Merge branch 'development' of https://github.com/TorqueGameEngines/Torque3D into alpha41/scriptStability

Engine/source/CMakeLists.txt

@@ -82,6 +82,9 @@ endif()
 torqueAddSourceDirectories("gfx" "gfx/Null" "gfx/test" "gfx/bitmap" "gfx/bitmap/loaders"
                              "gfx/util" "gfx/video" "gfx/sim" )
 
+# add the stb headers
+set(TORQUE_INCLUDE_DIRECTORIES ${TORQUE_INCLUDE_DIRECTORIES} "gfx/bitmap/loaders/stb")
+
 if (TORQUE_OPENGL)
   torqueAddSourceDirectories("gfx/gl" "gfx/gl/sdl" "gfx/gl/tGL")
 endif (TORQUE_OPENGL)

Engine/source/T3D/rigid.cpp

@@ -156,7 +156,7 @@ bool Rigid::resolveCollision(const Point3F& p, const Point3F &normal, Rigid* rig
       return false;
 
    // Compute impulse
-   F32 d, n = -nv * (2.0f + restitution * rigid->restitution);
+   F32 d, n = -nv * (1.0+(restitution + rigid->restitution)*0.5);
    Point3F a1,b1,c1;
    mCross(r1,normal,&a1);
    invWorldInertia.mulV(a1,&b1);

Engine/source/T3D/rigidShape.cpp

@@ -900,9 +900,11 @@ bool RigidShape::onNewDataBlock(GameBaseData* dptr, bool reload)
    mRigid.restitution = mDataBlock->body.restitution;
    mRigid.setCenterOfMass(mDataBlock->massCenter);
 
-   // Ignores massBox, just set sphere for now. Derived objects
-   // can set what they want.
-   mRigid.setObjectInertia();
+   // Set inertial tensor, default for the RigidShape is sphere
+   if (mDataBlock->massBox.x > 0 && mDataBlock->massBox.y > 0 && mDataBlock->massBox.z > 0)
+      mRigid.setObjectInertia(mDataBlock->massBox);
+   else
+      mRigid.setObjectInertia(mObjBox.maxExtents - mObjBox.minExtents);
 
    scriptOnNewDataBlock();
 
@@ -1115,9 +1117,9 @@ void RigidShape::updatePos(F32 dt)
       if (mCollisionList.getCount())
       {
          F32 k = mRigid.getKineticEnergy();
-         F32 G = mNetGravity * dt;
+         F32 G = mNetGravity;
          F32 Kg = 0.5 * mRigid.mass * G * G;
-         if (k < sRestTol * Kg && ++restCount > sRestCount)
+         if (k < sRestTol * Kg* dt && ++restCount > sRestCount)
             mRigid.setAtRest();
       }
       else
@@ -1205,8 +1207,12 @@ void RigidShape::updatePos(F32 dt)
 
 void RigidShape::updateForces(F32 dt)
 {
-   if (mDisableMove) return;
-
+   if (mDisableMove)
+   {
+      mRigid.linVelocity = Point3F::Zero;
+      mRigid.angMomentum = Point3F::Zero;
+      return;
+   }
    Point3F torque(0, 0, 0);
    Point3F force(0, 0, mRigid.mass * mNetGravity);
 
@@ -1222,10 +1228,6 @@ void RigidShape::updateForces(F32 dt)
 
    mRigid.force  = force;
    mRigid.torque = torque;
-
-   // If we're still atRest, make sure we're not accumulating anything
-   if ((force.lenSquared() < mDataBlock->contactTol)&& (force.lenSquared() < mDataBlock->contactTol))
-      mRigid.setAtRest();
 }
 
 
@@ -1260,8 +1262,7 @@ bool RigidShape::updateCollision(F32 dt)
    }
 
    // Resolve collisions
-   bool collided = resolveCollision(mRigid,mCollisionList);
-   resolveContacts(mRigid,mCollisionList,dt);
+   bool collided = resolveCollision(mRigid,mCollisionList, dt);
    return collided;
 }
 
@@ -1271,7 +1272,7 @@ bool RigidShape::updateCollision(F32 dt)
 Handle collision impacts, as opposed to contacts. Impulses are calculated based
 on standard collision resolution formulas.
 */
-bool RigidShape::resolveCollision(Rigid&  ns,CollisionList& cList)
+bool RigidShape::resolveCollision(Rigid&  ns,CollisionList& cList, F32 dt)
 {
    PROFILE_SCOPE(RigidShape_resolveCollision);
    // Apply impulses to resolve collision
@@ -1279,6 +1280,11 @@ bool RigidShape::resolveCollision(Rigid&  ns,CollisionList& cList)
    for (S32 i = 0; i < cList.getCount(); i++)
    {
       Collision& c = cList[i];
+      if (c.object == this)
+      {
+         //Con::printf("IMPOSSIBLE!!!!--------------------------------> Self-collision event?");
+         continue;
+      }
       if (c.distance < mDataBlock->collisionTol)
       {
          // Velocity into surface
@@ -1288,9 +1294,7 @@ bool RigidShape::resolveCollision(Rigid&  ns,CollisionList& cList)
          F32 vn = mDot(v, c.normal);
 
          // Only interested in velocities greater than sContactTol,
-         // velocities less than that will be dealt with as contacts
-         // "constraints".
-         if (vn < -mDataBlock->contactTol)
+         if (mFabs(vn) > mDataBlock->contactTol)
          {
 
             // Apply impulses to the rigid body to keep it from
@@ -1313,45 +1317,15 @@ bool RigidShape::resolveCollision(Rigid&  ns,CollisionList& cList)
                queueCollision(col, v - col->getVelocity());
             }
          }
-      }
-   }
-
-   return collided;
-}
-
-//----------------------------------------------------------------------------
-/** Resolve contact forces
-Resolve contact forces using the "penalty" method. Forces are generated based
-on the depth of penetration and the moment of inertia at the point of contact.
-*/
-bool RigidShape::resolveContacts(Rigid& ns,CollisionList& cList,F32 dt)
-{
-   PROFILE_SCOPE(RigidShape_resolveContacts);
-   // Use spring forces to manage contact constraints.
-   bool collided = false;
-   Point3F t,p(0,0,0),l(0,0,0);
-   for (S32 i = 0; i < cList.getCount(); i++) 
-   {
-      const Collision& c = cList[i];
-      if (c.distance < mDataBlock->collisionTol) 
-      {
-
-         // Velocity into the surface
-         Point3F v,r;
-         ns.getOriginVector(c.point,&r);
-         ns.getVelocity(r,&v);
-         F32 vn = mDot(v,c.normal);
-
-         // Only interested in velocities less than mDataBlock->contactTol,
-         // velocities greater than that are dealt with as collisions.
-         if (mFabs(vn) < mDataBlock->contactTol) 
+         // velocities less than that will be dealt with as contacts
+         // "constraints".
+         else
          {
-            collided = true;
-
+            Point3F t;
             // Penetration force. This is actually a spring which
             // will seperate the body from the collision surface.
-            F32 zi = 2 * mFabs(mRigid.getZeroImpulse(r,c.normal));
-            F32 s = (mDataBlock->collisionTol - c.distance) * zi - ((vn / mDataBlock->contactTol) * zi);
+            F32 zi = 2 * mFabs(mRigid.getZeroImpulse(r, c.normal) / dt);
+            F32 s = mMax((mDataBlock->collisionTol - c.distance) * zi - ((vn / 2.0) * zi),0.0f);
             Point3F f = c.normal * s;
 
             // Friction impulse, calculated as a function of the
@@ -1359,10 +1333,10 @@ bool RigidShape::resolveContacts(Rigid& ns,CollisionList& cList,F32 dt)
             // perpendicular to the normal.
             Point3F uv = v - (c.normal * vn);
             F32 ul = uv.len();
-            if (s > 0 && ul) 
+            if (s > 0 && ul)
             {
                uv /= -ul;
-               F32 u = ul * ns.getZeroImpulse(r,uv);
+               F32 u = ul * ns.getZeroImpulse(r, uv) / dt;
                s *= mRigid.friction;
                if (u > s)
                   u = s;
@@ -1370,20 +1344,17 @@ bool RigidShape::resolveContacts(Rigid& ns,CollisionList& cList,F32 dt)
             }
 
             // Accumulate forces
-            p += f;
-            mCross(r,f,&t);
-            l += t;
+            mCross(r, f, &t);
+            ns.linMomentum += f * dt;
+            ns.angMomentum += t * dt;
+            ns.updateVelocity();
          }
       }
    }
 
-   // Contact constraint forces act over time...
-   ns.linMomentum += p * dt;
-   ns.angMomentum += l * dt;
-   ns.updateVelocity();
-   return true;
-}
 
+   return collided;
+}
 
 //----------------------------------------------------------------------------
 
@@ -1693,6 +1664,8 @@ void RigidShape::initPersistFields()
    docsURL;
    addField("disableMove", TypeBool, Offset(mDisableMove, RigidShape),
       "When this flag is set, the vehicle will ignore throttle changes.");
+   addField("isAtRest", TypeBool, Offset(mRigid.atRest, RigidShape),
+      "Debug read of the rest state. do not set");   
    Parent::initPersistFields();
 }
 

Engine/source/T3D/rigidShape.h

@@ -216,8 +216,7 @@ class RigidShape: public ShapeBase
    bool onNewDataBlock( GameBaseData *dptr, bool reload );
    void updatePos(F32 dt);
    bool updateCollision(F32 dt);
-   bool resolveCollision(Rigid& ns,CollisionList& cList);
-   bool resolveContacts(Rigid& ns,CollisionList& cList,F32 dt);
+   bool resolveCollision(Rigid& ns,CollisionList& cList, F32 dt);
    bool resolveDisplacement(Rigid& ns,CollisionState *state,F32 dt);
    void checkTriggers();
    static void findCallback(SceneObject* obj,void * key);

Engine/source/T3D/shapeBase.h

@@ -91,7 +91,7 @@ class SFXProfile;
 
 typedef void* Light;
 
-const F32 gGravity = -20;
+const F32 gGravity = -9.8f;
 
 //--------------------------------------------------------------------------
 

Engine/source/T3D/vehicles/vehicle.cpp

@@ -807,9 +807,9 @@ void Vehicle::updatePos(F32 dt)
       if (mCollisionList.getCount()) 
       {
          F32 k = mRigid.getKineticEnergy();
-         F32 G = mNetGravity * dt;
+         F32 G = mNetGravity;
          F32 Kg = 0.5 * mRigid.mass * G * G;
-         if (k < sRestTol * Kg && ++restCount > sRestCount)
+         if (k < sRestTol * Kg * dt && ++restCount > sRestCount)
             mRigid.setAtRest();
       }
       else

Engine/source/console/simObject.cpp

@@ -40,6 +40,7 @@
 #include "core/fileObject.h"
 #include "persistence/taml/tamlCustom.h"
 #include "gui/editor/guiInspector.h"
+#include "console/script.h"
 
 #include "sim/netObject.h"
 
@@ -90,7 +91,7 @@ SimObject::SimObject()
    mNameSpace    = NULL;
    mNotifyList   = NULL;
    mFlags.set( ModStaticFields | ModDynamicFields );
-
+   mPrototype = true;
    mProgenitorFile = StringTable->EmptyString();
 
    mFieldDictionary = NULL;
@@ -159,7 +160,8 @@ void SimObject::initPersistFields()
 
       addProtectedField("inheritFrom", TypeString, Offset(mInheritFrom, SimObject), &setInheritFrom, &defaultProtectedGetFn,
          "Optional Name of object to inherit from as a parent.");
-                  
+
+      addProtectedField("Prototype", TypeBool, Offset(mPrototype, SimObject), &_doPrototype, &defaultProtectedGetFn, "Prototype Methods", AbstractClassRep::FieldFlags::FIELD_ComponentInspectors);
    endGroup( "Ungrouped" );
 
    addGroup( "Object" );
@@ -212,6 +214,15 @@ void SimObject::initPersistFields()
 }
 
 //-----------------------------------------------------------------------------
+bool SimObject::_doPrototype(void* object, const char* index, const char* data)
+{
+   if (!Con::isFunction("PrototypeClass")) return false;
+   if (dAtoi(data) != 1) return false;
+   SimObject* obj = reinterpret_cast<SimObject*>(object);
+   String command = String("PrototypeClass(") + (obj->getName()? String(obj->getName()) : String::ToString(obj->getId())) + ");";
+   Con::evaluate(command.c_str());
+   return false;
+}
 
 String SimObject::describeSelf() const
 {
@@ -2707,6 +2718,59 @@ DefineEngineMethod(SimObject, getMethodSigs, ArrayObject*, (bool commands), (fal
 
    return dictionary;
 }
+
+DefineEngineFunction(getMethodSigsNS, ArrayObject*, (StringTableEntry className, bool commands), (false),
+   "List the methods defined on this object.\n\n"
+   "Each description is a newline-separated vector with the following elements:\n"
+   "- method prototype string.\n"
+   "- Documentation string (not including prototype).  This takes up the remainder of the vector.\n"
+   "@return An ArrayObject populated with (name,description) pairs of all methods defined on the object.")
+{
+   
+   Namespace* ns = Con::lookupNamespace(className);
+   if (!ns)
+      return 0;
+
+   ArrayObject* dictionary = new ArrayObject();
+   dictionary->registerObject();
+
+   VectorPtr<Namespace::Entry*> vec(__FILE__, __LINE__);
+   ns->getEntryList(&vec);
+   for (Vector< Namespace::Entry* >::iterator j = vec.begin(); j != vec.end(); j++)
+   {
+      Namespace::Entry* e = *j;
+
+      if (commands)
+      {
+         if ((e->mType < Namespace::Entry::ConsoleFunctionType))
+            continue;
+      }
+      else
+      {
+         if ((e->mType > Namespace::Entry::ScriptCallbackType))
+            continue;
+      }
+      StringBuilder str;
+      str.append("function ");
+      str.append(ns->getName());
+      str.append("::");
+      str.append(e->getPrototypeSig());
+      str.append('\n');
+      str.append("{");
+      String docs = e->getDocString();
+      if (!docs.isEmpty())
+      {
+         str.append("\n/*");
+         str.append(docs);
+         str.append("\n*/");
+      }
+      str.append('\n');
+      str.append("}");
+      dictionary->push_back(e->mFunctionName, str.end());
+   }
+
+   return dictionary;
+}
 //-----------------------------------------------------------------------------
 
 namespace {
@@ -3262,6 +3326,33 @@ DefineEngineMethod( SimObject, getFieldCount, S32, (),,
    return list.size() - numDummyEntries;
 }
 
+DefineEngineFunction(getFieldCountNS, S32, (StringTableEntry className), ,
+   "Get the number of static fields on the name space.\n"
+   "@return The number of static fields defined on the object.")
+{
+   Namespace* ns = Con::lookupNamespace(className);
+   if (!ns)
+      return 0;
+   AbstractClassRep* rep = ns->mClassRep;
+   if (!rep)
+      return 0;
+
+   const AbstractClassRep::FieldList& list = rep->mFieldList;
+   const AbstractClassRep::Field* f;
+   U32 numDummyEntries = 0;
+
+   for (S32 i = 0; i < list.size(); i++)
+   {
+      f = &list[i];
+
+      // The special field types do not need to be counted.
+      if (f->type >= AbstractClassRep::ARCFirstCustomField || f->flag.test(AbstractClassRep::FieldFlags::FIELD_ComponentInspectors))
+         numDummyEntries++;
+   }
+
+   return list.size() - numDummyEntries;
+}
+
 //-----------------------------------------------------------------------------
 
 DefineEngineMethod( SimObject, getField, const char*, ( S32 index ),,
@@ -3293,6 +3384,42 @@ DefineEngineMethod( SimObject, getField, const char*, ( S32 index ),,
    return "";
 }
 
+DefineEngineFunction(getFieldNS, const char*, (StringTableEntry className,S32 index), ,
+   "Retrieve the value of a static field by index.\n"
+   "@param index The index of the static field.\n"
+   "@return The value of the static field with the given index or \"\".")
+{
+   Namespace* ns = Con::lookupNamespace(className);
+   if (!ns)
+      return 0;
+   AbstractClassRep* rep = ns->mClassRep;
+   if (!rep)
+      return 0;
+
+   const AbstractClassRep::FieldList& list = rep->mFieldList;
+   if ((index < 0) || (index >= list.size()))
+      return "";
+
+   const AbstractClassRep::Field* f;
+   S32 currentField = 0;
+   for (U32 i = 0; i < list.size() && currentField <= index; i++)
+   {
+      f = &list[i];
+
+      // The special field types can be skipped.
+      if (f->type >= AbstractClassRep::ARCFirstCustomField || f->flag.test(AbstractClassRep::FieldFlags::FIELD_ComponentInspectors))
+         continue;
+
+      if (currentField == index)
+         return f->pFieldname;
+
+      currentField++;
+   }
+
+   // if we found nada, return nada.
+   return "";
+}
+
 DefineEngineFunction(getClassHierarchy, const char*, (const char* name), ,
    "Returns the inheritance hierarchy for a given class.")
 {

Engine/source/console/simObject.h

@@ -300,7 +300,7 @@ class SimObject: public ConsoleObject, public TamlCallbacks
       SimObject*       nextIdObject;
 
       StringTableEntry mInheritFrom;
-
+      bool    mPrototype;
       /// SimGroup we're contained in, if any.
       SimGroup*   mGroup;
       
@@ -388,7 +388,7 @@ class SimObject: public ConsoleObject, public TamlCallbacks
    public:
       inline void setProgenitorFile(const char* pFile) { mProgenitorFile = StringTable->insert(pFile); }
       inline StringTableEntry getProgenitorFile(void) const { return mProgenitorFile; }
-
+      static bool _doPrototype(void* object, const char* index, const char* data);
    protected:
       /// Taml callbacks.
       virtual void onTamlPreWrite(void) {}

Engine/source/gfx/bitmap/ddsFile.cpp

@@ -565,15 +565,10 @@ void DDSFile::SurfaceData::dumpImage(DDSFile *dds, U32 mip, const char *file)
 
    // Copy our data in.
    dMemcpy(foo->getWritableBits(), mMips[mip], dds->getSurfaceSize(dds->mHeight, dds->mWidth, mip) );
-   
-   FileStream  stream;
-
-   stream.open( file, Torque::FS::File::Write );
 
-   if ( stream.getStatus() == Stream::Ok )
+   if(!foo->writeBitmap("png", file))
    {
-      // Write it out.
-      foo->writeBitmap("png", stream);
+      Con::errorf("DDSFile::SurfaceData::dumpImage() - Error writing %s !", file);
    }
 
    // Clean up.

Engine/source/gfx/bitmap/gBitmap.cpp

@@ -1192,7 +1192,7 @@ bool GBitmap::write(Stream& io_rStream) const
 //-------------------------------------- Persistent I/O
 //
 
-bool  GBitmap::readBitmap( const String &bmType, Stream &ioStream )
+bool  GBitmap::readBitmap(const String& bmType, const Torque::Path& path)
 {
    PROFILE_SCOPE(ResourceGBitmap_readBitmap);
    const GBitmap::Registration   *regInfo = GBitmap::sFindRegInfo( bmType );
@@ -1203,11 +1203,22 @@ bool  GBitmap::readBitmap( const String &bmType, Stream &ioStream )
       return false;
    }
 
-   return regInfo->readFunc( ioStream, this );
+   return regInfo->readFunc(path, this);
 }
 
-bool  GBitmap::writeBitmap( const String &bmType, Stream &ioStream, U32 compressionLevel )
+bool  GBitmap::writeBitmap( const String &bmType, const Torque::Path& path, U32 compressionLevel )
 {
+   FileStream stream;
+   if (!stream.open(path, Torque::FS::File::Write))
+   {
+      Con::errorf("GBitmap::writeBitmap failed to open path %s", path.getFullFileName().c_str());
+      stream.close();
+      return false;
+   }
+
+   // free file for stb
+   stream.close();
+
    const GBitmap::Registration   *regInfo = GBitmap::sFindRegInfo( bmType );
 
    if ( regInfo == NULL )
@@ -1216,7 +1227,7 @@ bool  GBitmap::writeBitmap( const String &bmType, Stream &ioStream, U32 compress
       return false;
    }
 
-   return regInfo->writeFunc( this, ioStream, (compressionLevel == U32_MAX) ? regInfo->defaultCompression : compressionLevel );
+   return regInfo->writeFunc(path, this, (compressionLevel == U32_MAX) ? regInfo->defaultCompression : compressionLevel );
 }
 
 template<> void *Resource<GBitmap>::create(const Torque::Path &path)
@@ -1239,7 +1250,7 @@ template<> void *Resource<GBitmap>::create(const Torque::Path &path)
 
    GBitmap *bmp = new GBitmap;
    const String extension = path.getExtension();
-   if( !bmp->readBitmap( extension, stream ) )
+   if( !bmp->readBitmap( extension, path ) )
    {
       Con::errorf( "Resource<GBitmap>::create - error reading '%s'", path.getFullPath().c_str() );
       delete bmp;
@@ -1431,21 +1442,14 @@ DefineEngineFunction(saveScaledImage, bool, (const char* bitmapSource, const cha
    Torque::Path destinationPath = Torque::Path(bitmapDest);
    destinationPath.setExtension("png");
 
-   // Open up the file on disk.
-   FileStream fs;
-   if (!fs.open(destinationPath.getFullPath(), Torque::FS::File::Write))
+   if(!image->writeBitmap("png", destinationPath.getFullPath()))
    {
-      Con::errorf("saveScaledImage() - Failed to open output file '%s'!", bitmapDest);
+      Con::errorf("saveScaledImage() - Error writing %s !", bitmapDest);
       delete image;
       return false;
    }
-   else
-   {
-      image->writeBitmap("png", fs);
-
-      fs.close();
-      delete image;
-   }
 
+   
+   delete image;
    return true;
 }

Engine/source/gfx/bitmap/gBitmap.h

@@ -71,10 +71,10 @@ public:
    struct Registration
    {
       /// The read function prototype.
-      typedef bool(*ReadFunc)(Stream &stream, GBitmap *bitmap);
+      typedef bool(*ReadFunc)(const Torque::Path& path, GBitmap* bitmap);
 
       /// The write function prototype.  Compression levels are image-specific - see their registration declaration for details.
-      typedef bool(*WriteFunc)(GBitmap *bitmap, Stream &stream, U32 compressionLevel);
+      typedef bool(*WriteFunc)(const Torque::Path& path, GBitmap* bitmap, U32 compressionLevel);
 
       /// Used to sort the registrations so that 
       /// lookups occur in a fixed order.
@@ -241,13 +241,16 @@ public:
    /// Read a bitmap from a stream
    /// @param bmType This is a file extension to describe the type of the data [i.e. "png" for PNG file, etc]
    /// @param ioStream The stream to read from
-   bool  readBitmap( const String &bmType, Stream &ioStream );
+   bool  readBitmap(const String& bmType, const Torque::Path& path);
 
    /// Write a bitmap to a stream
    /// @param bmType This is a file extension to describe the type of the data [i.e. "png" for PNG file, etc]
    /// @param ioStream The stream to read from
-   /// @param compressionLevel Image format-specific compression level.  If set to U32_MAX, we use the default compression defined when the format was registered.
-   bool  writeBitmap( const String &bmType, Stream &ioStream, U32 compressionLevel = U32_MAX );
+   /// @param compressionLevel Image format specific compression level. For JPEG sets the quality level percentage, range 0 to 100.
+   /// For PNG compression level is 0 - 10
+   /// Not used for other image formats.
+   
+   bool  writeBitmap( const String &bmType, const Torque::Path& path, U32 compressionLevel = U32_MAX );
 
    bool readMNG(Stream& io_rStream);               // located in bitmapMng.cc
    bool writeMNG(Stream& io_rStream) const;

Engine/source/gfx/bitmap/loaders/bitmapBmp.cpp

@@ -1,246 +0,0 @@
-//-----------------------------------------------------------------------------
-// Copyright (c) 2012 GarageGames, LLC
-//
-// 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.
-//-----------------------------------------------------------------------------
-
-#include "core/stream/stream.h"
-
-#include "gfx/bitmap/gBitmap.h"
-
-
-static bool sReadBMP(Stream &stream, GBitmap *bitmap);
-static bool sWriteBMP(GBitmap *bitmap, Stream &stream, U32 compressionLevel);
-
-static struct _privateRegisterBMP
-{
-   _privateRegisterBMP()
-   {
-      GBitmap::Registration reg;
-
-      reg.extensions.push_back( "bmp" );
-
-      reg.readFunc = sReadBMP;
-      reg.writeFunc = sWriteBMP;
-
-      GBitmap::sRegisterFormat( reg );
-   }
-} sStaticRegisterBMP;
-
-
-// structures mirror those defined by the win32 API
-
-struct RGBQUAD
-{
-   U8 rgbBlue;
-   U8 rgbGreen;
-   U8 rgbRed;
-   U8 rgbReserved;
-};
-
-struct BITMAPFILEHEADER
-{
-   U16 bfType;
-   U32 bfSize;
-   U16 bfReserved1;
-   U16 bfReserved2;
-   U32 bfOffBits;
-};
-
-struct BITMAPINFOHEADER
-{
-   U32 biSize;
-   S32 biWidth;
-   S32 biHeight;
-   U16 biPlanes;
-   U16 biBitCount;
-   U32 biCompression;
-   U32 biSizeImage;
-   S32 biXPelsPerMeter;
-   S32 biYPelsPerMeter;
-   U32 biClrUsed;
-   U32 biClrImportant;
-};
-
-// constants for the biCompression field
-#define BI_RGB        0L
-#define BI_RLE8       1L
-#define BI_RLE4       2L
-#define BI_BITFIELDS  3L
-
-
-//------------------------------------------------------------------------------
-//-------------------------------------- Supplementary I/O (Partially located in
-//                                                          bitmapPng.cc)
-//
-
-static bool sReadBMP(Stream &stream, GBitmap *bitmap)
-{
-   PROFILE_SCOPE(sReadBMP);
-   BITMAPINFOHEADER  bi;
-   BITMAPFILEHEADER  bf;
-   RGBQUAD           rgb[256];
-
-   stream.read(&bf.bfType);
-   stream.read(&bf.bfSize);
-   stream.read(&bf.bfReserved1);
-   stream.read(&bf.bfReserved2);
-   stream.read(&bf.bfOffBits);
-
-   stream.read(&bi.biSize);
-   stream.read(&bi.biWidth);
-   stream.read(&bi.biHeight);
-   stream.read(&bi.biPlanes);
-   stream.read(&bi.biBitCount);
-   stream.read(&bi.biCompression);
-   stream.read(&bi.biSizeImage);
-   stream.read(&bi.biXPelsPerMeter);
-   stream.read(&bi.biYPelsPerMeter);
-   stream.read(&bi.biClrUsed);
-   stream.read(&bi.biClrImportant);
-
-   GFXFormat fmt = GFXFormatR8G8B8;
-   if(bi.biBitCount == 8)
-   {
-      // read in texture palette
-      if(!bi.biClrUsed)
-         bi.biClrUsed = 256;
-      stream.read(sizeof(RGBQUAD) * bi.biClrUsed, rgb);
-   }
-   bitmap->allocateBitmap(bi.biWidth, bi.biHeight, false, fmt);
-   U32   width  = bitmap->getWidth();
-   U32   height = bitmap->getHeight();
-   U32   bytesPerPixel = bitmap->getBytesPerPixel();
-
-   for(U32 i = 0; i < bi.biHeight; i++)
-   {
-      U8 *rowDest = bitmap->getAddress(0, height - i - 1);
-      if (bi.biBitCount == 8)
-      {
-         // use palette...don't worry about being slow
-         for (S32 j=0; j<width; j++)
-         {
-            U8 palIdx;
-            stream.read(&palIdx);
-            U8 * pixelLocation = &rowDest[j*bytesPerPixel];
-            pixelLocation[0] = rgb[palIdx].rgbRed;
-            pixelLocation[1] = rgb[palIdx].rgbGreen;
-            pixelLocation[2] = rgb[palIdx].rgbBlue;
-            if (bytesPerPixel==3)
-               pixelLocation[3] = 255;
-         }
-      }
-      else
-         stream.read(bytesPerPixel * width, rowDest);
-   }
-
-   if(bytesPerPixel == 3 && bi.biBitCount != 8) // do BGR swap
-   {
-      U8 *ptr = bitmap->getAddress(0,0);
-      for(S32 i = 0; i < width * height; i++)
-      {
-         U8 tmp = ptr[0];
-         ptr[0] = ptr[2];
-         ptr[2] = tmp;
-         ptr += 3;
-      }
-   }
-
-   // We know BMP's don't have any transparency
-   bitmap->setHasTransparency(false);
-
-   return true;
-}
-
-static bool sWriteBMP(GBitmap *bitmap, Stream &stream, U32 compressionLevel)
-{
-   TORQUE_UNUSED( compressionLevel );  // BMP does not use compression
-
-   BITMAPINFOHEADER  bi;
-   BITMAPFILEHEADER  bf;
-
-   bi.biSize            = sizeof(BITMAPINFOHEADER);
-   bi.biWidth           = bitmap->getWidth();
-   bi.biHeight          = bitmap->getHeight();         //our data is top-down
-   bi.biPlanes = 1;
-
-   if(bitmap->getFormat() == GFXFormatR8G8B8)
-   {
-      bi.biBitCount = 24;
-      bi.biCompression = BI_RGB;
-      bi.biClrUsed = 0;
-   }
-   else
-   {
-      bi.biBitCount = 0;
-      bi.biCompression = BI_RGB; // Removes warning C4701 on line
-      AssertISV(false, "GBitmap::writeMSBmp - only support R8G8B8 formats!");
-   }
-
-   U32   width  = bitmap->getWidth();
-   U32   height = bitmap->getHeight();
-
-   U32 bytesPP = bi.biBitCount >> 3;
-   bi.biSizeImage       = width * height * bytesPP;
-   bi.biXPelsPerMeter   = 0;
-   bi.biYPelsPerMeter   = 0;
-   bi.biClrUsed         = 0;
-   bi.biClrImportant    = 0;
-
-   bf.bfType   = makeFourCCTag('B','M',0,0);     //Type of file 'BM'
-   bf.bfOffBits= sizeof(BITMAPINFOHEADER)
-               + sizeof(BITMAPFILEHEADER)
-               + (sizeof(RGBQUAD)*bi.biClrUsed);
-   bf.bfSize            = bf.bfOffBits + bi.biSizeImage;
-   bf.bfReserved1       = 0;
-   bf.bfReserved2       = 0;
-
-   stream.write(bf.bfType);
-   stream.write(bf.bfSize);
-   stream.write(bf.bfReserved1);
-   stream.write(bf.bfReserved2);
-   stream.write(bf.bfOffBits);
-
-   stream.write(bi.biSize);
-   stream.write(bi.biWidth);
-   stream.write(bi.biHeight);
-   stream.write(bi.biPlanes);
-   stream.write(bi.biBitCount);
-   stream.write(bi.biCompression);
-   stream.write(bi.biSizeImage);
-   stream.write(bi.biXPelsPerMeter);
-   stream.write(bi.biYPelsPerMeter);
-   stream.write(bi.biClrUsed);
-   stream.write(bi.biClrImportant);
-
-   //write the bitmap bits
-   U8* pMSUpsideDownBits = new U8[bi.biSizeImage];
-   for (U32 i = 0; i < height; i++)
-   {
-      const U8* pSrc = bitmap->getAddress(0, i);
-      U8* pDst = pMSUpsideDownBits + (height - i - 1) * width * bytesPP;
-
-      dMemcpy(pDst, pSrc, width * bytesPP);
-   }
-
-   stream.write(bi.biSizeImage, pMSUpsideDownBits);
-   delete [] pMSUpsideDownBits;
-
-   return stream.getStatus() == Stream::Ok;
-}

Engine/source/gfx/bitmap/loaders/bitmapJpeg.cpp

@@ -1,251 +0,0 @@
-//-----------------------------------------------------------------------------
-// Copyright (c) 2012 GarageGames, LLC
-//
-// 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.
-//-----------------------------------------------------------------------------
-
-#include "ljpeg/jpeglib.h"
-
-#include "core/stream/stream.h"
-
-#include "gfx/bitmap/gBitmap.h"
-
-
-static bool sReadJPG(Stream &stream, GBitmap *bitmap);
-static bool sWriteJPG(GBitmap *bitmap, Stream &stream, U32 compressionLevel);
-
-static struct _privateRegisterJPG
-{
-   _privateRegisterJPG()
-   {
-      GBitmap::Registration reg;
-
-      reg.priority = 50;
-      reg.extensions.push_back( "jpeg" );
-      reg.extensions.push_back( "jpg" );
-
-      reg.readFunc = sReadJPG;
-      reg.writeFunc = sWriteJPG;
-
-      GBitmap::sRegisterFormat( reg );
-   }
-} sStaticRegisterJPG;
-
-//-------------------------------------- Replacement I/O for standard LIBjpeg
-//                                        functions.  we don't wanna use
-//                                        FILE*'s...
-static S32 jpegReadDataFn(void *client_data, U8 *data, S32 length)
-{
-   Stream *stream = (Stream*)client_data;
-   AssertFatal(stream != NULL, "jpegReadDataFn::No stream.");
-   S32 pos = stream->getPosition();
-   if (stream->read(length, data))
-      return length;
-
-   if (stream->getStatus() == Stream::EOS)
-      return (stream->getPosition()-pos);
-   else
-      return 0;
-}
-
-
-//--------------------------------------
-static S32 jpegWriteDataFn(void *client_data, U8 *data, S32 length)
-{
-   Stream *stream = (Stream*)client_data;
-   AssertFatal(stream != NULL, "jpegWriteDataFn::No stream.");
-   if (stream->write(length, data))
-      return length;
-   else
-      return 0;
-}
-
-
-//--------------------------------------
-static S32 jpegFlushDataFn(void *)
-{
-   // do nothing since we can't flush the stream object
-   return 0;
-}
-
-
-//--------------------------------------
-static S32 jpegErrorFn(void *client_data)
-{
-   Stream *stream = (Stream*)client_data;
-   AssertFatal(stream != NULL, "jpegErrorFn::No stream.");
-   return (stream->getStatus() != Stream::Ok);
-}
-
-
-//--------------------------------------
-static bool sReadJPG(Stream &stream, GBitmap *bitmap)
-{
-   PROFILE_SCOPE(sReadJPG);
-   JFREAD  = jpegReadDataFn;
-   JFERROR = jpegErrorFn;
-
-   jpeg_decompress_struct cinfo;
-   jpeg_error_mgr jerr;
-
-   // We set up the normal JPEG error routines, then override error_exit.
-   //cinfo.err = jpeg_std_error(&jerr.pub);
-   //jerr.pub.error_exit = my_error_exit;
-
-   // if (setjmp(jerr.setjmp_buffer))
-   // {
-   //    // If we get here, the JPEG code has signaled an error.
-   //    // We need to clean up the JPEG object, close the input file, and return.
-   //    jpeg_destroy_decompress(&cinfo);
-   //    return false;
-   // }
-
-
-   cinfo.err = jpeg_std_error(&jerr);    // set up the normal JPEG error routines.
-   cinfo.client_data = (void*)&stream;       // set the stream into the client_data
-
-   // Now we can initialize the JPEG decompression object.
-   jpeg_create_decompress(&cinfo);
-
-   jpeg_stdio_src(&cinfo);
-
-   // Read file header, set default decompression parameters
-   jpeg_read_header(&cinfo, true);
-
-   GFXFormat format;
-   switch (cinfo.out_color_space)
-   {
-      case JCS_GRAYSCALE:  format = GFXFormatA8; break;
-      case JCS_RGB:        format = GFXFormatR8G8B8;   break;
-      default:
-         jpeg_destroy_decompress(&cinfo);
-         return false;
-   }
-
-   // Start decompressor
-   jpeg_start_decompress(&cinfo);
-
-   // allocate the bitmap space and init internal variables...
-   bitmap->allocateBitmap(cinfo.output_width, cinfo.output_height, false, format);
-
-   // Set up the row pointers...
-   U32 rowBytes = cinfo.output_width * cinfo.output_components;
-
-   U8* pBase = (U8*)bitmap->getBits();
-   for (U32 i = 0; i < bitmap->getHeight(); i++)
-   {
-      JSAMPROW rowPointer = pBase + (U64)(i * rowBytes);
-      jpeg_read_scanlines(&cinfo, &rowPointer, 1);
-   }
-
-   // Finish decompression
-   jpeg_finish_decompress(&cinfo);
-
-   // Release JPEG decompression object
-   // This is an important step since it will release a good deal of memory.
-   jpeg_destroy_decompress(&cinfo);
-
-   // We know JPEG's don't have any transparency
-   bitmap->setHasTransparency(false);
-
-   return true;
-}
-
-
-//--------------------------------------------------------------------------
-static bool sWriteJPG(GBitmap *bitmap, Stream &stream, U32 compressionLevel)
-{
-   TORQUE_UNUSED(compressionLevel); // compression level not currently hooked up
-
-   GFXFormat   format = bitmap->getFormat();
-
-   // JPEG format does not support transparency so any image
-   // in Alpha format should be saved as a grayscale which coincides
-   // with how the readJPEG function will read-in a JPEG. So the
-   // only formats supported are RGB and Alpha, not RGBA.
-   AssertFatal(format == GFXFormatR8G8B8 || format == GFXFormatA8,
-	            "GBitmap::writeJPEG: ONLY RGB bitmap writing supported at this time.");
-   if (format != GFXFormatR8G8B8 && format != GFXFormatA8)
-      return false;
-
-   // maximum image size allowed
-   #define MAX_HEIGHT 4096
-   if (bitmap->getHeight() > MAX_HEIGHT)
-      return false;
-
-   // Bind our own stream writing, error, and memory flush functions
-   // to the jpeg library interface
-   JFWRITE = jpegWriteDataFn;
-   JFFLUSH = jpegFlushDataFn;
-   JFERROR = jpegErrorFn;
-
-   // Allocate and initialize our jpeg compression structure and error manager
-   jpeg_compress_struct cinfo;
-   jpeg_error_mgr jerr;
-
-   cinfo.err = jpeg_std_error(&jerr);    // set up the normal JPEG error routines.
-   cinfo.client_data = (void*)&stream;   // set the stream into the client_data
-   jpeg_create_compress(&cinfo);         // allocates a small amount of memory
-
-   // specify the destination for the compressed data(our stream)
-   jpeg_stdio_dest(&cinfo);
-
-   // set the image properties
-   cinfo.image_width = bitmap->getWidth();           // image width
-   cinfo.image_height = bitmap->getHeight();         // image height
-   cinfo.input_components = bitmap->getBytesPerPixel();   // samples per pixel(RGB:3, Alpha:1)
-   
-   switch (format)
-   {
-      case GFXFormatA8:  // no alpha support in JPEG format, so turn it into a grayscale
-         cinfo.in_color_space = JCS_GRAYSCALE;
-         break;
-      case GFXFormatR8G8B8:    // otherwise we are writing in RGB format
-         cinfo.in_color_space = JCS_RGB;
-         break;
-      default:
-         AssertFatal( false, "Format not handled in GBitmap::writeJPEG() switch" );
-         break;
-   }
-   // use default compression params(75% compression)
-   jpeg_set_defaults(&cinfo);
-
-   // begin JPEG compression cycle
-   jpeg_start_compress(&cinfo, true);
-
-   // Set up the row pointers...
-   U32 rowBytes = cinfo.image_width * cinfo.input_components;
-
-   U8* pBase = (U8*)bitmap->getBits();
-   for (U32 i = 0; i < bitmap->getHeight(); i++)
-   {
-      // write the image data
-      JSAMPROW rowPointer = pBase + (i * rowBytes);
-	   jpeg_write_scanlines(&cinfo, &rowPointer, 1);
-   }
-
-   // complete the compression cycle
-   jpeg_finish_compress(&cinfo);
-
-   // release the JPEG compression object
-   jpeg_destroy_compress(&cinfo);
-
-   // return success
-   return true;
-}

