end_seg_with_known_loc;
-   assert(end_pos >= 0);
-
-   for (;;) {
-      for (i = end_pos; i > 0; --i)
-         if (f->segments[i-1] != 255)
-            break;
-
-      start_seg_with_known_loc = i;
-
-      if (start_seg_with_known_loc > 0 || !(f->page_flag & PAGEFLAG_continued_packet))
-         break;
-
-      // (untested) the final packet begins on an earlier page
-      if (!go_to_page_before(f, page_start))
-         goto error;
-
-      page_start = stb_vorbis_get_file_offset(f);
-      if (!start_page(f)) goto error;
-      end_pos = f->segment_count - 1;
-   }
-
-   // prepare to start decoding
-   f->current_loc_valid = FALSE;
-   f->last_seg = FALSE;
-   f->valid_bits = 0;
-   f->packet_bytes = 0;
-   f->bytes_in_seg = 0;
-   f->previous_length = 0;
-   f->next_seg = start_seg_with_known_loc;
-
-   for (i = 0; i < start_seg_with_known_loc; i++)
-      skip(f, f->segments[i]);
-
-   // start decoding (optimizable - this frame is generally discarded)
-   vorbis_pump_first_frame(f);
-   return 1;
-
-error:
-   // try to restore the file to a valid state
-   stb_vorbis_seek_start(f);
-   return error(f, VORBIS_seek_failed);
-}
-
-// the same as vorbis_decode_initial, but without advancing
-static int peek_decode_initial(vorb *f, int *p_left_start, int *p_left_end, int *p_right_start, int *p_right_end, int *mode)
-{
-   int bits_read, bytes_read;
-
-   if (!vorbis_decode_initial(f, p_left_start, p_left_end, p_right_start, p_right_end, mode))
-      return 0;
-
-   // either 1 or 2 bytes were read, figure out which so we can rewind
-   bits_read = 1 + ilog(f->mode_count-1);
-   if (f->mode_config[*mode].blockflag)
-      bits_read += 2;
-   bytes_read = (bits_read + 7) / 8;
-
-   f->bytes_in_seg += bytes_read;
-   f->packet_bytes -= bytes_read;
-   skip(f, -bytes_read);
-   if (f->next_seg == -1)
-      f->next_seg = f->segment_count - 1;
-   else
-      f->next_seg--;
-   f->valid_bits = 0;
-
-   return 1;
-}
-
-int stb_vorbis_seek_frame(stb_vorbis *f, unsigned int sample_number)
-{
-   uint32 max_frame_samples;
-
-   if (IS_PUSH_MODE(f)) return error(f, VORBIS_invalid_api_mixing);
-
-   // fast page-level search
-   if (!seek_to_sample_coarse(f, sample_number))
-      return 0;
-
-   assert(f->current_loc_valid);
-   assert(f->current_loc <= sample_number);
-
-   // linear search for the relevant packet
-   max_frame_samples = (f->blocksize_1*3 - f->blocksize_0) >> 2;
-   while (f->current_loc < sample_number) {
-      int left_start, left_end, right_start, right_end, mode, frame_samples;
-      if (!peek_decode_initial(f, &left_start, &left_end, &right_start, &right_end, &mode))
-         return error(f, VORBIS_seek_failed);
-      // calculate the number of samples returned by the next frame
-      frame_samples = right_start - left_start;
-      if (f->current_loc + frame_samples > sample_number) {
-         return 1; // the next frame will contain the sample
-      } else if (f->current_loc + frame_samples + max_frame_samples > sample_number) {
-         // there's a chance the frame after this could contain the sample
-         vorbis_pump_first_frame(f);
-      } else {
-         // this frame is too early to be relevant
-         f->current_loc += frame_samples;
-         f->previous_length = 0;
-         maybe_start_packet(f);
-         flush_packet(f);
-      }
-   }
-   // the next frame will start with the sample
-   assert(f->current_loc == sample_number);
-   return 1;
-}
-
-int stb_vorbis_seek(stb_vorbis *f, unsigned int sample_number)
-{
-   if (!stb_vorbis_seek_frame(f, sample_number))
-      return 0;
-
-   if (sample_number != f->current_loc) {
-      int n;
-      uint32 frame_start = f->current_loc;
-      stb_vorbis_get_frame_float(f, &n, NULL);
-      assert(sample_number > frame_start);
-      assert(f->channel_buffer_start + (int) (sample_number-frame_start) <= f->channel_buffer_end);
-      f->channel_buffer_start += (sample_number - frame_start);
-   }
-
-   return 1;
-}
-
-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;
-
-     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, const stb_vorbis_alloc *alloc, unsigned int length)
-{
-   stb_vorbis *f, p;
-   vorbis_init(&p, alloc);
-   p.f = file;
-   p.f_start = (uint32) 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, const stb_vorbis_alloc *alloc)
-{
-   unsigned int len, start;
-   start = (unsigned int) ftell(file);
-   fseek(file, 0, SEEK_END);
-   len = (unsigned int) (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, const 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, const 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
-    1.09    - 2016/04/04 - back out 'avoid discarding last frame' fix from previous version
-    1.08    - 2016/04/02 - fixed multiple warnings; fix setup memory leaks;
-                           avoid discarding last frame of audio data
-    1.07    - 2015/01/16 - fixed some warnings, fix mingw, const-correct API
-                           some more crash fixes when out of memory or with corrupt files
-    1.06    - 2015/08/31 - full, correct support for seeking API (Dougall Johnson)
-                           some crash fixes when out of memory or with corrupt files
-    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

@@ -1,4 +1,4 @@
-// Ogg Vorbis audio decoder - v1.09 - public domain
+// Ogg Vorbis audio decoder - v1.10 - public domain
 // http://nothings.org/stb_vorbis/
 //
 // Original version written by Sean Barrett in 2007.
@@ -9,12 +9,7 @@
 //
 // LICENSE
 //
-//   This software is dual-licensed to the public domain and under the following
-//   license: you are granted a perpetual, irrevocable license to copy, modify,
-//   publish, and distribute this file as you see fit.
-//
-// 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.
+//   See end of file for license information.
 //
 // Limitations:
 //
@@ -37,6 +32,7 @@
 //    manxorist@github   saga musix
 //
 // Partial history:
+//    1.10    - 2017/03/03 - more robust seeking; fix negative ilog(); clear error in open_memory
 //    1.09    - 2016/04/04 - back out 'truncation of last frame' fix from previous version
 //    1.08    - 2016/04/02 - warnings; setup memory leaks; truncation of last frame
 //    1.07    - 2015/01/16 - fixes for crashes on invalid files; warning fixes; const
@@ -275,7 +271,7 @@ extern int stb_vorbis_seek(stb_vorbis *f, unsigned int sample_number);
 // do not need to seek to EXACTLY the target sample when using get_samples_*,
 // you can also use seek_frame().
 
-extern void stb_vorbis_seek_start(stb_vorbis *f);
+extern int stb_vorbis_seek_start(stb_vorbis *f);
 // this function is equivalent to stb_vorbis_seek(f,0)
 
 extern unsigned int stb_vorbis_stream_length_in_samples(stb_vorbis *f);
@@ -386,3 +382,5066 @@ enum STBVorbisError
 //  HEADER ENDS HERE
 //
 //////////////////////////////////////////////////////////////////////////////
+
+#ifndef STB_VORBIS_HEADER_ONLY
+
+// global configuration settings (e.g. set these in the project/makefile),
+// or just set them in this file at the top (although ideally the first few
+// should be visible when the header file is compiled too, although it's not
+// crucial)
+
+// STB_VORBIS_NO_PUSHDATA_API
+//     does not compile the code for the various stb_vorbis_*_pushdata()
+//     functions
+// #define STB_VORBIS_NO_PUSHDATA_API
+
+// STB_VORBIS_NO_PULLDATA_API
+//     does not compile the code for the non-pushdata APIs
+// #define STB_VORBIS_NO_PULLDATA_API
+
+// STB_VORBIS_NO_STDIO
+//     does not compile the code for the APIs that use FILE *s internally
+//     or externally (implied by STB_VORBIS_NO_PULLDATA_API)
+// #define STB_VORBIS_NO_STDIO
+
+// STB_VORBIS_NO_INTEGER_CONVERSION
+//     does not compile the code for converting audio sample data from
+//     float to integer (implied by STB_VORBIS_NO_PULLDATA_API)
+// #define STB_VORBIS_NO_INTEGER_CONVERSION
+
+// STB_VORBIS_NO_FAST_SCALED_FLOAT
+//      does not use a fast float-to-int trick to accelerate float-to-int on
+//      most platforms which requires endianness be defined correctly.
+//#define STB_VORBIS_NO_FAST_SCALED_FLOAT
+
+
+// STB_VORBIS_MAX_CHANNELS [number]
+//     globally define this to the maximum number of channels you need.
+//     The spec does not put a restriction on channels except that
+//     the count is stored in a byte, so 255 is the hard limit.
+//     Reducing this saves about 16 bytes per value, so using 16 saves
+//     (255-16)*16 or around 4KB. Plus anything other memory usage
+//     I forgot to account for. Can probably go as low as 8 (7.1 audio),
+//     6 (5.1 audio), or 2 (stereo only).
+#ifndef STB_VORBIS_MAX_CHANNELS
+#define STB_VORBIS_MAX_CHANNELS    16  // enough for anyone?
+#endif
+
+// STB_VORBIS_PUSHDATA_CRC_COUNT [number]
+//     after a flush_pushdata(), stb_vorbis begins scanning for the
+//     next valid page, without backtracking. when it finds something
+//     that looks like a page, it streams through it and verifies its
+//     CRC32. Should that validation fail, it keeps scanning. But it's
+//     possible that _while_ streaming through to check the CRC32 of
+//     one candidate page, it sees another candidate page. This #define
+//     determines how many "overlapping" candidate pages it can search
+//     at once. Note that "real" pages are typically ~4KB to ~8KB, whereas
+//     garbage pages could be as big as 64KB, but probably average ~16KB.
+//     So don't hose ourselves by scanning an apparent 64KB page and
+//     missing a ton of real ones in the interim; so minimum of 2
+#ifndef STB_VORBIS_PUSHDATA_CRC_COUNT
+#define STB_VORBIS_PUSHDATA_CRC_COUNT  4
+#endif
+
+// STB_VORBIS_FAST_HUFFMAN_LENGTH [number]
+//     sets the log size of the huffman-acceleration table.  Maximum
+//     supported value is 24. with larger numbers, more decodings are O(1),
+//     but the table size is larger so worse cache missing, so you'll have
+//     to probe (and try multiple ogg vorbis files) to find the sweet spot.
+#ifndef STB_VORBIS_FAST_HUFFMAN_LENGTH
+#define STB_VORBIS_FAST_HUFFMAN_LENGTH   10
+#endif
+
+// STB_VORBIS_FAST_BINARY_LENGTH [number]
+//     sets the log size of the binary-search acceleration table. this
+//     is used in similar fashion to the fast-huffman size to set initial
+//     parameters for the binary search
+
+// STB_VORBIS_FAST_HUFFMAN_INT
+//     The fast huffman tables are much more efficient if they can be
+//     stored as 16-bit results instead of 32-bit results. This restricts
+//     the codebooks to having only 65535 possible outcomes, though.
+//     (At least, accelerated by the huffman table.)
+#ifndef STB_VORBIS_FAST_HUFFMAN_INT
+#define STB_VORBIS_FAST_HUFFMAN_SHORT
+#endif
+
+// STB_VORBIS_NO_HUFFMAN_BINARY_SEARCH
+//     If the 'fast huffman' search doesn't succeed, then stb_vorbis falls
+//     back on binary searching for the correct one. This requires storing
+//     extra tables with the huffman codes in sorted order. Defining this
+//     symbol trades off space for speed by forcing a linear search in the
+//     non-fast case, except for "sparse" codebooks.
+// #define STB_VORBIS_NO_HUFFMAN_BINARY_SEARCH
+
+// STB_VORBIS_DIVIDES_IN_RESIDUE
+//     stb_vorbis precomputes the result of the scalar residue decoding
+//     that would otherwise require a divide per chunk. you can trade off
+//     space for time by defining this symbol.
+// #define STB_VORBIS_DIVIDES_IN_RESIDUE
+
+// STB_VORBIS_DIVIDES_IN_CODEBOOK
+//     vorbis VQ codebooks can be encoded two ways: with every case explicitly
+//     stored, or with all elements being chosen from a small range of values,
+//     and all values possible in all elements. By default, stb_vorbis expands
+//     this latter kind out to look like the former kind for ease of decoding,
+//     because otherwise an integer divide-per-vector-element is required to
+//     unpack the index. If you define STB_VORBIS_DIVIDES_IN_CODEBOOK, you can
+//     trade off storage for speed.
+//#define STB_VORBIS_DIVIDES_IN_CODEBOOK
+
+#ifdef STB_VORBIS_CODEBOOK_SHORTS
+#error "STB_VORBIS_CODEBOOK_SHORTS is no longer supported as it produced incorrect results for some input formats"
+#endif
+
+// STB_VORBIS_DIVIDE_TABLE
+//     this replaces small integer divides in the floor decode loop with
+//     table lookups. made less than 1% difference, so disabled by default.
+
+// STB_VORBIS_NO_INLINE_DECODE
+//     disables the inlining of the scalar codebook fast-huffman decode.
+//     might save a little codespace; useful for debugging
+// #define STB_VORBIS_NO_INLINE_DECODE
+
+// STB_VORBIS_NO_DEFER_FLOOR
+//     Normally we only decode the floor without synthesizing the actual
+//     full curve. We can instead synthesize the curve immediately. This
+//     requires more memory and is very likely slower, so I don't think
+//     you'd ever want to do it except for debugging.
+// #define STB_VORBIS_NO_DEFER_FLOOR
+
+
+
+
+//////////////////////////////////////////////////////////////////////////////
+
+#ifdef STB_VORBIS_NO_PULLDATA_API
+   #define STB_VORBIS_NO_INTEGER_CONVERSION
+   #define STB_VORBIS_NO_STDIO
+#endif
+
+#if defined(STB_VORBIS_NO_CRT) && !defined(STB_VORBIS_NO_STDIO)
+   #define STB_VORBIS_NO_STDIO 1
+#endif
+
+#ifndef STB_VORBIS_NO_INTEGER_CONVERSION
+#ifndef STB_VORBIS_NO_FAST_SCALED_FLOAT
+
+   // only need endianness for fast-float-to-int, which we don't
+   // use for pushdata
+
+   #ifndef STB_VORBIS_BIG_ENDIAN
+     #define STB_VORBIS_ENDIAN  0
+   #else
+     #define STB_VORBIS_ENDIAN  1
+   #endif
+
+#endif
+#endif
+
+
+#ifndef STB_VORBIS_NO_STDIO
+#include <stdio.h>
+#endif
+
+#ifndef STB_VORBIS_NO_CRT
+   #include <stdlib.h>
+   #include <string.h>
+   #include <assert.h>
+   #include <math.h>
+
+   // find definition of alloca if it's not in stdlib.h:
+   #ifdef _MSC_VER
+      #include <malloc.h>
+   #endif
+   #if defined(__linux__) || defined(__linux) || defined(__EMSCRIPTEN__)
+      #include <alloca.h>
+   #endif
+#else // STB_VORBIS_NO_CRT
+   #define NULL 0
+   #define malloc(s)   0
+   #define free(s)     ((void) 0)
+   #define realloc(s)  0
+#endif // STB_VORBIS_NO_CRT
+
+#include <limits.h>
+
+#ifdef __MINGW32__
+   // eff you mingw:
+   //     "fixed":
+   //         http://sourceforge.net/p/mingw-w64/mailman/message/32882927/
+   //     "no that broke the build, reverted, who cares about C":
+   //         http://sourceforge.net/p/mingw-w64/mailman/message/32890381/
+   #ifdef __forceinline
+   #undef __forceinline
+   #endif
+   #define __forceinline
+#elif !defined(_MSC_VER)
+   #if __GNUC__
+      #define __forceinline inline
+   #else
+      #define __forceinline
+   #endif
+#endif
+
+#if STB_VORBIS_MAX_CHANNELS > 256
+#error "Value of STB_VORBIS_MAX_CHANNELS outside of allowed range"
+#endif
+
+#if STB_VORBIS_FAST_HUFFMAN_LENGTH > 24
+#error "Value of STB_VORBIS_FAST_HUFFMAN_LENGTH outside of allowed range"
+#endif
+
+
+#if 0
+#include <crtdbg.h>
+#define CHECK(f)   _CrtIsValidHeapPointer(f->channel_buffers[1])
+#else
+#define CHECK(f)   ((void) 0)
+#endif
+
+#define MAX_BLOCKSIZE_LOG  13   // from specification
+#define MAX_BLOCKSIZE      (1 << MAX_BLOCKSIZE_LOG)
+
+
+typedef unsigned char  uint8;
+typedef   signed char   int8;
+typedef unsigned short uint16;
+typedef   signed short  int16;
+typedef unsigned int   uint32;
+typedef   signed int    int32;
+
+#ifndef TRUE
+#define TRUE 1
+#define FALSE 0
+#endif
+
+typedef float codetype;
+
+// @NOTE
+//
+// Some arrays below are tagged "//varies", which means it's actually
+// a variable-sized piece of data, but rather than malloc I assume it's
+// small enough it's better to just allocate it all together with the
+// main thing
+//
+// Most of the variables are specified with the smallest size I could pack
+// them into. It might give better performance to make them all full-sized
+// integers. It should be safe to freely rearrange the structures or change
+// the sizes larger--nothing relies on silently truncating etc., nor the
+// order of variables.
+
+#define FAST_HUFFMAN_TABLE_SIZE   (1 << STB_VORBIS_FAST_HUFFMAN_LENGTH)
+#define FAST_HUFFMAN_TABLE_MASK   (FAST_HUFFMAN_TABLE_SIZE - 1)
+
+typedef struct
+{
+   int dimensions, entries;
+   uint8 *codeword_lengths;
+   float  minimum_value;
+   float  delta_value;
+   uint8  value_bits;
+   uint8  lookup_type;
+   uint8  sequence_p;
+   uint8  sparse;
+   uint32 lookup_values;
+   codetype *multiplicands;
+   uint32 *codewords;
+   #ifdef STB_VORBIS_FAST_HUFFMAN_SHORT
+    int16  fast_huffman[FAST_HUFFMAN_TABLE_SIZE];
+   #else
+    int32  fast_huffman[FAST_HUFFMAN_TABLE_SIZE];
+   #endif
+   uint32 *sorted_codewords;
+   int    *sorted_values;
+   int     sorted_entries;
+} Codebook;
+
+typedef struct
+{
+   uint8 order;
+   uint16 rate;
+   uint16 bark_map_size;
+   uint8 amplitude_bits;
+   uint8 amplitude_offset;
+   uint8 number_of_books;
+   uint8 book_list[16]; // varies
+} Floor0;
+
+typedef struct
+{
+   uint8 partitions;
+   uint8 partition_class_list[32]; // varies
+   uint8 class_dimensions[16]; // varies
+   uint8 class_subclasses[16]; // varies
+   uint8 class_masterbooks[16]; // varies
+   int16 subclass_books[16][8]; // varies
+   uint16 Xlist[31*8+2]; // varies
+   uint8 sorted_order[31*8+2];
+   uint8 neighbors[31*8+2][2];
+   uint8 floor1_multiplier;
+   uint8 rangebits;
+   int values;
+} Floor1;
+
+typedef union
+{
+   Floor0 floor0;
+   Floor1 floor1;
+} Floor;
+
+typedef struct
+{
+   uint32 begin, end;
+   uint32 part_size;
+   uint8 classifications;
+   uint8 classbook;
+   uint8 **classdata;
+   int16 (*residue_books)[8];
+} Residue;
+
+typedef struct
+{
+   uint8 magnitude;
+   uint8 angle;
+   uint8 mux;
+} MappingChannel;
+
+typedef struct
+{
+   uint16 coupling_steps;
+   MappingChannel *chan;
+   uint8  submaps;
+   uint8  submap_floor[15]; // varies
+   uint8  submap_residue[15]; // varies
+} Mapping;
+
+typedef struct
+{
+   uint8 blockflag;
+   uint8 mapping;
+   uint16 windowtype;
+   uint16 transformtype;
+} Mode;
+
+typedef struct
+{
+   uint32  goal_crc;    // expected crc if match
+   int     bytes_left;  // bytes left in packet
+   uint32  crc_so_far;  // running crc
+   int     bytes_done;  // bytes processed in _current_ chunk
+   uint32  sample_loc;  // granule pos encoded in page
+} CRCscan;
+
+typedef struct
+{
+   uint32 page_start, page_end;
+   uint32 last_decoded_sample;
+} ProbedPage;
+
+struct stb_vorbis
+{
+  // user-accessible info
+   unsigned int sample_rate;
+   int channels;
+
+   unsigned int setup_memory_required;
+   unsigned int temp_memory_required;
+   unsigned int setup_temp_memory_required;
+
+  // 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;
+};
+
+#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 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=(uint32) i << 24, j=0; j < 8; ++j)
+         s = (s << 1) ^ (s >= (1U<<31) ? CRC32_POLY : 0);
+      crc_table[i] = s;
+   }
+}
+
+static __forceinline uint32 crc32_update(uint32 crc, uint8 byte)
+{
+   return (crc << 8) ^ crc_table[byte ^ (crc >> 24)];
+}
+
+
+// used in setup, and for huffman that doesn't go fast path
+static unsigned int bit_reverse(unsigned int n)
+{
+  n = ((n & 0xAAAAAAAA) >>  1) | ((n & 0x55555555) << 1);
+  n = ((n & 0xCCCCCCCC) >>  2) | ((n & 0x33333333) << 2);
+  n = ((n & 0xF0F0F0F0) >>  4) | ((n & 0x0F0F0F0F) << 4);
+  n = ((n & 0xFF00FF00) >>  8) | ((n & 0x00FF00FF) << 8);
+  return (n >> 16) | (n << 16);
+}
+
+static float square(float x)
+{
+   return x*x;
+}
+
+// this is a weird definition of log2() for which log2(1) = 1, log2(2) = 2, log2(4) = 3
+// as required by the specification. fast(?) implementation from stb.h
+// @OPTIMIZE: called multiple times per-packet with "constants"; move to setup
+static int ilog(int32 n)
+{
+   static signed char log2_4[16] = { 0,1,2,2,3,3,3,3,4,4,4,4,4,4,4,4 };
+
+   if (n < 0) return 0; // signed n returns 0
+
+   // 2 compares if n < 16, 3 compares otherwise (4 if signed or n > 1<<29)
+   if (n < (1 << 14))
+        if (n < (1 <<  4))            return  0 + log2_4[n      ];
+        else if (n < (1 <<  9))       return  5 + log2_4[n >>  5];
+             else                     return 10 + log2_4[n >> 10];
+   else if (n < (1 << 24))
+             if (n < (1 << 19))       return 15 + log2_4[n >> 15];
+             else                     return 20 + log2_4[n >> 20];
+        else if (n < (1 << 29))       return 25 + log2_4[n >> 25];
+             else                     return 30 + log2_4[n >> 30];
+}
+
+#ifndef M_PI
+  #define M_PI  3.14159265358979323846264f  // from CRC
+#endif
+
+// code length assigned to a value with no huffman encoding
+#define NO_CODE   255
+
+/////////////////////// LEAF SETUP FUNCTIONS //////////////////////////
+//
+// these functions are only called at setup, and only a few times
+// per file
+
+static float float32_unpack(uint32 x)
+{
+   // from the specification
+   uint32 mantissa = x & 0x1fffff;
+   uint32 sign = x & 0x80000000;
+   uint32 exp = (x & 0x7fe00000) >> 21;
+   double res = sign ? -(double)mantissa : (double)mantissa;
+   return (float) ldexp((float)res, 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] = 1U << (32-i);
+   // note that the above code treats the first case specially,
+   // but it's really the same as the following code, so they
+   // could probably be combined (except the initial code is 0,
+   // and I use 0 in available[] to mean 'empty')
+   for (i=k+1; i < n; ++i) {
+      uint32 res;
+      int z = len[i], y;
+      if (z == NO_CODE) continue;
+      // find lowest available leaf (should always be earliest,
+      // which is what the specification calls for)
+      // note that this property, and the fact we can never have
+      // more than one free leaf at a given level, isn't totally
+      // trivial to prove, but it seems true and the assert never
+      // fires, so!