Engine/source/gfx/bitmap/loaders/bitmapPng.cpp

@@ -1,645 +0,0 @@
-//-----------------------------------------------------------------------------
-// Copyright (c) 2012 GarageGames, LLC
-//
-// 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.
-//-----------------------------------------------------------------------------
-
-#include "platform/platform.h"
-
-#include "core/stream/fileStream.h"
-#include "core/stream/memStream.h"
-#include "core/strings/stringFunctions.h"
-
-#include "gfx/bitmap/gBitmap.h"
-#include "gfx/bitmap/pngUtils.h"
-
-#define PNG_INTERNAL 1
-#include <time.h>
-#include "lpng/png.h"
-#include "zlib/zlib.h"
-
-#ifdef NULL
-#undef NULL
-#define NULL 0
-#endif
-
-
-static bool sReadPNG(Stream &stream, GBitmap *bitmap);
-
-/// Compression levels for PNGs range from 0-9.
-/// A value outside that range will cause the write routine to look for the best compression for a given PNG.  This can be slow.
-static bool sWritePNG(GBitmap *bitmap, Stream &stream, U32 compressionLevel);
-static bool _writePNG(GBitmap *bitmap, Stream &stream, U32 compressionLevel, U32 strategy, U32 filter);
-
-static struct _privateRegisterPNG
-{
-   _privateRegisterPNG()
-   {
-      GBitmap::Registration reg;
-
-      reg.priority = 100;
-      reg.extensions.push_back( "png" );
-
-      reg.readFunc = sReadPNG;
-      reg.writeFunc = sWritePNG;
-      reg.defaultCompression = 6;
-
-      GBitmap::sRegisterFormat( reg );
-   }
-} sStaticRegisterPNG;
-
-
-//-------------------------------------- Replacement I/O for standard LIBPng
-//                                        functions.  we don't wanna use
-//                                        FILE*'s...
-static void pngReadDataFn(png_structp png_ptr,
-                    png_bytep   data,
-                    png_size_t  length)
-{
-   AssertFatal(png_get_io_ptr(png_ptr) != NULL, "No stream?");
-
-   Stream *strm = (Stream*)png_get_io_ptr(png_ptr);
-   bool success = strm->read((U32)length, data);
-   AssertFatal(success, "pngReadDataFn - failed to read from stream!");
-}
-
-
-//--------------------------------------
-static void pngWriteDataFn(png_structp png_ptr,
-                     png_bytep   data,
-                     png_size_t  length)
-{
-   AssertFatal(png_get_io_ptr(png_ptr) != NULL, "No stream?");
-
-   Stream *strm = (Stream*)png_get_io_ptr(png_ptr);
-   bool success = strm->write((U32)length, data);
-   AssertFatal(success, "pngWriteDataFn - failed to write to stream!");
-}
-
-
-//--------------------------------------
-static void pngFlushDataFn(png_structp /*png_ptr*/)
-{
-   //
-}
-
-static png_voidp pngMallocFn(png_structp /*png_ptr*/, png_size_t size)
-{
-   return FrameAllocator::alloc((U32)size);
-}
-
-static void pngFreeFn(png_structp /*png_ptr*/, png_voidp /*mem*/)
-{
-}
-
-static png_voidp pngRealMallocFn(png_structp /*png_ptr*/, png_size_t size)
-{
-   return (png_voidp)dMalloc(size);
-}
-
-static void pngRealFreeFn(png_structp /*png_ptr*/, png_voidp mem)
-{
-   dFree(mem);
-}
-
-//--------------------------------------
-static void pngFatalErrorFn(png_structp     /*png_ptr*/,
-                     png_const_charp pMessage)
-{
-   AssertISV(false, avar("Error reading PNG file:\n %s", pMessage));
-}
-
-
-//--------------------------------------
-static void pngWarningFn(png_structp, png_const_charp /*pMessage*/)
-{
-   //   AssertWarn(false, avar("Warning reading PNG file:\n %s", pMessage));
-}
-
-
-//--------------------------------------
-static bool sReadPNG(Stream &stream, GBitmap *bitmap)
-{
-   PROFILE_SCOPE(sReadPNG);
-   static const U32 cs_headerBytesChecked = 8;
-
-   U8 header[cs_headerBytesChecked];
-   stream.read(cs_headerBytesChecked, header);
-
-   bool isPng = png_check_sig(header, cs_headerBytesChecked) != 0;
-   if (!isPng)
-   {
-      AssertWarn(false, "GBitmap::readPNG: stream doesn't contain a PNG");
-      return false;
-   }
-
-   U32 prevWaterMark = FrameAllocator::getWaterMark();
-   png_structp png_ptr = png_create_read_struct_2(PNG_LIBPNG_VER_STRING,
-      NULL,
-      pngFatalErrorFn,
-      pngWarningFn,
-      NULL,
-      pngRealMallocFn,
-      pngRealFreeFn);
-
-   if (png_ptr == NULL) 
-   {
-      FrameAllocator::setWaterMark(prevWaterMark);
-      return false;
-   }
-
-   png_infop info_ptr = png_create_info_struct(png_ptr);
-   if (info_ptr == NULL) 
-   {
-      png_destroy_read_struct(&png_ptr,
-         (png_infopp)NULL,
-         (png_infopp)NULL);
-
-      FrameAllocator::setWaterMark(prevWaterMark);
-      return false;
-   }
-
-   png_infop end_info = png_create_info_struct(png_ptr);
-   if (end_info == NULL) 
-   {
-      png_destroy_read_struct(&png_ptr,
-         &info_ptr,
-         (png_infopp)NULL);
-
-      FrameAllocator::setWaterMark(prevWaterMark);
-      return false;
-   }
-
-   png_set_read_fn(png_ptr, &stream, pngReadDataFn);
-
-   // Read off the info on the image.
-   png_set_sig_bytes(png_ptr, cs_headerBytesChecked);
-   png_read_info(png_ptr, info_ptr);
-
-   // OK, at this point, if we have reached it ok, then we can reset the
-   //  image to accept the new data...
-   //
-   bitmap->deleteImage();
-
-   png_uint_32 width;
-   png_uint_32 height;
-   S32 bit_depth;
-   S32 color_type;
-
-   png_get_IHDR(png_ptr, info_ptr,
-      &width, &height,             // obv.
-      &bit_depth, &color_type,     // obv.
-      NULL,                        // interlace
-      NULL,                        // compression_type
-      NULL);                       // filter_type
-
-   // First, handle the color transformations.  We need this to read in the
-   //  data as RGB or RGBA, _always_, with a maximal channel width of 8 bits.
-   //
-   bool transAlpha     = false;
-   GFXFormat format = GFXFormatR8G8B8;
-
-   // Strip off any 16 bit info
-   //
-   if (bit_depth == 16 && color_type != PNG_COLOR_TYPE_GRAY) 
-   {
-      png_set_strip_16(png_ptr);
-   }
-
-   // Expand a transparency channel into a full alpha channel...
-   //
-   if (png_get_valid(png_ptr, info_ptr, PNG_INFO_tRNS)) 
-   {
-      png_set_expand(png_ptr);
-      transAlpha = true;
-   }
-
-   if (color_type == PNG_COLOR_TYPE_PALETTE) 
-   {
-      png_set_expand(png_ptr);
-      format = transAlpha ? GFXFormatR8G8B8A8 : GFXFormatR8G8B8;
-   }
-   else if (color_type == PNG_COLOR_TYPE_GRAY) 
-   {
-      png_set_expand(png_ptr);
-
-      if (bit_depth == 16)
-         format = GFXFormatL16;
-      else
-         format = GFXFormatA8;
-   }
-   else if (color_type == PNG_COLOR_TYPE_GRAY_ALPHA) 
-   {
-      png_set_expand(png_ptr);
-      png_set_gray_to_rgb(png_ptr);
-      format = GFXFormatR8G8B8A8;
-   }
-   else if (color_type == PNG_COLOR_TYPE_RGB) 
-   {
-      format = transAlpha ? GFXFormatR8G8B8A8 : GFXFormatR8G8B8;
-      png_set_expand(png_ptr);
-   }
-   else if (color_type == PNG_COLOR_TYPE_RGB_ALPHA) 
-   {
-      png_set_expand(png_ptr);
-      format = GFXFormatR8G8B8A8;
-   }
-
-   // Update the info pointer with the result of the transformations
-   //  above...
-   png_read_update_info(png_ptr, info_ptr);
-
-   png_uint_32 rowBytes = (png_uint_32)png_get_rowbytes(png_ptr, info_ptr);
-   if (format == GFXFormatR8G8B8) 
-   {
-      AssertFatal(rowBytes == width * 3,
-         "Error, our rowbytes are incorrect for this transform... (3)");
-   }
-   else if (format == GFXFormatR8G8B8A8) 
-   {
-      AssertFatal(rowBytes == width * 4,
-         "Error, our rowbytes are incorrect for this transform... (4)");
-   }
-   else if (format == GFXFormatL16)
-   {
-      AssertFatal(rowBytes == width * 2,
-         "Error, our rowbytes are incorrect for this transform... (2)");
-   }
-
-   // actually allocate the bitmap space...
-   bitmap->allocateBitmap(width, height,
-      false,            // don't extrude miplevels...
-      format);          // use determined format...
-
-   // Set up the row pointers...
-   png_bytep* rowPointers = new png_bytep[ height ];
-   U8* pBase = (U8*)bitmap->getBits();
-   
-   for (U32 i = 0; i < height; i++)
-      rowPointers[i] = pBase + (i * rowBytes);
-
-   // And actually read the image!
-   png_read_image(png_ptr, rowPointers);
-
-   // We're outta here, destroy the png structs, and release the lock
-   //  as quickly as possible...
-   //png_read_end(png_ptr, end_info);
-   delete [] rowPointers;
-   png_read_end(png_ptr, NULL);
-   png_destroy_read_struct(&png_ptr, &info_ptr, &end_info);
-
-   // Ok, the image is read in, now we need to finish up the initialization,
-   //  which means: setting up the detailing members, init'ing the palette
-   //  key, etc...
-   //
-   // actually, all of that was handled by allocateBitmap, so we're outta here
-   //
-
-   // Check this bitmap for transparency
-   bitmap->checkForTransparency();
-
-   FrameAllocator::setWaterMark(prevWaterMark);
-
-   return true;
-}
-
-//--------------------------------------------------------------------------
-static bool _writePNG(GBitmap *bitmap, Stream &stream, U32 compressionLevel, U32 strategy, U32 filter)
-{
-   GFXFormat   format = bitmap->getFormat();
-
-   // ONLY RGB bitmap writing supported at this time!
-   AssertFatal(   format == GFXFormatR8G8B8 || 
-                  format == GFXFormatR8G8B8A8 || 
-                  format == GFXFormatR8G8B8X8 || 
-                  format == GFXFormatA8 ||
-                  format == GFXFormatR5G6B5 ||
-                  format == GFXFormatR8G8B8A8_LINEAR_FORCE, "_writePNG: ONLY RGB bitmap writing supported at this time.");
-
-   if (  format != GFXFormatR8G8B8 && 
-         format != GFXFormatR8G8B8A8 && 
-         format != GFXFormatR8G8B8X8 && 
-         format != GFXFormatA8 &&
-         format != GFXFormatR5G6B5 && format != GFXFormatR8G8B8A8_LINEAR_FORCE)
-      return false;
-
-   png_structp png_ptr = png_create_write_struct_2(PNG_LIBPNG_VER_STRING,
-      NULL,
-      pngFatalErrorFn,
-      pngWarningFn,
-      NULL,
-      pngMallocFn,
-      pngFreeFn);
-   if (png_ptr == NULL)
-      return (false);
-
-   png_infop info_ptr = png_create_info_struct(png_ptr);
-   if (info_ptr == NULL)
-   {
-      png_destroy_write_struct(&png_ptr, (png_infopp)NULL);
-      return false;
-   }
-
-   png_set_write_fn(png_ptr, &stream, pngWriteDataFn, pngFlushDataFn);
-
-   // Set the compression level and image filters
-   png_set_compression_window_bits(png_ptr, 15);
-   png_set_compression_level(png_ptr, compressionLevel);
-   png_set_filter(png_ptr, 0, filter);
-
-   // Set the image information here.  Width and height are up to 2^31,
-   // bit_depth is one of 1, 2, 4, 8, or 16, but valid values also depend on
-   // the color_type selected. color_type is one of PNG_COLOR_TYPE_GRAY,
-   // PNG_COLOR_TYPE_GRAY_ALPHA, PNG_COLOR_TYPE_PALETTE, PNG_COLOR_TYPE_RGB,
-   // or PNG_COLOR_TYPE_RGB_ALPHA.  interlace is either PNG_INTERLACE_NONE or
-   // PNG_INTERLACE_ADAM7, and the compression_type and filter_type MUST
-   // currently be PNG_COMPRESSION_TYPE_BASE and PNG_FILTER_TYPE_BASE. REQUIRED
-
-   U32   width = bitmap->getWidth();
-   U32   height = bitmap->getHeight();
-
-   if (format == GFXFormatR8G8B8)
-   {
-      png_set_IHDR(png_ptr, info_ptr,
-         width, height,               // the width & height
-         8, PNG_COLOR_TYPE_RGB,       // bit_depth, color_type,
-         NULL,                        // no interlace
-         NULL,                        // compression type
-         NULL);                       // filter type
-   }
-   else if (format == GFXFormatR8G8B8A8 || format == GFXFormatR8G8B8X8 || format == GFXFormatR8G8B8A8_LINEAR_FORCE)
-   {
-      png_set_IHDR(png_ptr, info_ptr,
-         width, height,               // the width & height
-         8, PNG_COLOR_TYPE_RGB_ALPHA, // bit_depth, color_type,
-         NULL,                        // no interlace
-         NULL,                        // compression type
-         NULL);                       // filter type
-   }
-   else if (format == GFXFormatA8)
-   {
-      png_set_IHDR(png_ptr, info_ptr,
-         width, height,               // the width & height
-         8, PNG_COLOR_TYPE_GRAY,      // bit_depth, color_type,
-         NULL,                        // no interlace
-         NULL,                        // compression type
-         NULL);                       // filter type
-   }
-   else if (format == GFXFormatR5G6B5) 
-   {
-      png_set_IHDR(png_ptr, info_ptr,
-         width, height,               // the width & height
-         16, PNG_COLOR_TYPE_GRAY,     // bit_depth, color_type,
-         PNG_INTERLACE_NONE,          // no interlace
-         PNG_COMPRESSION_TYPE_DEFAULT,   // compression type
-         PNG_FILTER_TYPE_DEFAULT);       // filter type
-      
-      png_color_8_struct sigBit = { 0 };
-      sigBit.gray = 16;
-      png_set_sBIT(png_ptr, info_ptr, &sigBit );
-
-      png_set_swap( png_ptr );
-   }
-
-   png_write_info(png_ptr, info_ptr);
-   FrameAllocatorMarker marker;
-   png_bytep* row_pointers = (png_bytep*)marker.alloc( height * sizeof( png_bytep ) );
-   for (U32 i=0; i<height; i++)
-      row_pointers[i] = const_cast<png_bytep>(bitmap->getAddress(0, i));
-
-   png_write_image(png_ptr, row_pointers);
-
-   // Write S3TC data if present...
-   // Write FXT1 data if present...
-
-   png_write_end(png_ptr, info_ptr);
-   png_destroy_write_struct(&png_ptr, (png_infopp)NULL);
-
-   return true;
-}
-
-
-//--------------------------------------------------------------------------
-static bool sWritePNG(GBitmap *bitmap, Stream &stream, U32 compressionLevel)
-{
-   U32 waterMark = FrameAllocator::getWaterMark();
-
-   if ( compressionLevel < 10 )
-   {
-      bool retVal = _writePNG(bitmap, stream, compressionLevel, 0, PNG_ALL_FILTERS);
-      FrameAllocator::setWaterMark(waterMark);
-      return retVal;
-   }
-
-   // check all our methods of compression to find the best one and use it
-   U8* buffer = new U8[1 << 22]; // 4 Megs.  Should be enough...
-   MemStream* pMemStream = new MemStream(1 << 22, buffer, false, true);
-
-   const U32 zStrategies[] = { Z_DEFAULT_STRATEGY,
-      Z_FILTERED };
-   const U32 pngFilters[]  = { PNG_FILTER_NONE,
-      PNG_FILTER_SUB,
-      PNG_FILTER_UP,
-      PNG_FILTER_AVG,
-      PNG_FILTER_PAETH,
-      PNG_ALL_FILTERS };
-
-   U32 minSize      = 0xFFFFFFFF;
-   U32 bestStrategy = 0xFFFFFFFF;
-   U32 bestFilter   = 0xFFFFFFFF;
-   U32 bestCLevel   = 0xFFFFFFFF;
-
-   for (U32 cl = 0; cl <=9; cl++) 
-   {
-      for (U32 zs = 0; zs < 2; zs++) 
-      {
-         for (U32 pf = 0; pf < 6; pf++) 
-         {
-            pMemStream->setPosition(0);
-
-            U32 waterMarkInner = FrameAllocator::getWaterMark();
-
-            if (_writePNG(bitmap, *pMemStream, cl, zStrategies[zs], pngFilters[pf]) == false)
-               AssertFatal(false, "Handle this error!");
-
-            FrameAllocator::setWaterMark(waterMarkInner);
-
-            if (pMemStream->getPosition() < minSize) 
-            {
-               minSize = pMemStream->getPosition();
-               bestStrategy = zs;
-               bestFilter   = pf;
-               bestCLevel   = cl;
-            }
-         }
-      }
-   }
-   AssertFatal(minSize != 0xFFFFFFFF, "Error, no best found?");
-
-   delete pMemStream;
-   delete [] buffer;
-
-
-   bool retVal = _writePNG(bitmap, stream,
-      bestCLevel,
-      zStrategies[bestStrategy],
-      pngFilters[bestFilter]);
-   FrameAllocator::setWaterMark(waterMark);
-
-   return retVal;
-}
-
-//--------------------------------------------------------------------------
-// Stores PNG stream data
-struct DeferredPNGWriterData {
-   png_structp png_ptr;
-   png_infop info_ptr;
-   U32 width;
-   U32 height;   
-};
-DeferredPNGWriter::DeferredPNGWriter() : 
-   mData( NULL ),
-   mActive(false)
-{
-   mData = new DeferredPNGWriterData();
-}
-DeferredPNGWriter::~DeferredPNGWriter()
-{
-   delete mData;
-}
-
-bool DeferredPNGWriter::begin( GFXFormat format, S32 width, S32 height, Stream &stream, U32 compressionLevel )
-{   
-   // ONLY RGB bitmap writing supported at this time!
-   AssertFatal(   format == GFXFormatR8G8B8 || 
-                  format == GFXFormatR8G8B8A8 || 
-                  format == GFXFormatR8G8B8X8 || 
-                  format == GFXFormatA8 ||
-                  format == GFXFormatR5G6B5, "_writePNG: ONLY RGB bitmap writing supported at this time.");
-
-   if (  format != GFXFormatR8G8B8 && 
-         format != GFXFormatR8G8B8A8 && 
-         format != GFXFormatR8G8B8X8 && 
-         format != GFXFormatA8 &&
-         format != GFXFormatR5G6B5 )
-      return false;
-
-   mData->png_ptr = png_create_write_struct_2(PNG_LIBPNG_VER_STRING,
-      NULL,
-      pngFatalErrorFn,
-      pngWarningFn,
-      NULL,
-      pngRealMallocFn,
-      pngRealFreeFn);
-   if (mData->png_ptr == NULL)
-      return (false);
-
-   mData->info_ptr = png_create_info_struct(mData->png_ptr);
-   if (mData->info_ptr == NULL)
-   {
-      png_destroy_write_struct(&mData->png_ptr, (png_infopp)NULL);
-      return false;
-   }
-
-   png_set_write_fn(mData->png_ptr, &stream, pngWriteDataFn, pngFlushDataFn);
-
-   // Set the compression level and image filters
-   png_set_compression_window_bits(mData->png_ptr, 15);
-   png_set_compression_level(mData->png_ptr, compressionLevel);
-   png_set_filter(mData->png_ptr, 0, PNG_ALL_FILTERS);
-
-   // Set the image information here.  Width and height are up to 2^31,
-   // bit_depth is one of 1, 2, 4, 8, or 16, but valid values also depend on
-   // the color_type selected. color_type is one of PNG_COLOR_TYPE_GRAY,
-   // PNG_COLOR_TYPE_GRAY_ALPHA, PNG_COLOR_TYPE_PALETTE, PNG_COLOR_TYPE_RGB,
-   // or PNG_COLOR_TYPE_RGB_ALPHA.  interlace is either PNG_INTERLACE_NONE or
-   // PNG_INTERLACE_ADAM7, and the compression_type and filter_type MUST
-   // currently be PNG_COMPRESSION_TYPE_BASE and PNG_FILTER_TYPE_BASE. REQUIRED
-   
-   if (format == GFXFormatR8G8B8)
-   {
-      png_set_IHDR(mData->png_ptr, mData->info_ptr,
-         width, height,               // the width & height
-         8, PNG_COLOR_TYPE_RGB,       // bit_depth, color_type,
-         NULL,                        // no interlace
-         NULL,                        // compression type
-         NULL);                       // filter type
-   }
-   else if (format == GFXFormatR8G8B8A8 || format == GFXFormatR8G8B8X8)
-   {
-      png_set_IHDR(mData->png_ptr, mData->info_ptr,
-         width, height,               // the width & height
-         8, PNG_COLOR_TYPE_RGB_ALPHA, // bit_depth, color_type,
-         NULL,                        // no interlace
-         NULL,                        // compression type
-         NULL);                       // filter type
-   }
-   else if (format == GFXFormatA8)
-   {
-      png_set_IHDR(mData->png_ptr, mData->info_ptr,
-         width, height,               // the width & height
-         8, PNG_COLOR_TYPE_GRAY,      // bit_depth, color_type,
-         NULL,                        // no interlace
-         NULL,                        // compression type
-         NULL);                       // filter type
-   }
-   else if (format == GFXFormatR5G6B5) 
-   {
-      png_set_IHDR(mData->png_ptr, mData->info_ptr,
-         width, height,               // the width & height
-         16, PNG_COLOR_TYPE_GRAY,     // bit_depth, color_type,
-         PNG_INTERLACE_NONE,          // no interlace
-         PNG_COMPRESSION_TYPE_DEFAULT,   // compression type
-         PNG_FILTER_TYPE_DEFAULT);       // filter type
-      
-      png_color_8_struct sigBit = { 0 };
-      sigBit.gray = 16;
-      png_set_sBIT(mData->png_ptr, mData->info_ptr, &sigBit );
-
-      png_set_swap( mData->png_ptr );
-   }
-
-   png_write_info(mData->png_ptr, mData->info_ptr);
-   
-   mActive = true;
-
-   return true;
-}
-void DeferredPNGWriter::append( GBitmap* bitmap, U32 rows)
-{
-   AssertFatal(mActive, "Cannot append to an inactive DeferredPNGWriter!");
-
-   U32 height = getMin( bitmap->getHeight(), rows);
-
-   FrameAllocatorMarker marker;
-   png_bytep* row_pointers = (png_bytep*)marker.alloc( height * sizeof( png_bytep ) );
-   for (U32 i=0; i<height; i++)
-      row_pointers[i] = const_cast<png_bytep>(bitmap->getAddress(0, i));
-
-   png_write_rows(mData->png_ptr, row_pointers, height);
-}
-void DeferredPNGWriter::end()
-{
-   AssertFatal(mActive, "Cannot end an inactive DeferredPNGWriter!");
-
-   png_write_end(mData->png_ptr, mData->info_ptr);
-   png_destroy_write_struct(&mData->png_ptr, (png_infopp)NULL);
-
-   mActive = false;
-}

Engine/source/gfx/bitmap/loaders/bitmapSTB.cpp

@@ -0,0 +1,271 @@
+//-----------------------------------------------------------------------------
+// Copyright (c) 2012 GarageGames, LLC
+//
+// 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.
+//-----------------------------------------------------------------------------
+
+#include "platform/platform.h"
+
+#include "core/stream/fileStream.h"
+#include "core/stream/memStream.h"
+#include "core/strings/stringFunctions.h"
+#include "gfx/bitmap/gBitmap.h"
+
+#ifdef __clang__
+#define STBIWDEF static inline
+#endif
+
+#define STB_IMAGE_IMPLEMENTATION
+#define STB_IMAGE_STATIC
+#include "stb_image.h"
+
+#define STB_IMAGE_WRITE_IMPLEMENTATION
+#define STB_IMAGE_WRITE_STATIC
+#include "stb_image_write.h"
+
+static bool sReadSTB(const Torque::Path& path, GBitmap* bitmap);
+static bool sWriteSTB(const Torque::Path& path, GBitmap* bitmap, U32 compressionLevel);
+
+static struct _privateRegisterSTB
+{
+   _privateRegisterSTB()
+   {
+      GBitmap::Registration reg;
+      reg.priority = 100;
+
+      reg.extensions.push_back("png");
+      reg.extensions.push_back("bmp");
+      reg.extensions.push_back("jpg");
+      reg.extensions.push_back("jpeg");
+      reg.extensions.push_back("psd");
+      reg.extensions.push_back("hdr");
+      reg.extensions.push_back("tga");
+
+      reg.readFunc = sReadSTB;
+
+      reg.writeFunc = sWriteSTB;
+
+      // for png only.
+      reg.defaultCompression = 6;
+
+      GBitmap::sRegisterFormat(reg);
+
+   }
+} sStaticRegisterSTB;
+
+bool sReadSTB(const Torque::Path& path, GBitmap* bitmap)
+{
+   PROFILE_SCOPE(sReadSTB);
+
+   S32 x, y, n, channels;
+   String ext = path.getExtension();
+
+   U32 prevWaterMark = FrameAllocator::getWaterMark();
+
+   if (!stbi_info(path.getFullPath().c_str(), &x, &y, &channels))
+   {
+      FrameAllocator::setWaterMark(prevWaterMark);
+      return false;
+   }
+
+   // do this to map 2 channels to 4, 2 channel not supported by gbitmap yet..
+   if (channels == 2)
+      channels = 4;
+   if (!ext.equal("png"))
+   {
+      if (stbi_is_16_bit(path.getFullPath().c_str()))
+      {
+         U16* data = stbi_load_16(path.getFullPath().c_str(), &x, &y, &n, channels);
+
+         // if succesful deal make the bitmap, else try other loaders.
+         if (data)
+         {
+            GFXFormat format;
+            if (n == 1)
+               format = GFXFormatL16;
+            else
+               format = GFXFormatR16G16B16A16; // not sure if this is correct.
+
+            bitmap->deleteImage();
+
+            // actually allocate the bitmap space...
+            bitmap->allocateBitmap(x, y,
+               false,            // don't extrude miplevels...
+               format);          // use determined format...
+
+            U16* pBase = (U16*)bitmap->getBits();
+
+            U32 rowBytes = bitmap->getByteSize();
+
+            dMemcpy(pBase, data, rowBytes);
+
+            stbi_image_free(data);
+
+            FrameAllocator::setWaterMark(prevWaterMark);
+
+            return true;
+         }
+      }
+   }
+
+   if (ext.equal("hdr"))
+   {
+      // force load to 4 channel.
+      float* data = stbi_loadf(path.getFullPath().c_str(), &x, &y, &n, 4);
+
+      unsigned char* dataChar = stbi__hdr_to_ldr(data, x, y, 4);
+      bitmap->deleteImage();
+      // actually allocate the bitmap space...
+      bitmap->allocateBitmap(x, y,
+         false,
+         GFXFormatR8G8B8A8);
+
+      U8* pBase = (U8*)bitmap->getBits();
+
+      U32 rowBytes = x * y * 4;
+
+      dMemcpy(pBase, dataChar, rowBytes);
+
+      stbi_image_free(dataChar);
+
+      FrameAllocator::setWaterMark(prevWaterMark);
+
+      return true;
+   }
+
+   unsigned char* data = stbi_load(path.getFullPath().c_str(), &x, &y, &n, channels);
+
+   bitmap->deleteImage();
+
+   GFXFormat format;
+
+   switch (channels) {
+   case  1:
+      format = GFXFormatA8;
+      break;
+   case 2:
+      format = GFXFormatA8L8;
+      break;
+   case 3:
+      format = GFXFormatR8G8B8;
+      break;
+   case 4:
+      format = GFXFormatR8G8B8A8;
+      break;
+   default:
+      FrameAllocator::setWaterMark(prevWaterMark);
+      return false;
+   }
+
+   // actually allocate the bitmap space...
+   bitmap->allocateBitmap(x, y,
+      false,            // don't extrude miplevels...
+      format);          // use determined format...
+
+   U8* pBase = (U8*)bitmap->getBits();
+
+   U32 rowBytes = bitmap->getByteSize();
+
+   dMemcpy(pBase, data, rowBytes);
+
+   stbi_image_free(data);
+   // Check this bitmap for transparency
+   if (channels == 4)
+      bitmap->checkForTransparency();
+
+   FrameAllocator::setWaterMark(prevWaterMark);
+
+   return true;
+}
+
+/**
+ * Write bitmap to an image file.
+ *
+ * @param[in]       path                        Destination image file path. File name extension determines image format.
+ *                                              ".bmp" for Microsoft Bitmap.
+ *                                              ".hdr" for High Dynamic Range (HDR).
+ *                                              ".jpg" or ".jpeg" for Joint Photographic Experts Group (JPEG).
+ *                                              ".png" for Portable Network Graphics (PNG).
+ *                                              ".tga" for Truevision TGA (TARGA).
+ *
+ *
+ * @param[in]       bitmap                      Source bitmap to encode image from.
+ * @param           compressionLevel            Image format specific compression level.
+ *                                              For JPEG sets the quality level percentage, range 0 to 100.
+ *                                              Not used for other image formats.
+ */
+bool sWriteSTB(const Torque::Path& path, GBitmap* bitmap, U32 compressionLevel)
+{
+   PROFILE_SCOPE(sWriteSTB);
+
+   // get our data to be saved.
+   U32 width = bitmap->getWidth();
+   U32 height = bitmap->getHeight();
+   U32 bytes = bitmap->getBytesPerPixel();
+   GFXFormat format = bitmap->getFormat();
+   String ext = path.getExtension();
+
+
+
+   U32 stride = width * bytes;
+   // we always have at least 1
+   U32 comp = 1;
+
+   if (format == GFXFormatR8G8B8)
+   {
+      comp = 3;
+   }
+   else if (format == GFXFormatR8G8B8A8 || format == GFXFormatR8G8B8X8 || format == GFXFormatR8G8B8A8_LINEAR_FORCE)
+   {
+      comp = 4;
+   }
+
+   if (ext.equal("png"))
+   {
+      stbi_write_png_compression_level = compressionLevel;
+      if (stbi_write_png(path.getFullPath().c_str(), width, height, comp, bitmap->getWritableBits(), 0))
+         return true;
+   }
+
+   if (ext.equal("tga"))
+   {
+      if (stbi_write_tga(path.getFullPath().c_str(), width, height, comp, bitmap->getWritableBits()))
+         return true;
+   }
+
+   if (ext.equal("bmp"))
+   {
+      if (stbi_write_bmp(path.getFullPath().c_str(), width, height, comp, bitmap->getWritableBits()))
+         return true;
+   }
+
+   if (ext.equal("jpg") || ext.equal("jpeg"))
+   {
+      if (stbi_write_jpg(path.getFullPath().c_str(), width, height, comp, bitmap->getWritableBits(), compressionLevel))
+         return true;
+   }
+
+   if (ext.equal("hdr"))
+   {
+      if (stbi_write_hdr(path.getFullPath().c_str(), width, height, comp, (const F32*)bitmap->getWritableBits()))
+         return true;
+   }
+
+   return false;
+}