+      while (z > 0 && !available[z]) --z;
+      if (z == 0) { return FALSE; }
+      res = available[z];
+      assert(z >= 0 && z < 32);
+      available[z] = 0;
+      add_entry(c, bit_reverse(res), i, m++, len[i], values);
+      // propogate availability up the tree
+      if (z != len[i]) {
+         assert(len[i] >= 0 && len[i] < 32);
+         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,id;
+} stbv__floor_ordering;
+
+static int STBV_CDECL point_compare(const void *p, const void *q)
+{
+   stbv__floor_ordering *a = (stbv__floor_ordering *) p;
+   stbv__floor_ordering *b = (stbv__floor_ordering *) q;
+   return a->x < b->x ? -1 : a->x > b->x;
+}
+
+//
+/////////////////////// END LEAF SETUP FUNCTIONS //////////////////////////
+
+
+#if defined(STB_VORBIS_NO_STDIO)
+   #define USE_MEMORY(z)    TRUE
+#else
+   #define USE_MEMORY(z)    ((z)->stream)
+#endif
+
+static uint8 get8(vorb *z)
+{
+   if (USE_MEMORY(z)) {
+      if (z->stream >= z->stream_end) { z->eof = TRUE; return 0; }
+      return *z->stream++;
+   }
+
+   #ifndef STB_VORBIS_NO_STDIO
+   {
+   int c = fgetc(z->f);
+   if (c == EOF) { z->eof = TRUE; return 0; }
+   return c;
+   }
+   #endif
+}
+
+static uint32 get32(vorb *f)
+{
+   uint32 x;
+   x = get8(f);
+   x += get8(f) << 8;
+   x += get8(f) << 16;
+   x += (uint32) get8(f) << 24;
+   return x;
+}
+
+static int getn(vorb *z, uint8 *data, int n)
+{
+   if (USE_MEMORY(z)) {
+      if (z->stream+n > z->stream_end) { z->eof = 1; return 0; }
+      memcpy(data, z->stream, n);
+      z->stream += n;
+      return 1;
+   }
+
+   #ifndef STB_VORBIS_NO_STDIO   
+   if (fread(data, n, 1, z->f) == 1)
+      return 1;
+   else {
+      z->eof = 1;
+      return 0;
+   }
+   #endif
+}
+
+static void skip(vorb *z, int n)
+{
+   if (USE_MEMORY(z)) {
+      z->stream += n;
+      if (z->stream >= z->stream_end) z->eof = 1;
+      return;
+   }
+   #ifndef STB_VORBIS_NO_STDIO
+   {
+      long x = ftell(z->f);
+      fseek(z->f, x+n, SEEK_SET);
+   }
+   #endif
+}
+
+static int set_file_offset(stb_vorbis *f, unsigned int loc)
+{
+   #ifndef STB_VORBIS_NO_PUSHDATA_API
+   if (f->push_mode) return 0;
+   #endif
+   f->eof = 0;
+   if (USE_MEMORY(f)) {
+      if (f->stream_start + loc >= f->stream_end || f->stream_start + loc < f->stream_start) {
+         f->stream = f->stream_end;
+         f->eof = 1;
+         return 0;
+      } else {
+         f->stream = f->stream_start + loc;
+         return 1;
+      }
+   }
+   #ifndef STB_VORBIS_NO_STDIO
+   if (loc + f->f_start < loc || loc >= 0x80000000) {
+      loc = 0x7fffffff;
+      f->eof = 1;
+   } else {
+      loc += f->f_start;
+   }
+   if (!fseek(f->f, loc, SEEK_SET))
+      return 1;
+   f->eof = 1;
+   fseek(f->f, f->f_start, SEEK_END);
+   return 0;
+   #endif
+}
+
+
+static uint8 ogg_page_header[4] = { 0x4f, 0x67, 0x67, 0x53 };
+
+static int capture_pattern(vorb *f)
+{
+   if (0x4f != get8(f)) return FALSE;
+   if (0x67 != get8(f)) return FALSE;
+   if (0x67 != get8(f)) return FALSE;
+   if (0x53 != get8(f)) return FALSE;
+   return TRUE;
+}
+
+#define PAGEFLAG_continued_packet   1
+#define PAGEFLAG_first_page         2
+#define PAGEFLAG_last_page          4
+
+static int start_page_no_capturepattern(vorb *f)
+{
+   uint32 loc0,loc1,n;
+   // 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.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 += (unsigned) z << f->valid_bits;
+         f->valid_bits += 8;
+      } while (f->valid_bits <= 24);
+   }
+}
+
+enum
+{
+   VORBIS_packet_id = 1,
+   VORBIS_packet_comment = 3,
+   VORBIS_packet_setup = 5
+};
+
+static int codebook_decode_scalar_raw(vorb *f, Codebook *c)
+{
+   int i;
+   prep_huffman(f);
+
+   if (c->codewords == NULL && c->sorted_codewords == NULL)
+      return -1;
+
+   // cases to use binary search: sorted_codewords && !c->codewords
+   //                             sorted_codewords && c->entries > 8
+   if (c->entries > 8 ? c->sorted_codewords!=NULL : !c->codewords) {
+      // binary search
+      uint32 code = bit_reverse(f->acc);
+      int x=0, n=c->sorted_entries, len;
+
+      while (n > 1) {
+         // invariant: sc[x] <= code < sc[x+n]
+         int m = x + (n >> 1);
+         if (c->sorted_codewords[m] <= code) {
+            x = m;
+            n -= (n>>1);
+         } else {
+            n >>= 1;
+         }
+      }
+      // x is now the sorted index
+      if (!c->sparse) x = c->sorted_values[x];
+      // x is now sorted index if sparse, or symbol otherwise
+      len = c->codeword_lengths[x];
+      if (f->valid_bits >= len) {
+         f->acc >>= len;
+         f->valid_bits -= len;
+         return x;
+      }
+
+      f->valid_bits = 0;
+      return -1;
+   }
+
+   // if small, linear search
+   assert(!c->sparse);
+   for (i=0; i < c->entries; ++i) {
+      if (c->codeword_lengths[i] == NO_CODE) continue;
+      if (c->codewords[i] == (f->acc & ((1 << c->codeword_lengths[i])-1))) {
+         if (f->valid_bits >= c->codeword_lengths[i]) {
+            f->acc >>= c->codeword_lengths[i];
+            f->valid_bits -= c->codeword_lengths[i];
+            return i;
+         }
+         f->valid_bits = 0;
+         return -1;
+      }
+   }
+
+   error(f, VORBIS_invalid_stream);
+   f->valid_bits = 0;
+   return -1;
+}
+
+#ifndef STB_VORBIS_NO_INLINE_DECODE
+
+#define DECODE_RAW(var, f,c)                                  \
+   if (f->valid_bits < STB_VORBIS_FAST_HUFFMAN_LENGTH)        \
+      prep_huffman(f);                                        \
+   var = f->acc & FAST_HUFFMAN_TABLE_MASK;                    \
+   var = c->fast_huffman[var];                                \
+   if (var >= 0) {                                            \
+      int n = c->codeword_lengths[var];                       \
+      f->acc >>= n;                                           \
+      f->valid_bits -= n;                                     \
+      if (f->valid_bits < 0) { f->valid_bits = 0; var = -1; } \
+   } else {                                                   \
+      var = codebook_decode_scalar_raw(f,c);                  \
+   }
+
+#else
+
+static int codebook_decode_scalar(vorb *f, Codebook *c)
+{
+   int i;
+   if (f->valid_bits < STB_VORBIS_FAST_HUFFMAN_LENGTH)
+      prep_huffman(f);
+   // fast huffman table lookup
+   i = f->acc & FAST_HUFFMAN_TABLE_MASK;
+   i = c->fast_huffman[i];
+   if (i >= 0) {
+      f->acc >>= c->codeword_lengths[i];
+      f->valid_bits -= c->codeword_lengths[i];
+      if (f->valid_bits < 0) { f->valid_bits = 0; return -1; }
+      return i;
+   }
+   return codebook_decode_scalar_raw(f,c);
+}
+
+#define DECODE_RAW(var,f,c)    var = codebook_decode_scalar(f,c);
+
+#endif
+
+#define DECODE(var,f,c)                                       \
+   DECODE_RAW(var,f,c)                                        \
+   if (c->sparse) var = c->sorted_values[var];
+
+#ifndef STB_VORBIS_DIVIDES_IN_CODEBOOK
+  #define DECODE_VQ(var,f,c)   DECODE_RAW(var,f,c)
+#else
+  #define DECODE_VQ(var,f,c)   DECODE(var,f,c)
+#endif
+
+
+
+
+
+
+// CODEBOOK_ELEMENT_FAST is an optimization for the CODEBOOK_FLOATS case
+// where we avoid one addition
+#define CODEBOOK_ELEMENT(c,off)          (c->multiplicands[off])
+#define CODEBOOK_ELEMENT_FAST(c,off)     (c->multiplicands[off])
+#define CODEBOOK_ELEMENT_BASE(c)         (0)
+
+static int codebook_decode_start(vorb *f, Codebook *c)
+{
+   int z = -1;
+
+   // type 0 is only legal in a scalar context
+   if (c->lookup_type == 0)
+      error(f, VORBIS_invalid_stream);
+   else {
+      DECODE_VQ(z,f,c);
+      if (c->sparse) assert(z < c->sorted_entries);
+      if (z < 0) {  // check for EOP
+         if (!f->bytes_in_seg)
+            if (f->last_seg)
+               return z;
+         error(f, VORBIS_invalid_stream);
+      }
+   }
+   return z;
+}
+
+static int codebook_decode(vorb *f, Codebook *c, float *output, int len)
+{
+   int i,z = codebook_decode_start(f,c);
+   if (z < 0) return FALSE;
+   if (len > c->dimensions) len = c->dimensions;
+
+#ifdef STB_VORBIS_DIVIDES_IN_CODEBOOK
+   if (c->lookup_type == 1) {
+      float last = CODEBOOK_ELEMENT_BASE(c);
+      int div = 1;
+      for (i=0; i < len; ++i) {
+         int off = (z / div) % c->lookup_values;
+         float val = CODEBOOK_ELEMENT_FAST(c,off) + last;
+         output[i] += val;
+         if (c->sequence_p) last = val + c->minimum_value;
+         div *= c->lookup_values;
+      }
+      return TRUE;
+   }
+#endif
+
+   z *= c->dimensions;
+   if (c->sequence_p) {
+      float last = CODEBOOK_ELEMENT_BASE(c);
+      for (i=0; i < len; ++i) {
+         float val = CODEBOOK_ELEMENT_FAST(c,z+i) + last;
+         output[i] += val;
+         last = val + c->minimum_value;
+      }
+   } else {
+      float last = CODEBOOK_ELEMENT_BASE(c);
+      for (i=0; i < len; ++i) {
+         output[i] += CODEBOOK_ELEMENT_FAST(c,z+i) + last;
+      }
+   }
+
+   return TRUE;
+}
+
+static int codebook_decode_step(vorb *f, Codebook *c, float *output, int len, int step)
+{
+   int i,z = codebook_decode_start(f,c);
+   float last = CODEBOOK_ELEMENT_BASE(c);
+   if (z < 0) return FALSE;
+   if (len > c->dimensions) len = c->dimensions;
+
+#ifdef STB_VORBIS_DIVIDES_IN_CODEBOOK
+   if (c->lookup_type == 1) {
+      int div = 1;
+      for (i=0; i < len; ++i) {
+         int off = (z / div) % c->lookup_values;
+         float val = CODEBOOK_ELEMENT_FAST(c,off) + last;
+         output[i*step] += val;
+         if (c->sequence_p) last = val;
+         div *= c->lookup_values;
+      }
+      return TRUE;
+   }
+#endif
+
+   z *= c->dimensions;
+   for (i=0; i < len; ++i) {
+      float val = CODEBOOK_ELEMENT_FAST(c,z+i) + last;
+      output[i*step] += val;
+      if (c->sequence_p) last = val;
+   }
+
+   return TRUE;
+}
+
+static int codebook_decode_deinterleave_repeat(vorb *f, Codebook *c, float **outputs, int ch, int *c_inter_p, int *p_inter_p, int len, int total_decode)
+{
+   int c_inter = *c_inter_p;
+   int p_inter = *p_inter_p;
+   int i,z, effective = c->dimensions;
+
+   // type 0 is only legal in a scalar context
+   if (c->lookup_type == 0)   return error(f, VORBIS_invalid_stream);
+
+   while (total_decode > 0) {
+      float last = CODEBOOK_ELEMENT_BASE(c);
+      DECODE_VQ(z,f,c);
+      #ifndef STB_VORBIS_DIVIDES_IN_CODEBOOK
+      assert(!c->sparse || z < c->sorted_entries);
+      #endif
+      if (z < 0) {
+         if (!f->bytes_in_seg)
+            if (f->last_seg) return FALSE;
+         return error(f, VORBIS_invalid_stream);
+      }
+
+      // if this will take us off the end of the buffers, stop short!
+      // we check by computing the length of the virtual interleaved
+      // buffer (len*ch), our current offset within it (p_inter*ch)+(c_inter),
+      // and the length we'll be using (effective)
+      if (c_inter + p_inter*ch + effective > len * ch) {
+         effective = len*ch - (p_inter*ch - c_inter);
+      }
+
+   #ifdef STB_VORBIS_DIVIDES_IN_CODEBOOK
+      if (c->lookup_type == 1) {
+         int div = 1;
+         for (i=0; i < effective; ++i) {
+            int off = (z / div) % c->lookup_values;
+            float val = CODEBOOK_ELEMENT_FAST(c,off) + last;
+            if (outputs[c_inter])
+               outputs[c_inter][p_inter] += val;
+            if (++c_inter == ch) { c_inter = 0; ++p_inter; }
+            if (c->sequence_p) last = val;
+            div *= c->lookup_values;
+         }
+      } else
+   #endif
+      {
+         z *= c->dimensions;
+         if (c->sequence_p) {
+            for (i=0; i < effective; ++i) {
+               float val = CODEBOOK_ELEMENT_FAST(c,z+i) + last;
+               if (outputs[c_inter])
+                  outputs[c_inter][p_inter] += val;
+               if (++c_inter == ch) { c_inter = 0; ++p_inter; }
+               last = val;
+            }
+         } else {
+            for (i=0; i < effective; ++i) {
+               float val = CODEBOOK_ELEMENT_FAST(c,z+i) + last;
+               if (outputs[c_inter])
+                  outputs[c_inter][p_inter] += val;
+               if (++c_inter == ch) { c_inter = 0; ++p_inter; }
+            }
+         }
+      }
+
+      total_decode -= effective;
+   }
+   *c_inter_p = c_inter;
+   *p_inter_p = p_inter;
+   return TRUE;
+}
+
+static int predict_point(int x, int x0, int x1, int y0, int y1)
+{
+   int dy = y1 - y0;
+   int adx = x1 - x0;
+   // @OPTIMIZE: force int division to round in the right direction... is this necessary on x86?