Engine/source/gfx/bitmap/loaders/bitmapTga.cpp

@@ -1,491 +0,0 @@
-//-----------------------------------------------------------------------------
-// Copyright (c) 2012 GarageGames, LLC
-//
-// 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.
-//-----------------------------------------------------------------------------
-
-#include "core/stream/stream.h"
-
-#include "gfx/bitmap/gBitmap.h"
-
-
-static bool sReadTGA(Stream &stream, GBitmap *bitmap);
-static bool sWriteTGA(GBitmap *bitmap, Stream &stream, U32 compressionLevel);
-
-static struct _privateRegisterTGA
-{
-   _privateRegisterTGA()
-   {
-      GBitmap::Registration reg;
-
-      reg.extensions.push_back( "tga" );
-
-      reg.readFunc = sReadTGA;
-      reg.writeFunc = sWriteTGA;
-
-      GBitmap::sRegisterFormat( reg );
-   }
-} sStaticRegisterTGA;
-
-
-//------------------------------------------------------------------------------
-//-------------------------------------- Supplementary I/O
-//
-
-enum eImageType
-{
-   TypeNoData        = 0,
-   TypeUncPaletted   = 1,
-   TypeUncTruecolor  = 2,
-   TypeUncGrayscale  = 3,
-   TypeRlePaletted   = 9,
-   TypeRleTruecolor  = 10,
-   TypeRleGrayscale  = 11
-};
-
-enum ePixelMap
-{
-   MapLowerLeft      = 0,
-   MapLowerRight     = 1,
-   MapUpperLeft      = 2,
-   MapUpperRight     = 3,
-};
-
-static void tga_write_pixel_to_mem( U8 * dat, U8 img_spec, U32 number, 
-                                   U32 w, U32 h, U32 pixel, U32 bppOut )
-{
-   // write the pixel to the data regarding how the
-   // header says the data is ordered.
-
-   U32 x, y;
-
-   switch( (img_spec & 0x30) >> 4 )
-   {
-   case MapLowerRight:
-      x = w - 1 - (number % w);
-      y = h - 1 - (number / w);
-      break;
-
-   case MapUpperLeft:
-      x = number % w;
-      y = number / w;
-      break;
-
-   case MapUpperRight:
-      x = w - 1 - (number % w);
-      y = number / w;
-      break;
-
-   case MapLowerLeft:
-   default:
-      x = number % w;
-      y = h - 1 - (number / w);
-      break;
-
-   }
-
-   U32 addy = (y * w + x) * bppOut;
-   for ( U32 j = 0; j < bppOut; j++ )
-      dat[addy + j] = (U8)((pixel >> (j * 8)) & 0xFF);
-}
-
-static U32 tga_get_pixel( Stream& stream, U8 bppIn, 
-                            U8 * colormap, U8 cmapBytesEntry )
-{
-   /* get the image data value out */
-
-   U32 tmp_int32 = 0;
-
-   for ( U32 j = 0; j < bppIn; j++ )
-   {
-      U8 tmp_byte;
-      if ( !stream.read( &tmp_byte ) )
-         tmp_int32 = 0;
-      else
-         tmp_int32 += tmp_byte << (j * 8);
-   }
-
-   /* byte-order correct the thing */
-   switch( bppIn )
-   {
-   case 2:
-      tmp_int32 = convertLEndianToHost( (U16)tmp_int32 );
-      break;
-
-   case 3: /* intentional fall-thru */
-   case 4:
-      tmp_int32 = convertLEndianToHost( tmp_int32 );
-      break;
-   }
-
-   U32 tmp_col;
-
-   if ( colormap )
-   {
-      /* need to look up value to get real color */
-      tmp_col = 0;
-      for ( U32 j = 0; j < cmapBytesEntry; j++ )
-         tmp_col += colormap[cmapBytesEntry * tmp_int32 + j] << (8 * j);
-   }
-   else
-   {
-      tmp_col = tmp_int32;
-   }
-
-   return tmp_col;
-}
-
-static U32 tga_convert_color( U32 pixel, U32 bppIn, U8 alphabits, U32 bppOut )
-{
-   // this is not only responsible for converting from different depths
-   // to other depths, it also switches BGR to RGB.
-
-   // this thing will also premultiply alpha, on a pixel by pixel basis.
-
-   U8 r, g, b, a;
-
-   switch( bppIn )
-   {
-   case 32:
-      if ( alphabits == 0 )
-         goto is_24_bit_in_disguise;
-      // 32-bit to 32-bit -- nop.
-      break;
-
-   case 24:
-is_24_bit_in_disguise:
-      // 24-bit to 32-bit; (only force alpha to full)
-      pixel |= 0xFF000000;
-      break;
-
-   case 15:
-is_15_bit_in_disguise:
-      r = (U8)(((F32)((pixel & 0x7C00) >> 10)) * 8.2258f);
-      g = (U8)(((F32)((pixel & 0x03E0) >> 5 )) * 8.2258f);
-      b = (U8)(((F32)(pixel & 0x001F)) * 8.2258f);
-      // 15-bit to 32-bit; (force alpha to full)
-      pixel = 0xFF000000 + (r << 16) + (g << 8) + b;
-      break;
-
-   case 16:
-      if ( alphabits == 1 )
-         goto is_15_bit_in_disguise;
-
-      // 16-bit to 32-bit; (force alpha to full)
-      r = (U8)(((F32)((pixel & 0xF800) >> 11)) * 8.2258f);
-      g = (U8)(((F32)((pixel & 0x07E0) >> 5 )) * 4.0476f);
-      b = (U8)(((F32)(pixel & 0x001F)) * 8.2258f);
-      pixel = 0xFF000000 + (r << 16) + (g << 8) + b;
-      break;
-   }
-
-   // convert the 32-bit pixel from BGR to RGB.
-   pixel = (pixel & 0xFF00FF00) + ((pixel & 0xFF) << 16) + ((pixel & 0xFF0000) >> 16);
-
-   r = pixel & 0x000000FF;
-   g = (pixel & 0x0000FF00) >> 8;
-   b = (pixel & 0x00FF0000) >> 16;
-   a = (pixel & 0xFF000000) >> 24;
-
-   // not premultiplied alpha -- multiply.
-   r = (U8)(((F32)r / 255.0f) * ((F32)a / 255.0f) * 255.0f);
-   g = (U8)(((F32)g / 255.0f) * ((F32)a / 255.0f) * 255.0f);
-   b = (U8)(((F32)b / 255.0f) * ((F32)a / 255.0f) * 255.0f);
-
-   pixel = r + (g << 8) + (b << 16) + (a << 24);
-
-   /* now convert from 32-bit to whatever they want. */
-   switch( bppOut )
-   {
-   case 4:
-      // 32 to 32 -- nop.
-      break;
-
-   case 3:
-      // 32 to 24 -- discard alpha.
-      pixel &= 0x00FFFFFF;
-      break;
-   }
-
-   return pixel;
-}
-
-static bool sReadTGA(Stream &stream, GBitmap *bitmap)
-{
-   PROFILE_SCOPE(sReadTGA);
-   struct Header
-   {
-      U8    idLength;         // length of the image_id string below.
-      U8    cmapType;         // paletted image <=> cmapType
-      U8    imageType;        // can be any of the IMG_TYPE constants above.
-      U16   cmapFirst;        // 
-      U16   cmapLength;       // how long the colormap is
-      U8    cmapEntrySize;    // how big a palette entry is.
-      U16   xOrigin;          // the x origin of the image in the image data.
-      U16   yOrigin;          // the y origin of the image in the image data.
-      U16   width;            // the width of the image.
-      U16   height;           // the height of the image.
-      U8    pixelDepth;       // the depth of a pixel in the image.
-      U8    imageDesc;        // the image descriptor.
-   };
-
-   // Read header
-   Header header;
-   stream.read( &header.idLength );
-   stream.read( &header.cmapType );
-   stream.read( &header.imageType );
-   stream.read( &header.cmapFirst );
-   stream.read( &header.cmapLength );
-   stream.read( &header.cmapEntrySize );
-   stream.read( &header.xOrigin );
-   stream.read( &header.yOrigin );
-   stream.read( &header.width );
-   stream.read( &header.height );
-   stream.read( &header.pixelDepth );
-   stream.read( &header.imageDesc );
-
-   U32 numPixels = header.width * header.height;
-   if ( numPixels == 0 )
-   {
-      //Con::errorf( "Texture has width and/or height set to 0" );
-      return false;
-   }
-
-   U8 alphabits = header.imageDesc & 0x0F;
-
-   /* seek past the image id, if there is one */
-   if ( header.idLength )
-   {
-      if ( !stream.setPosition( stream.getPosition() + header.idLength ) )
-      {
-         //Con::errorf( "Unexpected end of stream encountered" );
-         return false;
-      }
-   }
-
-   /* if this is a 'nodata' image, just jump out. */
-   if ( header.imageType == TypeNoData )
-   {
-      //Con::errorf( "Texture contains no data" );
-      return false;
-   }
-
-   /* deal with the colormap, if there is one. */
-   U8* colormap = NULL;
-   U32 cmapBytes = 0;
-   U8 cmapBytesEntry = 0;
-
-   if ( header.cmapType )
-   {
-      switch( header.imageType )
-      {
-      case TypeUncPaletted:
-      case TypeRlePaletted:
-         break;
-
-      case TypeUncTruecolor:
-      case TypeRleTruecolor:
-         // this should really be an error, but some really old
-         // crusty targas might actually be like this (created by TrueVision, no less!)
-         // so, we'll hack our way through it.
-         break;
-
-      case TypeUncGrayscale:
-      case TypeRleGrayscale:
-         //Con::errorf( "Found colormap for a grayscale image" );
-         return false;
-      }
-
-      /* ensure colormap entry size is something we support */
-      if ( !(header.cmapEntrySize == 15 || 
-          header.cmapEntrySize == 16 ||
-          header.cmapEntrySize == 24 ||
-          header.cmapEntrySize == 32) )
-      {
-         //Con::errorf( "Unsupported colormap entry size" );
-         return false;
-      }
-
-      /* allocate memory for a colormap */
-      if ( header.cmapEntrySize & 0x07 )
-         cmapBytesEntry = (((8 - (header.cmapEntrySize & 0x07)) + header.cmapEntrySize) >> 3);
-      else
-         cmapBytesEntry = (header.cmapEntrySize >> 3);
-
-      cmapBytes = cmapBytesEntry * header.cmapLength;
-      colormap = new U8[ cmapBytes ];
-
-      for ( U32 i = 0; i < header.cmapLength; i++ )
-      {
-         /* seek ahead to first entry used */
-         if ( header.cmapFirst != 0 )
-            stream.setPosition( stream.getPosition() + header.cmapFirst * cmapBytesEntry );
-
-         U32 tmp_int32 = 0;
-         for ( U32 j = 0; j < cmapBytesEntry; j++ )
-         {
-            U8 tmp_byte;
-            if ( !stream.read( &tmp_byte ) )
-            {
-               delete [] colormap;
-               //Con::errorf( "Bad colormap" );
-               return false;
-            }
-            tmp_int32 += tmp_byte << (j * 8);
-         }
-
-         // byte order correct.
-         tmp_int32 = convertLEndianToHost( tmp_int32 );
-
-         for ( U32 j = 0; j < cmapBytesEntry; j++ )
-            colormap[i * cmapBytesEntry + j] = (tmp_int32 >> (8 * j)) & 0xFF;
-      }
-   }
-
-   // compute number of bytes in an image data unit (either index or BGR triple)
-   U8 inBytesPerPixel = 0;
-   if ( header.pixelDepth & 0x07 )
-      inBytesPerPixel = (((8 - (header.pixelDepth & 0x07)) + header.pixelDepth) >> 3);
-   else
-      inBytesPerPixel = (header.pixelDepth >> 3);
-
-   /* assume that there's one byte per pixel */
-   if ( inBytesPerPixel == 0 )
-      inBytesPerPixel = 1;
-
-   GFXFormat gfxFmt;
-   U32 outBytesPerPixel;
-   switch ( header.pixelDepth )
-   {
-   case 32:
-      gfxFmt = GFXFormatR8G8B8A8;
-      outBytesPerPixel = 4;
-      break;
-
-   case 24:
-   default:
-      gfxFmt = GFXFormatR8G8B8;
-      outBytesPerPixel = 3;
-      break;
-   }
-
-   bitmap->allocateBitmap( header.width, header.height, false, gfxFmt );
-
-   // compute the true number of bits per pixel
-   U8 trueBitsPerPixel = header.cmapType ? header.cmapEntrySize : header.pixelDepth;
-
-   // Override the number of alpha bits if necessary
-   // Some apps generate transparent TGAs with alphabits set to 0 in the image descriptor
-   if ( ( trueBitsPerPixel == 32 ) && ( alphabits == 0 ) )
-      alphabits = 8;
-
-   switch( header.imageType )
-   {
-   case TypeUncTruecolor:
-   case TypeUncGrayscale:
-   case TypeUncPaletted:
-
-      /* FIXME: support grayscale */
-
-      for ( U32 i = 0; i < numPixels; i++ )
-      {
-         // get the color value.
-         U32 tmp_col = tga_get_pixel( stream, inBytesPerPixel, colormap, cmapBytesEntry );
-         tmp_col = tga_convert_color( tmp_col, trueBitsPerPixel, alphabits, outBytesPerPixel );
-
-         // now write the data out.
-         tga_write_pixel_to_mem( bitmap->getAddress( 0, 0 ), header.imageDesc, 
-            i, header.width, header.height, tmp_col, outBytesPerPixel );
-      }
-      break;
-
-   case TypeRleTruecolor:
-   case TypeRleGrayscale:
-   case TypeRlePaletted:
-
-      // FIXME: handle grayscale..
-
-      for ( U32 i = 0; i < numPixels; )
-      {
-         /* a bit of work to do to read the data.. */
-         U8 packet_header;
-         if ( !stream.read( 1, &packet_header ) )
-         {
-            // well, just let them fill the rest with null pixels then...
-            packet_header = 1;
-         }
-
-         if ( packet_header & 0x80 )
-         {
-            /* run length packet */
-            U32 tmp_col = tga_get_pixel( stream, inBytesPerPixel, colormap, cmapBytesEntry );
-            tmp_col = tga_convert_color( tmp_col, trueBitsPerPixel, alphabits, outBytesPerPixel );
-
-            U8 repcount = (packet_header & 0x7F) + 1;
-
-            /* write all the data out */
-            for ( U32 j = 0; j < repcount; j++ )
-            {
-               tga_write_pixel_to_mem( bitmap->getAddress( 0, 0 ), header.imageDesc, 
-                   i + j, header.width, header.height, tmp_col, outBytesPerPixel );
-            }
-
-            i += repcount;
-
-         }
-         else
-         {
-            /* raw packet */
-            /* get pixel from file */
-            U8 repcount = (packet_header & 0x7F) + 1;
-
-            for ( U32 j = 0; j < repcount; j++ )
-            {
-               U32 tmp_col = tga_get_pixel( stream, inBytesPerPixel, colormap, cmapBytesEntry );
-               tmp_col = tga_convert_color( tmp_col, trueBitsPerPixel, alphabits, outBytesPerPixel );
-
-               tga_write_pixel_to_mem( bitmap->getAddress( 0, 0 ), header.imageDesc, 
-                   i + j, header.width, header.height, tmp_col, outBytesPerPixel );
-            }
-
-            i += repcount;
-         }
-      }
-      break;
-
-   default:
-       //Con::errorf( "Unknown image type" );
-      delete[] colormap;
-      return false;
-   }
-
-   delete [] colormap;
-
-   // 32-bit tgas have an alpha channel
-   bitmap->setHasTransparency( header.pixelDepth == 32 );
-
-   return true;
-}
-
-static bool sWriteTGA(GBitmap *bitmap, Stream &stream, U32 compressionLevel)
-{
-   AssertISV(false, "GBitmap::writeTGA - doesn't support writing tga files!");
-
-   return false;
-}

Engine/source/gfx/bitmap/loaders/stb/stb_image.h

@@ -0,0 +1,7987 @@
+/* stb_image - v2.28 - public domain image loader - http://nothings.org/stb
+                                  no warranty implied; use at your own risk
+
+   Do this:
+      #define STB_IMAGE_IMPLEMENTATION
+   before you include this file in *one* C or C++ file to create the implementation.
+
+   // i.e. it should look like this:
+   #include ...
+   #include ...
+   #include ...
+   #define STB_IMAGE_IMPLEMENTATION
+   #include "stb_image.h"
+
+   You can #define STBI_ASSERT(x) before the #include to avoid using assert.h.
+   And #define STBI_MALLOC, STBI_REALLOC, and STBI_FREE to avoid using malloc,realloc,free
+
+
+   QUICK NOTES:
+      Primarily of interest to game developers and other people who can
+          avoid problematic images and only need the trivial interface
+
+      JPEG baseline & progressive (12 bpc/arithmetic not supported, same as stock IJG lib)
+      PNG 1/2/4/8/16-bit-per-channel
+
+      TGA (not sure what subset, if a subset)
+      BMP non-1bpp, non-RLE
+      PSD (composited view only, no extra channels, 8/16 bit-per-channel)
+
+      GIF (*comp always reports as 4-channel)
+      HDR (radiance rgbE format)
+      PIC (Softimage PIC)
+      PNM (PPM and PGM binary only)
+
+      Animated GIF still needs a proper API, but here's one way to do it:
+          http://gist.github.com/urraka/685d9a6340b26b830d49
+
+      - decode from memory or through FILE (define STBI_NO_STDIO to remove code)
+      - decode from arbitrary I/O callbacks
+      - SIMD acceleration on x86/x64 (SSE2) and ARM (NEON)
+
+   Full documentation under "DOCUMENTATION" below.
+
+
+LICENSE
+
+  See end of file for license information.
+
+RECENT REVISION HISTORY:
+
+      2.28  (2023-01-29) many error fixes, security errors, just tons of stuff
+      2.27  (2021-07-11) document stbi_info better, 16-bit PNM support, bug fixes
+      2.26  (2020-07-13) many minor fixes
+      2.25  (2020-02-02) fix warnings
+      2.24  (2020-02-02) fix warnings; thread-local failure_reason and flip_vertically
+      2.23  (2019-08-11) fix clang static analysis warning
+      2.22  (2019-03-04) gif fixes, fix warnings
+      2.21  (2019-02-25) fix typo in comment
+      2.20  (2019-02-07) support utf8 filenames in Windows; fix warnings and platform ifdefs
+      2.19  (2018-02-11) fix warning
+      2.18  (2018-01-30) fix warnings
+      2.17  (2018-01-29) bugfix, 1-bit BMP, 16-bitness query, fix warnings
+      2.16  (2017-07-23) all functions have 16-bit variants; optimizations; bugfixes
+      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;
+                         allocate large structures on the stack;
+                         correct channel count for PNG & BMP
+      2.10  (2016-01-22) avoid warning introduced in 2.09
+      2.09  (2016-01-16) 16-bit TGA; comments in PNM files; STBI_REALLOC_SIZED
+
+   See end of file for full revision history.
+
+
+ ============================    Contributors    =========================
+
+ Image formats                          Extensions, features
+    Sean Barrett (jpeg, png, bmp)          Jetro Lauha (stbi_info)
+    Nicolas Schulz (hdr, psd)              Martin "SpartanJ" Golini (stbi_info)
+    Jonathan Dummer (tga)                  James "moose2000" Brown (iPhone PNG)
+    Jean-Marc Lienher (gif)                Ben "Disch" Wenger (io callbacks)
+    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)
+    github:urraka (animated gif)           Junggon Kim (PNM comments)
+    Christopher Forseth (animated gif)     Daniel Gibson (16-bit TGA)
+                                           socks-the-fox (16-bit PNG)
+                                           Jeremy Sawicki (handle all ImageNet JPGs)
+ Optimizations & bugfixes                  Mikhail Morozov (1-bit BMP)
+    Fabian "ryg" Giesen                    Anael Seghezzi (is-16-bit query)
+    Arseny Kapoulkine                      Simon Breuss (16-bit PNM)
+    John-Mark Allen
+    Carmelo J Fdez-Aguera
+
+ Bug & warning fixes
+    Marc LeBlanc            David Woo          Guillaume George     Martins Mozeiko
+    Christpher Lloyd        Jerry Jansson      Joseph Thomson       Blazej Dariusz Roszkowski
+    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                         github:rlyeh
+    Janez Zemva             John Bartholomew   Michal Cichon        github:romigrou
+    Jonathan Blow           Ken Hamada         Tero Hanninen        github:svdijk
+    Eugene Golushkov        Laurent Gomila     Cort Stratton        github:snagar
+    Aruelien Pocheville     Sergio Gonzalez    Thibault Reuille     github:Zelex
+    Cass Everitt            Ryamond Barbiero                        github:grim210
+    Paul Du Bois            Engin Manap        Aldo Culquicondor    github:sammyhw
+    Philipp Wiesemann       Dale Weiler        Oriol Ferrer Mesia   github:phprus
+    Josh Tobin              Neil Bickford      Matthew Gregan       github:poppolopoppo
+    Julian Raschke          Gregory Mullen     Christian Floisand   github:darealshinji
+    Baldur Karlsson         Kevin Schmidt      JR Smith             github:Michaelangel007
+                            Brad Weinberger    Matvey Cherevko      github:mosra
+    Luca Sas                Alexander Veselov  Zack Middleton       [reserved]
+    Ryan C. Gordon          [reserved]                              [reserved]
+                     DO NOT ADD YOUR NAME HERE
+
+                     Jacko Dirks
+
+  To add your name to the credits, pick a random blank space in the middle and fill it.
+  80% of merge conflicts on stb PRs are due to people adding their name at the end
+  of the credits.
+*/
+
+#ifndef STBI_INCLUDE_STB_IMAGE_H
+#define STBI_INCLUDE_STB_IMAGE_H
+
+// DOCUMENTATION
+//
+// Limitations:
+//    - no 12-bit-per-channel JPEG
+//    - no JPEGs with arithmetic coding
+//    - GIF always returns *comp=4
+//
+// Basic usage (see HDR discussion below for HDR usage):
+//    int x,y,n;
+//    unsigned char *data = stbi_load(filename, &x, &y, &n, 0);
+//    // ... process data if not NULL ...
+//    // ... x = width, y = height, n = # 8-bit components per pixel ...
+//    // ... replace '0' with '1'..'4' to force that many components per pixel
+//    // ... but 'n' will always be the number that it would have been if you said 0
+//    stbi_image_free(data);
+//
+// Standard parameters:
+//    int *x                 -- outputs image width in pixels
+//    int *y                 -- outputs image height in pixels
+//    int *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
+// corrupt or invalid. The pixel data consists of *y scanlines of *x pixels,
+// with each pixel consisting of N interleaved 8-bit components; the first
+// pixel pointed to is top-left-most in the image. There is no padding between
+// image scanlines or between pixels, regardless of format. The number of
+// components N is 'desired_channels' if desired_channels is non-zero, or
+// *channels_in_file otherwise. If desired_channels is non-zero,
+// *channels_in_file has the number of components that _would_ have been
+// output otherwise. E.g. if you set desired_channels to 4, you will always
+// get RGBA output, but you can check *channels_in_file to see if it's trivially
+// opaque because e.g. there were only 3 channels in the source image.
+//
+// An output image with N components has the following components interleaved
+// in this order in each pixel:
+//
+//     N=#comp     components
+//       1           grey
+//       2           grey, alpha
+//       3           red, green, blue
+//       4           red, green, blue, alpha
+//
+// If image loading fails for any reason, the return value will be NULL,
+// and *x, *y, *channels_in_file will be unchanged. The function
+// stbi_failure_reason() can be queried for an extremely brief, end-user
+// unfriendly explanation of why the load failed. Define STBI_NO_FAILURE_STRINGS
+// to avoid compiling these strings at all, and STBI_FAILURE_USERMSG to get slightly
+// more user-friendly ones.
+//
+// Paletted PNG, BMP, GIF, and PIC images are automatically depalettized.
+//
+// To query the width, height and component count of an image without having to
+// decode the full file, you can use the stbi_info family of functions:
+//
+//   int x,y,n,ok;
+//   ok = stbi_info(filename, &x, &y, &n);
+//   // returns ok=1 and sets x, y, n if image is a supported format,
+//   // 0 otherwise.
+//
+// Note that stb_image pervasively uses ints in its public API for sizes,
+// including sizes of memory buffers. This is now part of the API and thus
+// hard to change without causing breakage. As a result, the various image
+// loaders all have certain limits on image size; these differ somewhat
+// by format but generally boil down to either just under 2GB or just under
+// 1GB. When the decoded image would be larger than this, stb_image decoding
+// will fail.
+//
+// Additionally, stb_image will reject image files that have any of their
+// dimensions set to a larger value than the configurable STBI_MAX_DIMENSIONS,
+// which defaults to 2**24 = 16777216 pixels. Due to the above memory limit,
+// the only way to have an image with such dimensions load correctly
+// is for it to have a rather extreme aspect ratio. Either way, the
+// assumption here is that such larger images are likely to be malformed
+// or malicious. If you do need to load an image with individual dimensions
+// larger than that, and it still fits in the overall size limit, you can
+// #define STBI_MAX_DIMENSIONS on your own to be something larger.
+//
+// ===========================================================================
+//
+// UNICODE:
+//
+//   If compiling for Windows and you wish to use Unicode filenames, compile
+//   with
+//       #define STBI_WINDOWS_UTF8
+//   and pass utf8-encoded filenames. Call stbi_convert_wchar_to_utf8 to convert
+//   Windows wchar_t filenames to utf8.
+//
+// ===========================================================================
+//
+// Philosophy
+//
+// stb libraries are designed with the following priorities:
+//
+//    1. easy to use
+//    2. easy to maintain
+//    3. good performance
+//
+// Sometimes I let "good performance" creep up in priority over "easy to maintain",
+// and for best performance I may provide less-easy-to-use APIs that give higher
+// performance, in addition to the easy-to-use ones. Nevertheless, it's important
+// to keep in mind that from the standpoint of you, a client of this library,
+// all you care about is #1 and #3, and stb libraries DO NOT emphasize #3 above all.
+//
+// Some secondary priorities arise directly from the first two, some of which
+// provide more explicit reasons why performance can't be emphasized.
+//
+//    - Portable ("ease of use")
+//    - Small source code footprint ("easy to maintain")
+//    - No dependencies ("ease of use")
+//
+// ===========================================================================
+//
+// I/O callbacks
+//
+// I/O callbacks allow you to read from arbitrary sources, like packaged
+// files or some other source. Data read from callbacks are processed
+// through a small internal buffer (currently 128 bytes) to try to reduce
+// overhead.
+//
+// The three functions you must define are "read" (reads some bytes of data),
+// "skip" (skips some bytes of data), "eof" (reports if the stream is at the end).
+//
+// ===========================================================================
+//
+// SIMD support
+//
+// The JPEG decoder will try to automatically use SIMD kernels on x86 when
+// supported by the compiler. For ARM Neon support, you must explicitly
+// request it.
+//
+// (The old do-it-yourself SIMD API is no longer supported in the current
+// code.)
+//
+// On x86, SSE2 will automatically be used when available based on a run-time
+// test; if not, the generic C versions are used as a fall-back. On ARM targets,
+// the typical path is to have separate builds for NEON and non-NEON devices
+// (at least this is true for iOS and Android). Therefore, the NEON support is
+// toggled by a build flag: define STBI_NEON to get NEON loops.
+//
+// If for some reason you do not want to use any of SIMD code, or if
+// you have issues compiling it, you can disable it entirely by
+// defining STBI_NO_SIMD.
+//
+// ===========================================================================
+//
+// HDR image support   (disable by defining STBI_NO_HDR)
+//
+// stb_image supports loading HDR images in general, and currently the Radiance
+// .HDR file format specifically. You can still load any file through the existing
+// interface; if you attempt to load an HDR file, it will be automatically remapped
+// to LDR, assuming gamma 2.2 and an arbitrary scale factor defaulting to 1;
+// both of these constants can be reconfigured through this interface:
+//
+//     stbi_hdr_to_ldr_gamma(2.2f);
+//     stbi_hdr_to_ldr_scale(1.0f);
+//
+// (note, do not use _inverse_ constants; stbi_image will invert them
+// appropriately).
+//
+// Additionally, there is a new, parallel interface for loading files as
+// (linear) floats to preserve the full dynamic range:
+//
+//    float *data = stbi_loadf(filename, &x, &y, &n, 0);
+//
+// If you load LDR images through this interface, those images will
+// be promoted to floating point values, run through the inverse of
+// constants corresponding to the above:
+//
+//     stbi_ldr_to_hdr_scale(1.0f);
+//     stbi_ldr_to_hdr_gamma(2.2f);
+//
+// Finally, given a filename (or an open file or memory block--see header
+// file for details) containing image data, you can query for the "most
+// appropriate" interface to use (that is, whether the image is HDR or
+// not), using:
+//
+//     stbi_is_hdr(char *filename);
+//
+// ===========================================================================
+//
+// iPhone PNG support:
+//
+// We optionally support converting iPhone-formatted PNGs (which store
+// premultiplied BGRA) back to RGB, even though they're internally encoded
+// differently. To enable this conversion, call
+// stbi_convert_iphone_png_to_rgb(1).
+//
+// Call stbi_set_unpremultiply_on_load(1) as well to force a divide per
+// pixel to remove any premultiplied alpha *only* if the image file explicitly
+// says there's premultiplied data (currently only happens in iPhone images,
+// and only if iPhone convert-to-rgb processing is on).
+//
+// ===========================================================================
+//
+// 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
+//
+//  - If you define STBI_MAX_DIMENSIONS, stb_image will reject images greater
+//    than that size (in either width or height) without further processing.
+//    This is to let programs in the wild set an upper bound to prevent
+//    denial-of-service attacks on untrusted data, as one could generate a
+//    valid image of gigantic dimensions and force stb_image to allocate a
+//    huge block of memory and spend disproportionate time decoding it. By
+//    default this is set to (1 << 24), which is 16777216, but that's still
+//    very big.
+
+#ifndef STBI_NO_STDIO
+#include <stdio.h>
+#endif // STBI_NO_STDIO
+
+#define STBI_VERSION 1
+
+enum
+{
+   STBI_default = 0, // only used for desired_channels
+
+   STBI_grey       = 1,
+   STBI_grey_alpha = 2,
+   STBI_rgb        = 3,
+   STBI_rgb_alpha  = 4
+};
+
+#include <stdlib.h>
+typedef unsigned char stbi_uc;
+typedef unsigned short stbi_us;
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#ifndef STBIDEF
+#ifdef STB_IMAGE_STATIC
+#define STBIDEF static
+#else
+#define STBIDEF extern
+#endif
+#endif
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// PRIMARY API - works on images of any type
+//
+
+//
+// load image by filename, open file, or memory buffer
+//
+
+typedef struct
+{
+   int      (*read)  (void *user,char *data,int size);   // fill 'data' with 'size' bytes.  return number of bytes actually read
+   void     (*skip)  (void *user,int n);                 // skip the next 'n' bytes, or 'unget' the last -n bytes if negative
+   int      (*eof)   (void *user);                       // returns nonzero if we are at end of file/data
+} stbi_io_callbacks;
+
+////////////////////////////////////
+//
+// 8-bits-per-channel interface
+//
+
+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            (char const *filename, int *x, int *y, int *channels_in_file, int desired_channels);
+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
+
+#ifndef STBI_NO_GIF
+STBIDEF stbi_uc *stbi_load_gif_from_memory(stbi_uc const *buffer, int len, int **delays, int *x, int *y, int *z, int *comp, int req_comp);
+#endif
+
+#ifdef STBI_WINDOWS_UTF8
+STBIDEF int stbi_convert_wchar_to_utf8(char *buffer, size_t bufferlen, const wchar_t* input);
+#endif
+
+////////////////////////////////////
+//
+// 16-bits-per-channel interface
+//
+
+STBIDEF stbi_us *stbi_load_16_from_memory   (stbi_uc const *buffer, int len, int *x, int *y, int *channels_in_file, int desired_channels);
+STBIDEF stbi_us *stbi_load_16_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_us *stbi_load_16          (char const *filename, int *x, int *y, int *channels_in_file, int desired_channels);
+STBIDEF stbi_us *stbi_load_from_file_16(FILE *f, int *x, int *y, int *channels_in_file, int desired_channels);
+#endif
+
+////////////////////////////////////
+//
+// float-per-channel interface
+//
+#ifndef STBI_NO_LINEAR
+   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            (char const *filename, int *x, int *y, int *channels_in_file, int desired_channels);
+   STBIDEF float *stbi_loadf_from_file  (FILE *f, int *x, int *y, int *channels_in_file, int desired_channels);
+   #endif
+#endif
+
+#ifndef STBI_NO_HDR
+   STBIDEF void   stbi_hdr_to_ldr_gamma(float gamma);
+   STBIDEF void   stbi_hdr_to_ldr_scale(float scale);
+#endif // STBI_NO_HDR
+
+#ifndef STBI_NO_LINEAR
+   STBIDEF void   stbi_ldr_to_hdr_gamma(float gamma);
+   STBIDEF void   stbi_ldr_to_hdr_scale(float scale);
+#endif // STBI_NO_LINEAR
+
+// stbi_is_hdr is always defined, but always returns false if STBI_NO_HDR
+STBIDEF int    stbi_is_hdr_from_callbacks(stbi_io_callbacks const *clbk, void *user);
+STBIDEF int    stbi_is_hdr_from_memory(stbi_uc const *buffer, int len);
+#ifndef STBI_NO_STDIO
+STBIDEF int      stbi_is_hdr          (char const *filename);
+STBIDEF int      stbi_is_hdr_from_file(FILE *f);
+#endif // STBI_NO_STDIO
+
+
+// get a VERY brief reason for failure
+// on most compilers (and ALL modern mainstream compilers) this is threadsafe
+STBIDEF const char *stbi_failure_reason  (void);
+
+// free the loaded image -- this is just free()
+STBIDEF void     stbi_image_free      (void *retval_from_stbi_load);
+
+// get image dimensions & components without fully decoding
+STBIDEF int      stbi_info_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp);
+STBIDEF int      stbi_info_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp);
+STBIDEF int      stbi_is_16_bit_from_memory(stbi_uc const *buffer, int len);
+STBIDEF int      stbi_is_16_bit_from_callbacks(stbi_io_callbacks const *clbk, void *user);
+
+#ifndef STBI_NO_STDIO
+STBIDEF int      stbi_info               (char const *filename,     int *x, int *y, int *comp);
+STBIDEF int      stbi_info_from_file     (FILE *f,                  int *x, int *y, int *comp);
+STBIDEF int      stbi_is_16_bit          (char const *filename);
+STBIDEF int      stbi_is_16_bit_from_file(FILE *f);
+#endif
+
+
+
+// for image formats that explicitly notate that they have premultiplied alpha,
+// we just return the colors as stored in the file. set this flag to force
+// unpremultiplication. results are undefined if the unpremultiply overflow.
+STBIDEF void stbi_set_unpremultiply_on_load(int flag_true_if_should_unpremultiply);
+
+// indicate whether we should process iphone images back to canonical format,
+// or just pass them through "as-is"
+STBIDEF void stbi_convert_iphone_png_to_rgb(int flag_true_if_should_convert);
+
+// flip the image vertically, so the first pixel in the output array is the bottom left
+STBIDEF void stbi_set_flip_vertically_on_load(int flag_true_if_should_flip);
+
+// as above, but only applies to images loaded on the thread that calls the function
+// this function is only available if your compiler supports thread-local variables;
+// calling it will fail to link if your compiler doesn't
+STBIDEF void stbi_set_unpremultiply_on_load_thread(int flag_true_if_should_unpremultiply);
+STBIDEF void stbi_convert_iphone_png_to_rgb_thread(int flag_true_if_should_convert);
+STBIDEF void stbi_set_flip_vertically_on_load_thread(int flag_true_if_should_flip);
+
+// ZLIB client - used by PNG, available for other purposes
+
+STBIDEF char *stbi_zlib_decode_malloc_guesssize(const char *buffer, int len, int initial_size, int *outlen);
+STBIDEF char *stbi_zlib_decode_malloc_guesssize_headerflag(const char *buffer, int len, int initial_size, int *outlen, int parse_header);
+STBIDEF char *stbi_zlib_decode_malloc(const char *buffer, int len, int *outlen);
+STBIDEF int   stbi_zlib_decode_buffer(char *obuffer, int olen, const char *ibuffer, int ilen);
+
+STBIDEF char *stbi_zlib_decode_noheader_malloc(const char *buffer, int len, int *outlen);
+STBIDEF int   stbi_zlib_decode_noheader_buffer(char *obuffer, int olen, const char *ibuffer, int ilen);
+
+
+#ifdef __cplusplus
+}
+#endif
+
+//
+//
+////   end header file   /////////////////////////////////////////////////////
+#endif // STBI_INCLUDE_STB_IMAGE_H
+
+#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, pow
+#endif
+
+#ifndef STBI_NO_STDIO
+#include <stdio.h>
+#endif
+
+#ifndef STBI_ASSERT
+#include <assert.h>
+#define STBI_ASSERT(x) assert(x)
+#endif
+
+#ifdef __cplusplus
+#define STBI_EXTERN extern "C"
+#else
+#define STBI_EXTERN extern
+#endif
+
+
+#ifndef _MSC_VER
+   #ifdef __cplusplus
+   #define stbi_inline inline
+   #else
+   #define stbi_inline
+   #endif
+#else
+   #define stbi_inline __forceinline
+#endif
+
+#ifndef STBI_NO_THREAD_LOCALS
+   #if defined(__cplusplus) &&  __cplusplus >= 201103L
+      #define STBI_THREAD_LOCAL       thread_local
+   #elif defined(__GNUC__) && __GNUC__ < 5
+      #define STBI_THREAD_LOCAL       __thread
+   #elif defined(_MSC_VER)
+      #define STBI_THREAD_LOCAL       __declspec(thread)
+   #elif defined (__STDC_VERSION__) && __STDC_VERSION__ >= 201112L && !defined(__STDC_NO_THREADS__)
+      #define STBI_THREAD_LOCAL       _Thread_local
+   #endif
+
+   #ifndef STBI_THREAD_LOCAL
+      #if defined(__GNUC__)
+        #define STBI_THREAD_LOCAL       __thread
+      #endif
+   #endif
+#endif
+
+#if defined(_MSC_VER) || defined(__SYMBIAN32__)
+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) >> (-(y) & 31)))
+#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
+
+#if !defined(STBI_NO_JPEG) && defined(STBI_SSE2)
+static int stbi__sse2_available(void)
+{
+   int info3 = stbi__cpuid3();
+   return ((info3 >> 26) & 1) != 0;
+}
+#endif
+
+#else // assume GCC-style if not VC++
+#define STBI_SIMD_ALIGN(type, name) type name __attribute__((aligned(16)))
+
+#if !defined(STBI_NO_JPEG) && defined(STBI_SSE2)
+static int stbi__sse2_available(void)
+{
+   // 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
+#endif
+
+// ARM NEON
+#if defined(STBI_NO_SIMD) && defined(STBI_NEON)
+#undef STBI_NEON
+#endif
+
+#ifdef STBI_NEON
+#include <arm_neon.h>
+#ifdef _MSC_VER
+#define STBI_SIMD_ALIGN(type, name) __declspec(align(16)) type name
+#else
+#define STBI_SIMD_ALIGN(type, name) type name __attribute__((aligned(16)))
+#endif
+#endif
+
+#ifndef STBI_SIMD_ALIGN
+#define STBI_SIMD_ALIGN(type, name) type name
+#endif
+
+#ifndef STBI_MAX_DIMENSIONS
+#define STBI_MAX_DIMENSIONS (1 << 24)
+#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];
+   int callback_already_read;
+
+   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->callback_already_read = 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->callback_already_read = 0;
+   s->img_buffer = 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)
+{
+   int ch;
+   fseek((FILE*) user, n, SEEK_CUR);
+   ch = fgetc((FILE*) user);  /* have to read a byte to reset feof()'s flag */
+   if (ch != EOF) {
+      ungetc(ch, (FILE *) user);  /* push byte back onto stream if valid. */
+   }
+}
+
+static int stbi__stdio_eof(void *user)
+{
+   return feof((FILE*) user) || ferror((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);
+static int      stbi__png_is16(stbi__context *s);
+#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);
+static int      stbi__psd_is16(stbi__context *s);
+#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 void    *stbi__load_gif_main(stbi__context *s, int **delays, int *x, int *y, int *z, 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 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);
+static int      stbi__pnm_is16(stbi__context *s);
+#endif
+
+static
+#ifdef STBI_THREAD_LOCAL
+STBI_THREAD_LOCAL
+#endif
+const char *stbi__g_failure_reason;
+
+STBIDEF const char *stbi_failure_reason(void)
+{
+   return stbi__g_failure_reason;
+}
+
+#ifndef STBI_NO_FAILURE_STRINGS
+static int stbi__err(const char *str)
+{
+   stbi__g_failure_reason = str;
+   return 0;
+}
+#endif
+
+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;
+}
+
+#if !defined(STBI_NO_JPEG) || !defined(STBI_NO_PNG) || !defined(STBI_NO_TGA) || !defined(STBI_NO_HDR)
+// 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);
+}
+#endif
+
+// 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
+#if !defined(STBI_NO_LINEAR) || !defined(STBI_NO_HDR) || !defined(STBI_NO_PNM)
+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);
+}
+#endif
+
+#if !defined(STBI_NO_JPEG) || !defined(STBI_NO_PNG) || !defined(STBI_NO_TGA) || !defined(STBI_NO_HDR)
+// 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);
+}
+#endif
+
+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);
+}
+
+#if !defined(STBI_NO_LINEAR) || !defined(STBI_NO_HDR) || !defined(STBI_NO_PNM)
+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);
+}
+#endif
+
+// returns 1 if the sum of two signed ints is valid (between -2^31 and 2^31-1 inclusive), 0 on overflow.
+static int stbi__addints_valid(int a, int b)
+{
+   if ((a >= 0) != (b >= 0)) return 1; // a and b have different signs, so no overflow
+   if (a < 0 && b < 0) return a >= INT_MIN - b; // same as a + b >= INT_MIN; INT_MIN - b cannot overflow since b < 0.
+   return a <= INT_MAX - b;
+}
+
+// returns 1 if the product of two signed shorts is valid, 0 on overflow.
+static int stbi__mul2shorts_valid(short a, short b)
+{
+   if (b == 0 || b == -1) return 1; // multiplication by 0 is always 0; check for -1 so SHRT_MIN/b doesn't overflow
+   if ((a >= 0) == (b >= 0)) return a <= SHRT_MAX/b; // product is positive, so similar to mul2sizes_valid
+   if (b < 0) return a <= SHRT_MIN / b; // same as a * b >= SHRT_MIN
+   return a >= SHRT_MIN / b;
+}
+
+// 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_global = 0;
+
+STBIDEF void stbi_set_flip_vertically_on_load(int flag_true_if_should_flip)
+{
+   stbi__vertically_flip_on_load_global = flag_true_if_should_flip;
+}
+
+#ifndef STBI_THREAD_LOCAL
+#define stbi__vertically_flip_on_load  stbi__vertically_flip_on_load_global
+#else
+static STBI_THREAD_LOCAL int stbi__vertically_flip_on_load_local, stbi__vertically_flip_on_load_set;
+
+STBIDEF void stbi_set_flip_vertically_on_load_thread(int flag_true_if_should_flip)
+{
+   stbi__vertically_flip_on_load_local = flag_true_if_should_flip;
+   stbi__vertically_flip_on_load_set = 1;
+}
+
+#define stbi__vertically_flip_on_load  (stbi__vertically_flip_on_load_set       \
+                                         ? stbi__vertically_flip_on_load_local  \
+                                         : stbi__vertically_flip_on_load_global)
+#endif // STBI_THREAD_LOCAL
+
+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;
+
+   // test the formats with a very explicit header first (at least a FOURCC
+   // or distinctive magic number first)
+   #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);
+   #else
+   STBI_NOTUSED(bpc);
+   #endif
+   #ifndef STBI_NO_PIC
+   if (stbi__pic_test(s))  return stbi__pic_load(s,x,y,comp,req_comp, ri);
+   #endif
+
+   // then the formats that can end up attempting to load with just 1 or 2
+   // bytes matching expectations; these are prone to false positives, so
+   // try them later
+   #ifndef STBI_NO_JPEG
+   if (stbi__jpeg_test(s)) return stbi__jpeg_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 void stbi__vertical_flip(void *image, int w, int h, int bytes_per_pixel)
+{
+   int row;
+   size_t bytes_per_row = (size_t)w * bytes_per_pixel;
+   stbi_uc temp[2048];
+   stbi_uc *bytes = (stbi_uc *)image;
+
+   for (row = 0; row < (h>>1); row++) {
+      stbi_uc *row0 = bytes + row*bytes_per_row;
+      stbi_uc *row1 = bytes + (h - row - 1)*bytes_per_row;
+      // swap row0 with row1
+      size_t bytes_left = bytes_per_row;
+      while (bytes_left) {
+         size_t bytes_copy = (bytes_left < sizeof(temp)) ? bytes_left : sizeof(temp);
+         memcpy(temp, row0, bytes_copy);
+         memcpy(row0, row1, bytes_copy);
+         memcpy(row1, temp, bytes_copy);
+         row0 += bytes_copy;
+         row1 += bytes_copy;
+         bytes_left -= bytes_copy;
+      }
+   }
+}
+
+#ifndef STBI_NO_GIF
+static void stbi__vertical_flip_slices(void *image, int w, int h, int z, int bytes_per_pixel)
+{
+   int slice;
+   int slice_size = w * h * bytes_per_pixel;
+
+   stbi_uc *bytes = (stbi_uc *)image;
+   for (slice = 0; slice < z; ++slice) {
+      stbi__vertical_flip(bytes, w, h, bytes_per_pixel);
+      bytes += slice_size;
+   }
+}
+#endif
+
+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;
+
+   // it is the responsibility of the loaders to make sure we get either 8 or 16 bit.
+   STBI_ASSERT(ri.bits_per_channel == 8 || ri.bits_per_channel == 16);
+
+   if (ri.bits_per_channel != 8) {
+      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 channels = req_comp ? req_comp : *comp;
+      stbi__vertical_flip(result, *x, *y, channels * sizeof(stbi_uc));
+   }
+
+   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;
+
+   // it is the responsibility of the loaders to make sure we get either 8 or 16 bit.
+   STBI_ASSERT(ri.bits_per_channel == 8 || ri.bits_per_channel == 16);
+
+   if (ri.bits_per_channel != 16) {
+      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 channels = req_comp ? req_comp : *comp;
+      stbi__vertical_flip(result, *x, *y, channels * sizeof(stbi__uint16));
+   }
+
+   return (stbi__uint16 *) result;
+}
+
+#if !defined(STBI_NO_HDR) && !defined(STBI_NO_LINEAR)
+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 channels = req_comp ? req_comp : *comp;
+      stbi__vertical_flip(result, *x, *y, channels * sizeof(float));
+   }
+}
+#endif
+
+#ifndef STBI_NO_STDIO
+
+#if defined(_WIN32) && defined(STBI_WINDOWS_UTF8)
+STBI_EXTERN __declspec(dllimport) int __stdcall MultiByteToWideChar(unsigned int cp, unsigned long flags, const char *str, int cbmb, wchar_t *widestr, int cchwide);
+STBI_EXTERN __declspec(dllimport) int __stdcall WideCharToMultiByte(unsigned int cp, unsigned long flags, const wchar_t *widestr, int cchwide, char *str, int cbmb, const char *defchar, int *used_default);
+#endif
+
+#if defined(_WIN32) && defined(STBI_WINDOWS_UTF8)
+STBIDEF int stbi_convert_wchar_to_utf8(char *buffer, size_t bufferlen, const wchar_t* input)
+{
+	return WideCharToMultiByte(65001 /* UTF8 */, 0, input, -1, buffer, (int) bufferlen, NULL, NULL);
+}
+#endif
+
+static FILE *stbi__fopen(char const *filename, char const *mode)
+{
+   FILE *f;
+#if defined(_WIN32) && defined(STBI_WINDOWS_UTF8)
+   wchar_t wMode[64];
+   wchar_t wFilename[1024];
+	if (0 == MultiByteToWideChar(65001 /* UTF8 */, 0, filename, -1, wFilename, sizeof(wFilename)/sizeof(*wFilename)))
+      return 0;
+
+	if (0 == MultiByteToWideChar(65001 /* UTF8 */, 0, mode, -1, wMode, sizeof(wMode)/sizeof(*wMode)))
+      return 0;
+
+#if defined(_MSC_VER) && _MSC_VER >= 1400
+	if (0 != _wfopen_s(&f, wFilename, wMode))
+		f = 0;
+#else
+   f = _wfopen(wFilename, wMode);
+#endif
+
+#elif 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_us *stbi_load_16_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *channels_in_file, int desired_channels)
+{
+   stbi__context s;
+   stbi__start_mem(&s,buffer,len);
+   return stbi__load_and_postprocess_16bit(&s,x,y,channels_in_file,desired_channels);
+}
+
+STBIDEF stbi_us *stbi_load_16_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *channels_in_file, int desired_channels)
+{
+   stbi__context s;
+   stbi__start_callbacks(&s, (stbi_io_callbacks *)clbk, user);
+   return stbi__load_and_postprocess_16bit(&s,x,y,channels_in_file,desired_channels);
+}
+
+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_GIF
+STBIDEF stbi_uc *stbi_load_gif_from_memory(stbi_uc const *buffer, int len, int **delays, int *x, int *y, int *z, int *comp, int req_comp)
+{
+   unsigned char *result;
+   stbi__context s;
+   stbi__start_mem(&s,buffer,len);
+
+   result = (unsigned char*) stbi__load_gif_main(&s, delays, x, y, z, comp, req_comp);
+   if (stbi__vertically_flip_on_load) {
+      stbi__vertical_flip_slices( result, *x, *y, *z, *comp );
+   }
+
+   return result;
+}
+#endif
+
+#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
+   long pos = ftell(f);
+   int res;
+   stbi__context s;
+   stbi__start_file(&s,f);
+   res = stbi__hdr_test(&s);
+   fseek(f, pos, SEEK_SET);
+   return res;
+   #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);
+   s->callback_already_read += (int) (s->img_buffer - s->img_buffer_original);
+   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;
+}
+
+#if defined(STBI_NO_JPEG) && defined(STBI_NO_HDR) && defined(STBI_NO_PIC) && defined(STBI_NO_PNM)
+// nothing
+#else
+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;
+}
+#endif
+
+#if defined(STBI_NO_JPEG) && defined(STBI_NO_PNG) && defined(STBI_NO_BMP) && defined(STBI_NO_PSD) && defined(STBI_NO_TGA) && defined(STBI_NO_GIF) && defined(STBI_NO_PIC)
+// nothing
+#else
+static void stbi__skip(stbi__context *s, int n)
+{
+   if (n == 0) return;  // already there!
+   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;
+}
+#endif
+
+#if defined(STBI_NO_PNG) && defined(STBI_NO_TGA) && defined(STBI_NO_HDR) && defined(STBI_NO_PNM)
+// nothing
+#else
+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;
+}
+#endif
+
+#if defined(STBI_NO_JPEG) && defined(STBI_NO_PNG) && defined(STBI_NO_PSD) && defined(STBI_NO_PIC)
+// nothing
+#else
+static int stbi__get16be(stbi__context *s)
+{
+   int z = stbi__get8(s);
+   return (z << 8) + stbi__get8(s);
+}
+#endif
+
+#if defined(STBI_NO_PNG) && defined(STBI_NO_PSD) && defined(STBI_NO_PIC)
+// nothing
+#else
+static stbi__uint32 stbi__get32be(stbi__context *s)
+{
+   stbi__uint32 z = stbi__get16be(s);
+   return (z << 16) + stbi__get16be(s);
+}
+#endif
+
+#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);
+   z += (stbi__uint32)stbi__get16le(s) << 16;
+   return z;
+}
+#endif
+
+#define STBI__BYTECAST(x)  ((stbi_uc) ((x) & 255))  // truncate int to byte without warnings
+
+#if defined(STBI_NO_JPEG) && defined(STBI_NO_PNG) && defined(STBI_NO_BMP) && defined(STBI_NO_PSD) && defined(STBI_NO_TGA) && defined(STBI_NO_GIF) && defined(STBI_NO_PIC) && defined(STBI_NO_PNM)
+// nothing
+#else
+//////////////////////////////////////////////////////////////////////////////
+//
+//  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);
+}
+#endif
+
+#if defined(STBI_NO_PNG) && defined(STBI_NO_BMP) && defined(STBI_NO_PSD) && defined(STBI_NO_TGA) && defined(STBI_NO_GIF) && defined(STBI_NO_PIC) && defined(STBI_NO_PNM)
+// nothing
+#else
+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); STBI_FREE(data); STBI_FREE(good); return stbi__errpuc("unsupported", "Unsupported format conversion");
+      }
+      #undef STBI__CASE
+   }
+
+   STBI_FREE(data);
+   return good;
+}
+#endif
+
+#if defined(STBI_NO_PNG) && defined(STBI_NO_PSD)
+// nothing
+#else
+static stbi__uint16 stbi__compute_y_16(int r, int g, int b)
+{
+   return (stbi__uint16) (((r*77) + (g*150) +  (29*b)) >> 8);
+}
+#endif
+
+#if defined(STBI_NO_PNG) && defined(STBI_NO_PSD)
+// nothing
+#else
+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); STBI_FREE(data); STBI_FREE(good); return (stbi__uint16*) stbi__errpuc("unsupported", "Unsupported format conversion");
+      }
+      #undef STBI__CASE
+   }
+
+   STBI_FREE(data);
+   return good;
+}
+#endif
+
+#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 (n < comp) {
+      for (i=0; i < x*y; ++i) {
+         output[i*comp + n] = data[i*comp + n]/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;
+   unsigned int 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);
+         if(k >= 257) return stbi__err("bad size list","Corrupt JPEG");
+      }
+   }
+   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 >= (1u << 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 * 256) + (run * 16) + (len + magbits));
+         }
+      }
+   }
+}
+
+static void stbi__grow_buffer_unsafe(stbi__jpeg *j)
+{
+   do {
+      unsigned 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 const 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];
+   if(c < 0 || c >= 256) // symbol id out of bounds!
+       return -1;
+   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 const int 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);
+   if (j->code_bits < n) return 0; // ran out of bits from stream, return 0s intead of continuing
+
+   sgn = j->code_buffer >> 31; // sign bit always in MSB; 0 if MSB clear (positive), 1 if MSB set (negative)
+   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__jbias[n] & (sgn - 1));
+}
+
+// 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);
+   if (j->code_bits < n) return 0; // ran out of bits from stream, return 0s intead of continuing
+   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);