+   int err = abs(dy) * (x - x0);
+   int off = err / adx;
+   return dy < 0 ? y0 - off : y0 + off;
+}
+
+// the following table is block-copied from the specification
+static float inverse_db_table[256] =
+{
+  1.0649863e-07f, 1.1341951e-07f, 1.2079015e-07f, 1.2863978e-07f, 
+  1.3699951e-07f, 1.4590251e-07f, 1.5538408e-07f, 1.6548181e-07f, 
+  1.7623575e-07f, 1.8768855e-07f, 1.9988561e-07f, 2.1287530e-07f, 
+  2.2670913e-07f, 2.4144197e-07f, 2.5713223e-07f, 2.7384213e-07f, 
+  2.9163793e-07f, 3.1059021e-07f, 3.3077411e-07f, 3.5226968e-07f, 
+  3.7516214e-07f, 3.9954229e-07f, 4.2550680e-07f, 4.5315863e-07f, 
+  4.8260743e-07f, 5.1396998e-07f, 5.4737065e-07f, 5.8294187e-07f, 
+  6.2082472e-07f, 6.6116941e-07f, 7.0413592e-07f, 7.4989464e-07f, 
+  7.9862701e-07f, 8.5052630e-07f, 9.0579828e-07f, 9.6466216e-07f, 
+  1.0273513e-06f, 1.0941144e-06f, 1.1652161e-06f, 1.2409384e-06f, 
+  1.3215816e-06f, 1.4074654e-06f, 1.4989305e-06f, 1.5963394e-06f, 
+  1.7000785e-06f, 1.8105592e-06f, 1.9282195e-06f, 2.0535261e-06f, 
+  2.1869758e-06f, 2.3290978e-06f, 2.4804557e-06f, 2.6416497e-06f, 
+  2.8133190e-06f, 2.9961443e-06f, 3.1908506e-06f, 3.3982101e-06f, 
+  3.6190449e-06f, 3.8542308e-06f, 4.1047004e-06f, 4.3714470e-06f, 
+  4.6555282e-06f, 4.9580707e-06f, 5.2802740e-06f, 5.6234160e-06f, 
+  5.9888572e-06f, 6.3780469e-06f, 6.7925283e-06f, 7.2339451e-06f, 
+  7.7040476e-06f, 8.2047000e-06f, 8.7378876e-06f, 9.3057248e-06f, 
+  9.9104632e-06f, 1.0554501e-05f, 1.1240392e-05f, 1.1970856e-05f, 
+  1.2748789e-05f, 1.3577278e-05f, 1.4459606e-05f, 1.5399272e-05f, 
+  1.6400004e-05f, 1.7465768e-05f, 1.8600792e-05f, 1.9809576e-05f, 
+  2.1096914e-05f, 2.2467911e-05f, 2.3928002e-05f, 2.5482978e-05f, 
+  2.7139006e-05f, 2.8902651e-05f, 3.0780908e-05f, 3.2781225e-05f, 
+  3.4911534e-05f, 3.7180282e-05f, 3.9596466e-05f, 4.2169667e-05f, 
+  4.4910090e-05f, 4.7828601e-05f, 5.0936773e-05f, 5.4246931e-05f, 
+  5.7772202e-05f, 6.1526565e-05f, 6.5524908e-05f, 6.9783085e-05f, 
+  7.4317983e-05f, 7.9147585e-05f, 8.4291040e-05f, 8.9768747e-05f, 
+  9.5602426e-05f, 0.00010181521f, 0.00010843174f, 0.00011547824f, 
+  0.00012298267f, 0.00013097477f, 0.00013948625f, 0.00014855085f, 
+  0.00015820453f, 0.00016848555f, 0.00017943469f, 0.00019109536f, 
+  0.00020351382f, 0.00021673929f, 0.00023082423f, 0.00024582449f, 
+  0.00026179955f, 0.00027881276f, 0.00029693158f, 0.00031622787f, 
+  0.00033677814f, 0.00035866388f, 0.00038197188f, 0.00040679456f, 
+  0.00043323036f, 0.00046138411f, 0.00049136745f, 0.00052329927f, 
+  0.00055730621f, 0.00059352311f, 0.00063209358f, 0.00067317058f, 
+  0.00071691700f, 0.00076350630f, 0.00081312324f, 0.00086596457f, 
+  0.00092223983f, 0.00098217216f, 0.0010459992f,  0.0011139742f, 
+  0.0011863665f,  0.0012634633f,  0.0013455702f,  0.0014330129f, 
+  0.0015261382f,  0.0016253153f,  0.0017309374f,  0.0018434235f, 
+  0.0019632195f,  0.0020908006f,  0.0022266726f,  0.0023713743f, 
+  0.0025254795f,  0.0026895994f,  0.0028643847f,  0.0030505286f, 
+  0.0032487691f,  0.0034598925f,  0.0036847358f,  0.0039241906f, 
+  0.0041792066f,  0.0044507950f,  0.0047400328f,  0.0050480668f, 
+  0.0053761186f,  0.0057254891f,  0.0060975636f,  0.0064938176f, 
+  0.0069158225f,  0.0073652516f,  0.0078438871f,  0.0083536271f, 
+  0.0088964928f,  0.009474637f,   0.010090352f,   0.010746080f, 
+  0.011444421f,   0.012188144f,   0.012980198f,   0.013823725f, 
+  0.014722068f,   0.015678791f,   0.016697687f,   0.017782797f, 
+  0.018938423f,   0.020169149f,   0.021479854f,   0.022875735f, 
+  0.024362330f,   0.025945531f,   0.027631618f,   0.029427276f, 
+  0.031339626f,   0.033376252f,   0.035545228f,   0.037855157f, 
+  0.040315199f,   0.042935108f,   0.045725273f,   0.048696758f, 
+  0.051861348f,   0.055231591f,   0.058820850f,   0.062643361f, 
+  0.066714279f,   0.071049749f,   0.075666962f,   0.080584227f, 
+  0.085821044f,   0.091398179f,   0.097337747f,   0.10366330f, 
+  0.11039993f,    0.11757434f,    0.12521498f,    0.13335215f, 
+  0.14201813f,    0.15124727f,    0.16107617f,    0.17154380f, 
+  0.18269168f,    0.19456402f,    0.20720788f,    0.22067342f, 
+  0.23501402f,    0.25028656f,    0.26655159f,    0.28387361f, 
+  0.30232132f,    0.32196786f,    0.34289114f,    0.36517414f, 
+  0.38890521f,    0.41417847f,    0.44109412f,    0.46975890f, 
+  0.50028648f,    0.53279791f,    0.56742212f,    0.60429640f, 
+  0.64356699f,    0.68538959f,    0.72993007f,    0.77736504f, 
+  0.82788260f,    0.88168307f,    0.9389798f,     1.0f
+};
+
+
+// @OPTIMIZE: if you want to replace this bresenham line-drawing routine,
+// note that you must produce bit-identical output to decode correctly;
+// this specific sequence of operations is specified in the spec (it's
+// drawing integer-quantized frequency-space lines that the encoder
+// expects to be exactly the same)
+//     ... also, isn't the whole point of Bresenham's algorithm to NOT
+// have to divide in the setup? sigh.
+#ifndef STB_VORBIS_NO_DEFER_FLOOR
+#define LINE_OP(a,b)   a *= b
+#else
+#define LINE_OP(a,b)   a = b
+#endif
+
+#ifdef STB_VORBIS_DIVIDE_TABLE
+#define DIVTAB_NUMER   32
+#define DIVTAB_DENOM   64
+int8 integer_divide_table[DIVTAB_NUMER][DIVTAB_DENOM]; // 2KB
+#endif
+
+static __forceinline void draw_line(float *output, int x0, int y0, int x1, int y1, int n)
+{
+   int dy = y1 - y0;
+   int adx = x1 - x0;
+   int ady = abs(dy);
+   int base;
+   int x=x0,y=y0;
+   int err = 0;
+   int sy;
+
+#ifdef STB_VORBIS_DIVIDE_TABLE
+   if (adx < DIVTAB_DENOM && ady < DIVTAB_NUMER) {
+      if (dy < 0) {
+         base = -integer_divide_table[ady][adx];
+         sy = base-1;
+      } else {
+         base =  integer_divide_table[ady][adx];
+         sy = base+1;
+      }
+   } else {
+      base = dy / adx;
+      if (dy < 0)
+         sy = base - 1;
+      else
+         sy = base+1;
+   }
+#else
+   base = dy / adx;
+   if (dy < 0)
+      sy = base - 1;
+   else
+      sy = base+1;
+#endif
+   ady -= abs(base) * adx;
+   if (x1 > n) x1 = n;
+   if (x < x1) {
+      LINE_OP(output[x], inverse_db_table[y]);
+      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
+
+   CHECK(f);
+
+   for (i=0; i < ch; ++i)
+      if (!do_not_decode[i])
+         memset(residue_buffers[i], 0, sizeof(float) * n);
+
+   if (rtype == 2 && ch != 1) {
+      for (j=0; j < ch; ++j)
+         if (!do_not_decode[j])
+            break;
+      if (j == ch)
+         goto done;
+
+      for (pass=0; pass < 8; ++pass) {
+         int pcount = 0, class_set = 0;
+         if (ch == 2) {
+            while (pcount < part_read) {
+               int z = r->begin + pcount*r->part_size;
+               int c_inter = (z & 1), p_inter = z>>1;
+               if (pass == 0) {
+                  Codebook *c = f->codebooks+r->classbook;
+                  int q;
+                  DECODE(q,f,c);
+                  if (q == EOP) goto done;
+                  #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE
+                  part_classdata[0][class_set] = r->classdata[q];
+                  #else
+                  for (i=classwords-1; i >= 0; --i) {
+                     classifications[0][i+pcount] = q % r->classifications;
+                     q /= r->classifications;
+                  }
+                  #endif
+               }
+               for (i=0; i < classwords && pcount < part_read; ++i, ++pcount) {
+                  int z = r->begin + pcount*r->part_size;
+                  #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE
+                  int c = part_classdata[0][class_set][i];
+                  #else
+                  int c = classifications[0][pcount];
+                  #endif
+                  int b = r->residue_books[c][pass];
+                  if (b >= 0) {
+                     Codebook *book = f->codebooks + b;
+                     #ifdef STB_VORBIS_DIVIDES_IN_CODEBOOK
+                     if (!codebook_decode_deinterleave_repeat(f, book, residue_buffers, ch, &c_inter, &p_inter, n, r->part_size))
+                        goto done;
+                     #else
+                     // saves 1%
+                     if (!codebook_decode_deinterleave_repeat(f, book, residue_buffers, ch, &c_inter, &p_inter, n, r->part_size))
+                        goto done;
+                     #endif
+                  } else {
+                     z += r->part_size;
+                     c_inter = z & 1;
+                     p_inter = z >> 1;
+                  }
+               }
+               #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE
+               ++class_set;
+               #endif
+            }
+         } else if (ch == 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;
+                     if (!codebook_decode_deinterleave_repeat(f, book, residue_buffers, ch, &c_inter, &p_inter, n, r->part_size))
+                        goto done;
+                  } else {
+                     z += r->part_size;
+                     c_inter = 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;
+                     if (!codebook_decode_deinterleave_repeat(f, book, residue_buffers, ch, &c_inter, &p_inter, n, r->part_size))
+                        goto done;
+                  } else {
+                     z += r->part_size;
+                     c_inter = z % ch;
+                     p_inter = z / ch;
+                  }
+               }
+               #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE
+               ++class_set;
+               #endif
+            }
+         }
+      }
+      goto done;
+   }
+   CHECK(f);
+
+   for (pass=0; pass < 8; ++pass) {
+      int pcount = 0, class_set=0;
+      while (pcount < part_read) {
+         if (pass == 0) {
+            for (j=0; j < ch; ++j) {
+               if (!do_not_decode[j]) {
+                  Codebook *c = f->codebooks+r->classbook;
+                  int temp;
+                  DECODE(temp,f,c);
+                  if (temp == EOP) goto done;
+                  #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE
+                  part_classdata[j][class_set] = r->classdata[temp];
+                  #else
+                  for (i=classwords-1; i >= 0; --i) {
+                     classifications[j][i+pcount] = temp % r->classifications;
+                     temp /= r->classifications;
+                  }
+                  #endif
+               }
+            }
+         }
+         for (i=0; i < classwords && pcount < part_read; ++i, ++pcount) {
+            for (j=0; j < ch; ++j) {
+               if (!do_not_decode[j]) {
+                  #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE
+                  int c = part_classdata[j][class_set][i];
+                  #else
+                  int c = classifications[j][pcount];
+                  #endif
+                  int b = r->residue_books[c][pass];
+                  if (b >= 0) {
+                     float *target = residue_buffers[j];
+                     int offset = r->begin + pcount * r->part_size;
+                     int n = r->part_size;
+                     Codebook *book = f->codebooks + b;
+                     if (!residue_decode(f, book, target, offset, n, rtype))
+                        goto done;
+                  }
+               }
+            }
+         }
+         #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE
+         ++class_set;
+         #endif
+      }
+   }
+  done:
+   CHECK(f);
+   #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE
+   temp_free(f,part_classdata);
+   #else
+   temp_free(f,classifications);
+   #endif
+   temp_alloc_restore(f,temp_alloc_point);
+}
+
+
+#if 0
+// slow way for debugging
+void inverse_mdct_slow(float *buffer, int n)
+{
+   int i,j;
+   int n2 = n >> 1;
+   float *x = (float *) malloc(sizeof(*x) * n2);
+   memcpy(x, buffer, sizeof(*x) * n2);
+   for (i=0; i < n; ++i) {
+      float acc = 0;
+      for (j=0; j < n2; ++j)
+         // formula from paper:
+         //acc += n/4.0f * x[j] * (float) cos(M_PI / 2 / n * (2 * i + 1 + n/2.0)*(2*j+1));
+         // formula from wikipedia
+         //acc += 2.0f / n2 * x[j] * (float) cos(M_PI/n2 * (i + 0.5 + n2/2)*(j + 0.5));
+         // these are equivalent, except the formula from the paper inverts the multiplier!
+         // however, what actually works is NO MULTIPLIER!?!
+         //acc += 64 * 2.0f / n2 * x[j] * (float) cos(M_PI/n2 * (i + 0.5 + n2/2)*(j + 0.5));
+         acc += x[j] * (float) cos(M_PI / 2 / n * (2 * i + 1 + n/2.0)*(2*j+1));
+      buffer[i] = acc;
+   }
+   free(x);
+}
+#elif 0
+// same as above, but just barely able to run in real time on modern machines
+void inverse_mdct_slow(float *buffer, int n, vorb *f, int blocktype)
+{
+   float mcos[16384];
+   int i,j;
+   int n2 = n >> 1, nmask = (n << 2) -1;
+   float *x = (float *) malloc(sizeof(*x) * n2);
+   memcpy(x, buffer, sizeof(*x) * n2);
+   for (i=0; i < 4*n; ++i)
+      mcos[i] = (float) cos(M_PI / 2 * i / n);
+
+   for (i=0; i < n; ++i) {
+      float acc = 0;
+      for (j=0; j < n2; ++j)
+         acc += x[j] * mcos[(2 * i + 1 + n2)*(2*j+1) & nmask];
+      buffer[i] = acc;
+   }
+   free(x);
+}
+#elif 0
+// transform to use a slow dct-iv; this is STILL basically trivial,
+// but only requires half as many ops
+void dct_iv_slow(float *buffer, int n)
+{
+   float mcos[16384];
+   float x[2048];
+   int i,j;
+   int n2 = n >> 1, nmask = (n << 3) - 1;
+   memcpy(x, buffer, sizeof(*x) * n);
+   for (i=0; i < 8*n; ++i)
+      mcos[i] = (float) cos(M_PI / 4 * i / n);
+   for (i=0; i < n; ++i) {
+      float acc = 0;
+      for (j=0; j < n; ++j)
+         acc += x[j] * mcos[((2 * i + 1)*(2*j+1)) & nmask];
+      buffer[i] = acc;
+   }
+}
+
+void inverse_mdct_slow(float *buffer, int n, vorb *f, int blocktype)
+{
+   int i, n4 = n >> 2, n2 = n >> 1, n3_4 = n - n4;
+   float temp[4096];
+
+   memcpy(temp, buffer, n2 * sizeof(float));
+   dct_iv_slow(temp, n2);  // returns -c'-d, a-b'
+
+   for (i=0; i < n4  ; ++i) buffer[i] = temp[i+n4];            // a-b'
+   for (   ; i < n3_4; ++i) buffer[i] = -temp[n3_4 - i - 1];   // b-a', c+d'
+   for (   ; i < n   ; ++i) buffer[i] = -temp[i - n3_4];       // c'+d
+}
+#endif
+
+#ifndef LIBVORBIS_MDCT
+#define LIBVORBIS_MDCT 0
+#endif
+
+#if LIBVORBIS_MDCT
+// directly call the vorbis MDCT using an interface documented
+// by Jeff Roberts... useful for performance comparison
+typedef struct 
+{
+  int n;
+  int log2n;
+  
+  float *trig;
+  int   *bitrev;
+
+  float scale;
+} mdct_lookup;
+
+extern void mdct_init(mdct_lookup *lookup, int n);
+extern void mdct_clear(mdct_lookup *l);
+extern void mdct_backward(mdct_lookup *init, float *in, float *out);
+
+mdct_lookup M1,M2;
+
+void inverse_mdct(float *buffer, int n, vorb *f, int blocktype)
+{
+   mdct_lookup *M;
+   if (M1.n == n) M = &M1;
+   else if (M2.n == n) M = &M2;
+   else if (M1.n == 0) { mdct_init(&M1, n); M = &M1; }
+   else { 
+      if (M2.n) __asm int 3;
+      mdct_init(&M2, n);
+      M = &M2;
+   }
+
+   mdct_backward(M, buffer, buffer);
+}
+#endif
+
+
+// the following were split out into separate functions while optimizing;
+// they could be pushed back up but eh. __forceinline showed no change;
+// they're probably already being inlined.
+static void imdct_step3_iter0_loop(int n, float *e, int i_off, int k_off, float *A)
+{
+   float *ee0 = e + i_off;
+   float *ee2 = ee0 + k_off;
+   int i;
+
+   assert((n & 3) == 0);
+   for (i=(n>>2); i > 0; --i) {
+      float k00_20, k01_21;
+      k00_20  = ee0[ 0] - ee2[ 0];
+      k01_21  = ee0[-1] - ee2[-1];
+      ee0[ 0] += ee2[ 0];//ee0[ 0] = ee0[ 0] + ee2[ 0];
+      ee0[-1] += ee2[-1];//ee0[-1] = ee0[-1] + ee2[-1];
+      ee2[ 0] = k00_20 * A[0] - k01_21 * A[1];
+      ee2[-1] = k01_21 * A[0] + k00_20 * A[1];
+      A += 8;
+
+      k00_20  = ee0[-2] - ee2[-2];
+      k01_21  = ee0[-3] - ee2[-3];
+      ee0[-2] += ee2[-2];//ee0[-2] = ee0[-2] + ee2[-2];
+      ee0[-3] += ee2[-3];//ee0[-3] = ee0[-3] + ee2[-3];
+      ee2[-2] = k00_20 * A[0] - k01_21 * A[1];
+      ee2[-3] = k01_21 * A[0] + k00_20 * A[1];
+      A += 8;
+
+      k00_20  = ee0[-4] - ee2[-4];
+      k01_21  = ee0[-5] - ee2[-5];
+      ee0[-4] += ee2[-4];//ee0[-4] = ee0[-4] + ee2[-4];
+      ee0[-5] += ee2[-5];//ee0[-5] = ee0[-5] + ee2[-5];
+      ee2[-4] = k00_20 * A[0] - k01_21 * A[1];
+      ee2[-5] = k01_21 * A[0] + k00_20 * A[1];
+      A += 8;
+
+      k00_20  = ee0[-6] - ee2[-6];
+      k01_21  = ee0[-7] - ee2[-7];
+      ee0[-6] += ee2[-6];//ee0[-6] = ee0[-6] + ee2[-6];
+      ee0[-7] += ee2[-7];//ee0[-7] = ee0[-7] + ee2[-7];
+      ee2[-6] = k00_20 * A[0] - k01_21 * A[1];
+      ee2[-7] = k01_21 * A[0] + k00_20 * A[1];
+      A += 8;
+      ee0 -= 8;
+      ee2 -= 8;
+   }
+}
+
+static void imdct_step3_inner_r_loop(int lim, float *e, int d0, int k_off, float *A, int k1)
+{
+   int i;
+   float k00_20, k01_21;
+
+   float *e0 = e + d0;
+   float *e2 = e0 + k_off;
+
+   for (i=lim >> 2; i > 0; --i) {
+      k00_20 = e0[-0] - e2[-0];
+      k01_21 = e0[-1] - e2[-1];
+      e0[-0] += e2[-0];//e0[-0] = e0[-0] + e2[-0];
+      e0[-1] += e2[-1];//e0[-1] = e0[-1] + e2[-1];
+      e2[-0] = (k00_20)*A[0] - (k01_21) * A[1];
+      e2[-1] = (k01_21)*A[0] + (k00_20) * A[1];
+
+      A += k1;
+
+      k00_20 = e0[-2] - e2[-2];
+      k01_21 = e0[-3] - e2[-3];
+      e0[-2] += e2[-2];//e0[-2] = e0[-2] + e2[-2];
+      e0[-3] += e2[-3];//e0[-3] = e0[-3] + e2[-3];
+      e2[-2] = (k00_20)*A[0] - (k01_21) * A[1];
+      e2[-3] = (k01_21)*A[0] + (k00_20) * A[1];
+
+      A += k1;
+
+      k00_20 = e0[-4] - e2[-4];
+      k01_21 = e0[-5] - e2[-5];
+      e0[-4] += e2[-4];//e0[-4] = e0[-4] + e2[-4];
+      e0[-5] += e2[-5];//e0[-5] = e0[-5] + e2[-5];
+      e2[-4] = (k00_20)*A[0] - (k01_21) * A[1];
+      e2[-5] = (k01_21)*A[0] + (k00_20) * A[1];
+
+      A += k1;
+
+      k00_20 = e0[-6] - e2[-6];
+      k01_21 = e0[-7] - e2[-7];
+      e0[-6] += e2[-6];//e0[-6] = e0[-6] + e2[-6];
+      e0[-7] += e2[-7];//e0[-7] = e0[-7] + e2[-7];
+      e2[-6] = (k00_20)*A[0] - (k01_21) * A[1];
+      e2[-7] = (k01_21)*A[0] + (k00_20) * A[1];
+
+      e0 -= 8;
+      e2 -= 8;
+
+      A += k1;
+   }
+}
+
+static void imdct_step3_inner_s_loop(int n, float *e, int i_off, int k_off, float *A, int a_off, int k0)
+{
+   int i;
+   float A0 = A[0];
+   float A1 = A[0+1];
+   float A2 = A[0+a_off];
+   float A3 = A[0+a_off+1];
+   float A4 = A[0+a_off*2+0];
+   float A5 = A[0+a_off*2+1];
+   float A6 = A[0+a_off*3+0];
+   float A7 = A[0+a_off*3+1];
+
+   float k00,k11;
+
+   float *ee0 = e  +i_off;
+   float *ee2 = ee0+k_off;
+
+   for (i=n; i > 0; --i) {
+      k00     = ee0[ 0] - ee2[ 0];
+      k11     = ee0[-1] - ee2[-1];
+      ee0[ 0] =  ee0[ 0] + ee2[ 0];
+      ee0[-1] =  ee0[-1] + ee2[-1];
+      ee2[ 0] = (k00) * A0 - (k11) * A1;
+      ee2[-1] = (k11) * A0 + (k00) * A1;
+
+      k00     = ee0[-2] - ee2[-2];
+      k11     = ee0[-3] - ee2[-3];
+      ee0[-2] =  ee0[-2] + ee2[-2];
+      ee0[-3] =  ee0[-3] + ee2[-3];
+      ee2[-2] = (k00) * A2 - (k11) * A3;
+      ee2[-3] = (k11) * A2 + (k00) * A3;
+
+      k00     = ee0[-4] - ee2[-4];
+      k11     = ee0[-5] - ee2[-5];
+      ee0[-4] =  ee0[-4] + ee2[-4];
+      ee0[-5] =  ee0[-5] + ee2[-5];
+      ee2[-4] = (k00) * A4 - (k11) * A5;
+      ee2[-5] = (k11) * A4 + (k00) * A5;
+
+      k00     = ee0[-6] - ee2[-6];
+      k11     = ee0[-7] - ee2[-7];
+      ee0[-6] =  ee0[-6] + ee2[-6];
+      ee0[-7] =  ee0[-7] + ee2[-7];
+      ee2[-6] = (k00) * A6 - (k11) * A7;
+      ee2[-7] = (k11) * A6 + (k00) * A7;
+
+      ee0 -= k0;
+      ee2 -= k0;
+   }
+}
+
+static __forceinline void iter_54(float *z)
+{
+   float k00,k11,k22,k33;
+   float y0,y1,y2,y3;
+
+   k00  = z[ 0] - z[-4];
+   y0   = z[ 0] + z[-4];
+   y2   = z[-2] + z[-6];
+   k22  = z[-2] - z[-6];
+
+   z[-0] = y0 + y2;      // z0 + z4 + z2 + z6
+   z[-2] = y0 - y2;      // z0 + z4 - z2 - z6
+
+   // done with y0,y2
+
+   k33  = z[-3] - z[-7];
+
+   z[-4] = k00 + k33;    // z0 - z4 + z3 - z7
+   z[-6] = k00 - k33;    // z0 - z4 - z3 + z7
+
+   // done with k33
+
+   k11  = z[-1] - z[-5];
+   y1   = z[-1] + z[-5];
+   y3   = z[-3] + z[-7];
+
+   z[-1] = y1 + y3;      // z1 + z5 + z3 + z7
+   z[-3] = y1 - y3;      // z1 + z5 - z3 - z7
+   z[-5] = k11 - k22;    // z1 - z5 + z2 - z6
+   z[-7] = k11 + k22;    // z1 - z5 - z2 + z6
+}
+
+static void imdct_step3_inner_s_loop_ld654(int n, float *e, int i_off, float *A, int base_n)
+{
+   int a_off = base_n >> 3;
+   float A2 = A[0+a_off];
+   float *z = e + i_off;
+   float *base = z - 16 * n;
+
+   while (z > base) {
+      float k00,k11;
+
+      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_free(f,buf2);
+   temp_alloc_restore(f,save_point);
+}
+
+#if 0
+// this is the original version of the above code, if you want to optimize it from scratch
+void inverse_mdct_naive(float *buffer, int n)
+{
+   float s;
+   float A[1 << 12], B[1 << 12], C[1 << 11];
+   int i,k,k2,k4, n2 = n >> 1, n4 = n >> 2, n8 = n >> 3, l;
+   int n3_4 = n - n4, ld;
+   // how can they claim this only uses N words?!