+   if (j->code_bits < 1) return 0; // ran out of bits from stream, return 0s intead of continuing
+   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 const 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 || t > 15) 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;
+   if (!stbi__addints_valid(j->img_comp[b].dc_pred, diff)) return stbi__err("bad delta","Corrupt JPEG");
+   dc = j->img_comp[b].dc_pred + diff;
+   j->img_comp[b].dc_pred = dc;
+   if (!stbi__mul2shorts_valid(dc, dequant[0])) return stbi__err("can't merge dc and ac", "Corrupt JPEG");
+   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
+         if (s > j->code_bits) return stbi__err("bad huffman code", "Combined length longer than code bits available");
+         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);
+      if (t < 0 || t > 15) return stbi__err("can't merge dc and ac", "Corrupt JPEG");
+      diff = t ? stbi__extend_receive(j, t) : 0;
+
+      if (!stbi__addints_valid(j->img_comp[b].dc_pred, diff)) return stbi__err("bad delta", "Corrupt JPEG");
+      dc = j->img_comp[b].dc_pred + diff;
+      j->img_comp[b].dc_pred = dc;
+      if (!stbi__mul2shorts_valid(dc, 1 << j->succ_low)) return stbi__err("can't merge dc and ac", "Corrupt JPEG");
+      data[0] = (short) (dc * (1 << 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
+            if (s > j->code_bits) return stbi__err("bad huffman code", "Combined length longer than code bits available");
+            j->code_buffer <<= s;
+            j->code_bits -= s;
+            zig = stbi__jpeg_dezigzag[k++];
+            data[zig] = (short) ((r >> 8) * (1 << 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) * (1 << 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) * 4096)
+
+// 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]*4;
+         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]] = (stbi__uint16)(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];
+            }
+            if(n > 256) return stbi__err("bad DHT header","Corrupt JPEG"); // Loop over i < n would write past end of values!
+            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
+   if (s->img_y > STBI_MAX_DIMENSIONS) return stbi__err("too large","Very large image (corrupt?)");
+   if (s->img_x > STBI_MAX_DIMENSIONS) return stbi__err("too large","Very large image (corrupt?)");
+   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 const 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;
+   }
+
+   // check that plane subsampling factors are integer ratios; our resamplers can't deal with fractional ratios
+   // and I've never seen a non-corrupted JPEG file actually use them
+   for (i=0; i < s->img_n; ++i) {
+      if (h_max % z->img_comp[i].h != 0) return stbi__err("bad H","Corrupt JPEG");
+      if (v_max % z->img_comp[i].v != 0) return stbi__err("bad V","Corrupt JPEG");
+   }
+
+   // 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;
+}
+
+static int stbi__skip_jpeg_junk_at_end(stbi__jpeg *j)
+{
+   // some JPEGs have junk at end, skip over it but if we find what looks
+   // like a valid marker, resume there
+   while (!stbi__at_eof(j->s)) {
+      int x = stbi__get8(j->s);
+      while (x == 255) { // might be a marker
+         if (stbi__at_eof(j->s)) return STBI__MARKER_none;
+         x = stbi__get8(j->s);
+         if (x != 0x00 && x != 0xff) {
+            // not a stuffed zero or lead-in to another marker, looks
+            // like an actual marker, return it
+            return x;
+         }
+         // stuffed zero has x=0 now which ends the loop, meaning we go
+         // back to regular scan loop.
+         // repeated 0xff keeps trying to read the next byte of the marker.
+      }
+   }
+   return STBI__MARKER_none;
+}
+
+// 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 ) {
+         j->marker = stbi__skip_jpeg_junk_at_end(j);
+            // if we reach eof without hitting a marker, stbi__get_marker() below will fail and we'll eventually return 0
+         }
+         m = stbi__get_marker(j);
+         if (STBI__RESTART(m))
+            m = stbi__get_marker(j);
+      } else if (stbi__DNL(m)) {
+         int Ld = stbi__get16be(j->s);
+         stbi__uint32 NL = stbi__get16be(j->s);
+         if (Ld != 4) return stbi__err("bad DNL len", "Corrupt JPEG");
+         if (NL != j->s->img_y) return stbi__err("bad DNL height", "Corrupt JPEG");
+         m = stbi__get_marker(j);
+      } else {
+         if (!stbi__process_marker(j, m)) return 1;
+         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;
+
+   // nothing to do if no components requested; check this now to avoid
+   // accessing uninitialized coutput[0] later
+   if (decode_n <= 0) { stbi__cleanup_jpeg(z); return NULL; }
+
+   // resample and color-convert
+   {
+      int k;
+      unsigned int i,j;
+      stbi_uc *output;
+      stbi_uc *coutput[4] = { NULL, NULL, NULL, NULL };
+
+      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 m = coutput[3][i];
+                     out[0] = stbi__blinn_8x8(coutput[0][i], m);
+                     out[1] = stbi__blinn_8x8(coutput[1][i], m);
+                     out[2] = stbi__blinn_8x8(coutput[2][i], m);
+                     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 m = coutput[3][i];
+                     out[0] = stbi__blinn_8x8(255 - out[0], m);
+                     out[1] = stbi__blinn_8x8(255 - out[1], m);
+                     out[2] = stbi__blinn_8x8(255 - out[2], m);
+                     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 m = coutput[3][i];
+                  stbi_uc r = stbi__blinn_8x8(coutput[0][i], m);
+                  stbi_uc g = stbi__blinn_8x8(coutput[1][i], m);
+                  stbi_uc b = stbi__blinn_8x8(coutput[2][i], m);
+                  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));
+   if (!j) return stbi__errpuc("outofmem", "Out of memory");
+   memset(j, 0, 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));
+   if (!j) return stbi__err("outofmem", "Out of memory");
+   memset(j, 0, 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)));
+   if (!j) return stbi__err("outofmem", "Out of memory");
+   memset(j, 0, 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)
+#define STBI__ZNSYMS 288 // number of symbols in literal/length alphabet
+
+// 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[STBI__ZNSYMS];
+   stbi__uint16 value[STBI__ZNSYMS];
+} 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 int stbi__zeof(stbi__zbuf *z)
+{
+   return (z->zbuffer >= z->zbuffer_end);
+}
+
+stbi_inline static stbi_uc stbi__zget8(stbi__zbuf *z)
+{
+   return stbi__zeof(z) ? 0 : *z->zbuffer++;
+}
+
+static void stbi__fill_bits(stbi__zbuf *z)
+{
+   do {
+      if (z->code_buffer >= (1U << z->num_bits)) {
+        z->zbuffer = z->zbuffer_end;  /* treat this as EOF so we fail. */
+        return;
+      }
+      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];
+   if (b >= STBI__ZNSYMS) return -1; // some data was corrupt somewhere!
+   if (z->size[b] != s) return -1;  // was originally an assert, but report failure instead.
+   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) {
+      if (stbi__zeof(a)) {
+         return -1;   /* report error for unexpected end of data. */
+      }
+      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;
+   unsigned int cur, limit, old_limit;
+   z->zout = zout;
+   if (!z->z_expandable) return stbi__err("output buffer limit","Corrupt PNG");
+   cur   = (unsigned int) (z->zout - z->zout_start);
+   limit = old_limit = (unsigned) (z->zout_end - z->zout_start);
+   if (UINT_MAX - cur < (unsigned) n) return stbi__err("outofmem", "Out of memory");
+   while (cur + n > limit) {
+      if(limit > UINT_MAX / 2) return stbi__err("outofmem", "Out of memory");
+      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 const 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 const 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 const 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 const 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;
+         }
+         if (z >= 286) return stbi__err("bad huffman code","Corrupt PNG"); // per DEFLATE, length codes 286 and 287 must not appear in compressed data
+         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 || z >= 30) return stbi__err("bad huffman code","Corrupt PNG"); // per DEFLATE, distance codes 30 and 31 must not appear in compressed data
+         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 const 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 if (c == 18) {
+            c = stbi__zreceive(a,7)+11;
+         } else {
+            return stbi__err("bad codelengths", "Corrupt PNG");
+         }
+         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;
+   }
+   if (a->num_bits < 0) return stbi__err("zlib corrupt","Corrupt PNG");
+   // 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 (stbi__zeof(a)) return stbi__err("bad zlib header","Corrupt PNG"); // zlib spec
+   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[STBI__ZNSYMS] =
+{
+   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  , STBI__ZNSYMS)) 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 const 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 const 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");
+
+   if (!stbi__mad3sizes_valid(img_n, x, depth, 7)) return stbi__err("too large", "Corrupt PNG");
+   img_width_bytes = (((img_n * x * depth) + 7) >> 3);
+   img_len = (img_width_bytes + 1) * y;
+
+   // we used to check for exact match between raw_len and img_len on non-interlaced PNGs,
+   // but issue #276 reported a PNG in the wild that had extra data at the end (all zeros),
+   // so just check for raw_len < img_len always.
+   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) {
+         if (img_width_bytes > x) return stbi__err("invalid width","Corrupt PNG");
+         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);
+   if (!final) return stbi__err("outofmem", "Out of memory");
+   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_global = 0;
+static int stbi__de_iphone_flag_global = 0;
+
+STBIDEF void stbi_set_unpremultiply_on_load(int flag_true_if_should_unpremultiply)
+{
+   stbi__unpremultiply_on_load_global = flag_true_if_should_unpremultiply;
+}
+
+STBIDEF void stbi_convert_iphone_png_to_rgb(int flag_true_if_should_convert)
+{
+   stbi__de_iphone_flag_global = flag_true_if_should_convert;
+}
+
+#ifndef STBI_THREAD_LOCAL
+#define stbi__unpremultiply_on_load  stbi__unpremultiply_on_load_global
+#define stbi__de_iphone_flag  stbi__de_iphone_flag_global
+#else
+static STBI_THREAD_LOCAL int stbi__unpremultiply_on_load_local, stbi__unpremultiply_on_load_set;
+static STBI_THREAD_LOCAL int stbi__de_iphone_flag_local, stbi__de_iphone_flag_set;
+
+STBIDEF void stbi_set_unpremultiply_on_load_thread(int flag_true_if_should_unpremultiply)
+{
+   stbi__unpremultiply_on_load_local = flag_true_if_should_unpremultiply;
+   stbi__unpremultiply_on_load_set = 1;
+}
+
+STBIDEF void stbi_convert_iphone_png_to_rgb_thread(int flag_true_if_should_convert)
+{
+   stbi__de_iphone_flag_local = flag_true_if_should_convert;
+   stbi__de_iphone_flag_set = 1;
+}
+
+#define stbi__unpremultiply_on_load  (stbi__unpremultiply_on_load_set           \
+                                       ? stbi__unpremultiply_on_load_local      \
+                                       : stbi__unpremultiply_on_load_global)
+#define stbi__de_iphone_flag  (stbi__de_iphone_flag_set                         \
+                                ? stbi__de_iphone_flag_local                    \
+                                : stbi__de_iphone_flag_global)
+#endif // STBI_THREAD_LOCAL
+
+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) {
+               stbi_uc half = a / 2;
+               p[0] = (p[2] * 255 + half) / a;
+               p[1] = (p[1] * 255 + half) / a;
+               p[2] = ( t   * 255 + half) / 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)  (((unsigned) (a) << 24) + ((unsigned) (b) << 16) + ((unsigned) (c) << 8) + (unsigned) (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]={0};
+   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);
+            s->img_y = stbi__get32be(s);
+            if (s->img_y > STBI_MAX_DIMENSIONS) return stbi__err("too large","Very large image (corrupt?)");
+            if (s->img_x > STBI_MAX_DIMENSIONS) 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");
+            } 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");
+            }
+            // even with 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;
+               // non-paletted with tRNS = constant alpha. if header-scanning, we can stop now.
+               if (scan == STBI__SCAN_header) { ++s->img_n; return 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) {
+               // header scan definitely stops at first IDAT
+               if (pal_img_n)
+                  s->img_n = pal_img_n;
+               return 1;
+            }
+            if (c.length > (1u << 30)) return stbi__err("IDAT size limit", "IDAT section larger than 2^30 bytes");
+            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;
+            } else if (has_trans) {
+               // non-paletted image with tRNS -> source image has (constant) alpha
+               ++s->img_n;
+            }
+            STBI_FREE(z->expanded); z->expanded = NULL;
+            // end of PNG chunk, read and skip CRC
+            stbi__get32be(s);
+            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 if (p->depth == 16)
+         ri->bits_per_channel = 16;
+      else
+         return stbi__errpuc("bad bits_per_channel", "PNG not supported: unsupported color 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);
+}
+
+static int stbi__png_is16(stbi__context *s)
+{
+   stbi__png p;
+   p.s = s;
+   if (!stbi__png_info_raw(&p, NULL, NULL, NULL))
+	   return 0;
+   if (p.depth != 16) {
+      stbi__rewind(p.s);
+      return 0;
+   }
+   return 1;
+}
+#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;/* >>=  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;
+}
+
+// extract an arbitrarily-aligned N-bit value (N=bits)
+// from v, and then make it 8-bits long and fractionally
+// extend it to full full range.
+static int stbi__shiftsigned(unsigned int v, int shift, int bits)
+{
+   static unsigned int mul_table[9] = {
+      0,
+      0xff/*0b11111111*/, 0x55/*0b01010101*/, 0x49/*0b01001001*/, 0x11/*0b00010001*/,
+      0x21/*0b00100001*/, 0x41/*0b01000001*/, 0x81/*0b10000001*/, 0x01/*0b00000001*/,
+   };
+   static unsigned int shift_table[9] = {
+      0, 0,0,1,0,2,4,6,0,
+   };
+   if (shift < 0)
+      v <<= -shift;
+   else
+      v >>= shift;
+   STBI_ASSERT(v < 256);
+   v >>= (8-bits);
+   STBI_ASSERT(bits >= 0 && bits <= 8);
+   return (int) ((unsigned) v * mul_table[bits]) >> shift_table[bits];
+}
+
+typedef struct
+{
+   int bpp, offset, hsz;
+   unsigned int mr,mg,mb,ma, all_a;
+   int extra_read;
+} stbi__bmp_data;
+
+static int stbi__bmp_set_mask_defaults(stbi__bmp_data *info, int compress)
+{
+   // BI_BITFIELDS specifies masks explicitly, don't override
+   if (compress == 3)
+      return 1;
+
+   if (compress == 0) {
+      if (info->bpp == 16) {
+         info->mr = 31u << 10;
+         info->mg = 31u <<  5;
+         info->mb = 31u <<  0;
+      } else 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 {
+         // otherwise, use defaults, which is all-0
+         info->mr = info->mg = info->mb = info->ma = 0;
+      }
+      return 1;
+   }
+   return 0; // error
+}
+
+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;
+   info->extra_read = 14;
+
+   if (info->offset < 0) return stbi__errpuc("bad BMP", "bad BMP");
+
+   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 (hsz != 12) {
+      int compress = stbi__get32le(s);
+      if (compress == 1 || compress == 2) return stbi__errpuc("BMP RLE", "BMP type not supported: RLE");
+      if (compress >= 4) return stbi__errpuc("BMP JPEG/PNG", "BMP type not supported: unsupported compression"); // this includes PNG/JPEG modes
+      if (compress == 3 && info->bpp != 16 && info->bpp != 32) return stbi__errpuc("bad BMP", "bad BMP"); // bitfields requires 16 or 32 bits/pixel
+      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) {
+               stbi__bmp_set_mask_defaults(info, compress);
+            } else if (compress == 3) {
+               info->mr = stbi__get32le(s);
+               info->mg = stbi__get32le(s);
+               info->mb = stbi__get32le(s);
+               info->extra_read += 12;
+               // 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 {
+         // V4/V5 header
+         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);
+         if (compress != 3) // override mr/mg/mb unless in BI_BITFIELDS mode, as per docs
+            stbi__bmp_set_mask_defaults(info, compress);
+         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);
+
+   if (s->img_y > STBI_MAX_DIMENSIONS) return stbi__errpuc("too large","Very large image (corrupt?)");
+   if (s->img_x > STBI_MAX_DIMENSIONS) return stbi__errpuc("too large","Very large image (corrupt?)");
+
+   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 - info.extra_read - 24) / 3;
+   } else {
+      if (info.bpp < 16)
+         psize = (info.offset - info.extra_read - info.hsz) >> 2;
+   }
+   if (psize == 0) {
+      // accept some number of extra bytes after the header, but if the offset points either to before
+      // the header ends or implies a large amount of extra data, reject the file as malformed
+      int bytes_read_so_far = s->callback_already_read + (int)(s->img_buffer - s->img_buffer_original);
+      int header_limit = 1024; // max we actually read is below 256 bytes currently.
+      int extra_data_limit = 256*4; // what ordinarily goes here is a palette; 256 entries*4 bytes is its max size.
+      if (bytes_read_so_far <= 0 || bytes_read_so_far > header_limit) {
+         return stbi__errpuc("bad header", "Corrupt BMP");
+      }
+      // we established that bytes_read_so_far is positive and sensible.
+      // the first half of this test rejects offsets that are either too small positives, or
+      // negative, and guarantees that info.offset >= bytes_read_so_far > 0. this in turn
+      // ensures the number computed in the second half of the test can't overflow.
+      if (info.offset < bytes_read_so_far || info.offset - bytes_read_so_far > extra_data_limit) {
+         return stbi__errpuc("bad offset", "Corrupt BMP");
+      } else {
+         stbi__skip(s, info.offset - bytes_read_so_far);
+      }
+   }
+
+   if (info.bpp == 24 && ma == 0xff000000)
+      s->img_n = 3;
+   else
+      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 - info.extra_read - info.hsz - psize * (info.hsz == 12 ? 3 : 4));
+      if (info.bpp == 1) width = (s->img_x + 7) >> 3;
+      else 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;
+      if (info.bpp == 1) {
+         for (j=0; j < (int) s->img_y; ++j) {
+            int bit_offset = 7, v = stbi__get8(s);
+            for (i=0; i < (int) s->img_x; ++i) {
+               int color = (v>>bit_offset)&0x1;
+               out[z++] = pal[color][0];
+               out[z++] = pal[color][1];
+               out[z++] = pal[color][2];
+               if (target == 4) out[z++] = 255;
+               if (i+1 == (int) s->img_x) break;
+               if((--bit_offset) < 0) {
+                  bit_offset = 7;
+                  v = stbi__get8(s);
+               }
+            }
+            stbi__skip(s, pad);
+         }
+      } else {
+         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 - info.extra_read - 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);
+         if (rcount > 8 || gcount > 8 || bcount > 8 || acount > 8) { STBI_FREE(out); return stbi__errpuc("bad masks", "Corrupt BMP"); }
+      }
+      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));
+               unsigned 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;
+               // fallthrough
+      case 15: if(is_rgb16) *is_rgb16 = 1;
+               return STBI_rgb;
+      case 24: // fallthrough
+      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);
+   STBI_NOTUSED(tga_x_origin); // @TODO
+   STBI_NOTUSED(tga_y_origin); // @TODO
+
+   if (tga_height > STBI_MAX_DIMENSIONS) return stbi__errpuc("too large","Very large image (corrupt?)");
+   if (tga_width > STBI_MAX_DIMENSIONS) return stbi__errpuc("too large","Very large image (corrupt?)");
+
+   //   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)
+      {
+         if (tga_palette_len == 0) {  /* you have to have at least one entry! */
+            STBI_FREE(tga_data);
+            return stbi__errpuc("bad palette", "Corrupt TGA");
+         }
+
+         //   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;
+   STBI_NOTUSED(tga_palette_start);
+   //   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);
+
+   if (h > STBI_MAX_DIMENSIONS) return stbi__errpuc("too large","Very large image (corrupt?)");
+   if (w > STBI_MAX_DIMENSIONS) return stbi__errpuc("too large","Very large image (corrupt?)");
+
+   // 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 preceded 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 (y > STBI_MAX_DIMENSIONS) return stbi__errpuc("too large","Very large image (corrupt?)");
+   if (x > STBI_MAX_DIMENSIONS) return stbi__errpuc("too large","Very large image (corrupt?)");
+
+   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);
+   if (!result) return stbi__errpuc("outofmem", "Out of memory");
+   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;                 // output buffer (always 4 components)
+   stbi_uc *background;          // The current "background" as far as a gif is concerned
+   stbi_uc *history;
+   int flags, bgindex, ratio, transparent, eflags;
+   stbi_uc  pal[256][4];
+   stbi_uc lpal[256][4];
+   stbi__gif_lzw codes[8192];
+   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;
+   int delay;
+} 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 (g->w > STBI_MAX_DIMENSIONS) return stbi__err("too large","Very large image (corrupt?)");
+   if (g->h > STBI_MAX_DIMENSIONS) return stbi__err("too large","Very large image (corrupt?)");
+
+   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 (!g) return stbi__err("outofmem", "Out of memory");
+   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;
+   int idx;
+
+   // 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;
+
+   idx = g->cur_x + g->cur_y;
+   p = &g->out[idx];
+   g->history[idx / 4] = 1;
+
+   c = &g->color_table[g->codes[code].suffix * 4];
+   if (c[3] > 128) { // don't render transparent pixels;
+      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 > 8192) {
+                  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");
+         }
+      }
+   }
+}
+
+// this function is designed to support animated gifs, although stb_image doesn't support it
+// two back is the image from two frames ago, used for a very specific disposal format
+static stbi_uc *stbi__gif_load_next(stbi__context *s, stbi__gif *g, int *comp, int req_comp, stbi_uc *two_back)
+{
+   int dispose;
+   int first_frame;
+   int pi;
+   int pcount;
+   STBI_NOTUSED(req_comp);
+
+   // on first frame, any non-written pixels get the background colour (non-transparent)
+   first_frame = 0;
+   if (g->out == 0) {
+      if (!stbi__gif_header(s, g, comp,0)) return 0; // stbi__g_failure_reason set by stbi__gif_header
+      if (!stbi__mad3sizes_valid(4, g->w, g->h, 0))
+         return stbi__errpuc("too large", "GIF image is too large");
+      pcount = g->w * g->h;
+      g->out = (stbi_uc *) stbi__malloc(4 * pcount);
+      g->background = (stbi_uc *) stbi__malloc(4 * pcount);
+      g->history = (stbi_uc *) stbi__malloc(pcount);
+      if (!g->out || !g->background || !g->history)
+         return stbi__errpuc("outofmem", "Out of memory");
+
+      // image is treated as "transparent" at the start - ie, nothing overwrites the current background;
+      // background colour is only used for pixels that are not rendered first frame, after that "background"
+      // color refers to the color that was there the previous frame.
+      memset(g->out, 0x00, 4 * pcount);
+      memset(g->background, 0x00, 4 * pcount); // state of the background (starts transparent)
+      memset(g->history, 0x00, pcount);        // pixels that were affected previous frame
+      first_frame = 1;
+   } else {
+      // second frame - how do we dispose of the previous one?
+      dispose = (g->eflags & 0x1C) >> 2;
+      pcount = g->w * g->h;
+
+      if ((dispose == 3) && (two_back == 0)) {
+         dispose = 2; // if I don't have an image to revert back to, default to the old background
+      }
+
+      if (dispose == 3) { // use previous graphic
+         for (pi = 0; pi < pcount; ++pi) {
+            if (g->history[pi]) {
+               memcpy( &g->out[pi * 4], &two_back[pi * 4], 4 );
+            }
+         }
+      } else if (dispose == 2) {
+         // restore what was changed last frame to background before that frame;
+         for (pi = 0; pi < pcount; ++pi) {
+            if (g->history[pi]) {
+               memcpy( &g->out[pi * 4], &g->background[pi * 4], 4 );
+            }
+         }
+      } else {
+         // This is a non-disposal case eithe way, so just
+         // leave the pixels as is, and they will become the new background
+         // 1: do not dispose
+         // 0:  not specified.
+      }
+
+      // background is what out is after the undoing of the previou frame;
+      memcpy( g->background, g->out, 4 * g->w * g->h );
+   }
+
+   // clear my history;
+   memset( g->history, 0x00, g->w * g->h );        // pixels that were affected previous frame
+
+   for (;;) {
+      int tag = stbi__get8(s);
+      switch (tag) {
+         case 0x2C: /* Image Descriptor */
+         {
+            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;
+
+            // if the width of the specified rectangle is 0, that means
+            // we may not see *any* pixels or the image is malformed;
+            // to make sure this is caught, move the current y down to
+            // max_y (which is what out_gif_code checks).
+            if (w == 0)
+               g->cur_y = g->max_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) {
+               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) return NULL;
+
+            // if this was the first frame,
+            pcount = g->w * g->h;
+            if (first_frame && (g->bgindex > 0)) {
+               // if first frame, any pixel not drawn to gets the background color
+               for (pi = 0; pi < pcount; ++pi) {
+                  if (g->history[pi] == 0) {
+                     g->pal[g->bgindex][3] = 255; // just in case it was made transparent, undo that; It will be reset next frame if need be;
+                     memcpy( &g->out[pi * 4], &g->pal[g->bgindex], 4 );
+                  }
+               }
+            }
+
+            return o;
+         }
+
+         case 0x21: // Comment Extension.
+         {
+            int len;
+            int ext = stbi__get8(s);
+            if (ext == 0xF9) { // Graphic Control Extension.
+               len = stbi__get8(s);
+               if (len == 4) {
+                  g->eflags = stbi__get8(s);
+                  g->delay = 10 * stbi__get16le(s); // delay - 1/100th of a second, saving as 1/1000ths.
+
+                  // unset old transparent
+                  if (g->transparent >= 0) {
+                     g->pal[g->transparent][3] = 255;
+                  }
+                  if (g->eflags & 0x01) {
+                     g->transparent = stbi__get8(s);
+                     if (g->transparent >= 0) {
+                        g->pal[g->transparent][3] = 0;
+                     }
+                  } else {
+                     // don't need transparent
+                     stbi__skip(s, 1);
+                     g->transparent = -1;
+                  }
+               } 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");
+      }
+   }
+}
+
+static void *stbi__load_gif_main_outofmem(stbi__gif *g, stbi_uc *out, int **delays)
+{
+   STBI_FREE(g->out);
+   STBI_FREE(g->history);
+   STBI_FREE(g->background);
+
+   if (out) STBI_FREE(out);
+   if (delays && *delays) STBI_FREE(*delays);
+   return stbi__errpuc("outofmem", "Out of memory");
+}
+
+static void *stbi__load_gif_main(stbi__context *s, int **delays, int *x, int *y, int *z, int *comp, int req_comp)
+{
+   if (stbi__gif_test(s)) {
+      int layers = 0;
+      stbi_uc *u = 0;
+      stbi_uc *out = 0;
+      stbi_uc *two_back = 0;
+      stbi__gif g;
+      int stride;
+      int out_size = 0;
+      int delays_size = 0;
+
+      STBI_NOTUSED(out_size);
+      STBI_NOTUSED(delays_size);
+
+      memset(&g, 0, sizeof(g));
+      if (delays) {
+         *delays = 0;
+      }
+
+      do {
+         u = stbi__gif_load_next(s, &g, comp, req_comp, two_back);
+         if (u == (stbi_uc *) s) u = 0;  // end of animated gif marker
+
+         if (u) {
+            *x = g.w;
+            *y = g.h;
+            ++layers;
+            stride = g.w * g.h * 4;
+
+            if (out) {
+               void *tmp = (stbi_uc*) STBI_REALLOC_SIZED( out, out_size, layers * stride );
+               if (!tmp)
+                  return stbi__load_gif_main_outofmem(&g, out, delays);
+               else {
+                   out = (stbi_uc*) tmp;
+                   out_size = layers * stride;
+               }
+
+               if (delays) {
+                  int *new_delays = (int*) STBI_REALLOC_SIZED( *delays, delays_size, sizeof(int) * layers );
+                  if (!new_delays)
+                     return stbi__load_gif_main_outofmem(&g, out, delays);
+                  *delays = new_delays;
+                  delays_size = layers * sizeof(int);
+               }
+            } else {
+               out = (stbi_uc*)stbi__malloc( layers * stride );
+               if (!out)
+                  return stbi__load_gif_main_outofmem(&g, out, delays);
+               out_size = layers * stride;
+               if (delays) {
+                  *delays = (int*) stbi__malloc( layers * sizeof(int) );
+                  if (!*delays)
+                     return stbi__load_gif_main_outofmem(&g, out, delays);
+                  delays_size = layers * sizeof(int);
+               }
+            }
+            memcpy( out + ((layers - 1) * stride), u, stride );
+            if (layers >= 2) {
+               two_back = out - 2 * stride;
+            }
+
+            if (delays) {
+               (*delays)[layers - 1U] = g.delay;
+            }
+         }
+      } while (u != 0);
+
+      // free temp buffer;
+      STBI_FREE(g.out);
+      STBI_FREE(g.history);
+      STBI_FREE(g.background);
+
+      // do the final conversion after loading everything;
+      if (req_comp && req_comp != 4)
+         out = stbi__convert_format(out, 4, req_comp, layers * g.w, g.h);
+
+      *z = layers;
+      return out;
+   } else {
+      return stbi__errpuc("not GIF", "Image was not as a gif type.");
+   }
+}
+
+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;
+   memset(&g, 0, sizeof(g));
+   STBI_NOTUSED(ri);
+
+   u = stbi__gif_load_next(s, &g, comp, req_comp, 0);
+   if (u == (stbi_uc *) s) u = 0;  // end of animated gif marker
+   if (u) {
+      *x = g.w;
+      *y = g.h;
+
+      // moved conversion to after successful load so that the same
+      // can be done for multiple frames.
+      if (req_comp && req_comp != 4)
+         u = stbi__convert_format(u, 4, req_comp, g.w, g.h);
+   } else if (g.out) {
+      // if there was an error and we allocated an image buffer, free it!
+      STBI_FREE(g.out);
+   }
+
+   // free buffers needed for multiple frame loading;
+   STBI_FREE(g.history);
+   STBI_FREE(g.background);
+
+   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);
+
+   if (height > STBI_MAX_DIMENSIONS) return stbi__errpf("too large","Very large image (corrupt?)");
+   if (width > STBI_MAX_DIMENSIONS) return stbi__errpf("too large","Very large image (corrupt?)");
+
+   *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 == 0) || (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 == 0) || (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);
+   if (p == NULL) {
+      stbi__rewind( s );
+      return 0;
+   }
+   if (x) *x = s->img_x;
+   if (y) *y = s->img_y;
+   if (comp) {
+      if (info.bpp == 24 && info.ma == 0xff000000)
+         *comp = 3;
+      else
+         *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, depth;
+   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);
+   depth = stbi__get16be(s);
+   if (depth != 8 && depth != 16) {
+       stbi__rewind( s );
+       return 0;
+   }
+   if (stbi__get16be(s) != 3) {
+       stbi__rewind( s );
+       return 0;
+   }
+   *comp = 4;
+   return 1;
+}
+
+static int stbi__psd_is16(stbi__context *s)
+{
+   int channelCount, depth;
+   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;
+   }
+   STBI_NOTUSED(stbi__get32be(s));
+   STBI_NOTUSED(stbi__get32be(s));
+   depth = stbi__get16be(s);
+   if (depth != 16) {
+       stbi__rewind( s );
+       return 0;
+   }
+   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)
+
+#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);
+
+   ri->bits_per_channel = stbi__pnm_info(s, (int *)&s->img_x, (int *)&s->img_y, (int *)&s->img_n);
+   if (ri->bits_per_channel == 0)
+      return 0;
+
+   if (s->img_y > STBI_MAX_DIMENSIONS) return stbi__errpuc("too large","Very large image (corrupt?)");
+   if (s->img_x > STBI_MAX_DIMENSIONS) return stbi__errpuc("too large","Very large image (corrupt?)");
+
+   *x = s->img_x;
+   *y = s->img_y;
+   if (comp) *comp = s->img_n;
+
+   if (!stbi__mad4sizes_valid(s->img_n, s->img_x, s->img_y, ri->bits_per_channel / 8, 0))
+      return stbi__errpuc("too large", "PNM too large");
+
+   out = (stbi_uc *) stbi__malloc_mad4(s->img_n, s->img_x, s->img_y, ri->bits_per_channel / 8, 0);
+   if (!out) return stbi__errpuc("outofmem", "Out of memory");
+   if (!stbi__getn(s, out, s->img_n * s->img_x * s->img_y * (ri->bits_per_channel / 8))) {
+      STBI_FREE(out);
+      return stbi__errpuc("bad PNM", "PNM file truncated");
+   }
+
+   if (req_comp && req_comp != s->img_n) {
+      if (ri->bits_per_channel == 16) {
+         out = (stbi_uc *) stbi__convert_format16((stbi__uint16 *) out, s->img_n, req_comp, s->img_x, s->img_y);
+      } else {
+         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);
+      if((value > 214748364) || (value == 214748364 && *c > '7'))
+          return stbi__err("integer parse overflow", "Parsing an integer in the PPM header overflowed a 32-bit int");
+   }
+
+   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
+   if(*x == 0)
+       return stbi__err("invalid width", "PPM image header had zero or overflowing width");
+   stbi__pnm_skip_whitespace(s, &c);
+
+   *y = stbi__pnm_getinteger(s, &c); // read height
+   if (*y == 0)
+       return stbi__err("invalid width", "PPM image header had zero or overflowing width");
+   stbi__pnm_skip_whitespace(s, &c);
+
+   maxv = stbi__pnm_getinteger(s, &c);  // read max value
+   if (maxv > 65535)
+      return stbi__err("max value > 65535", "PPM image supports only 8-bit and 16-bit images");
+   else if (maxv > 255)
+      return 16;
+   else
+      return 8;
+}
+
+static int stbi__pnm_is16(stbi__context *s)
+{
+   if (stbi__pnm_info(s, NULL, NULL, NULL) == 16)
+	   return 1;
+   return 0;
+}
+#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");
+}
+
+static int stbi__is_16_main(stbi__context *s)
+{
+   #ifndef STBI_NO_PNG
+   if (stbi__png_is16(s))  return 1;
+   #endif
+
+   #ifndef STBI_NO_PSD
+   if (stbi__psd_is16(s))  return 1;
+   #endif
+
+   #ifndef STBI_NO_PNM
+   if (stbi__pnm_is16(s))  return 1;
+   #endif
+   return 0;
+}
+
+#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;
+}
+
+STBIDEF int stbi_is_16_bit(char const *filename)
+{
+    FILE *f = stbi__fopen(filename, "rb");
+    int result;
+    if (!f) return stbi__err("can't fopen", "Unable to open file");
+    result = stbi_is_16_bit_from_file(f);
+    fclose(f);
+    return result;
+}
+
+STBIDEF int stbi_is_16_bit_from_file(FILE *f)
+{
+   int r;
+   stbi__context s;
+   long pos = ftell(f);
+   stbi__start_file(&s, f);
+   r = stbi__is_16_main(&s);
+   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);
+}
+
+STBIDEF int stbi_is_16_bit_from_memory(stbi_uc const *buffer, int len)
+{
+   stbi__context s;
+   stbi__start_mem(&s,buffer,len);
+   return stbi__is_16_main(&s);
+}
+
+STBIDEF int stbi_is_16_bit_from_callbacks(stbi_io_callbacks const *c, void *user)
+{
+   stbi__context s;
+   stbi__start_callbacks(&s, (stbi_io_callbacks *) c, user);
+   return stbi__is_16_main(&s);
+}
+
+#endif // STB_IMAGE_IMPLEMENTATION
+
+/*
+   revision history:
+      2.20  (2019-02-07) support utf8 filenames in Windows; fix warnings and platform ifdefs
+      2.19  (2018-02-11) fix warning
+      2.18  (2018-01-30) fix warnings
+      2.17  (2018-01-29) change sbti__shiftsigned to avoid clang -O2 bug
+                         1-bit BMP
+                         *_is_16_bit api
+                         avoid warnings
+      2.16  (2017-07-23) all functions have 16-bit variants;
+                         STBI_NO_STDIO works again;
+                         compilation fixes;
+                         fix rounding in unpremultiply;
+                         optimize vertical flip;
+                         disable raw_len validation;
+                         documentation fixes
+      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.
+------------------------------------------------------------------------------
+*/