+   // oh, because they're only used sparsely, whoops
+   float u[1 << 13], X[1 << 13], v[1 << 13], w[1 << 13];
+   // set up twiddle factors
+
+   for (k=k2=0; k < n4; ++k,k2+=2) {
+      A[k2  ] = (float)  cos(4*k*M_PI/n);
+      A[k2+1] = (float) -sin(4*k*M_PI/n);
+      B[k2  ] = (float)  cos((k2+1)*M_PI/n/2);
+      B[k2+1] = (float)  sin((k2+1)*M_PI/n/2);
+   }
+   for (k=k2=0; k < n8; ++k,k2+=2) {
+      C[k2  ] = (float)  cos(2*(k2+1)*M_PI/n);
+      C[k2+1] = (float) -sin(2*(k2+1)*M_PI/n);
+   }
+
+   // IMDCT algorithm from "The use of multirate filter banks for coding of high quality digital audio"
+   // Note there are bugs in that pseudocode, presumably due to them attempting
+   // to rename the arrays nicely rather than representing the way their actual
+   // implementation bounces buffers back and forth. As a result, even in the
+   // "some formulars corrected" version, a direct implementation fails. These
+   // are noted below as "paper bug".
+
+   // copy and reflect spectral data
+   for (k=0; k < n2; ++k) u[k] = buffer[k];
+   for (   ; k < n ; ++k) u[k] = -buffer[n - k - 1];
+   // kernel from paper
+   // step 1
+   for (k=k2=k4=0; k < n4; k+=1, k2+=2, k4+=4) {
+      v[n-k4-1] = (u[k4] - u[n-k4-1]) * A[k2]   - (u[k4+2] - u[n-k4-3])*A[k2+1];
+      v[n-k4-3] = (u[k4] - u[n-k4-1]) * A[k2+1] + (u[k4+2] - u[n-k4-3])*A[k2];
+   }
+   // step 2
+   for (k=k4=0; k < n8; k+=1, k4+=4) {
+      w[n2+3+k4] = v[n2+3+k4] + v[k4+3];
+      w[n2+1+k4] = v[n2+1+k4] + v[k4+1];
+      w[k4+3]    = (v[n2+3+k4] - v[k4+3])*A[n2-4-k4] - (v[n2+1+k4]-v[k4+1])*A[n2-3-k4];
+      w[k4+1]    = (v[n2+1+k4] - v[k4+1])*A[n2-4-k4] + (v[n2+3+k4]-v[k4+3])*A[n2-3-k4];
+   }
+   // step 3
+   ld = ilog(n) - 1; // ilog is off-by-one from normal definitions
+   for (l=0; l < ld-3; ++l) {
+      int k0 = n >> (l+2), k1 = 1 << (l+3);
+      int rlim = n >> (l+4), r4, r;
+      int s2lim = 1 << (l+2), s2;
+      for (r=r4=0; r < rlim; r4+=4,++r) {
+         for (s2=0; s2 < s2lim; s2+=2) {
+            u[n-1-k0*s2-r4] = w[n-1-k0*s2-r4] + w[n-1-k0*(s2+1)-r4];
+            u[n-3-k0*s2-r4] = w[n-3-k0*s2-r4] + w[n-3-k0*(s2+1)-r4];
+            u[n-1-k0*(s2+1)-r4] = (w[n-1-k0*s2-r4] - w[n-1-k0*(s2+1)-r4]) * A[r*k1]
+                                - (w[n-3-k0*s2-r4] - w[n-3-k0*(s2+1)-r4]) * A[r*k1+1];
+            u[n-3-k0*(s2+1)-r4] = (w[n-3-k0*s2-r4] - w[n-3-k0*(s2+1)-r4]) * A[r*k1]
+                                + (w[n-1-k0*s2-r4] - w[n-1-k0*(s2+1)-r4]) * A[r*k1+1];
+         }
+      }
+      if (l+1 < ld-3) {
+         // paper bug: ping-ponging of u&w here is omitted
+         memcpy(w, u, sizeof(u));
+      }
+   }
+
+   // step 4
+   for (i=0; i < n8; ++i) {
+      int j = bit_reverse(i) >> (32-ld+3);
+      assert(j < n8);
+      if (i == j) {
+         // paper bug: original code probably swapped in place; if copying,
+         //            need to directly copy in this case
+         int i8 = i << 3;
+         v[i8+1] = u[i8+1];
+         v[i8+3] = u[i8+3];
+         v[i8+5] = u[i8+5];
+         v[i8+7] = u[i8+7];
+      } else if (i < j) {
+         int i8 = i << 3, j8 = j << 3;
+         v[j8+1] = u[i8+1], v[i8+1] = u[j8 + 1];
+         v[j8+3] = u[i8+3], v[i8+3] = u[j8 + 3];
+         v[j8+5] = u[i8+5], v[i8+5] = u[j8 + 5];
+         v[j8+7] = u[i8+7], v[i8+7] = u[j8 + 7];
+      }
+   }
+   // step 5
+   for (k=0; k < n2; ++k) {
+      w[k] = v[k*2+1];
+   }
+   // step 6
+   for (k=k2=k4=0; k < n8; ++k, k2 += 2, k4 += 4) {
+      u[n-1-k2] = w[k4];
+      u[n-2-k2] = w[k4+1];
+      u[n3_4 - 1 - k2] = w[k4+2];
+      u[n3_4 - 2 - k2] = w[k4+3];
+   }
+   // step 7
+   for (k=k2=0; k < n8; ++k, k2 += 2) {
+      v[n2 + k2 ] = ( u[n2 + k2] + u[n-2-k2] + C[k2+1]*(u[n2+k2]-u[n-2-k2]) + C[k2]*(u[n2+k2+1]+u[n-2-k2+1]))/2;
+      v[n-2 - k2] = ( u[n2 + k2] + u[n-2-k2] - C[k2+1]*(u[n2+k2]-u[n-2-k2]) - C[k2]*(u[n2+k2+1]+u[n-2-k2+1]))/2;
+      v[n2+1+ k2] = ( u[n2+1+k2] - u[n-1-k2] + C[k2+1]*(u[n2+1+k2]+u[n-1-k2]) - C[k2]*(u[n2+k2]-u[n-2-k2]))/2;
+      v[n-1 - k2] = (-u[n2+1+k2] + u[n-1-k2] + C[k2+1]*(u[n2+1+k2]+u[n-1-k2]) - C[k2]*(u[n2+k2]-u[n-2-k2]))/2;
+   }
+   // step 8
+   for (k=k2=0; k < n4; ++k,k2 += 2) {
+      X[k]      = v[k2+n2]*B[k2  ] + v[k2+1+n2]*B[k2+1];
+      X[n2-1-k] = v[k2+n2]*B[k2+1] - v[k2+1+n2]*B[k2  ];
+   }
+
+   // decode kernel to output
+   // determined the following value experimentally
+   // (by first figuring out what made inverse_mdct_slow work); then matching that here
+   // (probably vorbis encoder premultiplies by n or n/2, to save it on the decoder?)
+   s = 0.5; // theoretically would be n4
+
+   // [[[ note! the s value of 0.5 is compensated for by the B[] in the current code,
+   //     so it needs to use the "old" B values to behave correctly, or else
+   //     set s to 1.0 ]]]
+   for (i=0; i < n4  ; ++i) buffer[i] = s * X[i+n4];
+   for (   ; i < n3_4; ++i) buffer[i] = -s * X[n3_4 - i - 1];
+   for (   ; i < n   ; ++i) buffer[i] = -s * X[i - n3_4];
+}
+#endif
+
+static float *get_window(vorb *f, int len)
+{
+   len <<= 1;
+   if (len == f->blocksize_0) return f->window[0];
+   if (len == f->blocksize_1) return f->window[1];
+   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];
+            if (lx != hx)
+               draw_line(target, lx,ly, hx,hy, n2);
+            CHECK(f);
+            lx = hx, ly = hy;
+         }
+      }
+      if (lx < n2) {
+         // optimization of: draw_line(target, lx,ly, n,ly, n2);
+         for (j=lx; j < n2; ++j)
+            LINE_OP(target[j], inverse_db_table[ly]);
+         CHECK(f);
+      }
+   }
+   return TRUE;
+}
+
+// The meaning of "left" and "right"
+//
+// For a given frame:
+//     we compute samples from 0..n
+//     window_center is n/2
+//     we'll window and mix the samples from left_start to left_end with data from the previous frame
+//     all of the samples from left_end to right_start can be output without mixing; however,
+//        this interval is 0-length except when transitioning between short and long frames
+//     all of the samples from right_start to right_end need to be mixed with the next frame,
+//        which we don't have, so those get saved in a buffer
+//     frame N's right_end-right_start, the number of samples to mix with the next frame,
+//        has to be the same as frame N+1's left_end-left_start (which they are by
+//        construction)
+
+static int vorbis_decode_initial(vorb *f, int *p_left_start, int *p_left_end, int *p_right_start, int *p_right_end, int *mode)
+{
+   Mode *m;
+   int i, n, prev, next, window_center;
+   f->channel_buffer_start = f->channel_buffer_end = 0;
+
+  retry:
+   if (f->eof) return FALSE;
+   if (!maybe_start_packet(f))
+      return FALSE;
+   // check packet type
+   if (get_bits(f,1) != 0) {
+      if (IS_PUSH_MODE(f))
+         return error(f,VORBIS_bad_packet_type);
+      while (EOP != get8_packet(f));
+      goto retry;
+   }
+
+   if (f->alloc.alloc_buffer)
+      assert(f->alloc.alloc_buffer_length_in_bytes == f->temp_offset);
+
+   i = get_bits(f, ilog(f->mode_count-1));
+   if (i == EOP) return FALSE;
+   if (i >= f->mode_count) return FALSE;
+   *mode = i;
+   m = f->mode_config + i;
+   if (m->blockflag) {
+      n = f->blocksize_1;
+      prev = get_bits(f,1);
+      next = get_bits(f,1);
+   } else {
+      prev = next = 0;
+      n = f->blocksize_0;
+   }
+
+// WINDOWING
+
+   window_center = n >> 1;
+   if (m->blockflag && !prev) {
+      *p_left_start = (n - f->blocksize_0) >> 2;
+      *p_left_end   = (n + f->blocksize_0) >> 2;
+   } else {
+      *p_left_start = 0;
+      *p_left_end   = window_center;
+   }
+   if (m->blockflag && !next) {
+      *p_right_start = (n*3 - f->blocksize_0) >> 2;
+      *p_right_end   = (n*3 + f->blocksize_0) >> 2;
+   } else {
+      *p_right_start = window_center;
+      *p_right_end   = n;
+   }
+
+   return TRUE;
+}
+
+static int vorbis_decode_packet_rest(vorb *f, int *len, Mode *m, int left_start, int left_end, int right_start, int right_end, int *p_left)
+{
+   Mapping *map;
+   int i,j,k,n,n2;
+   int zero_channel[256];
+   int really_zero_channel[256];
+
+// WINDOWING
+
+   n = f->blocksize[m->blockflag];
+   map = &f->mapping[m->mapping];
+
+// FLOORS
+   n2 = n >> 1;
+
+   CHECK(f);
+
+   for (i=0; i < f->channels; ++i) {
+      int s = map->chan[i].mux, floor;
+      zero_channel[i] = FALSE;
+      floor = map->submap_floor[s];
+      if (f->floor_types[floor] == 0) {
+         return error(f, VORBIS_invalid_stream);
+      } else {
+         Floor1 *g = &f->floor_config[floor].floor1;
+         if (get_bits(f, 1)) {
+            short *finalY;
+            uint8 step2_flag[256];
+            static int range_list[4] = { 256, 128, 86, 64 };
+            int range = range_list[g->floor1_multiplier-1];
+            int offset = 2;
+            finalY = f->finalY[i];
+            finalY[0] = get_bits(f, ilog(range)-1);
+            finalY[1] = get_bits(f, ilog(range)-1);
+            for (j=0; j < g->partitions; ++j) {
+               int pclass = g->partition_class_list[j];
+               int cdim = g->class_dimensions[pclass];
+               int cbits = g->class_subclasses[pclass];
+               int csub = (1 << cbits)-1;
+               int cval = 0;
+               if (cbits) {
+                  Codebook *c = f->codebooks + g->class_masterbooks[pclass];
+                  DECODE(cval,f,c);
+               }
+               for (k=0; k < cdim; ++k) {
+                  int book = g->subclass_books[pclass][cval & csub];
+                  cval = cval >> cbits;
+                  if (book >= 0) {
+                     int temp;
+                     Codebook *c = f->codebooks + book;
+                     DECODE(temp,f,c);
+                     finalY[offset++] = temp;
+                  } else
+                     finalY[offset++] = 0;
+               }
+            }
+            if (f->valid_bits == INVALID_BITS) goto error; // behavior according to spec
+            step2_flag[0] = step2_flag[1] = 1;
+            for (j=2; j < g->values; ++j) {
+               int low, high, pred, highroom, lowroom, room, val;
+               low = g->neighbors[j][0];
+               high = g->neighbors[j][1];
+               //neighbors(g->Xlist, j, &low, &high);
+               pred = predict_point(g->Xlist[j], g->Xlist[low], g->Xlist[high], finalY[low], finalY[high]);
+               val = finalY[j];
+               highroom = range - pred;
+               lowroom = pred;
+               if (highroom < lowroom)
+                  room = highroom * 2;
+               else
+                  room = lowroom * 2;
+               if (val) {
+                  step2_flag[low] = step2_flag[high] = 1;
+                  step2_flag[j] = 1;
+                  if (val >= room)
+                     if (highroom > lowroom)
+                        finalY[j] = val - lowroom + pred;
+                     else
+                        finalY[j] = pred - val + highroom - 1;
+                  else
+                     if (val & 1)
+                        finalY[j] = pred - ((val+1)>>1);
+                     else
+                        finalY[j] = pred + (val>>1);
+               } else {
+                  step2_flag[j] = 0;
+                  finalY[j] = pred;
+               }
+            }
+
+#ifdef STB_VORBIS_NO_DEFER_FLOOR
+            do_floor(f, map, i, n, f->floor_buffers[i], finalY, step2_flag);
+#else
+            // defer final floor computation until _after_ residue
+            for (j=0; j < g->values; ++j) {
+               if (!step2_flag[j])
+                  finalY[j] = -1;
+            }
+#endif
+         } else {
+           error:
+            zero_channel[i] = TRUE;
+         }
+         // So we just defer everything else to later
+
+         // at this point we've decoded the floor into buffer
+      }
+   }
+   CHECK(f);
+   // 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;
+      }
+
+   CHECK(f);
+// 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);
+   CHECK(f);
+
+// INVERSE COUPLING
+   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;
+      }
+   }
+   CHECK(f);
+
+   // finish decoding the floors
+#ifndef STB_VORBIS_NO_DEFER_FLOOR
+   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
+   CHECK(f);
+   for (i=0; i < f->channels; ++i)
+      inverse_mdct(f->channel_buffers[i], n, f, m->blockflag);
+   CHECK(f);
+
+   // 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) {
+      if (f->discard_samples_deferred >= right_start - left_start) {
+         f->discard_samples_deferred -= (right_start - left_start);
+         left_start = right_start;
+         *p_left = left_start;
+      } else {
+         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-left_start)) {
+            if (current_end < f->current_loc) {
+               // negative truncation, that's impossible!
+               *len = 0;
+            } else {
+               *len = current_end - f->current_loc;
+            }
+            *len += left_start;
+            if (*len > right_end) *len = right_end; // this should never happen
+            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
+   CHECK(f);
+
+   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 int vorbis_pump_first_frame(stb_vorbis *f)
+{
+   int len, right, left, res;
+   res = vorbis_decode_packet(f, &len, &left, &right);
+   if (res)
+      vorbis_finish_frame(f, len, left, right);
+   return res;
+}
+
+#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;
+      CHECK(f);
+      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->dimensions == 0 && c->entries != 0)    return error(f, VORBIS_invalid_setup);
+
+      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;
+               if (lengths[j] == 32)
+                  return error(f, VORBIS_invalid_setup);
+            } 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);
+         if (c->codeword_lengths == NULL) return error(f, VORBIS_outofmem);
+         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;
+
+      CHECK(f);
+      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));
+         if (c->sorted_codewords == NULL) return error(f, VORBIS_outofmem);
+         // 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 == NULL) return error(f, VORBIS_outofmem);
+         ++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);
+
+      CHECK(f);
+      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;
+         }
+         if (c->lookup_values == 0) return error(f, VORBIS_invalid_setup);
+         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;
+            float last=0;
+            // 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) {
+               unsigned int z = sparse ? c->sorted_values[j] : j;
+               unsigned int div=1;
+               for (k=0; k < c->dimensions; ++k) {
+                  int off = (z / div) % c->lookup_values;
+                  float val = mults[off];
+                  val = mults[off]*c->delta_value + c->minimum_value + last;
+                  c->multiplicands[j*c->dimensions + k] = val;
+                  if (c->sequence_p)
+                     last = val;
+                  if (k+1 < c->dimensions) {
+                     if (div > UINT_MAX / (unsigned int) c->lookup_values) {
+                        setup_temp_free(f, mults,sizeof(mults[0])*c->lookup_values);
+                        return error(f, VORBIS_invalid_setup);
+                     }
+                     div *= c->lookup_values;
+                  }
+               }
+            }
+            c->lookup_type = 2;
+         }
+         else
+#endif
+         {
+            float last=0;
+            CHECK(f);
+            c->multiplicands = (codetype *) setup_malloc(f, sizeof(c->multiplicands[0]) * c->lookup_values);
+            if (c->multiplicands == NULL) { setup_temp_free(f, mults,sizeof(mults[0])*c->lookup_values); return error(f, VORBIS_outofmem); }
+            for (j=0; j < (int) c->lookup_values; ++j) {
+               float val = mults[j] * c->delta_value + c->minimum_value + last;
+               c->multiplicands[j] = val;
+               if (c->sequence_p)
+                  last = val;
+            }
+         }
+#ifndef STB_VORBIS_DIVIDES_IN_CODEBOOK
+        skip:;
+#endif
+         setup_temp_free(f, mults, sizeof(mults[0])*c->lookup_values);
+
+         CHECK(f);
+      }
+      CHECK(f);
+   }
+
+   // 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));
+   if (f->floor_config == NULL) return error(f, VORBIS_outofmem);
+   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 {
+         stbv__floor_ordering 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].id = j;
+         }
+         qsort(p, g->values, sizeof(p[0]), point_compare);
+         for (j=0; j < g->values; ++j)
+            g->sorted_order[j] = (uint8) p[j].id;
+         // 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[0]));
+   if (f->residue_config == NULL) return error(f, VORBIS_outofmem);
+   memset(f->residue_config, 0, f->residue_count * sizeof(f->residue_config[0]));
+   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);
+      if (r->end < r->begin) return error(f, VORBIS_invalid_setup);
+      r->part_size = get_bits(f,24)+1;
+      r->classifications = get_bits(f,6)+1;
+      r->classbook = get_bits(f,8);
+      if (r->classbook >= f->codebook_count) return error(f, VORBIS_invalid_setup);
+      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);
+      if (r->residue_books == NULL) return error(f, VORBIS_outofmem);
+      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);
+         if (r->classdata[j] == NULL) return error(f, VORBIS_outofmem);
+         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));
+   if (f->mapping == NULL) return error(f, VORBIS_outofmem);
+   memset(f->mapping, 0, 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 (m->chan == NULL) return error(f, VORBIS_outofmem);
+      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);
+      if (f->channel_buffers[i] == NULL || f->previous_window[i] == NULL || f->finalY[i] == NULL) return error(f, VORBIS_outofmem);
+      #ifdef STB_VORBIS_NO_DEFER_FLOOR
+      f->floor_buffers[i]   = (float *) setup_malloc(f, sizeof(float) * f->blocksize_1/2);
+      if (f->floor_buffers[i] == NULL) return error(f, VORBIS_outofmem);
+      #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;
+   if (p->residue_config) {
+      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) {
+      CHECK(p);
+      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);
+   if (p->mapping) {
+      for (i=0; i < p->mapping_count; ++i)
+         setup_free(p, p->mapping[i].chan);
+      setup_free(p, p->mapping);
+   }
+   CHECK(p);
+   for (i=0; i < p->channels && i < STB_VORBIS_MAX_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, const 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
+         const 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, (uint8 *) data, data_len);
+   }
+
+   f->stream     = (uint8 *) data;
+   f->stream_end = (uint8 *) 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 (int) (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 (int) (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 (int) (f->stream - data);
+}
+
+stb_vorbis *stb_vorbis_open_pushdata(
+         const unsigned char *data, int data_len, // the memory available for decoding
+         int *data_used,              // only defined if result is not NULL
+         int *error, const stb_vorbis_alloc *alloc)
+{
+   stb_vorbis *f, p;
+   vorbis_init(&p, alloc);
+   p.stream     = (uint8 *) data;
+   p.stream_end = (uint8 *) 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 = (int) (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 (unsigned int) (f->stream - f->stream_start);
+   #ifndef STB_VORBIS_NO_STDIO
+   return (unsigned int) (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 candidate
+         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);
+      }
+   }
+}
+
+
+#define SAMPLE_unknown  0xffffffff
+
+// seeking is implemented with a binary search, which narrows down the range to
+// 64K, before using a linear search (because finding the synchronization
+// pattern can be expensive, and the chance we'd find the end page again is
+// relatively high for small ranges)
+//
+// two initial interpolation-style probes are used at the start of the search
+// to try to bound either side of the binary search sensibly, while still
+// working in O(log n) time if they fail.
+
+static int get_seek_page_info(stb_vorbis *f, ProbedPage *z)
+{
+   uint8 header[27], lacing[255];
+   int i,len;
+
+   // record where the page starts
+   z->page_start = stb_vorbis_get_file_offset(f);
+
+   // parse the header
+   getn(f, header, 27);
+   if (header[0] != 'O' || header[1] != 'g' || header[2] != 'g' || header[3] != 'S')
+      return 0;
+   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] << 24);
+
+   // restore file state to where we were
+   set_file_offset(f, z->page_start);
+   return 1;
+}
+
+// rarely used function to seek back to the preceeding page while finding the
+// start of a packet
+static int go_to_page_before(stb_vorbis *f, unsigned int limit_offset)
+{
+   unsigned int previous_safe, end;
+
+   // now we want to seek back 64K from the limit
+   if (limit_offset >= 65536 && limit_offset-65536 >= f->first_audio_page_offset)
+      previous_safe = limit_offset - 65536;
+   else
+      previous_safe = f->first_audio_page_offset;
+
+   set_file_offset(f, previous_safe);
+
+   while (vorbis_find_page(f, &end, NULL)) {
+      if (end >= limit_offset && stb_vorbis_get_file_offset(f) < limit_offset)
+         return 1;
+      set_file_offset(f, end);
+   }
+
+   return 0;
+}
+
+// implements the search logic for finding a page and starting decoding. if
+// the function succeeds, current_loc_valid will be true and current_loc will
+// be less than or equal to the provided sample number (the closer the
+// better).
+static int seek_to_sample_coarse(stb_vorbis *f, uint32 sample_number)
+{
+   ProbedPage left, right, mid;
+   int i, start_seg_with_known_loc, end_pos, page_start;
+   uint32 delta, stream_length, padding;
+   double offset, bytes_per_sample;
+   int probe = 0;
+
+   // find the last page and validate the target sample
+   stream_length = stb_vorbis_stream_length_in_samples(f);
+   if (stream_length == 0)            return error(f, VORBIS_seek_without_length);
+   if (sample_number > stream_length) return error(f, VORBIS_seek_invalid);
+
+   // this is the maximum difference between the window-center (which is the
+   // actual granule position value), and the right-start (which the spec
+   // indicates should be the granule position (give or take one)).