Engine/source/gfx/bitmap/loaders/stb/stb_image_resize2.h

@@ -0,0 +1,10303 @@
+/* stb_image_resize2 - v2.01 - public domain image resizing
+   
+   by Jeff Roberts (v2) and Jorge L Rodriguez 
+   http://github.com/nothings/stb
+
+   Can be threaded with the extended API. SSE2, AVX, Neon and WASM SIMD support. Only 
+   scaling and translation is supported, no rotations or shears.
+
+   COMPILING & LINKING
+      In one C/C++ file that #includes this file, do this:
+         #define STB_IMAGE_RESIZE_IMPLEMENTATION
+      before the #include. That will create the implementation in that file.
+
+   PORTING FROM VERSION 1
+
+      The API has changed. You can continue to use the old version of stb_image_resize.h,
+      which is available in the "deprecated/" directory.
+
+      If you're using the old simple-to-use API, porting is straightforward.
+      (For more advanced APIs, read the documentation.)
+
+        stbir_resize_uint8():
+          - call `stbir_resize_uint8_linear`, cast channel count to `stbir_pixel_layout`
+
+        stbir_resize_float():
+          - call `stbir_resize_float_linear`, cast channel count to `stbir_pixel_layout`
+
+        stbir_resize_uint8_srgb():
+          - function name is unchanged
+          - cast channel count to `stbir_pixel_layout`
+          - above is sufficient unless your image has alpha and it's not RGBA/BGRA
+            - in that case, follow the below instructions for stbir_resize_uint8_srgb_edgemode
+
+        stbir_resize_uint8_srgb_edgemode()
+          - switch to the "medium complexity" API
+          - stbir_resize(), very similar API but a few more parameters:
+            - pixel_layout: cast channel count to `stbir_pixel_layout`
+            - data_type:    STBIR_TYPE_UINT8_SRGB
+            - edge:         unchanged (STBIR_EDGE_WRAP, etc.)
+            - filter:       STBIR_FILTER_DEFAULT
+          - which channel is alpha is specified in stbir_pixel_layout, see enum for details
+
+   EASY API CALLS:
+     Easy API downsamples w/Mitchell filter, upsamples w/cubic interpolation, clamps to edge.
+
+     stbir_resize_uint8_srgb( input_pixels,  input_w,  input_h,  input_stride_in_bytes,
+                              output_pixels, output_w, output_h, output_stride_in_bytes,
+                              pixel_layout_enum )
+
+     stbir_resize_uint8_linear( input_pixels,  input_w,  input_h,  input_stride_in_bytes,
+                                output_pixels, output_w, output_h, output_stride_in_bytes,
+                                pixel_layout_enum )
+
+     stbir_resize_float_linear( input_pixels,  input_w,  input_h,  input_stride_in_bytes,
+                                output_pixels, output_w, output_h, output_stride_in_bytes,
+                                pixel_layout_enum )
+
+     If you pass NULL or zero for the output_pixels, we will allocate the output buffer
+     for you and return it from the function (free with free() or STBIR_FREE).
+     As a special case, XX_stride_in_bytes of 0 means packed continuously in memory.
+
+   API LEVELS
+      There are three levels of API - easy-to-use, medium-complexity and extended-complexity.
+
+      See the "header file" section of the source for API documentation.
+
+   ADDITIONAL DOCUMENTATION
+
+      MEMORY ALLOCATION
+         By default, we use malloc and free for memory allocation.  To override the 
+         memory allocation, before the implementation #include, add a:
+
+            #define STBIR_MALLOC(size,user_data) ...
+            #define STBIR_FREE(ptr,user_data)   ...
+
+         Each resize makes exactly one call to malloc/free (unless you use the 
+         extended API where you can do one allocation for many resizes). Under
+         address sanitizer, we do separate allocations to find overread/writes.
+
+      PERFORMANCE
+         This library was written with an emphasis on performance. When testing
+         stb_image_resize with RGBA, the fastest mode is STBIR_4CHANNEL with 
+         STBIR_TYPE_UINT8 pixels and CLAMPed edges (which is what many other resize
+         libs do by default). Also, make sure SIMD is turned on of course (default 
+         for 64-bit targets). Avoid WRAP edge mode if you want the fastest speed.
+
+         This library also comes with profiling built-in. If you define STBIR_PROFILE,
+         you can use the advanced API and get low-level profiling information by 
+         calling stbir_resize_extended_profile_info() or stbir_resize_split_profile_info()
+         after a resize.
+
+      SIMD
+         Most of the routines have optimized SSE2, AVX, NEON and WASM versions. 
+
+         On Microsoft compilers, we automatically turn on SIMD for 64-bit x64 and 
+         ARM; for 32-bit x86 and ARM, you select SIMD mode by defining STBIR_SSE2 or 
+         STBIR_NEON. For AVX and AVX2, we auto-select it by detecting the /arch:AVX
+         or /arch:AVX2 switches. You can also always manually turn SSE2, AVX or AVX2 
+         support on by defining STBIR_SSE2, STBIR_AVX or STBIR_AVX2.
+
+         On Linux, SSE2 and Neon is on by default for 64-bit x64 or ARM64. For 32-bit,
+         we select x86 SIMD mode by whether you have -msse2, -mavx or -mavx2 enabled
+         on the command line. For 32-bit ARM, you must pass -mfpu=neon-vfpv4 for both
+         clang and GCC, but GCC also requires an additional -mfp16-format=ieee to 
+         automatically enable NEON.
+
+         On x86 platforms, you can also define STBIR_FP16C to turn on FP16C instructions
+         for converting back and forth to half-floats. This is autoselected when we
+         are using AVX2. Clang and GCC also require the -mf16c switch. ARM always uses 
+         the built-in half float hardware NEON instructions. 
+
+         You can also tell us to use multiply-add instructions with STBIR_USE_FMA. 
+         Because x86 doesn't always have fma, we turn it off by default to maintain
+         determinism across all platforms. If you don't care about non-FMA determinism
+         and are willing to restrict yourself to more recent x86 CPUs (around the AVX 
+         timeframe), then fma will give you around a 15% speedup.
+
+         You can force off SIMD in all cases by defining STBIR_NO_SIMD. You can turn
+         off AVX or AVX2 specifically with STBIR_NO_AVX or STBIR_NO_AVX2. AVX is 10%
+         to 40% faster, and AVX2 is generally another 12%.
+        
+      ALPHA CHANNEL
+         Most of the resizing functions provide the ability to control how the alpha 
+         channel of an image is processed.
+
+         When alpha represents transparency, it is important that when combining
+         colors with filtering, the pixels should not be treated equally; they
+         should use a weighted average based on their alpha values. For example,
+         if a pixel is 1% opaque bright green and another pixel is 99% opaque
+         black and you average them, the average will be 50% opaque, but the
+         unweighted average and will be a middling green color, while the weighted
+         average will be nearly black. This means the unweighted version introduced
+         green energy that didn't exist in the source image.
+
+         (If you want to know why this makes sense, you can work out the math for
+         the following: consider what happens if you alpha composite a source image
+         over a fixed color and then average the output, vs. if you average the
+         source image pixels and then composite that over the same fixed color.
+         Only the weighted average produces the same result as the ground truth
+         composite-then-average result.)
+
+         Therefore, it is in general best to "alpha weight" the pixels when applying
+         filters to them. This essentially means multiplying the colors by the alpha
+         values before combining them, and then dividing by the alpha value at the
+         end.
+
+         The computer graphics industry introduced a technique called "premultiplied
+         alpha" or "associated alpha" in which image colors are stored in image files
+         already multiplied by their alpha. This saves some math when compositing,
+         and also avoids the need to divide by the alpha at the end (which is quite
+         inefficient). However, while premultiplied alpha is common in the movie CGI
+         industry, it is not commonplace in other industries like videogames, and most
+         consumer file formats are generally expected to contain not-premultiplied
+         colors. For example, Photoshop saves PNG files "unpremultiplied", and web
+         browsers like Chrome and Firefox expect PNG images to be unpremultiplied.
+
+         Note that there are three possibilities that might describe your image
+         and resize expectation:
+
+             1. images are not premultiplied, alpha weighting is desired
+             2. images are not premultiplied, alpha weighting is not desired
+             3. images are premultiplied
+
+         Both case #2 and case #3 require the exact same math: no alpha weighting
+         should be applied or removed. Only case 1 requires extra math operations;
+         the other two cases can be handled identically.
+
+         stb_image_resize expects case #1 by default, applying alpha weighting to
+         images, expecting the input images to be unpremultiplied. This is what the
+         COLOR+ALPHA buffer types tell the resizer to do. 
+
+         When you use the pixel layouts STBIR_RGBA, STBIR_BGRA, STBIR_ARGB, 
+         STBIR_ABGR, STBIR_RX, or STBIR_XR you are telling us that the pixels are 
+         non-premultiplied. In these cases, the resizer will alpha weight the colors 
+         (effectively creating the premultiplied image), do the filtering, and then 
+         convert back to non-premult on exit.
+
+         When you use the pixel layouts STBIR_RGBA_PM, STBIR_RGBA_PM, STBIR_RGBA_PM,
+         STBIR_RGBA_PM, STBIR_RX_PM or STBIR_XR_PM, you are telling that the pixels 
+         ARE premultiplied. In this case, the resizer doesn't have to do the 
+         premultipling - it can filter directly on the input. This about twice as 
+         fast as the non-premultiplied case, so it's the right option if your data is 
+         already setup correctly.
+
+         When you use the pixel layout STBIR_4CHANNEL or STBIR_2CHANNEL, you are 
+         telling us that there is no channel that represents transparency; it may be 
+         RGB and some unrelated fourth channel that has been stored in the alpha 
+         channel, but it is actually not alpha. No special processing will be 
+         performed. 
+
+         The difference between the generic 4 or 2 channel layouts, and the 
+         specialized _PM versions is with the _PM versions you are telling us that
+         the data *is* alpha, just don't premultiply it. That's important when
+         using SRGB pixel formats, we need to know where the alpha is, because
+         it is converted linearly (rather than with the SRGB converters).
+   
+         Because alpha weighting produces the same effect as premultiplying, you
+         even have the option with non-premultiplied inputs to let the resizer
+         produce a premultiplied output. Because the intially computed alpha-weighted
+         output image is effectively premultiplied, this is actually more performant
+         than the normal path which un-premultiplies the output image as a final step.
+
+         Finally, when converting both in and out of non-premulitplied space (for
+         example, when using STBIR_RGBA), we go to somewhat heroic measures to 
+         ensure that areas with zero alpha value pixels get something reasonable 
+         in the RGB values. If you don't care about the RGB values of zero alpha 
+         pixels, you can call the stbir_set_non_pm_alpha_speed_over_quality() 
+         function - this runs a premultiplied resize about 25% faster. That said,
+         when you really care about speed, using premultiplied pixels for both in
+         and out (STBIR_RGBA_PM, etc) much faster than both of these premultiplied
+         options.
+
+      PIXEL LAYOUT CONVERSION
+         The resizer can convert from some pixel layouts to others. When using the
+         stbir_set_pixel_layouts(), you can, for example, specify STBIR_RGBA
+         on input, and STBIR_ARGB on output, and it will re-organize the channels
+         during the resize. Currently, you can only convert between two pixel
+         layouts with the same number of channels.
+
+      DETERMINISM
+         We commit to being deterministic (from x64 to ARM to scalar to SIMD, etc). 
+         This requires compiling with fast-math off (using at least /fp:precise). 
+         Also, you must turn off fp-contracting (which turns mult+adds into fmas)!
+         We attempt to do this with pragmas, but with Clang, you usually want to add 
+         -ffp-contract=off to the command line as well.
+
+         For 32-bit x86, you must use SSE and SSE2 codegen for determinism. That is, 
+         if the scalar x87 unit gets used at all, we immediately lose determinism. 
+         On Microsoft Visual Studio 2008 and earlier, from what we can tell there is
+         no way to be deterministic in 32-bit x86 (some x87 always leaks in, even 
+         with fp:strict). On 32-bit x86 GCC, determinism requires both -msse2 and 
+         -fpmath=sse.
+
+         Note that we will not be deterministic with float data containing NaNs -
+         the NaNs will propagate differently on different SIMD and platforms. 
+
+         If you turn on STBIR_USE_FMA, then we will be deterministic with other 
+         fma targets, but we will differ from non-fma targets (this is unavoidable, 
+         because a fma isn't simply an add with a mult - it also introduces a 
+         rounding difference compared to non-fma instruction sequences. 
+
+      FLOAT PIXEL FORMAT RANGE
+         Any range of values can be used for the non-alpha float data that you pass 
+         in (0 to 1, -1 to 1, whatever). However, if you are inputting float values 
+         but *outputting* bytes or shorts, you must use a range of 0 to 1 so that we 
+         scale back properly. The alpha channel must also be 0 to 1 for any format 
+         that does premultiplication prior to resizing. 
+
+         Note also that with float output, using filters with negative lobes, the 
+         output filtered values might go slightly out of range. You can define 
+         STBIR_FLOAT_LOW_CLAMP and/or STBIR_FLOAT_HIGH_CLAMP to specify the range 
+         to clamp to on output, if that's important. 
+
+      MAX/MIN SCALE FACTORS
+         The input pixel resolutions are in integers, and we do the internal pointer
+         resolution in size_t sized integers. However, the scale ratio from input
+         resolution to output resolution is calculated in float form. This means
+         the effective possible scale ratio is limited to 24 bits (or 16 million
+         to 1). As you get close to the size of the float resolution (again, 16
+         million pixels wide or high), you might start seeing float inaccuracy
+         issues in general in the pipeline. If you have to do extreme resizes,
+         you can usually do this is multiple stages (using float intermediate
+         buffers).
+
+      FLIPPED IMAGES
+         Stride is just the delta from one scanline to the next. This means you can 
+         use a negative stride to handle inverted images (point to the final 
+         scanline and use a negative stride). You can invert the input or output,
+         using negative strides.
+
+      DEFAULT FILTERS
+         For functions which don't provide explicit control over what filters to 
+         use, you can change the compile-time defaults with:
+
+            #define STBIR_DEFAULT_FILTER_UPSAMPLE     STBIR_FILTER_something
+            #define STBIR_DEFAULT_FILTER_DOWNSAMPLE   STBIR_FILTER_something
+
+         See stbir_filter in the header-file section for the list of filters.
+
+      NEW FILTERS
+         A number of 1D filter kernels are supplied. For a list of supported 
+         filters, see the stbir_filter enum. You can install your own filters by 
+         using the stbir_set_filter_callbacks function.
+
+      PROGRESS
+         For interactive use with slow resize operations, you can use the the 
+         scanline callbacks in the extended API. It would have to be a *very* large 
+         image resample to need progress though - we're very fast.
+
+      CEIL and FLOOR
+         In scalar mode, the only functions we use from math.h are ceilf and floorf, 
+         but if you have your own versions, you can define the STBIR_CEILF(v) and 
+         STBIR_FLOORF(v) macros and we'll use them instead. In SIMD, we just use
+         our own versions.
+
+      ASSERT
+         Define STBIR_ASSERT(boolval) to override assert() and not use assert.h
+
+      FUTURE TODOS
+        *  For polyphase integral filters, we just memcpy the coeffs to dupe
+           them, but we should indirect and use the same coeff memory.
+        *  Add pixel layout conversions for sensible different channel counts
+           (maybe, 1->3/4, 3->4, 4->1, 3->1).
+         * For SIMD encode and decode scanline routines, do any pre-aligning
+           for bad input/output buffer alignments and pitch?
+         * For very wide scanlines, we should we do vertical strips to stay within
+           L2 cache. Maybe do chunks of 1K pixels at a time. There would be 
+           some pixel reconversion, but probably dwarfed by things falling out
+           of cache. Probably also something possible with alternating between
+           scattering and gathering at high resize scales?
+         * Rewrite the coefficient generator to do many at once.
+         * AVX-512 vertical kernels - worried about downclocking here.
+         * Convert the reincludes to macros when we know they aren't changing.
+         * Experiment with pivoting the horizontal and always using the
+           vertical filters (which are faster, but perhaps not enough to overcome
+           the pivot cost and the extra memory touches). Need to buffer the whole
+           image so have to balance memory use.
+         * Most of our code is internally function pointers, should we compile
+           all the SIMD stuff always and dynamically dispatch? 
+
+   CONTRIBUTORS
+      Jeff Roberts: 2.0 implementation, optimizations, SIMD
+      Martins Mozeiko: NEON simd, WASM simd, clang and GCC whisperer.
+      Fabian Giesen: half float and srgb converters
+      Sean Barrett: API design, optimizations
+      Jorge L Rodriguez: Original 1.0 implementation
+      Aras Pranckevicius: bugfixes for 1.0
+      Nathan Reed: warning fixes for 1.0
+
+   REVISIONS
+      2.00 (2022-02-20) mostly new source: new api, optimizations, simd, vertical-first, etc 
+                       (2x-5x faster without simd, 4x-12x faster with simd)
+                       (in some cases, 20x to 40x faster - resizing to very small for example)
+      0.96 (2019-03-04) fixed warnings
+      0.95 (2017-07-23) fixed warnings
+      0.94 (2017-03-18) fixed warnings
+      0.93 (2017-03-03) fixed bug with certain combinations of heights
+      0.92 (2017-01-02) fix integer overflow on large (>2GB) images
+      0.91 (2016-04-02) fix warnings; fix handling of subpixel regions
+      0.90 (2014-09-17) first released version
+
+   LICENSE
+     See end of file for license information.
+*/
+
+#if !defined(STB_IMAGE_RESIZE_DO_HORIZONTALS) && !defined(STB_IMAGE_RESIZE_DO_VERTICALS) && !defined(STB_IMAGE_RESIZE_DO_CODERS)   // for internal re-includes
+
+#ifndef STBIR_INCLUDE_STB_IMAGE_RESIZE2_H
+#define STBIR_INCLUDE_STB_IMAGE_RESIZE2_H
+
+#include <stddef.h>
+#ifdef _MSC_VER
+typedef unsigned char    stbir_uint8;
+typedef unsigned short   stbir_uint16;
+typedef unsigned int     stbir_uint32;
+typedef unsigned __int64 stbir_uint64;
+#else
+#include <stdint.h>
+typedef uint8_t  stbir_uint8;
+typedef uint16_t stbir_uint16;
+typedef uint32_t stbir_uint32;
+typedef uint64_t stbir_uint64;
+#endif
+
+#ifdef _M_IX86_FP
+#if ( _M_IX86_FP >= 1 )
+#ifndef STBIR_SSE
+#define STBIR_SSE
+#endif
+#endif
+#endif 
+
+#if defined(_x86_64) || defined( __x86_64__ ) || defined( _M_X64 ) || defined(__x86_64) || defined(_M_AMD64) || defined(__SSE2__) || defined(STBIR_SSE) || defined(STBIR_SSE2)
+  #ifndef STBIR_SSE2
+    #define STBIR_SSE2
+  #endif
+  #if defined(__AVX__) || defined(STBIR_AVX2)
+    #ifndef STBIR_AVX
+      #ifndef STBIR_NO_AVX
+        #define STBIR_AVX
+      #endif
+    #endif
+  #endif
+  #if defined(__AVX2__) || defined(STBIR_AVX2)
+    #ifndef STBIR_NO_AVX2
+      #ifndef STBIR_AVX2  
+        #define STBIR_AVX2
+      #endif
+      #if defined( _MSC_VER ) && !defined(__clang__)
+        #ifndef STBIR_FP16C  // FP16C instructions are on all AVX2 cpus, so we can autoselect it here on microsoft - clang needs -m16c
+          #define STBIR_FP16C
+        #endif
+      #endif
+    #endif
+  #endif
+  #ifdef __F16C__
+    #ifndef STBIR_FP16C  // turn on FP16C instructions if the define is set (for clang and gcc)
+      #define STBIR_FP16C
+    #endif
+  #endif
+#endif
+
+#if defined( _M_ARM64 ) || defined( __aarch64__ ) || defined( __arm64__ ) || defined(_M_ARM) || (__ARM_NEON_FP & 4) != 0 &&  __ARM_FP16_FORMAT_IEEE != 0
+#ifndef STBIR_NEON
+#define STBIR_NEON
+#endif
+#endif
+
+#if defined(_M_ARM)
+#ifdef STBIR_USE_FMA
+#undef STBIR_USE_FMA // no FMA for 32-bit arm on MSVC 
+#endif
+#endif
+
+#if defined(__wasm__) && defined(__wasm_simd128__)
+#ifndef STBIR_WASM
+#define STBIR_WASM
+#endif
+#endif
+
+#ifndef STBIRDEF
+#ifdef STB_IMAGE_RESIZE_STATIC
+#define STBIRDEF static
+#else
+#ifdef __cplusplus
+#define STBIRDEF extern "C"
+#else
+#define STBIRDEF extern
+#endif
+#endif
+#endif
+
+//////////////////////////////////////////////////////////////////////////////
+////   start "header file" ///////////////////////////////////////////////////
+//
+// Easy-to-use API:
+//
+//     * stride is the offset between successive rows of image data 
+//        in memory, in bytes. specify 0 for packed continuously in memory
+//     * colorspace is linear or sRGB as specified by function name
+//     * Uses the default filters
+//     * Uses edge mode clamped
+//     * returned result is 1 for success or 0 in case of an error.
+
+
+// stbir_pixel_layout specifies:
+//   number of channels
+//   order of channels
+//   whether color is premultiplied by alpha
+// for back compatibility, you can cast the old channel count to an stbir_pixel_layout
+typedef enum 
+{
+  STBIR_BGR      = 0,               // 3-chan, with order specified (for channel flipping)
+  STBIR_1CHANNEL = 1,              
+  STBIR_2CHANNEL = 2,
+  STBIR_RGB      = 3,               // 3-chan, with order specified (for channel flipping) 
+  STBIR_RGBA     = 4,               // alpha formats, alpha is NOT premultiplied into color channels
+
+  STBIR_4CHANNEL = 5,
+  STBIR_BGRA = 6,
+  STBIR_ARGB = 7,
+  STBIR_ABGR = 8,
+  STBIR_RA   = 9,
+  STBIR_AR   = 10,
+
+  STBIR_RGBA_PM = 11,               // alpha formats, alpha is premultiplied into color channels
+  STBIR_BGRA_PM = 12,
+  STBIR_ARGB_PM = 13,
+  STBIR_ABGR_PM = 14,
+  STBIR_RA_PM   = 15,
+  STBIR_AR_PM   = 16,
+} stbir_pixel_layout;
+
+//===============================================================
+//  Simple-complexity API
+//
+//    If output_pixels is NULL (0), then we will allocate the buffer and return it to you.
+//--------------------------------
+
+STBIRDEF unsigned char * stbir_resize_uint8_srgb( const unsigned char *input_pixels , int input_w , int input_h, int input_stride_in_bytes,
+                                                        unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+                                                        stbir_pixel_layout pixel_type );
+
+STBIRDEF unsigned char * stbir_resize_uint8_linear( const unsigned char *input_pixels , int input_w , int input_h, int input_stride_in_bytes,
+                                                          unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+                                                          stbir_pixel_layout pixel_type );
+
+STBIRDEF float * stbir_resize_float_linear( const float *input_pixels , int input_w , int input_h, int input_stride_in_bytes,
+                                                  float *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+                                                  stbir_pixel_layout pixel_type );
+//===============================================================
+
+//===============================================================
+// Medium-complexity API
+//
+// This extends the easy-to-use API as follows:
+//
+//     * Can specify the datatype - U8, U8_SRGB, U16, FLOAT, HALF_FLOAT
+//     * Edge wrap can selected explicitly
+//     * Filter can be selected explicitly
+//--------------------------------
+
+typedef enum
+{
+  STBIR_EDGE_CLAMP   = 0,
+  STBIR_EDGE_REFLECT = 1,
+  STBIR_EDGE_WRAP    = 2,  // this edge mode is slower and uses more memory
+  STBIR_EDGE_ZERO    = 3,
+} stbir_edge;
+
+typedef enum
+{
+  STBIR_FILTER_DEFAULT      = 0,  // use same filter type that easy-to-use API chooses
+  STBIR_FILTER_BOX          = 1,  // A trapezoid w/1-pixel wide ramps, same result as box for integer scale ratios
+  STBIR_FILTER_TRIANGLE     = 2,  // On upsampling, produces same results as bilinear texture filtering
+  STBIR_FILTER_CUBICBSPLINE = 3,  // The cubic b-spline (aka Mitchell-Netrevalli with B=1,C=0), gaussian-esque
+  STBIR_FILTER_CATMULLROM   = 4,  // An interpolating cubic spline
+  STBIR_FILTER_MITCHELL     = 5,  // Mitchell-Netrevalli filter with B=1/3, C=1/3
+  STBIR_FILTER_POINT_SAMPLE = 6,  // Simple point sampling
+  STBIR_FILTER_OTHER        = 7,  // User callback specified
+} stbir_filter;
+
+typedef enum
+{
+  STBIR_TYPE_UINT8            = 0,
+  STBIR_TYPE_UINT8_SRGB       = 1,
+  STBIR_TYPE_UINT8_SRGB_ALPHA = 2,  // alpha channel, when present, should also be SRGB (this is very unusual)
+  STBIR_TYPE_UINT16           = 3,
+  STBIR_TYPE_FLOAT            = 4,
+  STBIR_TYPE_HALF_FLOAT       = 5
+} stbir_datatype;
+
+// medium api
+STBIRDEF void *  stbir_resize( const void *input_pixels , int input_w , int input_h, int input_stride_in_bytes,
+                                     void *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+                               stbir_pixel_layout pixel_layout, stbir_datatype data_type,
+                               stbir_edge edge, stbir_filter filter );
+//===============================================================
+
+
+
+//===============================================================
+// Extended-complexity API
+//
+// This API exposes all resize functionality.
+//
+//     * Separate filter types for each axis
+//     * Separate edge modes for each axis
+//     * Separate input and output data types
+//     * Can specify regions with subpixel correctness
+//     * Can specify alpha flags
+//     * Can specify a memory callback 
+//     * Can specify a callback data type for pixel input and output 
+//     * Can be threaded for a single resize
+//     * Can be used to resize many frames without recalculating the sampler info
+//
+//  Use this API as follows:
+//     1) Call the stbir_resize_init function on a local STBIR_RESIZE structure
+//     2) Call any of the stbir_set functions
+//     3) Optionally call stbir_build_samplers() if you are going to resample multiple times
+//        with the same input and output dimensions (like resizing video frames)
+//     4) Resample by calling stbir_resize_extended().
+//     5) Call stbir_free_samplers() if you called stbir_build_samplers()
+//--------------------------------
+
+
+// Types:
+
+// INPUT CALLBACK: this callback is used for input scanlines
+typedef void const * stbir_input_callback( void * optional_output, void const * input_ptr, int num_pixels, int x, int y, void * context );
+
+// OUTPUT CALLBACK: this callback is used for output scanlines
+typedef void stbir_output_callback( void const * output_ptr, int num_pixels, int y, void * context );
+
+// callbacks for user installed filters
+typedef float stbir__kernel_callback( float x, float scale, void * user_data ); // centered at zero
+typedef float stbir__support_callback( float scale, void * user_data );
+
+// internal structure with precomputed scaling
+typedef struct stbir__info stbir__info; 
+
+typedef struct STBIR_RESIZE  // use the stbir_resize_init and stbir_override functions to set these values for future compatibility
+{
+  void * user_data;
+  void const * input_pixels;
+  int input_w, input_h;
+  double input_s0, input_t0, input_s1, input_t1;
+  stbir_input_callback * input_cb;
+  void * output_pixels;
+  int output_w, output_h;
+  int output_subx, output_suby, output_subw, output_subh;
+  stbir_output_callback * output_cb;
+  int input_stride_in_bytes;
+  int output_stride_in_bytes;
+  int splits;
+  int fast_alpha;
+  int needs_rebuild;
+  int called_alloc;
+  stbir_pixel_layout input_pixel_layout_public;
+  stbir_pixel_layout output_pixel_layout_public;
+  stbir_datatype input_data_type;
+  stbir_datatype output_data_type;
+  stbir_filter horizontal_filter, vertical_filter;
+  stbir_edge horizontal_edge, vertical_edge;
+  stbir__kernel_callback * horizontal_filter_kernel; stbir__support_callback * horizontal_filter_support;
+  stbir__kernel_callback * vertical_filter_kernel; stbir__support_callback * vertical_filter_support;
+  stbir__info * samplers;      
+} STBIR_RESIZE;
+
+// extended complexity api
+
+
+// First off, you must ALWAYS call stbir_resize_init on your resize structure before any of the other calls!
+STBIRDEF void stbir_resize_init( STBIR_RESIZE * resize,
+                                 const void *input_pixels,  int input_w,  int input_h, int input_stride_in_bytes, // stride can be zero
+                                       void *output_pixels, int output_w, int output_h, int output_stride_in_bytes, // stride can be zero
+                                 stbir_pixel_layout pixel_layout, stbir_datatype data_type );
+
+//===============================================================
+// You can update these parameters any time after resize_init and there is no cost
+//--------------------------------
+
+STBIRDEF void stbir_set_datatypes( STBIR_RESIZE * resize, stbir_datatype input_type, stbir_datatype output_type );         
+STBIRDEF void stbir_set_pixel_callbacks( STBIR_RESIZE * resize, stbir_input_callback * input_cb, stbir_output_callback * output_cb );   // no callbacks by default
+STBIRDEF void stbir_set_user_data( STBIR_RESIZE * resize, void * user_data );                                               // pass back STBIR_RESIZE* by default
+STBIRDEF void stbir_set_buffer_ptrs( STBIR_RESIZE * resize, const void * input_pixels, int input_stride_in_bytes, void * output_pixels, int output_stride_in_bytes );
+
+//===============================================================
+
+
+//===============================================================
+// If you call any of these functions, you will trigger a sampler rebuild!
+//--------------------------------
+
+STBIRDEF int stbir_set_pixel_layouts( STBIR_RESIZE * resize, stbir_pixel_layout input_pixel_layout, stbir_pixel_layout output_pixel_layout );  // sets new buffer layouts
+STBIRDEF int stbir_set_edgemodes( STBIR_RESIZE * resize, stbir_edge horizontal_edge, stbir_edge vertical_edge );       // CLAMP by default
+
+STBIRDEF int stbir_set_filters( STBIR_RESIZE * resize, stbir_filter horizontal_filter, stbir_filter vertical_filter ); // STBIR_DEFAULT_FILTER_UPSAMPLE/DOWNSAMPLE by default
+STBIRDEF int stbir_set_filter_callbacks( STBIR_RESIZE * resize, stbir__kernel_callback * horizontal_filter, stbir__support_callback * horizontal_support, stbir__kernel_callback * vertical_filter, stbir__support_callback * vertical_support ); 
+
+STBIRDEF int stbir_set_pixel_subrect( STBIR_RESIZE * resize, int subx, int suby, int subw, int subh );        // sets both sub-regions (full regions by default)
+STBIRDEF int stbir_set_input_subrect( STBIR_RESIZE * resize, double s0, double t0, double s1, double t1 );    // sets input sub-region (full region by default)
+STBIRDEF int stbir_set_output_pixel_subrect( STBIR_RESIZE * resize, int subx, int suby, int subw, int subh ); // sets output sub-region (full region by default)
+
+// when inputting AND outputting non-premultiplied alpha pixels, we use a slower but higher quality technique
+//   that fills the zero alpha pixel's RGB values with something plausible.  If you don't care about areas of
+//   zero alpha, you can call this function to get about a 25% speed improvement for STBIR_RGBA to STBIR_RGBA
+//   types of resizes.
+STBIRDEF int stbir_set_non_pm_alpha_speed_over_quality( STBIR_RESIZE * resize, int non_pma_alpha_speed_over_quality );
+//===============================================================
+
+
+//===============================================================
+// You can call build_samplers to prebuild all the internal data we need to resample.
+//   Then, if you call resize_extended many times with the same resize, you only pay the
+//   cost once.
+// If you do call build_samplers, you MUST call free_samplers eventually.
+//--------------------------------
+
+// This builds the samplers and does one allocation
+STBIRDEF int stbir_build_samplers( STBIR_RESIZE * resize ); 
+
+// You MUST call this, if you call stbir_build_samplers or stbir_build_samplers_with_splits
+STBIRDEF void stbir_free_samplers( STBIR_RESIZE * resize );
+//===============================================================
+
+
+// And this is the main function to perform the resize synchronously on one thread.
+STBIRDEF int stbir_resize_extended( STBIR_RESIZE * resize );
+
+
+//===============================================================
+// Use these functions for multithreading.
+//   1) You call stbir_build_samplers_with_splits first on the main thread
+//   2) Then stbir_resize_with_split on each thread
+//   3) stbir_free_samplers when done on the main thread
+//--------------------------------
+
+// This will build samplers for threading.
+//   You can pass in the number of threads you'd like to use (try_splits).
+//   It returns the number of splits (threads) that you can call it with.
+///  It might be less if the image resize can't be split up that many ways.
+
+STBIRDEF int stbir_build_samplers_with_splits( STBIR_RESIZE * resize, int try_splits );             
+
+// This function does a split of the resizing (you call this fuction for each
+// split, on multiple threads). A split is a piece of the output resize pixel space.
+
+// Note that you MUST call stbir_build_samplers_with_splits before stbir_resize_extended_split!
+
+// Usually, you will always call stbir_resize_split with split_start as the thread_index
+//   and "1" for the split_count.
+// But, if you have a weird situation where you MIGHT want 8 threads, but sometimes
+//   only 4 threads, you can use 0,2,4,6 for the split_start's and use "2" for the 
+//   split_count each time to turn in into a 4 thread resize. (This is unusual).
+
+STBIRDEF int stbir_resize_extended_split( STBIR_RESIZE * resize, int split_start, int split_count );         
+//===============================================================
+
+
+//===============================================================
+// Pixel Callbacks info:
+//--------------------------------
+
+//   The input callback is super flexible - it calls you with the input address
+//   (based on the stride and base pointer), it gives you an optional_output
+//   pointer that you can fill, or you can just return your own pointer into
+//   your own data. 
+//
+//   You can also do conversion from non-supported data types if necessary - in 
+//   this case, you ignore the input_ptr and just use the x and y parameters to 
+//   calculate your own input_ptr based on the size of each non-supported pixel.
+//   (Something like the third example below.)
+//
+//   You can also install just an input or just an output callback by setting the
+//   callback that you don't want to zero.
+//
+//     First example, progress: (getting a callback that you can monitor the progress):
+//        void const * my_callback( void * optional_output, void const * input_ptr, int num_pixels, int x, int y, void * context )
+//        {
+//           percentage_done = y / input_height;
+//           return input_ptr;  // use buffer from call
+//        }
+//
+//     Next example, copying: (copy from some other buffer or stream):  
+//        void const * my_callback( void * optional_output, void const * input_ptr, int num_pixels, int x, int y, void * context )
+//        {
+//           CopyOrStreamData( optional_output, other_data_src, num_pixels * pixel_width_in_bytes );
+//           return optional_output;  // return the optional buffer that we filled
+//        }
+//
+//     Third example, input another buffer without copying: (zero-copy from other buffer):  
+//        void const * my_callback( void * optional_output, void const * input_ptr, int num_pixels, int x, int y, void * context )
+//        {
+//           void * pixels = ( (char*) other_image_base ) + ( y * other_image_stride ) + ( x * other_pixel_width_in_bytes );
+//           return pixels;       // return pointer to your data without copying
+//        }
+//
+//
+//   The output callback is considerably simpler - it just calls you so that you can dump
+//   out each scanline. You could even directly copy out to disk if you have a simple format
+//   like TGA or BMP. You can also convert to other output types here if you want.
+//
+//   Simple example:
+//        void const * my_output( void * output_ptr, int num_pixels, int y, void * context )
+//        {
+//           percentage_done = y / output_height;
+//           fwrite( output_ptr, pixel_width_in_bytes, num_pixels, output_file );
+//        }
+//===============================================================
+
+
+
+
+//===============================================================
+// optional built-in profiling API
+//--------------------------------
+
+#ifdef STBIR_PROFILE
+
+typedef struct STBIR_PROFILE_INFO 
+{
+  stbir_uint64 total_clocks;
+
+  // how many clocks spent (of total_clocks) in the various resize routines, along with a string description
+  //    there are "resize_count" number of zones
+  stbir_uint64 clocks[ 8 ];
+  char const ** descriptions;
+  
+  // count of clocks and descriptions
+  stbir_uint32 count;
+} STBIR_PROFILE_INFO;
+
+// use after calling stbir_resize_extended (or stbir_build_samplers or stbir_build_samplers_with_splits)
+STBIRDEF void stbir_resize_build_profile_info( STBIR_PROFILE_INFO * out_info, STBIR_RESIZE const * resize );
+
+// use after calling stbir_resize_extended
+STBIRDEF void stbir_resize_extended_profile_info( STBIR_PROFILE_INFO * out_info, STBIR_RESIZE const * resize );
+
+// use after calling stbir_resize_extended_split
+STBIRDEF void stbir_resize_split_profile_info( STBIR_PROFILE_INFO * out_info, STBIR_RESIZE const * resize, int split_start, int split_num );
+
+//===============================================================
+
+#endif
+
+
+////   end header file   /////////////////////////////////////////////////////
+#endif // STBIR_INCLUDE_STB_IMAGE_RESIZE2_H
+
+#if defined(STB_IMAGE_RESIZE_IMPLEMENTATION) || defined(STB_IMAGE_RESIZE2_IMPLEMENTATION)
+
+#ifndef STBIR_ASSERT
+#include <assert.h>
+#define STBIR_ASSERT(x) assert(x)
+#endif
+
+#ifndef STBIR_MALLOC
+#include <stdlib.h>
+#define STBIR_MALLOC(size,user_data) ((void)(user_data), malloc(size))
+#define STBIR_FREE(ptr,user_data)    ((void)(user_data), free(ptr))
+// (we used the comma operator to evaluate user_data, to avoid "unused parameter" warnings)
+#endif
+
+#ifdef _MSC_VER
+
+#define stbir__inline __forceinline
+
+#else
+
+#define stbir__inline __inline__
+
+// Clang address sanitizer
+#if defined(__has_feature)
+  #if __has_feature(address_sanitizer) || __has_feature(memory_sanitizer)
+    #ifndef STBIR__SEPARATE_ALLOCATIONS
+      #define STBIR__SEPARATE_ALLOCATIONS
+    #endif
+  #endif
+#endif
+
+#endif
+
+// GCC and MSVC
+#if defined(__SANITIZE_ADDRESS__)
+  #ifndef STBIR__SEPARATE_ALLOCATIONS
+    #define STBIR__SEPARATE_ALLOCATIONS
+  #endif
+#endif
+
+// Always turn off automatic FMA use - use STBIR_USE_FMA if you want.
+// Otherwise, this is a determinism disaster.
+#ifndef STBIR_DONT_CHANGE_FP_CONTRACT  // override in case you don't want this behavior
+#if defined(_MSC_VER) && !defined(__clang__)
+#if _MSC_VER > 1200
+#pragma fp_contract(off)
+#endif
+#elif defined(__GNUC__) &&  !defined(__clang__)
+#pragma GCC optimize("fp-contract=off")
+#else
+#pragma STDC FP_CONTRACT OFF
+#endif
+#endif
+
+#ifdef _MSC_VER
+#define STBIR__UNUSED(v)  (void)(v)
+#else
+#define STBIR__UNUSED(v)  (void)sizeof(v)
+#endif
+
+#define STBIR__ARRAY_SIZE(a) (sizeof((a))/sizeof((a)[0]))
+
+
+#ifndef STBIR_DEFAULT_FILTER_UPSAMPLE
+#define STBIR_DEFAULT_FILTER_UPSAMPLE    STBIR_FILTER_CATMULLROM
+#endif
+
+#ifndef STBIR_DEFAULT_FILTER_DOWNSAMPLE
+#define STBIR_DEFAULT_FILTER_DOWNSAMPLE  STBIR_FILTER_MITCHELL
+#endif
+
+
+#ifndef STBIR__HEADER_FILENAME
+#define STBIR__HEADER_FILENAME "stb_image_resize2.h"
+#endif
+
+// the internal pixel layout enums are in a different order, so we can easily do range comparisons of types
+//   the public pixel layout is ordered in a way that if you cast num_channels (1-4) to the enum, you get something sensible 
+typedef enum 
+{
+  STBIRI_1CHANNEL = 0,
+  STBIRI_2CHANNEL = 1,
+  STBIRI_RGB      = 2,
+  STBIRI_BGR      = 3,
+  STBIRI_4CHANNEL = 4,
+  
+  STBIRI_RGBA = 5,
+  STBIRI_BGRA = 6,
+  STBIRI_ARGB = 7,
+  STBIRI_ABGR = 8,
+  STBIRI_RA   = 9,
+  STBIRI_AR   = 10,
+
+  STBIRI_RGBA_PM = 11,
+  STBIRI_BGRA_PM = 12,
+  STBIRI_ARGB_PM = 13,
+  STBIRI_ABGR_PM = 14,
+  STBIRI_RA_PM   = 15,
+  STBIRI_AR_PM   = 16,
+} stbir_internal_pixel_layout;
+
+// define the public pixel layouts to not compile inside the implementation (to avoid accidental use)
+#define STBIR_BGR bad_dont_use_in_implementation
+#define STBIR_1CHANNEL STBIR_BGR
+#define STBIR_2CHANNEL STBIR_BGR
+#define STBIR_RGB STBIR_BGR
+#define STBIR_RGBA STBIR_BGR
+#define STBIR_4CHANNEL STBIR_BGR
+#define STBIR_BGRA STBIR_BGR
+#define STBIR_ARGB STBIR_BGR
+#define STBIR_ABGR STBIR_BGR
+#define STBIR_RA STBIR_BGR
+#define STBIR_AR STBIR_BGR
+#define STBIR_RGBA_PM STBIR_BGR
+#define STBIR_BGRA_PM STBIR_BGR
+#define STBIR_ARGB_PM STBIR_BGR
+#define STBIR_ABGR_PM STBIR_BGR
+#define STBIR_RA_PM STBIR_BGR
+#define STBIR_AR_PM STBIR_BGR
+
+// must match stbir_datatype
+static unsigned char stbir__type_size[] = {
+  1,1,1,2,4,2 // STBIR_TYPE_UINT8,STBIR_TYPE_UINT8_SRGB,STBIR_TYPE_UINT8_SRGB_ALPHA,STBIR_TYPE_UINT16,STBIR_TYPE_FLOAT,STBIR_TYPE_HALF_FLOAT
+};
+
+// When gathering, the contributors are which source pixels contribute.
+// When scattering, the contributors are which destination pixels are contributed to.
+typedef struct
+{
+  int n0; // First contributing pixel
+  int n1; // Last contributing pixel
+} stbir__contributors;
+
+typedef struct
+{
+  int lowest;    // First sample index for whole filter
+  int highest;   // Last sample index for whole filter
+  int widest;    // widest single set of samples for an output
+} stbir__filter_extent_info;
+
+typedef struct
+{
+  int n0; // First pixel of decode buffer to write to
+  int n1; // Last pixel of decode that will be written to
+  int pixel_offset_for_input;  // Pixel offset into input_scanline
+} stbir__span;
+
+typedef struct stbir__scale_info
+{
+  int input_full_size;
+  int output_sub_size;
+  float scale;
+  float inv_scale;
+  float pixel_shift; // starting shift in output pixel space (in pixels)
+  int scale_is_rational;
+  stbir_uint32 scale_numerator, scale_denominator;
+} stbir__scale_info;
+
+typedef struct
+{
+  stbir__contributors * contributors;
+  float* coefficients;
+  stbir__contributors * gather_prescatter_contributors;
+  float * gather_prescatter_coefficients;
+  stbir__scale_info scale_info;
+  float support;
+  stbir_filter filter_enum;
+  stbir__kernel_callback * filter_kernel;
+  stbir__support_callback * filter_support;
+  stbir_edge edge;
+  int coefficient_width;
+  int filter_pixel_width;
+  int filter_pixel_margin;
+  int num_contributors;
+  int contributors_size;
+  int coefficients_size;
+  stbir__filter_extent_info extent_info;
+  int is_gather;  // 0 = scatter, 1 = gather with scale >= 1, 2 = gather with scale < 1
+  int gather_prescatter_num_contributors;
+  int gather_prescatter_coefficient_width;
+  int gather_prescatter_contributors_size;
+  int gather_prescatter_coefficients_size;
+} stbir__sampler;
+
+typedef struct
+{
+  stbir__contributors conservative;
+  int edge_sizes[2];    // this can be less than filter_pixel_margin, if the filter and scaling falls off
+  stbir__span spans[2]; // can be two spans, if doing input subrect with clamp mode WRAP
+} stbir__extents;
+
+typedef struct 
+{
+#ifdef STBIR_PROFILE
+  union
+  {
+    struct { stbir_uint64 total, looping, vertical, horizontal, decode, encode, alpha, unalpha; } named;
+    stbir_uint64 array[8];
+  } profile;
+  stbir_uint64 * current_zone_excluded_ptr;
+#endif
+  float* decode_buffer;
+
+  int ring_buffer_first_scanline;
+  int ring_buffer_last_scanline;
+  int ring_buffer_begin_index;    // first_scanline is at this index in the ring buffer
+  int start_output_y, end_output_y;
+  int start_input_y, end_input_y;  // used in scatter only
+
+  #ifdef STBIR__SEPARATE_ALLOCATIONS
+    float** ring_buffers; // one pointer for each ring buffer
+  #else
+    float* ring_buffer;  // one big buffer that we index into
+  #endif
+
+  float* vertical_buffer;
+
+  char no_cache_straddle[64];
+} stbir__per_split_info;
+
+typedef void stbir__decode_pixels_func( float * decode, int width_times_channels, void const * input );
+typedef void stbir__alpha_weight_func( float * decode_buffer, int width_times_channels );
+typedef void stbir__horizontal_gather_channels_func( float * output_buffer, unsigned int output_sub_size, float const * decode_buffer, 
+  stbir__contributors const * horizontal_contributors, float const * horizontal_coefficients, int coefficient_width );
+typedef void stbir__alpha_unweight_func(float * encode_buffer, int width_times_channels );
+typedef void stbir__encode_pixels_func( void * output, int width_times_channels, float const * encode );
+
+struct stbir__info
+{
+#ifdef STBIR_PROFILE
+  union
+  {
+    struct { stbir_uint64 total, build, alloc, horizontal, vertical, cleanup, pivot; } named;
+    stbir_uint64 array[7];
+  } profile;
+  stbir_uint64 * current_zone_excluded_ptr;
+#endif
+  stbir__sampler horizontal;
+  stbir__sampler vertical;
+
+  void const * input_data;
+  void * output_data;
+
+  int input_stride_bytes;
+  int output_stride_bytes;
+  int ring_buffer_length_bytes;   // The length of an individual entry in the ring buffer. The total number of ring buffers is stbir__get_filter_pixel_width(filter)
+  int ring_buffer_num_entries;    // Total number of entries in the ring buffer.
+
+  stbir_datatype input_type;
+  stbir_datatype output_type;
+
+  stbir_input_callback * in_pixels_cb;
+  void * user_data;
+  stbir_output_callback * out_pixels_cb;
+
+  stbir__extents scanline_extents;
+
+  void * alloced_mem;
+  stbir__per_split_info * split_info;  // by default 1, but there will be N of these allocated based on the thread init you did
+
+  stbir__decode_pixels_func * decode_pixels;
+  stbir__alpha_weight_func * alpha_weight;
+  stbir__horizontal_gather_channels_func * horizontal_gather_channels;
+  stbir__alpha_unweight_func * alpha_unweight;
+  stbir__encode_pixels_func * encode_pixels;
+  
+  int alloced_total;
+  int splits; // count of splits
+  
+  stbir_internal_pixel_layout input_pixel_layout_internal;
+  stbir_internal_pixel_layout output_pixel_layout_internal;
+
+  int input_color_and_type;
+  int offset_x, offset_y; // offset within output_data
+  int vertical_first;
+  int channels;
+  int effective_channels; // same as channels, except on RGBA/ARGB (7), or XA/AX (3)
+  int alloc_ring_buffer_num_entries;    // Number of entries in the ring buffer that will be allocated
+};
+
+
+#define stbir__max_uint8_as_float             255.0f
+#define stbir__max_uint16_as_float            65535.0f
+#define stbir__max_uint8_as_float_inverted    (1.0f/255.0f)
+#define stbir__max_uint16_as_float_inverted   (1.0f/65535.0f)
+#define stbir__small_float ((float)1 / (1 << 20) / (1 << 20) / (1 << 20) / (1 << 20) / (1 << 20) / (1 << 20))
+
+// min/max friendly
+#define STBIR_CLAMP(x, xmin, xmax) do { \
+  if ( (x) < (xmin) ) (x) = (xmin);     \
+  if ( (x) > (xmax) ) (x) = (xmax);     \
+} while (0)
+
+static stbir__inline int stbir__min(int a, int b)
+{
+  return a < b ? a : b;
+}
+
+static stbir__inline int stbir__max(int a, int b)
+{
+  return a > b ? a : b;
+}
+
+static float stbir__srgb_uchar_to_linear_float[256] = {
+  0.000000f, 0.000304f, 0.000607f, 0.000911f, 0.001214f, 0.001518f, 0.001821f, 0.002125f, 0.002428f, 0.002732f, 0.003035f,
+  0.003347f, 0.003677f, 0.004025f, 0.004391f, 0.004777f, 0.005182f, 0.005605f, 0.006049f, 0.006512f, 0.006995f, 0.007499f,
+  0.008023f, 0.008568f, 0.009134f, 0.009721f, 0.010330f, 0.010960f, 0.011612f, 0.012286f, 0.012983f, 0.013702f, 0.014444f,
+  0.015209f, 0.015996f, 0.016807f, 0.017642f, 0.018500f, 0.019382f, 0.020289f, 0.021219f, 0.022174f, 0.023153f, 0.024158f,
+  0.025187f, 0.026241f, 0.027321f, 0.028426f, 0.029557f, 0.030713f, 0.031896f, 0.033105f, 0.034340f, 0.035601f, 0.036889f,
+  0.038204f, 0.039546f, 0.040915f, 0.042311f, 0.043735f, 0.045186f, 0.046665f, 0.048172f, 0.049707f, 0.051269f, 0.052861f,
+  0.054480f, 0.056128f, 0.057805f, 0.059511f, 0.061246f, 0.063010f, 0.064803f, 0.066626f, 0.068478f, 0.070360f, 0.072272f,
+  0.074214f, 0.076185f, 0.078187f, 0.080220f, 0.082283f, 0.084376f, 0.086500f, 0.088656f, 0.090842f, 0.093059f, 0.095307f,
+  0.097587f, 0.099899f, 0.102242f, 0.104616f, 0.107023f, 0.109462f, 0.111932f, 0.114435f, 0.116971f, 0.119538f, 0.122139f,
+  0.124772f, 0.127438f, 0.130136f, 0.132868f, 0.135633f, 0.138432f, 0.141263f, 0.144128f, 0.147027f, 0.149960f, 0.152926f,
+  0.155926f, 0.158961f, 0.162029f, 0.165132f, 0.168269f, 0.171441f, 0.174647f, 0.177888f, 0.181164f, 0.184475f, 0.187821f,
+  0.191202f, 0.194618f, 0.198069f, 0.201556f, 0.205079f, 0.208637f, 0.212231f, 0.215861f, 0.219526f, 0.223228f, 0.226966f,
+  0.230740f, 0.234551f, 0.238398f, 0.242281f, 0.246201f, 0.250158f, 0.254152f, 0.258183f, 0.262251f, 0.266356f, 0.270498f,
+  0.274677f, 0.278894f, 0.283149f, 0.287441f, 0.291771f, 0.296138f, 0.300544f, 0.304987f, 0.309469f, 0.313989f, 0.318547f,
+  0.323143f, 0.327778f, 0.332452f, 0.337164f, 0.341914f, 0.346704f, 0.351533f, 0.356400f, 0.361307f, 0.366253f, 0.371238f,
+  0.376262f, 0.381326f, 0.386430f, 0.391573f, 0.396755f, 0.401978f, 0.407240f, 0.412543f, 0.417885f, 0.423268f, 0.428691f,
+  0.434154f, 0.439657f, 0.445201f, 0.450786f, 0.456411f, 0.462077f, 0.467784f, 0.473532f, 0.479320f, 0.485150f, 0.491021f,
+  0.496933f, 0.502887f, 0.508881f, 0.514918f, 0.520996f, 0.527115f, 0.533276f, 0.539480f, 0.545725f, 0.552011f, 0.558340f,
+  0.564712f, 0.571125f, 0.577581f, 0.584078f, 0.590619f, 0.597202f, 0.603827f, 0.610496f, 0.617207f, 0.623960f, 0.630757f,
+  0.637597f, 0.644480f, 0.651406f, 0.658375f, 0.665387f, 0.672443f, 0.679543f, 0.686685f, 0.693872f, 0.701102f, 0.708376f,
+  0.715694f, 0.723055f, 0.730461f, 0.737911f, 0.745404f, 0.752942f, 0.760525f, 0.768151f, 0.775822f, 0.783538f, 0.791298f,
+  0.799103f, 0.806952f, 0.814847f, 0.822786f, 0.830770f, 0.838799f, 0.846873f, 0.854993f, 0.863157f, 0.871367f, 0.879622f,
+  0.887923f, 0.896269f, 0.904661f, 0.913099f, 0.921582f, 0.930111f, 0.938686f, 0.947307f, 0.955974f, 0.964686f, 0.973445f,
+  0.982251f, 0.991102f, 1.0f
+};
+
+typedef union
+{
+  unsigned int u;
+  float f;
+} stbir__FP32;
+
+// From https://gist.github.com/rygorous/2203834
+
+static const stbir_uint32 fp32_to_srgb8_tab4[104] = {
+  0x0073000d, 0x007a000d, 0x0080000d, 0x0087000d, 0x008d000d, 0x0094000d, 0x009a000d, 0x00a1000d,
+  0x00a7001a, 0x00b4001a, 0x00c1001a, 0x00ce001a, 0x00da001a, 0x00e7001a, 0x00f4001a, 0x0101001a,
+  0x010e0033, 0x01280033, 0x01410033, 0x015b0033, 0x01750033, 0x018f0033, 0x01a80033, 0x01c20033,
+  0x01dc0067, 0x020f0067, 0x02430067, 0x02760067, 0x02aa0067, 0x02dd0067, 0x03110067, 0x03440067,
+  0x037800ce, 0x03df00ce, 0x044600ce, 0x04ad00ce, 0x051400ce, 0x057b00c5, 0x05dd00bc, 0x063b00b5,
+  0x06970158, 0x07420142, 0x07e30130, 0x087b0120, 0x090b0112, 0x09940106, 0x0a1700fc, 0x0a9500f2,
+  0x0b0f01cb, 0x0bf401ae, 0x0ccb0195, 0x0d950180, 0x0e56016e, 0x0f0d015e, 0x0fbc0150, 0x10630143,
+  0x11070264, 0x1238023e, 0x1357021d, 0x14660201, 0x156601e9, 0x165a01d3, 0x174401c0, 0x182401af,
+  0x18fe0331, 0x1a9602fe, 0x1c1502d2, 0x1d7e02ad, 0x1ed4028d, 0x201a0270, 0x21520256, 0x227d0240,
+  0x239f0443, 0x25c003fe, 0x27bf03c4, 0x29a10392, 0x2b6a0367, 0x2d1d0341, 0x2ebe031f, 0x304d0300,
+  0x31d105b0, 0x34a80555, 0x37520507, 0x39d504c5, 0x3c37048b, 0x3e7c0458, 0x40a8042a, 0x42bd0401,
+  0x44c20798, 0x488e071e, 0x4c1c06b6, 0x4f76065d, 0x52a50610, 0x55ac05cc, 0x5892058f, 0x5b590559,
+  0x5e0c0a23, 0x631c0980, 0x67db08f6, 0x6c55087f, 0x70940818, 0x74a007bd, 0x787d076c, 0x7c330723,
+};
+ 
+static stbir__inline stbir_uint8 stbir__linear_to_srgb_uchar(float in)
+{
+  static const stbir__FP32 almostone = { 0x3f7fffff }; // 1-eps
+  static const stbir__FP32 minval = { (127-13) << 23 };
+  stbir_uint32 tab,bias,scale,t;
+  stbir__FP32 f;
+
+  // Clamp to [2^(-13), 1-eps]; these two values map to 0 and 1, respectively.
+  // The tests are carefully written so that NaNs map to 0, same as in the reference
+  // implementation.
+  if (!(in > minval.f)) // written this way to catch NaNs
+      return 0;
+  if (in > almostone.f)
+      return 255;
+
+  // Do the table lookup and unpack bias, scale
+  f.f = in;
+  tab = fp32_to_srgb8_tab4[(f.u - minval.u) >> 20];
+  bias = (tab >> 16) << 9;
+  scale = tab & 0xffff;
+
+  // Grab next-highest mantissa bits and perform linear interpolation
+  t = (f.u >> 12) & 0xff;
+  return (unsigned char) ((bias + scale*t) >> 16);
+}
+
+#ifndef STBIR_FORCE_GATHER_FILTER_SCANLINES_AMOUNT
+#define STBIR_FORCE_GATHER_FILTER_SCANLINES_AMOUNT 32 // when downsampling and <= 32 scanlines of buffering, use gather. gather used down to 1/8th scaling for 25% win.
+#endif
+
+// restrict pointers for the output pointers
+#if defined( _MSC_VER ) && !defined(__clang__)
+  #define STBIR_STREAMOUT_PTR( star ) star __restrict
+  #define STBIR_NO_UNROLL( ptr ) __assume(ptr) // this oddly keeps msvc from unrolling a loop
+#elif defined(  __clang__ )
+  #define STBIR_STREAMOUT_PTR( star ) star __restrict__
+  #define STBIR_NO_UNROLL( ptr ) __asm__ (""::"r"(ptr))
+#elif defined(  __GNUC__ )
+  #define STBIR_STREAMOUT_PTR( star ) star __restrict__
+  #define STBIR_NO_UNROLL( ptr ) __asm__ (""::"r"(ptr))
+#else
+  #define STBIR_STREAMOUT_PTR( star ) star
+  #define STBIR_NO_UNROLL( ptr )
+#endif
+
+#ifdef STBIR_NO_SIMD // force simd off for whatever reason
+
+// force simd off overrides everything else, so clear it all
+
+#ifdef STBIR_SSE2
+#undef STBIR_SSE2
+#endif
+
+#ifdef STBIR_AVX
+#undef STBIR_AVX
+#endif
+
+#ifdef STBIR_NEON
+#undef STBIR_NEON
+#endif
+
+#ifdef STBIR_AVX2
+#undef STBIR_AVX2
+#endif
+
+#ifdef STBIR_FP16C
+#undef STBIR_FP16C
+#endif
+
+#ifdef STBIR_WASM
+#undef STBIR_WASM
+#endif
+
+#ifdef STBIR_SIMD
+#undef STBIR_SIMD
+#endif
+
+#else // STBIR_SIMD
+
+#ifdef STBIR_SSE2
+  #include <emmintrin.h>
+  
+  #define stbir__simdf __m128
+  #define stbir__simdi __m128i
+
+  #define stbir_simdi_castf( reg ) _mm_castps_si128(reg)
+  #define stbir_simdf_casti( reg ) _mm_castsi128_ps(reg)
+
+  #define stbir__simdf_load( reg, ptr ) (reg) = _mm_loadu_ps( (float const*)(ptr) )
+  #define stbir__simdi_load( reg, ptr ) (reg) = _mm_loadu_si128 ( (stbir__simdi const*)(ptr) )
+  #define stbir__simdf_load1( out, ptr ) (out) = _mm_load_ss( (float const*)(ptr) )  // top values can be random (not denormal or nan for perf)
+  #define stbir__simdi_load1( out, ptr ) (out) = _mm_castps_si128( _mm_load_ss( (float const*)(ptr) ))
+  #define stbir__simdf_load1z( out, ptr ) (out) = _mm_load_ss( (float const*)(ptr) )  // top values must be zero
+  #define stbir__simdf_frep4( fvar ) _mm_set_ps1( fvar )
+  #define stbir__simdf_load1frep4( out, fvar ) (out) = _mm_set_ps1( fvar )
+  #define stbir__simdf_load2( out, ptr ) (out) = _mm_castsi128_ps( _mm_loadl_epi64( (__m128i*)(ptr)) ) // top values can be random (not denormal or nan for perf)
+  #define stbir__simdf_load2z( out, ptr ) (out) = _mm_castsi128_ps( _mm_loadl_epi64( (__m128i*)(ptr)) ) // top values must be zero
+  #define stbir__simdf_load2hmerge( out, reg, ptr ) (out) = _mm_castpd_ps(_mm_loadh_pd( _mm_castps_pd(reg), (double*)(ptr) ))
+
+  #define stbir__simdf_zeroP() _mm_setzero_ps()
+  #define stbir__simdf_zero( reg ) (reg) = _mm_setzero_ps()
+
+  #define stbir__simdf_store( ptr, reg )  _mm_storeu_ps( (float*)(ptr), reg )
+  #define stbir__simdf_store1( ptr, reg ) _mm_store_ss( (float*)(ptr), reg )
+  #define stbir__simdf_store2( ptr, reg ) _mm_storel_epi64( (__m128i*)(ptr), _mm_castps_si128(reg) )
+  #define stbir__simdf_store2h( ptr, reg ) _mm_storeh_pd( (double*)(ptr), _mm_castps_pd(reg) )
+
+  #define stbir__simdi_store( ptr, reg )  _mm_storeu_si128( (__m128i*)(ptr), reg )
+  #define stbir__simdi_store1( ptr, reg ) _mm_store_ss( (float*)(ptr), _mm_castsi128_ps(reg) )
+  #define stbir__simdi_store2( ptr, reg ) _mm_storel_epi64( (__m128i*)(ptr), (reg) )
+
+  #define stbir__prefetch( ptr ) _mm_prefetch((char*)(ptr), _MM_HINT_T0 )
+ 
+  #define stbir__simdi_expand_u8_to_u32(out0,out1,out2,out3,ireg) \
+  { \
+    stbir__simdi zero = _mm_setzero_si128(); \
+    out2 = _mm_unpacklo_epi8( ireg, zero ); \
+    out3 = _mm_unpackhi_epi8( ireg, zero ); \
+    out0 = _mm_unpacklo_epi16( out2, zero ); \
+    out1 = _mm_unpackhi_epi16( out2, zero ); \
+    out2 = _mm_unpacklo_epi16( out3, zero ); \
+    out3 = _mm_unpackhi_epi16( out3, zero ); \
+  }
+
+#define stbir__simdi_expand_u8_to_1u32(out,ireg) \
+  { \
+    stbir__simdi zero = _mm_setzero_si128(); \
+    out = _mm_unpacklo_epi8( ireg, zero ); \
+    out = _mm_unpacklo_epi16( out, zero ); \
+  }
+
+  #define stbir__simdi_expand_u16_to_u32(out0,out1,ireg) \
+  { \
+    stbir__simdi zero = _mm_setzero_si128(); \
+    out0 = _mm_unpacklo_epi16( ireg, zero ); \
+    out1 = _mm_unpackhi_epi16( ireg, zero ); \
+  }
+
+  #define stbir__simdf_convert_float_to_i32( i, f ) (i) = _mm_cvttps_epi32(f)
+  #define stbir__simdf_convert_float_to_int( f ) _mm_cvtt_ss2si(f)
+  #define stbir__simdf_convert_float_to_uint8( f ) ((unsigned char)_mm_cvtsi128_si32(_mm_cvttps_epi32(_mm_max_ps(_mm_min_ps(f,STBIR__CONSTF(STBIR_max_uint8_as_float)),_mm_setzero_ps()))))
+  #define stbir__simdf_convert_float_to_short( f ) ((unsigned short)_mm_cvtsi128_si32(_mm_cvttps_epi32(_mm_max_ps(_mm_min_ps(f,STBIR__CONSTF(STBIR_max_uint16_as_float)),_mm_setzero_ps()))))
+
+  #define stbir__simdi_to_int( i ) _mm_cvtsi128_si32(i) 
+  #define stbir__simdi_convert_i32_to_float(out, ireg) (out) = _mm_cvtepi32_ps( ireg )
+  #define stbir__simdf_add( out, reg0, reg1 ) (out) = _mm_add_ps( reg0, reg1 )
+  #define stbir__simdf_mult( out, reg0, reg1 ) (out) = _mm_mul_ps( reg0, reg1 )
+  #define stbir__simdf_mult_mem( out, reg, ptr ) (out) = _mm_mul_ps( reg, _mm_loadu_ps( (float const*)(ptr) ) )
+  #define stbir__simdf_mult1_mem( out, reg, ptr ) (out) = _mm_mul_ss( reg, _mm_load_ss( (float const*)(ptr) ) )
+  #define stbir__simdf_add_mem( out, reg, ptr ) (out) = _mm_add_ps( reg, _mm_loadu_ps( (float const*)(ptr) ) )
+  #define stbir__simdf_add1_mem( out, reg, ptr ) (out) = _mm_add_ss( reg, _mm_load_ss( (float const*)(ptr) ) )
+
+  #ifdef STBIR_USE_FMA           // not on by default to maintain bit identical simd to non-simd
+  #include <immintrin.h>
+  #define stbir__simdf_madd( out, add, mul1, mul2 ) (out) = _mm_fmadd_ps( mul1, mul2, add )
+  #define stbir__simdf_madd1( out, add, mul1, mul2 ) (out) = _mm_fmadd_ss( mul1, mul2, add )
+  #define stbir__simdf_madd_mem( out, add, mul, ptr ) (out) = _mm_fmadd_ps( mul, _mm_loadu_ps( (float const*)(ptr) ), add )
+  #define stbir__simdf_madd1_mem( out, add, mul, ptr ) (out) = _mm_fmadd_ss( mul, _mm_load_ss( (float const*)(ptr) ), add )
+  #else
+  #define stbir__simdf_madd( out, add, mul1, mul2 ) (out) = _mm_add_ps( add, _mm_mul_ps( mul1, mul2 ) )
+  #define stbir__simdf_madd1( out, add, mul1, mul2 ) (out) = _mm_add_ss( add, _mm_mul_ss( mul1, mul2 ) )
+  #define stbir__simdf_madd_mem( out, add, mul, ptr ) (out) = _mm_add_ps( add, _mm_mul_ps( mul, _mm_loadu_ps( (float const*)(ptr) ) ) )
+  #define stbir__simdf_madd1_mem( out, add, mul, ptr ) (out) = _mm_add_ss( add, _mm_mul_ss( mul, _mm_load_ss( (float const*)(ptr) ) ) )
+  #endif
+
+  #define stbir__simdf_add1( out, reg0, reg1 ) (out) = _mm_add_ss( reg0, reg1 )
+  #define stbir__simdf_mult1( out, reg0, reg1 ) (out) = _mm_mul_ss( reg0, reg1 )
+
+  #define stbir__simdf_and( out, reg0, reg1 ) (out) = _mm_and_ps( reg0, reg1 )
+  #define stbir__simdf_or( out, reg0, reg1 ) (out) = _mm_or_ps( reg0, reg1 )
+
+  #define stbir__simdf_min( out, reg0, reg1 ) (out) = _mm_min_ps( reg0, reg1 )
+  #define stbir__simdf_max( out, reg0, reg1 ) (out) = _mm_max_ps( reg0, reg1 )
+  #define stbir__simdf_min1( out, reg0, reg1 ) (out) = _mm_min_ss( reg0, reg1 )
+  #define stbir__simdf_max1( out, reg0, reg1 ) (out) = _mm_max_ss( reg0, reg1 )
+
+  #define stbir__simdf_0123ABCDto3ABx( out, reg0, reg1 ) (out)=_mm_castsi128_ps( _mm_shuffle_epi32( _mm_castps_si128( _mm_shuffle_ps( reg1,reg0, (0<<0) + (1<<2) + (2<<4) + (3<<6) )), (3<<0) + (0<<2) + (1<<4) + (2<<6) ) )
+  #define stbir__simdf_0123ABCDto23Ax( out, reg0, reg1 ) (out)=_mm_castsi128_ps( _mm_shuffle_epi32( _mm_castps_si128( _mm_shuffle_ps( reg1,reg0, (0<<0) + (1<<2) + (2<<4) + (3<<6) )), (2<<0) + (3<<2) + (0<<4) + (1<<6) ) )
+
+  static const stbir__simdf STBIR_zeroones = { 0.0f,1.0f,0.0f,1.0f };
+  static const stbir__simdf STBIR_onezeros = { 1.0f,0.0f,1.0f,0.0f };
+  #define stbir__simdf_aaa1( out, alp, ones ) (out)=_mm_castsi128_ps( _mm_shuffle_epi32( _mm_castps_si128( _mm_movehl_ps( ones, alp ) ), (1<<0) + (1<<2) + (1<<4) + (2<<6) ) )
+  #define stbir__simdf_1aaa( out, alp, ones ) (out)=_mm_castsi128_ps( _mm_shuffle_epi32( _mm_castps_si128( _mm_movelh_ps( ones, alp ) ), (0<<0) + (2<<2) + (2<<4) + (2<<6) ) )
+  #define stbir__simdf_a1a1( out, alp, ones) (out) = _mm_or_ps( _mm_castsi128_ps( _mm_srli_epi64( _mm_castps_si128(alp), 32 ) ), STBIR_zeroones )
+  #define stbir__simdf_1a1a( out, alp, ones) (out) = _mm_or_ps( _mm_castsi128_ps( _mm_slli_epi64( _mm_castps_si128(alp), 32 ) ), STBIR_onezeros )
+
+  #define stbir__simdf_swiz( reg, one, two, three, four ) _mm_castsi128_ps( _mm_shuffle_epi32( _mm_castps_si128( reg ), (one<<0) + (two<<2) + (three<<4) + (four<<6) ) )
+
+  #define stbir__simdi_and( out, reg0, reg1 ) (out) = _mm_and_si128( reg0, reg1 )
+  #define stbir__simdi_or( out, reg0, reg1 ) (out) = _mm_or_si128( reg0, reg1 )
+  #define stbir__simdi_16madd( out, reg0, reg1 ) (out) = _mm_madd_epi16( reg0, reg1 )
+
+  #define stbir__simdf_pack_to_8bytes(out,aa,bb) \
+  { \
+    stbir__simdf af,bf; \
+    stbir__simdi a,b; \
+    af = _mm_min_ps( aa, STBIR_max_uint8_as_float ); \
+    bf = _mm_min_ps( bb, STBIR_max_uint8_as_float ); \
+    af = _mm_max_ps( af, _mm_setzero_ps() ); \
+    bf = _mm_max_ps( bf, _mm_setzero_ps() ); \
+    a = _mm_cvttps_epi32( af ); \
+    b = _mm_cvttps_epi32( bf ); \
+    a = _mm_packs_epi32( a, b ); \
+    out = _mm_packus_epi16( a, a ); \
+  }
+
+  #define stbir__simdf_load4_transposed( o0, o1, o2, o3, ptr ) \
+      stbir__simdf_load( o0, (ptr) );    \
+      stbir__simdf_load( o1, (ptr)+4 );  \
+      stbir__simdf_load( o2, (ptr)+8 );  \
+      stbir__simdf_load( o3, (ptr)+12 ); \
+      {                                  \
+        __m128 tmp0, tmp1, tmp2, tmp3;   \
+        tmp0 = _mm_unpacklo_ps(o0, o1);  \
+        tmp2 = _mm_unpacklo_ps(o2, o3);  \
+        tmp1 = _mm_unpackhi_ps(o0, o1);  \
+        tmp3 = _mm_unpackhi_ps(o2, o3);  \
+        o0 = _mm_movelh_ps(tmp0, tmp2);  \
+        o1 = _mm_movehl_ps(tmp2, tmp0);  \
+        o2 = _mm_movelh_ps(tmp1, tmp3);  \
+        o3 = _mm_movehl_ps(tmp3, tmp1);  \
+      }
+
+  #define stbir__interleave_pack_and_store_16_u8( ptr, r0, r1, r2, r3 ) \
+      r0 = _mm_packs_epi32( r0, r1 ); \
+      r2 = _mm_packs_epi32( r2, r3 ); \
+      r1 = _mm_unpacklo_epi16( r0, r2 ); \
+      r3 = _mm_unpackhi_epi16( r0, r2 ); \
+      r0 = _mm_unpacklo_epi16( r1, r3 ); \
+      r2 = _mm_unpackhi_epi16( r1, r3 ); \
+      r0 = _mm_packus_epi16( r0, r2 ); \
+      stbir__simdi_store( ptr, r0 ); \
+
+  #define stbir__simdi_32shr( out, reg, imm ) out = _mm_srli_epi32( reg, imm )
+
+  #if defined(_MSC_VER) && !defined(__clang__)
+    // msvc inits with 8 bytes
+    #define STBIR__CONST_32_TO_8( v ) (char)(unsigned char)((v)&255),(char)(unsigned char)(((v)>>8)&255),(char)(unsigned char)(((v)>>16)&255),(char)(unsigned char)(((v)>>24)&255)
+    #define STBIR__CONST_4_32i( v ) STBIR__CONST_32_TO_8( v ), STBIR__CONST_32_TO_8( v ), STBIR__CONST_32_TO_8( v ), STBIR__CONST_32_TO_8( v )
+    #define STBIR__CONST_4d_32i( v0, v1, v2, v3 ) STBIR__CONST_32_TO_8( v0 ), STBIR__CONST_32_TO_8( v1 ), STBIR__CONST_32_TO_8( v2 ), STBIR__CONST_32_TO_8( v3 )
+  #else
+    // everything else inits with long long's
+    #define STBIR__CONST_4_32i( v ) (long long)((((stbir_uint64)(stbir_uint32)(v))<<32)|((stbir_uint64)(stbir_uint32)(v))),(long long)((((stbir_uint64)(stbir_uint32)(v))<<32)|((stbir_uint64)(stbir_uint32)(v)))
+    #define STBIR__CONST_4d_32i( v0, v1, v2, v3 ) (long long)((((stbir_uint64)(stbir_uint32)(v1))<<32)|((stbir_uint64)(stbir_uint32)(v0))),(long long)((((stbir_uint64)(stbir_uint32)(v3))<<32)|((stbir_uint64)(stbir_uint32)(v2)))
+  #endif
+
+  #define STBIR__SIMDF_CONST(var, x) stbir__simdf var = { x, x, x, x }
+  #define STBIR__SIMDI_CONST(var, x) stbir__simdi var = { STBIR__CONST_4_32i(x) }
+  #define STBIR__CONSTF(var) (var)
+  #define STBIR__CONSTI(var) (var)
+
+  #if defined(STBIR_AVX) || defined(__SSE4_1__)
+    #include <smmintrin.h>
+    #define stbir__simdf_pack_to_8words(out,reg0,reg1) out = _mm_packus_epi32(_mm_cvttps_epi32(_mm_max_ps(_mm_min_ps(reg0,STBIR__CONSTF(STBIR_max_uint16_as_float)),_mm_setzero_ps())), _mm_cvttps_epi32(_mm_max_ps(_mm_min_ps(reg1,STBIR__CONSTF(STBIR_max_uint16_as_float)),_mm_setzero_ps())))
+  #else
+    STBIR__SIMDI_CONST(stbir__s32_32768, 32768);
+    STBIR__SIMDI_CONST(stbir__s16_32768, ((32768<<16)|32768));
+
+    #define stbir__simdf_pack_to_8words(out,reg0,reg1) \
+      { \
+        stbir__simdi tmp0,tmp1; \
+        tmp0 = _mm_cvttps_epi32(_mm_max_ps(_mm_min_ps(reg0,STBIR__CONSTF(STBIR_max_uint16_as_float)),_mm_setzero_ps())); \
+        tmp1 = _mm_cvttps_epi32(_mm_max_ps(_mm_min_ps(reg1,STBIR__CONSTF(STBIR_max_uint16_as_float)),_mm_setzero_ps())); \
+        tmp0 = _mm_sub_epi32( tmp0, stbir__s32_32768 ); \
+        tmp1 = _mm_sub_epi32( tmp1, stbir__s32_32768 ); \
+        out = _mm_packs_epi32( tmp0, tmp1 ); \
+        out = _mm_sub_epi16( out, stbir__s16_32768 ); \
+      }
+
+  #endif
+
+  #define STBIR_SIMD
+
+  // if we detect AVX, set the simd8 defines
+  #ifdef STBIR_AVX
+    #include <immintrin.h>
+    #define STBIR_SIMD8
+    #define stbir__simdf8 __m256
+    #define stbir__simdi8 __m256i
+    #define stbir__simdf8_load( out, ptr ) (out) = _mm256_loadu_ps( (float const *)(ptr) )
+    #define stbir__simdi8_load( out, ptr ) (out) = _mm256_loadu_si256( (__m256i const *)(ptr) )
+    #define stbir__simdf8_mult( out, a, b ) (out) = _mm256_mul_ps( (a), (b) )
+    #define stbir__simdf8_store( ptr, out ) _mm256_storeu_ps( (float*)(ptr), out )
+    #define stbir__simdi8_store( ptr, reg )  _mm256_storeu_si256( (__m256i*)(ptr), reg )
+    #define stbir__simdf8_frep8( fval ) _mm256_set1_ps( fval )
+
+    #define stbir__simdf8_min( out, reg0, reg1 ) (out) = _mm256_min_ps( reg0, reg1 )
+    #define stbir__simdf8_max( out, reg0, reg1 ) (out) = _mm256_max_ps( reg0, reg1 )
+
+    #define stbir__simdf8_add4halves( out, bot4, top8 ) (out) = _mm_add_ps( bot4, _mm256_extractf128_ps( top8, 1 ) )
+    #define stbir__simdf8_mult_mem( out, reg, ptr ) (out) = _mm256_mul_ps( reg, _mm256_loadu_ps( (float const*)(ptr) ) )
+    #define stbir__simdf8_add_mem( out, reg, ptr ) (out) = _mm256_add_ps( reg, _mm256_loadu_ps( (float const*)(ptr) ) )
+    #define stbir__simdf8_add( out, a, b ) (out) = _mm256_add_ps( a, b )
+    #define stbir__simdf8_load1b( out, ptr ) (out) = _mm256_broadcast_ss( ptr )
+    #define stbir__simdf_load1rep4( out, ptr ) (out) = _mm_broadcast_ss( ptr )  // avx load instruction
+
+    #define stbir__simdi8_convert_i32_to_float(out, ireg) (out) = _mm256_cvtepi32_ps( ireg )
+    #define stbir__simdf8_convert_float_to_i32( i, f ) (i) = _mm256_cvttps_epi32(f)
+  
+    #define stbir__simdf8_bot4s( out, a, b ) (out) = _mm256_permute2f128_ps(a,b, (0<<0)+(2<<4) )
+    #define stbir__simdf8_top4s( out, a, b ) (out) = _mm256_permute2f128_ps(a,b, (1<<0)+(3<<4) )
+    
+    #define stbir__simdf8_gettop4( reg ) _mm256_extractf128_ps(reg,1)
+
+    #ifdef STBIR_AVX2
+
+    #define stbir__simdi8_expand_u8_to_u32(out0,out1,ireg) \
+    { \
+      stbir__simdi8 a, zero  =_mm256_setzero_si256();\
+      a = _mm256_permute4x64_epi64( _mm256_unpacklo_epi8( _mm256_permute4x64_epi64(_mm256_castsi128_si256(ireg),(0<<0)+(2<<2)+(1<<4)+(3<<6)), zero ),(0<<0)+(2<<2)+(1<<4)+(3<<6)); \
+      out0 = _mm256_unpacklo_epi16( a, zero ); \
+      out1 = _mm256_unpackhi_epi16( a, zero ); \
+    }
+
+    #define stbir__simdf8_pack_to_16bytes(out,aa,bb) \
+    { \
+      stbir__simdi8 t; \
+      stbir__simdf8 af,bf; \
+      stbir__simdi8 a,b; \
+      af = _mm256_min_ps( aa, STBIR_max_uint8_as_floatX ); \
+      bf = _mm256_min_ps( bb, STBIR_max_uint8_as_floatX ); \
+      af = _mm256_max_ps( af, _mm256_setzero_ps() ); \
+      bf = _mm256_max_ps( bf, _mm256_setzero_ps() ); \
+      a = _mm256_cvttps_epi32( af ); \
+      b = _mm256_cvttps_epi32( bf ); \
+      t = _mm256_permute4x64_epi64( _mm256_packs_epi32( a, b ), (0<<0)+(2<<2)+(1<<4)+(3<<6) ); \
+      out = _mm256_castsi256_si128( _mm256_permute4x64_epi64( _mm256_packus_epi16( t, t ), (0<<0)+(2<<2)+(1<<4)+(3<<6) ) ); \
+    }
+
+    #define stbir__simdi8_expand_u16_to_u32(out,ireg) out = _mm256_unpacklo_epi16( _mm256_permute4x64_epi64(_mm256_castsi128_si256(ireg),(0<<0)+(2<<2)+(1<<4)+(3<<6)), _mm256_setzero_si256() ); 
+  
+    #define stbir__simdf8_pack_to_16words(out,aa,bb) \
+      { \
+        stbir__simdf8 af,bf; \
+        stbir__simdi8 a,b; \
+        af = _mm256_min_ps( aa, STBIR_max_uint16_as_floatX ); \
+        bf = _mm256_min_ps( bb, STBIR_max_uint16_as_floatX ); \
+        af = _mm256_max_ps( af, _mm256_setzero_ps() ); \
+        bf = _mm256_max_ps( bf, _mm256_setzero_ps() ); \
+        a = _mm256_cvttps_epi32( af ); \
+        b = _mm256_cvttps_epi32( bf ); \
+        (out) = _mm256_permute4x64_epi64( _mm256_packus_epi32(a, b), (0<<0)+(2<<2)+(1<<4)+(3<<6) ); \
+      }
+
+    #else
+
+    #define stbir__simdi8_expand_u8_to_u32(out0,out1,ireg) \
+    { \
+      stbir__simdi a,zero = _mm_setzero_si128(); \
+      a = _mm_unpacklo_epi8( ireg, zero ); \
+      out0 = _mm256_setr_m128i( _mm_unpacklo_epi16( a, zero ), _mm_unpackhi_epi16( a, zero ) ); \
+      a = _mm_unpackhi_epi8( ireg, zero ); \
+      out1 = _mm256_setr_m128i( _mm_unpacklo_epi16( a, zero ), _mm_unpackhi_epi16( a, zero ) ); \
+    }
+  
+    #define stbir__simdf8_pack_to_16bytes(out,aa,bb) \
+    { \
+      stbir__simdi t; \
+      stbir__simdf8 af,bf; \
+      stbir__simdi8 a,b; \
+      af = _mm256_min_ps( aa, STBIR_max_uint8_as_floatX ); \
+      bf = _mm256_min_ps( bb, STBIR_max_uint8_as_floatX ); \
+      af = _mm256_max_ps( af, _mm256_setzero_ps() ); \
+      bf = _mm256_max_ps( bf, _mm256_setzero_ps() ); \
+      a = _mm256_cvttps_epi32( af ); \
+      b = _mm256_cvttps_epi32( bf ); \
+      out = _mm_packs_epi32( _mm256_castsi256_si128(a), _mm256_extractf128_si256( a, 1 ) ); \
+      out = _mm_packus_epi16( out, out ); \
+      t = _mm_packs_epi32( _mm256_castsi256_si128(b), _mm256_extractf128_si256( b, 1 ) ); \
+      t = _mm_packus_epi16( t, t ); \
+      out = _mm_castps_si128( _mm_shuffle_ps( _mm_castsi128_ps(out), _mm_castsi128_ps(t), (0<<0)+(1<<2)+(0<<4)+(1<<6) ) ); \
+    }
+  
+    #define stbir__simdi8_expand_u16_to_u32(out,ireg) \
+    { \
+      stbir__simdi a,b,zero = _mm_setzero_si128(); \
+      a = _mm_unpacklo_epi16( ireg, zero ); \
+      b = _mm_unpackhi_epi16( ireg, zero ); \
+      out = _mm256_insertf128_si256( _mm256_castsi128_si256( a ), b, 1 ); \
+    }
+