+   padding = ((f->blocksize_1 - f->blocksize_0) >> 2);
+   if (sample_number < padding)
+      sample_number = 0;
+   else
+      sample_number -= padding;
+
+   left = f->p_first;
+   while (left.last_decoded_sample == ~0U) {
+      // (untested) the first page does not have a 'last_decoded_sample'
+      set_file_offset(f, left.page_end);
+      if (!get_seek_page_info(f, &left)) goto error;
+   }
+
+   right = f->p_last;
+   assert(right.last_decoded_sample != ~0U);
+
+   // starting from the start is handled differently
+   if (sample_number <= left.last_decoded_sample) {
+      if (stb_vorbis_seek_start(f))
+         return 1;
+      return 0;
+   }
+
+   while (left.page_end != right.page_start) {
+      assert(left.page_end < right.page_start);
+      // search range in bytes
+      delta = right.page_start - left.page_end;
+      if (delta <= 65536) {
+         // there's only 64K left to search - handle it linearly
+         set_file_offset(f, left.page_end);
+      } else {
+         if (probe < 2) {
+            if (probe == 0) {
+               // first probe (interpolate)
+               double data_bytes = right.page_end - left.page_start;
+               bytes_per_sample = data_bytes / right.last_decoded_sample;
+               offset = left.page_start + bytes_per_sample * (sample_number - left.last_decoded_sample);
+            } else {
+               // second probe (try to bound the other side)
+               double error = ((double) sample_number - mid.last_decoded_sample) * bytes_per_sample;
+               if (error >= 0 && error <  8000) error =  8000;
+               if (error <  0 && error > -8000) error = -8000;
+               offset += error * 2;
+            }
+
+            // ensure the offset is valid
+            if (offset < left.page_end)
+               offset = left.page_end;
+            if (offset > right.page_start - 65536)
+               offset = right.page_start - 65536;
+
+            set_file_offset(f, (unsigned int) offset);
+         } else {
+            // binary search for large ranges (offset by 32K to ensure
+            // we don't hit the right page)
+            set_file_offset(f, left.page_end + (delta / 2) - 32768);
+         }
+
+         if (!vorbis_find_page(f, NULL, NULL)) goto error;
+      }
+
+      for (;;) {
+         if (!get_seek_page_info(f, &mid)) goto error;
+         if (mid.last_decoded_sample != ~0U) break;
+         // (untested) no frames end on this page
+         set_file_offset(f, mid.page_end);
+         assert(mid.page_start < right.page_start);
+      }
+
+      // if we've just found the last page again then we're in a tricky file,
+      // and we're close enough.
+      if (mid.page_start == right.page_start)
+         break;
+
+      if (sample_number < mid.last_decoded_sample)
+         right = mid;
+      else
+         left = mid;
+
+      ++probe;
+   }
+
+   // seek back to start of the last packet
+   page_start = left.page_start;
+   set_file_offset(f, page_start);
+   if (!start_page(f)) return error(f, VORBIS_seek_failed);
+   end_pos = f->end_seg_with_known_loc;
+   assert(end_pos >= 0);
+
+   for (;;) {
+      for (i = end_pos; i > 0; --i)
+         if (f->segments[i-1] != 255)
+            break;
+
+      start_seg_with_known_loc = i;
+
+      if (start_seg_with_known_loc > 0 || !(f->page_flag & PAGEFLAG_continued_packet))
+         break;
+
+      // (untested) the final packet begins on an earlier page
+      if (!go_to_page_before(f, page_start))
+         goto error;
+
+      page_start = stb_vorbis_get_file_offset(f);
+      if (!start_page(f)) goto error;
+      end_pos = f->segment_count - 1;
+   }
+
+   // prepare to start decoding
+   f->current_loc_valid = FALSE;
+   f->last_seg = FALSE;
+   f->valid_bits = 0;
+   f->packet_bytes = 0;
+   f->bytes_in_seg = 0;
+   f->previous_length = 0;
+   f->next_seg = start_seg_with_known_loc;
+
+   for (i = 0; i < start_seg_with_known_loc; i++)
+      skip(f, f->segments[i]);
+
+   // start decoding (optimizable - this frame is generally discarded)
+   if (!vorbis_pump_first_frame(f))
+      return 0;
+   if (f->current_loc > sample_number)
+      return error(f, VORBIS_seek_failed);
+   return 1;
+
+error:
+   // try to restore the file to a valid state
+   stb_vorbis_seek_start(f);
+   return error(f, VORBIS_seek_failed);
+}
+
+// the same as vorbis_decode_initial, but without advancing
+static int peek_decode_initial(vorb *f, int *p_left_start, int *p_left_end, int *p_right_start, int *p_right_end, int *mode)
+{
+   int bits_read, bytes_read;
+
+   if (!vorbis_decode_initial(f, p_left_start, p_left_end, p_right_start, p_right_end, mode))
+      return 0;
+
+   // either 1 or 2 bytes were read, figure out which so we can rewind
+   bits_read = 1 + ilog(f->mode_count-1);
+   if (f->mode_config[*mode].blockflag)
+      bits_read += 2;
+   bytes_read = (bits_read + 7) / 8;
+
+   f->bytes_in_seg += bytes_read;
+   f->packet_bytes -= bytes_read;
+   skip(f, -bytes_read);
+   if (f->next_seg == -1)
+      f->next_seg = f->segment_count - 1;
+   else
+      f->next_seg--;
+   f->valid_bits = 0;
+
+   return 1;
+}
+
+int stb_vorbis_seek_frame(stb_vorbis *f, unsigned int sample_number)
+{
+   uint32 max_frame_samples;
+
+   if (IS_PUSH_MODE(f)) return error(f, VORBIS_invalid_api_mixing);
+
+   // fast page-level search
+   if (!seek_to_sample_coarse(f, sample_number))
+      return 0;
+
+   assert(f->current_loc_valid);
+   assert(f->current_loc <= sample_number);
+
+   // linear search for the relevant packet
+   max_frame_samples = (f->blocksize_1*3 - f->blocksize_0) >> 2;
+   while (f->current_loc < sample_number) {
+      int left_start, left_end, right_start, right_end, mode, frame_samples;
+      if (!peek_decode_initial(f, &left_start, &left_end, &right_start, &right_end, &mode))
+         return error(f, VORBIS_seek_failed);
+      // calculate the number of samples returned by the next frame
+      frame_samples = right_start - left_start;
+      if (f->current_loc + frame_samples > sample_number) {
+         return 1; // the next frame will contain the sample
+      } else if (f->current_loc + frame_samples + max_frame_samples > sample_number) {
+         // there's a chance the frame after this could contain the sample
+         vorbis_pump_first_frame(f);
+      } else {
+         // this frame is too early to be relevant
+         f->current_loc += frame_samples;
+         f->previous_length = 0;
+         maybe_start_packet(f);
+         flush_packet(f);
+      }
+   }
+   // the next frame will start with the sample
+   assert(f->current_loc == sample_number);
+   return 1;
+}
+
+int stb_vorbis_seek(stb_vorbis *f, unsigned int sample_number)
+{
+   if (!stb_vorbis_seek_frame(f, sample_number))
+      return 0;
+
+   if (sample_number != f->current_loc) {
+      int n;
+      uint32 frame_start = f->current_loc;
+      stb_vorbis_get_frame_float(f, &n, NULL);
+      assert(sample_number > frame_start);
+      assert(f->channel_buffer_start + (int) (sample_number-frame_start) <= f->channel_buffer_end);
+      f->channel_buffer_start += (sample_number - frame_start);
+   }
+
+   return 1;
+}
+
+int stb_vorbis_seek_start(stb_vorbis *f)
+{
+   if (IS_PUSH_MODE(f)) { return error(f, VORBIS_invalid_api_mixing); }
+   set_file_offset(f, f->first_audio_page_offset);
+   f->previous_length = 0;
+   f->first_decode = TRUE;
+   f->next_seg = -1;
+   return 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;
+
+     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, const stb_vorbis_alloc *alloc, unsigned int length)
+{
+   stb_vorbis *f, p;
+   vorbis_init(&p, alloc);
+   p.f = file;
+   p.f_start = (uint32) 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, const stb_vorbis_alloc *alloc)
+{
+   unsigned int len, start;
+   start = (unsigned int) ftell(file);
+   fseek(file, 0, SEEK_END);
+   len = (unsigned int) (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, const 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, const 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);
+         if (error) *error = VORBIS__no_error;
+         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
+    1.10    - 2017/03/03 - more robust seeking; fix negative ilog(); clear error in open_memory
+    1.09    - 2016/04/04 - back out 'avoid discarding last frame' fix from previous version
+    1.08    - 2016/04/02 - fixed multiple warnings; fix setup memory leaks;
+                           avoid discarding last frame of audio data
+    1.07    - 2015/01/16 - fixed some warnings, fix mingw, const-correct API
+                           some more crash fixes when out of memory or with corrupt files 
+    1.06    - 2015/08/31 - full, correct support for seeking API (Dougall Johnson)
+                           some crash fixes when out of memory or with corrupt files
+    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
+
+
+/*
+------------------------------------------------------------------------------
+This software is available under 2 licenses -- choose whichever you prefer.
+------------------------------------------------------------------------------
+ALTERNATIVE A - MIT License
+Copyright (c) 2017 Sean Barrett
+Permission is hereby granted, free of charge, to any person obtaining a copy of 
+this software and associated documentation files (the "Software"), to deal in 
+the Software without restriction, including without limitation the rights to 
+use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies 
+of the Software, and to permit persons to whom the Software is furnished to do 
+so, subject to the following conditions:
+The above copyright notice and this permission notice shall be included in all 
+copies or substantial portions of the Software.
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 
+SOFTWARE.
+------------------------------------------------------------------------------
+ALTERNATIVE B - Public Domain (www.unlicense.org)
+This is free and unencumbered software released into the public domain.
+Anyone is free to copy, modify, publish, use, compile, sell, or distribute this 
+software, either in source code form or as a compiled binary, for any purpose, 
+commercial or non-commercial, and by any means.
+In jurisdictions that recognize copyright laws, the author or authors of this 
+software dedicate any and all copyright interest in the software to the public 
+domain. We make this dedication for the benefit of the public at large and to 
+the detriment of our heirs and successors. We intend this dedication to be an 
+overt act of relinquishment in perpetuity of all present and future rights to 
+this software under copyright law.
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 
+AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 
+ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 
+WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+------------------------------------------------------------------------------
+*/

Source/ThirdParty/imgui/.travis.yml

@@ -1,18 +0,0 @@
-language: cpp
-
-os:
-  - linux
-  - osx
-
-compiler:
-  - gcc
-  - clang
-
-before_install:
-  - if [ $TRAVIS_OS_NAME == linux ]; then sudo add-apt-repository -y ppa:pyglfw/pyglfw && sudo apt-get update -qq && sudo apt-get install -y --no-install-recommends libglfw3-dev libxrandr-dev libxi-dev libxxf86vm-dev; fi
-  - if [ $TRAVIS_OS_NAME == osx ]; then brew update && brew install glfw3; fi
-
-script: 
-  - make -C examples/opengl2_example
-  - make -C examples/opengl3_example
-

Source/ThirdParty/imgui/imgui.cpp

@@ -3138,6 +3138,14 @@ bool ImGui::IsMouseHoveringAnyWindow()
     return g.HoveredWindow != NULL;
 }
 
+// ATOMIC BEGIN
+bool ImGui::IsAnyWindowFocused()
+{
+    ImGuiContext& g = *GImGui;
+    return g.FocusedWindow != 0;
+}
+// ATOMIC END
+
 bool ImGui::IsPosHoveringAnyWindow(const ImVec2& pos)
 {
     return FindHoveredWindow(pos, false) != NULL;

Source/ThirdParty/imgui/imgui.h

@@ -437,6 +437,9 @@ namespace ImGui
     IMGUI_API bool          IsMouseReleased(int button);                                        // did mouse button released (went from Down to !Down)
     IMGUI_API bool          IsMouseHoveringWindow();                                            // is mouse hovering current window ("window" in API names always refer to current window). disregarding of any consideration of being blocked by a popup. (unlike IsWindowHovered() this will return true even if the window is blocked because of a popup)
     IMGUI_API bool          IsMouseHoveringAnyWindow();                                         // is mouse hovering any visible window
+// ATOMIC BEGIN
+    IMGUI_API bool          IsAnyWindowFocused();                                               // is any window (visible or not) focused
+// ATOMIC END
     IMGUI_API bool          IsMouseHoveringRect(const ImVec2& r_min, const ImVec2& r_max, bool clip = true);  // is mouse hovering given bounding rect (in screen space). clipped by current clipping settings. disregarding of consideration of focus/window ordering/blocked by a popup.
     IMGUI_API bool          IsMouseDragging(int button = 0, float lock_threshold = -1.0f);      // is mouse dragging. if lock_threshold < -1.0f uses io.MouseDraggingThreshold
     IMGUI_API ImVec2        GetMousePos();                                                      // shortcut to ImGui::GetIO().MousePos provided by user, to be consistent with other calls