+    #define stbir__simdf8_pack_to_16words(out,aa,bb) \
+      { \
+        stbir__simdi t0,t1; \
+        stbir__simdf8 af,bf; \
+        stbir__simdi8 a,b; \
+        af = _mm256_min_ps( aa, STBIR_max_uint16_as_floatX ); \
+        bf = _mm256_min_ps( bb, STBIR_max_uint16_as_floatX ); \
+        af = _mm256_max_ps( af, _mm256_setzero_ps() ); \
+        bf = _mm256_max_ps( bf, _mm256_setzero_ps() ); \
+        a = _mm256_cvttps_epi32( af ); \
+        b = _mm256_cvttps_epi32( bf ); \
+        t0 = _mm_packus_epi32( _mm256_castsi256_si128(a), _mm256_extractf128_si256( a, 1 ) ); \
+        t1 = _mm_packus_epi32( _mm256_castsi256_si128(b), _mm256_extractf128_si256( b, 1 ) ); \
+        out = _mm256_setr_m128i( t0, t1 ); \
+      }
+
+    #endif
+
+    static __m256i stbir_00001111 = { STBIR__CONST_4d_32i( 0, 0, 0, 0 ), STBIR__CONST_4d_32i( 1, 1, 1, 1 ) };
+    #define stbir__simdf8_0123to00001111( out, in ) (out) = _mm256_permutevar_ps ( in, stbir_00001111 )
+
+    static __m256i stbir_22223333 = { STBIR__CONST_4d_32i( 2, 2, 2, 2 ), STBIR__CONST_4d_32i( 3, 3, 3, 3 ) };
+    #define stbir__simdf8_0123to22223333( out, in ) (out) = _mm256_permutevar_ps ( in, stbir_22223333 )
+
+    #define stbir__simdf8_0123to2222( out, in ) (out) = stbir__simdf_swiz(_mm256_castps256_ps128(in), 2,2,2,2 )
+
+    #define stbir__simdf8_load2( out, ptr ) (out) = _mm256_castsi256_ps(_mm256_castsi128_si256( _mm_loadl_epi64( (__m128i*)(ptr)) )) // top values can be random (not denormal or nan for perf)
+    #define stbir__simdf8_load4b( out, ptr ) (out) = _mm256_broadcast_ps( (__m128 const *)(ptr) )
+
+    static __m256i stbir_00112233 = { STBIR__CONST_4d_32i( 0, 0, 1, 1 ), STBIR__CONST_4d_32i( 2, 2, 3, 3 ) };
+    #define stbir__simdf8_0123to00112233( out, in ) (out) = _mm256_permutevar_ps ( in, stbir_00112233 )
+    #define stbir__simdf8_add4( out, a8, b ) (out) = _mm256_add_ps( a8,  _mm256_castps128_ps256( b ) )
+
+    static __m256i stbir_load6 = { STBIR__CONST_4_32i( 0x80000000 ), STBIR__CONST_4d_32i(  0x80000000,  0x80000000, 0, 0 ) };
+    #define stbir__simdf8_load6z( out, ptr ) (out) = _mm256_maskload_ps( ptr, stbir_load6 )
+
+    #define stbir__simdf8_0123to00000000( out, in ) (out) =  _mm256_shuffle_ps ( in, in, (0<<0)+(0<<2)+(0<<4)+(0<<6) )
+    #define stbir__simdf8_0123to11111111( out, in ) (out) =  _mm256_shuffle_ps ( in, in, (1<<0)+(1<<2)+(1<<4)+(1<<6) )
+    #define stbir__simdf8_0123to22222222( out, in ) (out) =  _mm256_shuffle_ps ( in, in, (2<<0)+(2<<2)+(2<<4)+(2<<6) )
+    #define stbir__simdf8_0123to33333333( out, in ) (out) =  _mm256_shuffle_ps ( in, in, (3<<0)+(3<<2)+(3<<4)+(3<<6) )
+    #define stbir__simdf8_0123to21032103( out, in ) (out) =  _mm256_shuffle_ps ( in, in, (2<<0)+(1<<2)+(0<<4)+(3<<6) )
+    #define stbir__simdf8_0123to32103210( out, in ) (out) =  _mm256_shuffle_ps ( in, in, (3<<0)+(2<<2)+(1<<4)+(0<<6) )
+    #define stbir__simdf8_0123to12301230( out, in ) (out) =  _mm256_shuffle_ps ( in, in, (1<<0)+(2<<2)+(3<<4)+(0<<6) )
+    #define stbir__simdf8_0123to10321032( out, in ) (out) =  _mm256_shuffle_ps ( in, in, (1<<0)+(0<<2)+(3<<4)+(2<<6) )
+    #define stbir__simdf8_0123to30123012( out, in ) (out) =  _mm256_shuffle_ps ( in, in, (3<<0)+(0<<2)+(1<<4)+(2<<6) )
+
+    #define stbir__simdf8_0123to11331133( out, in ) (out) =  _mm256_shuffle_ps ( in, in, (1<<0)+(1<<2)+(3<<4)+(3<<6) )
+    #define stbir__simdf8_0123to00220022( out, in ) (out) =  _mm256_shuffle_ps ( in, in, (0<<0)+(0<<2)+(2<<4)+(2<<6) )
+
+    #define stbir__simdf8_aaa1( out, alp, ones ) (out) = _mm256_blend_ps( alp, ones, (1<<0)+(1<<1)+(1<<2)+(0<<3)+(1<<4)+(1<<5)+(1<<6)+(0<<7)); (out)=_mm256_shuffle_ps( out,out, (3<<0) + (3<<2) + (3<<4) + (0<<6) )
+    #define stbir__simdf8_1aaa( out, alp, ones ) (out) = _mm256_blend_ps( alp, ones, (0<<0)+(1<<1)+(1<<2)+(1<<3)+(0<<4)+(1<<5)+(1<<6)+(1<<7)); (out)=_mm256_shuffle_ps( out,out, (1<<0) + (0<<2) + (0<<4) + (0<<6) )
+    #define stbir__simdf8_a1a1( out, alp, ones) (out) = _mm256_blend_ps( alp, ones, (1<<0)+(0<<1)+(1<<2)+(0<<3)+(1<<4)+(0<<5)+(1<<6)+(0<<7)); (out)=_mm256_shuffle_ps( out,out, (1<<0) + (0<<2) + (3<<4) + (2<<6) )
+    #define stbir__simdf8_1a1a( out, alp, ones) (out) = _mm256_blend_ps( alp, ones, (0<<0)+(1<<1)+(0<<2)+(1<<3)+(0<<4)+(1<<5)+(0<<6)+(1<<7)); (out)=_mm256_shuffle_ps( out,out, (1<<0) + (0<<2) + (3<<4) + (2<<6) )
+
+    #define stbir__simdf8_zero( reg ) (reg) = _mm256_setzero_ps()
+
+    #ifdef STBIR_USE_FMA           // not on by default to maintain bit identical simd to non-simd
+    #define stbir__simdf8_madd( out, add, mul1, mul2 ) (out) = _mm256_fmadd_ps( mul1, mul2, add )
+    #define stbir__simdf8_madd_mem( out, add, mul, ptr ) (out) = _mm256_fmadd_ps( mul, _mm256_loadu_ps( (float const*)(ptr) ), add )
+    #define stbir__simdf8_madd_mem4( out, add, mul, ptr ) (out) = _mm256_fmadd_ps( _mm256_castps128_ps256( mul ), _mm256_castps128_ps256( _mm_loadu_ps( (float const*)(ptr) ) ), add )
+    #else
+    #define stbir__simdf8_madd( out, add, mul1, mul2 ) (out) = _mm256_add_ps( add, _mm256_mul_ps( mul1, mul2 ) )
+    #define stbir__simdf8_madd_mem( out, add, mul, ptr ) (out) = _mm256_add_ps( add, _mm256_mul_ps( mul, _mm256_loadu_ps( (float const*)(ptr) ) ) )
+    #define stbir__simdf8_madd_mem4( out, add, mul, ptr ) (out) = _mm256_add_ps( add, _mm256_castps128_ps256( _mm_mul_ps( mul, _mm_loadu_ps( (float const*)(ptr) ) ) ) )
+    #endif
+    #define stbir__if_simdf8_cast_to_simdf4( val ) _mm256_castps256_ps128( val )
+
+  #endif
+
+  #ifdef STBIR_FLOORF
+  #undef STBIR_FLOORF
+  #endif
+  #define STBIR_FLOORF stbir_simd_floorf
+  static stbir__inline float stbir_simd_floorf(float x)  // martins floorf
+  {
+    #if defined(STBIR_AVX) || defined(__SSE4_1__) || defined(STBIR_SSE41)
+    __m128 t = _mm_set_ss(x);
+    return _mm_cvtss_f32( _mm_floor_ss(t, t) );
+    #else
+    __m128 f = _mm_set_ss(x);
+    __m128 t = _mm_cvtepi32_ps(_mm_cvttps_epi32(f));
+    __m128 r = _mm_add_ss(t, _mm_and_ps(_mm_cmplt_ss(f, t), _mm_set_ss(-1.0f)));
+    return _mm_cvtss_f32(r);
+    #endif
+  }
+
+  #ifdef STBIR_CEILF
+  #undef STBIR_CEILF
+  #endif
+  #define STBIR_CEILF stbir_simd_ceilf
+  static stbir__inline float stbir_simd_ceilf(float x)  // martins ceilf
+  {
+    #if defined(STBIR_AVX) || defined(__SSE4_1__) || defined(STBIR_SSE41)
+    __m128 t = _mm_set_ss(x);
+    return _mm_cvtss_f32( _mm_ceil_ss(t, t) );
+    #else
+    __m128 f = _mm_set_ss(x);
+    __m128 t = _mm_cvtepi32_ps(_mm_cvttps_epi32(f));
+    __m128 r = _mm_add_ss(t, _mm_and_ps(_mm_cmplt_ss(t, f), _mm_set_ss(1.0f)));
+    return _mm_cvtss_f32(r);
+    #endif
+  }
+
+#elif defined(STBIR_NEON)
+  
+  #include <arm_neon.h>
+
+  #define stbir__simdf float32x4_t
+  #define stbir__simdi uint32x4_t
+
+  #define stbir_simdi_castf( reg ) vreinterpretq_u32_f32(reg)
+  #define stbir_simdf_casti( reg ) vreinterpretq_f32_u32(reg)
+
+  #define stbir__simdf_load( reg, ptr ) (reg) = vld1q_f32( (float const*)(ptr) )
+  #define stbir__simdi_load( reg, ptr ) (reg) = vld1q_u32( (uint32_t const*)(ptr) )
+  #define stbir__simdf_load1( out, ptr ) (out) = vld1q_dup_f32( (float const*)(ptr) ) // top values can be random (not denormal or nan for perf)
+  #define stbir__simdi_load1( out, ptr ) (out) = vld1q_dup_u32( (uint32_t const*)(ptr) )
+  #define stbir__simdf_load1z( out, ptr ) (out) = vld1q_lane_f32( (float const*)(ptr), vdupq_n_f32(0), 0 ) // top values must be zero
+  #define stbir__simdf_frep4( fvar ) vdupq_n_f32( fvar )
+  #define stbir__simdf_load1frep4( out, fvar ) (out) = vdupq_n_f32( fvar )
+  #define stbir__simdf_load2( out, ptr ) (out) = vcombine_f32( vld1_f32( (float const*)(ptr) ), vcreate_f32(0) ) // top values can be random (not denormal or nan for perf)
+  #define stbir__simdf_load2z( out, ptr ) (out) = vcombine_f32( vld1_f32( (float const*)(ptr) ), vcreate_f32(0) )  // top values must be zero
+  #define stbir__simdf_load2hmerge( out, reg, ptr ) (out) = vcombine_f32( vget_low_f32(reg), vld1_f32( (float const*)(ptr) ) )
+
+  #define stbir__simdf_zeroP() vdupq_n_f32(0)
+  #define stbir__simdf_zero( reg ) (reg) = vdupq_n_f32(0)
+
+  #define stbir__simdf_store( ptr, reg )  vst1q_f32( (float*)(ptr), reg )
+  #define stbir__simdf_store1( ptr, reg ) vst1q_lane_f32( (float*)(ptr), reg, 0)
+  #define stbir__simdf_store2( ptr, reg ) vst1_f32( (float*)(ptr), vget_low_f32(reg) )
+  #define stbir__simdf_store2h( ptr, reg ) vst1_f32( (float*)(ptr), vget_high_f32(reg) )
+
+  #define stbir__simdi_store( ptr, reg )  vst1q_u32( (uint32_t*)(ptr), reg )
+  #define stbir__simdi_store1( ptr, reg ) vst1q_lane_u32( (uint32_t*)(ptr), reg, 0 )
+  #define stbir__simdi_store2( ptr, reg ) vst1_u32( (uint32_t*)(ptr), vget_low_u32(reg) )
+
+  #define stbir__prefetch( ptr )
+
+  #define stbir__simdi_expand_u8_to_u32(out0,out1,out2,out3,ireg) \
+  { \
+    uint16x8_t l = vmovl_u8( vget_low_u8 ( vreinterpretq_u8_u32(ireg) ) ); \
+    uint16x8_t h = vmovl_u8( vget_high_u8( vreinterpretq_u8_u32(ireg) ) ); \
+    out0 = vmovl_u16( vget_low_u16 ( l ) ); \
+    out1 = vmovl_u16( vget_high_u16( l ) ); \
+    out2 = vmovl_u16( vget_low_u16 ( h ) ); \
+    out3 = vmovl_u16( vget_high_u16( h ) ); \
+  }
+
+  #define stbir__simdi_expand_u8_to_1u32(out,ireg) \
+  { \
+    uint16x8_t tmp = vmovl_u8( vget_low_u8( vreinterpretq_u8_u32(ireg) ) ); \
+    out = vmovl_u16( vget_low_u16( tmp ) ); \
+  }
+
+  #define stbir__simdi_expand_u16_to_u32(out0,out1,ireg) \
+  { \
+    uint16x8_t tmp = vreinterpretq_u16_u32(ireg); \
+    out0 = vmovl_u16( vget_low_u16 ( tmp ) ); \
+    out1 = vmovl_u16( vget_high_u16( tmp ) ); \
+  }
+
+  #define stbir__simdf_convert_float_to_i32( i, f ) (i) = vreinterpretq_u32_s32( vcvtq_s32_f32(f) )
+  #define stbir__simdf_convert_float_to_int( f ) vgetq_lane_s32(vcvtq_s32_f32(f), 0)
+  #define stbir__simdi_to_int( i ) (int)vgetq_lane_u32(i, 0) 
+  #define stbir__simdf_convert_float_to_uint8( f ) ((unsigned char)vgetq_lane_s32(vcvtq_s32_f32(vmaxq_f32(vminq_f32(f,STBIR__CONSTF(STBIR_max_uint8_as_float)),vdupq_n_f32(0))), 0))
+  #define stbir__simdf_convert_float_to_short( f ) ((unsigned short)vgetq_lane_s32(vcvtq_s32_f32(vmaxq_f32(vminq_f32(f,STBIR__CONSTF(STBIR_max_uint16_as_float)),vdupq_n_f32(0))), 0))
+  #define stbir__simdi_convert_i32_to_float(out, ireg) (out) = vcvtq_f32_s32( vreinterpretq_s32_u32(ireg) )
+  #define stbir__simdf_add( out, reg0, reg1 ) (out) = vaddq_f32( reg0, reg1 )
+  #define stbir__simdf_mult( out, reg0, reg1 ) (out) = vmulq_f32( reg0, reg1 )
+  #define stbir__simdf_mult_mem( out, reg, ptr ) (out) = vmulq_f32( reg, vld1q_f32( (float const*)(ptr) ) )
+  #define stbir__simdf_mult1_mem( out, reg, ptr ) (out) = vmulq_f32( reg, vld1q_dup_f32( (float const*)(ptr) ) )
+  #define stbir__simdf_add_mem( out, reg, ptr ) (out) = vaddq_f32( reg, vld1q_f32( (float const*)(ptr) ) )
+  #define stbir__simdf_add1_mem( out, reg, ptr ) (out) = vaddq_f32( reg, vld1q_dup_f32( (float const*)(ptr) ) )
+
+  #ifdef STBIR_USE_FMA           // not on by default to maintain bit identical simd to non-simd (and also x64 no madd to arm madd)
+  #define stbir__simdf_madd( out, add, mul1, mul2 ) (out) = vfmaq_f32( add, mul1, mul2 )
+  #define stbir__simdf_madd1( out, add, mul1, mul2 ) (out) = vfmaq_f32( add, mul1, mul2 )
+  #define stbir__simdf_madd_mem( out, add, mul, ptr ) (out) = vfmaq_f32( add, mul, vld1q_f32( (float const*)(ptr) ) )
+  #define stbir__simdf_madd1_mem( out, add, mul, ptr ) (out) = vfmaq_f32( add, mul, vld1q_dup_f32( (float const*)(ptr) ) )
+  #else
+  #define stbir__simdf_madd( out, add, mul1, mul2 ) (out) = vaddq_f32( add, vmulq_f32( mul1, mul2 ) )
+  #define stbir__simdf_madd1( out, add, mul1, mul2 ) (out) = vaddq_f32( add, vmulq_f32( mul1, mul2 ) )
+  #define stbir__simdf_madd_mem( out, add, mul, ptr ) (out) = vaddq_f32( add, vmulq_f32( mul, vld1q_f32( (float const*)(ptr) ) ) )
+  #define stbir__simdf_madd1_mem( out, add, mul, ptr ) (out) = vaddq_f32( add, vmulq_f32( mul, vld1q_dup_f32( (float const*)(ptr) ) ) )
+  #endif
+
+  #define stbir__simdf_add1( out, reg0, reg1 ) (out) = vaddq_f32( reg0, reg1 )
+  #define stbir__simdf_mult1( out, reg0, reg1 ) (out) = vmulq_f32( reg0, reg1 )
+
+  #define stbir__simdf_and( out, reg0, reg1 ) (out) = vreinterpretq_f32_u32( vandq_u32( vreinterpretq_u32_f32(reg0), vreinterpretq_u32_f32(reg1) ) )
+  #define stbir__simdf_or( out, reg0, reg1 ) (out) = vreinterpretq_f32_u32( vorrq_u32( vreinterpretq_u32_f32(reg0), vreinterpretq_u32_f32(reg1) ) )
+
+  #define stbir__simdf_min( out, reg0, reg1 ) (out) = vminq_f32( reg0, reg1 )
+  #define stbir__simdf_max( out, reg0, reg1 ) (out) = vmaxq_f32( reg0, reg1 )
+  #define stbir__simdf_min1( out, reg0, reg1 ) (out) = vminq_f32( reg0, reg1 )
+  #define stbir__simdf_max1( out, reg0, reg1 ) (out) = vmaxq_f32( reg0, reg1 )
+
+  #define stbir__simdf_0123ABCDto3ABx( out, reg0, reg1 ) (out) = vextq_f32( reg0, reg1, 3 )
+  #define stbir__simdf_0123ABCDto23Ax( out, reg0, reg1 ) (out) = vextq_f32( reg0, reg1, 2 )
+
+  #define stbir__simdf_a1a1( out, alp, ones ) (out) = vzipq_f32(vuzpq_f32(alp, alp).val[1], ones).val[0]
+  #define stbir__simdf_1a1a( out, alp, ones ) (out) = vzipq_f32(ones, vuzpq_f32(alp, alp).val[0]).val[0]
+
+  #if defined( _M_ARM64 ) || defined( __aarch64__ ) || defined( __arm64__ )
+
+    #define stbir__simdf_aaa1( out, alp, ones ) (out) = vcopyq_laneq_f32(vdupq_n_f32(vgetq_lane_f32(alp, 3)), 3, ones, 3)
+    #define stbir__simdf_1aaa( out, alp, ones ) (out) = vcopyq_laneq_f32(vdupq_n_f32(vgetq_lane_f32(alp, 0)), 0, ones, 0)
+
+    #if defined( _MSC_VER ) && !defined(__clang__)
+      #define stbir_make16(a,b,c,d) vcombine_u8( \
+        vcreate_u8( (4*a+0) | ((4*a+1)<<8) | ((4*a+2)<<16) | ((4*a+3)<<24) | \
+          ((stbir_uint64)(4*b+0)<<32) | ((stbir_uint64)(4*b+1)<<40) | ((stbir_uint64)(4*b+2)<<48) | ((stbir_uint64)(4*b+3)<<56)), \
+        vcreate_u8( (4*c+0) | ((4*c+1)<<8) | ((4*c+2)<<16) | ((4*c+3)<<24) | \
+          ((stbir_uint64)(4*d+0)<<32) | ((stbir_uint64)(4*d+1)<<40) | ((stbir_uint64)(4*d+2)<<48) | ((stbir_uint64)(4*d+3)<<56) ) )
+    #else
+      #define stbir_make16(a,b,c,d) (uint8x16_t){4*a+0,4*a+1,4*a+2,4*a+3,4*b+0,4*b+1,4*b+2,4*b+3,4*c+0,4*c+1,4*c+2,4*c+3,4*d+0,4*d+1,4*d+2,4*d+3}
+    #endif
+
+    #define stbir__simdf_swiz( reg, one, two, three, four ) vreinterpretq_f32_u8( vqtbl1q_u8( vreinterpretq_u8_f32(reg), stbir_make16(one, two, three, four) ) )
+  
+    #define stbir__simdi_16madd( out, reg0, reg1 ) \
+    { \
+      int16x8_t r0 = vreinterpretq_s16_u32(reg0); \
+      int16x8_t r1 = vreinterpretq_s16_u32(reg1); \
+      int32x4_t tmp0 = vmull_s16( vget_low_s16(r0), vget_low_s16(r1) ); \
+      int32x4_t tmp1 = vmull_s16( vget_high_s16(r0), vget_high_s16(r1) ); \
+      (out) = vreinterpretq_u32_s32( vpaddq_s32(tmp0, tmp1) ); \
+    }
+
+  #else
+
+    #define stbir__simdf_aaa1( out, alp, ones ) (out) = vsetq_lane_f32(1.0f, vdupq_n_f32(vgetq_lane_f32(alp, 3)), 3)
+    #define stbir__simdf_1aaa( out, alp, ones ) (out) = vsetq_lane_f32(1.0f, vdupq_n_f32(vgetq_lane_f32(alp, 0)), 0)
+
+    #if defined( _MSC_VER ) && !defined(__clang__)
+      static stbir__inline uint8x8x2_t stbir_make8x2(float32x4_t reg)
+      {
+        uint8x8x2_t r = { { vget_low_u8(vreinterpretq_u8_f32(reg)), vget_high_u8(vreinterpretq_u8_f32(reg)) } };
+        return r;
+      }
+      #define stbir_make8(a,b) vcreate_u8( \
+        (4*a+0) | ((4*a+1)<<8) | ((4*a+2)<<16) | ((4*a+3)<<24) | \
+        ((stbir_uint64)(4*b+0)<<32) | ((stbir_uint64)(4*b+1)<<40) | ((stbir_uint64)(4*b+2)<<48) | ((stbir_uint64)(4*b+3)<<56) )
+    #else
+      #define stbir_make8x2(reg) (uint8x8x2_t){ { vget_low_u8(vreinterpretq_u8_f32(reg)), vget_high_u8(vreinterpretq_u8_f32(reg)) } }
+      #define stbir_make8(a,b) (uint8x8_t){4*a+0,4*a+1,4*a+2,4*a+3,4*b+0,4*b+1,4*b+2,4*b+3}
+    #endif
+
+    #define stbir__simdf_swiz( reg, one, two, three, four ) vreinterpretq_f32_u8( vcombine_u8( \
+        vtbl2_u8( stbir_make8x2( reg ), stbir_make8( one, two ) ), \
+        vtbl2_u8( stbir_make8x2( reg ), stbir_make8( three, four ) ) ) )
+
+    #define stbir__simdi_16madd( out, reg0, reg1 ) \
+    { \
+      int16x8_t r0 = vreinterpretq_s16_u32(reg0); \
+      int16x8_t r1 = vreinterpretq_s16_u32(reg1); \
+      int32x4_t tmp0 = vmull_s16( vget_low_s16(r0), vget_low_s16(r1) ); \
+      int32x4_t tmp1 = vmull_s16( vget_high_s16(r0), vget_high_s16(r1) ); \
+      int32x2_t out0 = vpadd_s32( vget_low_s32(tmp0), vget_high_s32(tmp0) ); \
+      int32x2_t out1 = vpadd_s32( vget_low_s32(tmp1), vget_high_s32(tmp1) ); \
+      (out) = vreinterpretq_u32_s32( vcombine_s32(out0, out1) ); \
+    }
+
+  #endif
+
+  #define stbir__simdi_and( out, reg0, reg1 ) (out) = vandq_u32( reg0, reg1 )
+  #define stbir__simdi_or( out, reg0, reg1 ) (out) = vorrq_u32( reg0, reg1 )
+
+  #define stbir__simdf_pack_to_8bytes(out,aa,bb) \
+  { \
+    float32x4_t af = vmaxq_f32( vminq_f32(aa,STBIR__CONSTF(STBIR_max_uint8_as_float) ), vdupq_n_f32(0) ); \
+    float32x4_t bf = vmaxq_f32( vminq_f32(bb,STBIR__CONSTF(STBIR_max_uint8_as_float) ), vdupq_n_f32(0) ); \
+    int16x4_t ai = vqmovn_s32( vcvtq_s32_f32( af ) ); \
+    int16x4_t bi = vqmovn_s32( vcvtq_s32_f32( bf ) ); \
+    uint8x8_t out8 = vqmovun_s16( vcombine_s16(ai, bi) ); \
+    out = vreinterpretq_u32_u8( vcombine_u8(out8, out8) ); \
+  }
+
+  #define stbir__simdf_pack_to_8words(out,aa,bb) \
+  { \
+    float32x4_t af = vmaxq_f32( vminq_f32(aa,STBIR__CONSTF(STBIR_max_uint16_as_float) ), vdupq_n_f32(0) ); \
+    float32x4_t bf = vmaxq_f32( vminq_f32(bb,STBIR__CONSTF(STBIR_max_uint16_as_float) ), vdupq_n_f32(0) ); \
+    int32x4_t ai = vcvtq_s32_f32( af ); \
+    int32x4_t bi = vcvtq_s32_f32( bf ); \
+    out = vreinterpretq_u32_u16( vcombine_u16(vqmovun_s32(ai), vqmovun_s32(bi)) ); \
+  }
+
+  #define stbir__interleave_pack_and_store_16_u8( ptr, r0, r1, r2, r3 ) \
+  { \
+    int16x4x2_t tmp0 = vzip_s16( vqmovn_s32(vreinterpretq_s32_u32(r0)), vqmovn_s32(vreinterpretq_s32_u32(r2)) ); \
+    int16x4x2_t tmp1 = vzip_s16( vqmovn_s32(vreinterpretq_s32_u32(r1)), vqmovn_s32(vreinterpretq_s32_u32(r3)) ); \
+    uint8x8x2_t out = \
+    { { \
+      vqmovun_s16( vcombine_s16(tmp0.val[0], tmp0.val[1]) ), \
+      vqmovun_s16( vcombine_s16(tmp1.val[0], tmp1.val[1]) ), \
+    } }; \
+    vst2_u8(ptr, out); \
+  }
+
+  #define stbir__simdf_load4_transposed( o0, o1, o2, o3, ptr ) \
+  { \
+    float32x4x4_t tmp = vld4q_f32(ptr); \
+    o0 = tmp.val[0]; \
+    o1 = tmp.val[1]; \
+    o2 = tmp.val[2]; \
+    o3 = tmp.val[3]; \
+  }
+
+  #define stbir__simdi_32shr( out, reg, imm ) out = vshrq_n_u32( reg, imm )
+
+  #if defined( _MSC_VER ) && !defined(__clang__)
+    #define STBIR__SIMDF_CONST(var, x) __declspec(align(8)) float var[] = { x, x, x, x }
+    #define STBIR__SIMDI_CONST(var, x) __declspec(align(8)) uint32_t var[] = { x, x, x, x }
+    #define STBIR__CONSTF(var) (*(const float32x4_t*)var)
+    #define STBIR__CONSTI(var) (*(const uint32x4_t*)var)
+  #else
+    #define STBIR__SIMDF_CONST(var, x) stbir__simdf var = { x, x, x, x }
+    #define STBIR__SIMDI_CONST(var, x) stbir__simdi var = { x, x, x, x }
+    #define STBIR__CONSTF(var) (var)
+    #define STBIR__CONSTI(var) (var)
+  #endif
+
+  #ifdef STBIR_FLOORF
+  #undef STBIR_FLOORF
+  #endif
+  #define STBIR_FLOORF stbir_simd_floorf
+  static stbir__inline float stbir_simd_floorf(float x)
+  {
+    #if defined( _M_ARM64 ) || defined( __aarch64__ ) || defined( __arm64__ )
+    return vget_lane_f32( vrndm_f32( vdup_n_f32(x) ), 0);
+    #else
+    float32x2_t f = vdup_n_f32(x);
+    float32x2_t t = vcvt_f32_s32(vcvt_s32_f32(f));
+    uint32x2_t a = vclt_f32(f, t);
+    uint32x2_t b = vreinterpret_u32_f32(vdup_n_f32(-1.0f));
+    float32x2_t r = vadd_f32(t, vreinterpret_f32_u32(vand_u32(a, b)));
+    return vget_lane_f32(r, 0);
+    #endif
+  }
+
+  #ifdef STBIR_CEILF
+  #undef STBIR_CEILF
+  #endif
+  #define STBIR_CEILF stbir_simd_ceilf
+  static stbir__inline float stbir_simd_ceilf(float x)
+  {
+    #if defined( _M_ARM64 ) || defined( __aarch64__ ) || defined( __arm64__ )
+    return vget_lane_f32( vrndp_f32( vdup_n_f32(x) ), 0);
+    #else
+    float32x2_t f = vdup_n_f32(x);
+    float32x2_t t = vcvt_f32_s32(vcvt_s32_f32(f));
+    uint32x2_t a = vclt_f32(t, f);
+    uint32x2_t b = vreinterpret_u32_f32(vdup_n_f32(1.0f));
+    float32x2_t r = vadd_f32(t, vreinterpret_f32_u32(vand_u32(a, b)));
+    return vget_lane_f32(r, 0);
+    #endif
+  }
+
+  #define STBIR_SIMD
+
+#elif defined(STBIR_WASM)
+
+  #include <wasm_simd128.h>
+
+  #define stbir__simdf v128_t
+  #define stbir__simdi v128_t
+
+  #define stbir_simdi_castf( reg ) (reg)
+  #define stbir_simdf_casti( reg ) (reg)
+
+  #define stbir__simdf_load( reg, ptr )             (reg) = wasm_v128_load( (void const*)(ptr) )
+  #define stbir__simdi_load( reg, ptr )             (reg) = wasm_v128_load( (void const*)(ptr) )
+  #define stbir__simdf_load1( out, ptr )            (out) = wasm_v128_load32_splat( (void const*)(ptr) ) // top values can be random (not denormal or nan for perf)
+  #define stbir__simdi_load1( out, ptr )            (out) = wasm_v128_load32_splat( (void const*)(ptr) )
+  #define stbir__simdf_load1z( out, ptr )           (out) = wasm_v128_load32_zero( (void const*)(ptr) ) // top values must be zero
+  #define stbir__simdf_frep4( fvar )                wasm_f32x4_splat( fvar )
+  #define stbir__simdf_load1frep4( out, fvar )      (out) = wasm_f32x4_splat( fvar )
+  #define stbir__simdf_load2( out, ptr )            (out) = wasm_v128_load64_splat( (void const*)(ptr) ) // top values can be random (not denormal or nan for perf)
+  #define stbir__simdf_load2z( out, ptr )           (out) = wasm_v128_load64_zero( (void const*)(ptr) ) // top values must be zero
+  #define stbir__simdf_load2hmerge( out, reg, ptr ) (out) = wasm_v128_load64_lane( (void const*)(ptr), reg, 1 )
+
+  #define stbir__simdf_zeroP() wasm_f32x4_const_splat(0)
+  #define stbir__simdf_zero( reg ) (reg) = wasm_f32x4_const_splat(0)
+
+  #define stbir__simdf_store( ptr, reg )   wasm_v128_store( (void*)(ptr), reg )
+  #define stbir__simdf_store1( ptr, reg )  wasm_v128_store32_lane( (void*)(ptr), reg, 0 )
+  #define stbir__simdf_store2( ptr, reg )  wasm_v128_store64_lane( (void*)(ptr), reg, 0 )
+  #define stbir__simdf_store2h( ptr, reg ) wasm_v128_store64_lane( (void*)(ptr), reg, 1 )
+
+  #define stbir__simdi_store( ptr, reg )  wasm_v128_store( (void*)(ptr), reg )
+  #define stbir__simdi_store1( ptr, reg ) wasm_v128_store32_lane( (void*)(ptr), reg, 0 )
+  #define stbir__simdi_store2( ptr, reg ) wasm_v128_store64_lane( (void*)(ptr), reg, 0 )
+
+  #define stbir__prefetch( ptr )
+
+  #define stbir__simdi_expand_u8_to_u32(out0,out1,out2,out3,ireg) \
+  { \
+    v128_t l = wasm_u16x8_extend_low_u8x16 ( ireg ); \
+    v128_t h = wasm_u16x8_extend_high_u8x16( ireg ); \
+    out0 = wasm_u32x4_extend_low_u16x8 ( l ); \
+    out1 = wasm_u32x4_extend_high_u16x8( l ); \
+    out2 = wasm_u32x4_extend_low_u16x8 ( h ); \
+    out3 = wasm_u32x4_extend_high_u16x8( h ); \
+  }
+
+  #define stbir__simdi_expand_u8_to_1u32(out,ireg) \
+  { \
+    v128_t tmp = wasm_u16x8_extend_low_u8x16(ireg); \
+    out = wasm_u32x4_extend_low_u16x8(tmp); \
+  }
+
+  #define stbir__simdi_expand_u16_to_u32(out0,out1,ireg) \
+  { \
+    out0 = wasm_u32x4_extend_low_u16x8 ( ireg ); \
+    out1 = wasm_u32x4_extend_high_u16x8( ireg ); \
+  }
+
+  #define stbir__simdf_convert_float_to_i32( i, f )    (i) = wasm_i32x4_trunc_sat_f32x4(f)
+  #define stbir__simdf_convert_float_to_int( f )       wasm_i32x4_extract_lane(wasm_i32x4_trunc_sat_f32x4(f), 0)
+  #define stbir__simdi_to_int( i )                     wasm_i32x4_extract_lane(i, 0) 
+  #define stbir__simdf_convert_float_to_uint8( f )     ((unsigned char)wasm_i32x4_extract_lane(wasm_i32x4_trunc_sat_f32x4(wasm_f32x4_max(wasm_f32x4_min(f,STBIR_max_uint8_as_float),wasm_f32x4_const_splat(0))), 0))
+  #define stbir__simdf_convert_float_to_short( f )     ((unsigned short)wasm_i32x4_extract_lane(wasm_i32x4_trunc_sat_f32x4(wasm_f32x4_max(wasm_f32x4_min(f,STBIR_max_uint16_as_float),wasm_f32x4_const_splat(0))), 0))
+  #define stbir__simdi_convert_i32_to_float(out, ireg) (out) = wasm_f32x4_convert_i32x4(ireg)
+  #define stbir__simdf_add( out, reg0, reg1 )          (out) = wasm_f32x4_add( reg0, reg1 )
+  #define stbir__simdf_mult( out, reg0, reg1 )         (out) = wasm_f32x4_mul( reg0, reg1 )
+  #define stbir__simdf_mult_mem( out, reg, ptr )       (out) = wasm_f32x4_mul( reg, wasm_v128_load( (void const*)(ptr) ) )
+  #define stbir__simdf_mult1_mem( out, reg, ptr )      (out) = wasm_f32x4_mul( reg, wasm_v128_load32_splat( (void const*)(ptr) ) )
+  #define stbir__simdf_add_mem( out, reg, ptr )        (out) = wasm_f32x4_add( reg, wasm_v128_load( (void const*)(ptr) ) )
+  #define stbir__simdf_add1_mem( out, reg, ptr )       (out) = wasm_f32x4_add( reg, wasm_v128_load32_splat( (void const*)(ptr) ) )
+
+  #define stbir__simdf_madd( out, add, mul1, mul2 )    (out) = wasm_f32x4_add( add, wasm_f32x4_mul( mul1, mul2 ) )
+  #define stbir__simdf_madd1( out, add, mul1, mul2 )   (out) = wasm_f32x4_add( add, wasm_f32x4_mul( mul1, mul2 ) )
+  #define stbir__simdf_madd_mem( out, add, mul, ptr )  (out) = wasm_f32x4_add( add, wasm_f32x4_mul( mul, wasm_v128_load( (void const*)(ptr) ) ) )
+  #define stbir__simdf_madd1_mem( out, add, mul, ptr ) (out) = wasm_f32x4_add( add, wasm_f32x4_mul( mul, wasm_v128_load32_splat( (void const*)(ptr) ) ) )
+
+  #define stbir__simdf_add1( out, reg0, reg1 )  (out) = wasm_f32x4_add( reg0, reg1 )
+  #define stbir__simdf_mult1( out, reg0, reg1 ) (out) = wasm_f32x4_mul( reg0, reg1 )
+
+  #define stbir__simdf_and( out, reg0, reg1 ) (out) = wasm_v128_and( reg0, reg1 )
+  #define stbir__simdf_or( out, reg0, reg1 )  (out) = wasm_v128_or( reg0, reg1 )
+
+  #define stbir__simdf_min( out, reg0, reg1 ) (out) = wasm_f32x4_min( reg0, reg1 )
+  #define stbir__simdf_max( out, reg0, reg1 ) (out) = wasm_f32x4_max( reg0, reg1 )
+  #define stbir__simdf_min1( out, reg0, reg1 ) (out) = wasm_f32x4_min( reg0, reg1 )
+  #define stbir__simdf_max1( out, reg0, reg1 ) (out) = wasm_f32x4_max( reg0, reg1 )
+
+  #define stbir__simdf_0123ABCDto3ABx( out, reg0, reg1 ) (out) = wasm_i32x4_shuffle( reg0, reg1, 3, 4, 5, -1 )
+  #define stbir__simdf_0123ABCDto23Ax( out, reg0, reg1 ) (out) = wasm_i32x4_shuffle( reg0, reg1, 2, 3, 4, -1 )
+
+  #define stbir__simdf_aaa1(out,alp,ones) (out) = wasm_i32x4_shuffle(alp, ones, 3, 3, 3, 4)
+  #define stbir__simdf_1aaa(out,alp,ones) (out) = wasm_i32x4_shuffle(alp, ones, 4, 0, 0, 0)
+  #define stbir__simdf_a1a1(out,alp,ones) (out) = wasm_i32x4_shuffle(alp, ones, 1, 4, 3, 4)
+  #define stbir__simdf_1a1a(out,alp,ones) (out) = wasm_i32x4_shuffle(alp, ones, 4, 0, 4, 2)
+
+  #define stbir__simdf_swiz( reg, one, two, three, four ) wasm_i32x4_shuffle(reg, reg, one, two, three, four)
+
+  #define stbir__simdi_and( out, reg0, reg1 )    (out) = wasm_v128_and( reg0, reg1 )
+  #define stbir__simdi_or( out, reg0, reg1 )     (out) = wasm_v128_or( reg0, reg1 )
+  #define stbir__simdi_16madd( out, reg0, reg1 ) (out) = wasm_i32x4_dot_i16x8( reg0, reg1 )
+
+  #define stbir__simdf_pack_to_8bytes(out,aa,bb) \
+  { \
+    v128_t af = wasm_f32x4_max( wasm_f32x4_min(aa, STBIR_max_uint8_as_float), wasm_f32x4_const_splat(0) ); \
+    v128_t bf = wasm_f32x4_max( wasm_f32x4_min(bb, STBIR_max_uint8_as_float), wasm_f32x4_const_splat(0) ); \
+    v128_t ai = wasm_i32x4_trunc_sat_f32x4( af ); \
+    v128_t bi = wasm_i32x4_trunc_sat_f32x4( bf ); \
+    v128_t out16 = wasm_i16x8_narrow_i32x4( ai, bi ); \
+    out = wasm_u8x16_narrow_i16x8( out16, out16 ); \
+  }
+
+  #define stbir__simdf_pack_to_8words(out,aa,bb) \
+  { \
+    v128_t af = wasm_f32x4_max( wasm_f32x4_min(aa, STBIR_max_uint16_as_float), wasm_f32x4_const_splat(0)); \
+    v128_t bf = wasm_f32x4_max( wasm_f32x4_min(bb, STBIR_max_uint16_as_float), wasm_f32x4_const_splat(0)); \
+    v128_t ai = wasm_i32x4_trunc_sat_f32x4( af ); \
+    v128_t bi = wasm_i32x4_trunc_sat_f32x4( bf ); \
+    out = wasm_u16x8_narrow_i32x4( ai, bi ); \
+  }
+
+  #define stbir__interleave_pack_and_store_16_u8( ptr, r0, r1, r2, r3 ) \
+  { \
+    v128_t tmp0 = wasm_i16x8_narrow_i32x4(r0, r1); \
+    v128_t tmp1 = wasm_i16x8_narrow_i32x4(r2, r3); \
+    v128_t tmp = wasm_u8x16_narrow_i16x8(tmp0, tmp1); \
+    tmp = wasm_i8x16_shuffle(tmp, tmp, 0, 4, 8, 12, 1, 5, 9, 13, 2, 6, 10, 14, 3, 7, 11, 15); \
+    wasm_v128_store( (void*)(ptr), tmp); \
+  }
+
+  #define stbir__simdf_load4_transposed( o0, o1, o2, o3, ptr ) \
+  { \
+    v128_t t0 = wasm_v128_load( ptr    ); \
+    v128_t t1 = wasm_v128_load( ptr+4  ); \
+    v128_t t2 = wasm_v128_load( ptr+8  ); \
+    v128_t t3 = wasm_v128_load( ptr+12 ); \
+    v128_t s0 = wasm_i32x4_shuffle(t0, t1, 0, 4, 2, 6); \
+    v128_t s1 = wasm_i32x4_shuffle(t0, t1, 1, 5, 3, 7); \
+    v128_t s2 = wasm_i32x4_shuffle(t2, t3, 0, 4, 2, 6); \
+    v128_t s3 = wasm_i32x4_shuffle(t2, t3, 1, 5, 3, 7); \
+    o0 = wasm_i32x4_shuffle(s0, s2, 0, 1, 4, 5); \
+    o1 = wasm_i32x4_shuffle(s1, s3, 0, 1, 4, 5); \
+    o2 = wasm_i32x4_shuffle(s0, s2, 2, 3, 6, 7); \
+    o3 = wasm_i32x4_shuffle(s1, s3, 2, 3, 6, 7); \
+  }
+
+  #define stbir__simdi_32shr( out, reg, imm ) out = wasm_u32x4_shr( reg, imm )
+
+  typedef float stbir__f32x4 __attribute__((__vector_size__(16), __aligned__(16)));
+  #define STBIR__SIMDF_CONST(var, x) stbir__simdf var = (v128_t)(stbir__f32x4){ x, x, x, x }
+  #define STBIR__SIMDI_CONST(var, x) stbir__simdi var = { x, x, x, x }
+  #define STBIR__CONSTF(var) (var)
+  #define STBIR__CONSTI(var) (var)
+
+  #ifdef STBIR_FLOORF
+  #undef STBIR_FLOORF
+  #endif
+  #define STBIR_FLOORF stbir_simd_floorf
+  static stbir__inline float stbir_simd_floorf(float x)
+  {
+    return wasm_f32x4_extract_lane( wasm_f32x4_floor( wasm_f32x4_splat(x) ), 0);
+  }
+
+  #ifdef STBIR_CEILF
+  #undef STBIR_CEILF
+  #endif
+  #define STBIR_CEILF stbir_simd_ceilf
+  static stbir__inline float stbir_simd_ceilf(float x)
+  {
+    return wasm_f32x4_extract_lane( wasm_f32x4_ceil( wasm_f32x4_splat(x) ), 0);
+  }
+
+  #define STBIR_SIMD
+
+#endif  // SSE2/NEON/WASM
+
+#endif // NO SIMD
+
+#ifdef STBIR_SIMD8
+  #define stbir__simdfX stbir__simdf8
+  #define stbir__simdiX stbir__simdi8
+  #define stbir__simdfX_load stbir__simdf8_load
+  #define stbir__simdiX_load stbir__simdi8_load
+  #define stbir__simdfX_mult stbir__simdf8_mult
+  #define stbir__simdfX_add_mem stbir__simdf8_add_mem
+  #define stbir__simdfX_madd_mem stbir__simdf8_madd_mem
+  #define stbir__simdfX_store stbir__simdf8_store
+  #define stbir__simdiX_store stbir__simdi8_store
+  #define stbir__simdf_frepX  stbir__simdf8_frep8
+  #define stbir__simdfX_madd stbir__simdf8_madd
+  #define stbir__simdfX_min stbir__simdf8_min
+  #define stbir__simdfX_max stbir__simdf8_max
+  #define stbir__simdfX_aaa1 stbir__simdf8_aaa1
+  #define stbir__simdfX_1aaa stbir__simdf8_1aaa
+  #define stbir__simdfX_a1a1 stbir__simdf8_a1a1
+  #define stbir__simdfX_1a1a stbir__simdf8_1a1a
+  #define stbir__simdfX_convert_float_to_i32 stbir__simdf8_convert_float_to_i32
+  #define stbir__simdfX_pack_to_words stbir__simdf8_pack_to_16words
+  #define stbir__simdfX_zero stbir__simdf8_zero
+  #define STBIR_onesX STBIR_ones8
+  #define STBIR_max_uint8_as_floatX STBIR_max_uint8_as_float8
+  #define STBIR_max_uint16_as_floatX STBIR_max_uint16_as_float8
+  #define STBIR_simd_point5X STBIR_simd_point58
+  #define stbir__simdfX_float_count 8
+  #define stbir__simdfX_0123to1230 stbir__simdf8_0123to12301230
+  #define stbir__simdfX_0123to2103 stbir__simdf8_0123to21032103
+  static const stbir__simdf8 STBIR_max_uint16_as_float_inverted8 = { stbir__max_uint16_as_float_inverted,stbir__max_uint16_as_float_inverted,stbir__max_uint16_as_float_inverted,stbir__max_uint16_as_float_inverted,stbir__max_uint16_as_float_inverted,stbir__max_uint16_as_float_inverted,stbir__max_uint16_as_float_inverted,stbir__max_uint16_as_float_inverted };
+  static const stbir__simdf8 STBIR_max_uint8_as_float_inverted8 = { stbir__max_uint8_as_float_inverted,stbir__max_uint8_as_float_inverted,stbir__max_uint8_as_float_inverted,stbir__max_uint8_as_float_inverted,stbir__max_uint8_as_float_inverted,stbir__max_uint8_as_float_inverted,stbir__max_uint8_as_float_inverted,stbir__max_uint8_as_float_inverted };
+  static const stbir__simdf8 STBIR_ones8 = { 1.0,1.0,1.0,1.0,1.0,1.0,1.0,1.0 };
+  static const stbir__simdf8 STBIR_simd_point58 = { 0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5 };
+  static const stbir__simdf8 STBIR_max_uint8_as_float8 = { stbir__max_uint8_as_float,stbir__max_uint8_as_float,stbir__max_uint8_as_float,stbir__max_uint8_as_float, stbir__max_uint8_as_float,stbir__max_uint8_as_float,stbir__max_uint8_as_float,stbir__max_uint8_as_float };
+  static const stbir__simdf8 STBIR_max_uint16_as_float8 = { stbir__max_uint16_as_float,stbir__max_uint16_as_float,stbir__max_uint16_as_float,stbir__max_uint16_as_float, stbir__max_uint16_as_float,stbir__max_uint16_as_float,stbir__max_uint16_as_float,stbir__max_uint16_as_float };
+#else
+  #define stbir__simdfX stbir__simdf
+  #define stbir__simdiX stbir__simdi
+  #define stbir__simdfX_load stbir__simdf_load
+  #define stbir__simdiX_load stbir__simdi_load
+  #define stbir__simdfX_mult stbir__simdf_mult
+  #define stbir__simdfX_add_mem stbir__simdf_add_mem
+  #define stbir__simdfX_madd_mem stbir__simdf_madd_mem
+  #define stbir__simdfX_store stbir__simdf_store
+  #define stbir__simdiX_store stbir__simdi_store
+  #define stbir__simdf_frepX  stbir__simdf_frep4
+  #define stbir__simdfX_madd stbir__simdf_madd
+  #define stbir__simdfX_min stbir__simdf_min
+  #define stbir__simdfX_max stbir__simdf_max
+  #define stbir__simdfX_aaa1 stbir__simdf_aaa1
+  #define stbir__simdfX_1aaa stbir__simdf_1aaa
+  #define stbir__simdfX_a1a1 stbir__simdf_a1a1
+  #define stbir__simdfX_1a1a stbir__simdf_1a1a
+  #define stbir__simdfX_convert_float_to_i32 stbir__simdf_convert_float_to_i32
+  #define stbir__simdfX_pack_to_words stbir__simdf_pack_to_8words
+  #define stbir__simdfX_zero stbir__simdf_zero
+  #define STBIR_onesX STBIR__CONSTF(STBIR_ones)
+  #define STBIR_simd_point5X STBIR__CONSTF(STBIR_simd_point5)
+  #define STBIR_max_uint8_as_floatX STBIR__CONSTF(STBIR_max_uint8_as_float)
+  #define STBIR_max_uint16_as_floatX STBIR__CONSTF(STBIR_max_uint16_as_float)
+  #define stbir__simdfX_float_count 4
+  #define stbir__if_simdf8_cast_to_simdf4( val ) ( val )
+  #define stbir__simdfX_0123to1230 stbir__simdf_0123to1230
+  #define stbir__simdfX_0123to2103 stbir__simdf_0123to2103
+#endif
+
+
+#if defined(STBIR_NEON) && !defined(_M_ARM)
+
+  #if defined( _MSC_VER ) && !defined(__clang__)
+  typedef __int16 stbir__FP16;
+  #else
+  typedef float16_t stbir__FP16;
+  #endif
+
+#else // no NEON, or 32-bit ARM for MSVC
+
+  typedef union stbir__FP16
+  {
+    unsigned short u;
+  } stbir__FP16;
+
+#endif
+
+#if !defined(STBIR_NEON) && !defined(STBIR_FP16C) || defined(STBIR_NEON) && defined(_M_ARM)
+
+  // Fabian's half float routines, see: https://gist.github.com/rygorous/2156668
+
+  static stbir__inline float stbir__half_to_float( stbir__FP16 h )
+  {
+    static const stbir__FP32 magic = { (254 - 15) << 23 };
+    static const stbir__FP32 was_infnan = { (127 + 16) << 23 };
+    stbir__FP32 o;
+
+    o.u = (h.u & 0x7fff) << 13;     // exponent/mantissa bits
+    o.f *= magic.f;                 // exponent adjust
+    if (o.f >= was_infnan.f)        // make sure Inf/NaN survive
+      o.u |= 255 << 23;
+    o.u |= (h.u & 0x8000) << 16;    // sign bit
+    return o.f;
+  }
+
+  static stbir__inline stbir__FP16 stbir__float_to_half(float val)
+  {
+    stbir__FP32 f32infty = { 255 << 23 };
+    stbir__FP32 f16max   = { (127 + 16) << 23 };
+    stbir__FP32 denorm_magic = { ((127 - 15) + (23 - 10) + 1) << 23 };
+    unsigned int sign_mask = 0x80000000u;
+    stbir__FP16 o = { 0 };
+    stbir__FP32 f;
+    unsigned int sign; 
+
+    f.f = val;
+    sign = f.u & sign_mask;
+    f.u ^= sign;
+
+    if (f.u >= f16max.u) // result is Inf or NaN (all exponent bits set)
+      o.u = (f.u > f32infty.u) ? 0x7e00 : 0x7c00; // NaN->qNaN and Inf->Inf
+    else // (De)normalized number or zero
+    {
+      if (f.u < (113 << 23)) // resulting FP16 is subnormal or zero
+      {
+        // use a magic value to align our 10 mantissa bits at the bottom of
+        // the float. as long as FP addition is round-to-nearest-even this
+        // just works.
+        f.f += denorm_magic.f;
+        // and one integer subtract of the bias later, we have our final float!
+        o.u = (unsigned short) ( f.u - denorm_magic.u );
+      }
+      else
+      {
+        unsigned int mant_odd = (f.u >> 13) & 1; // resulting mantissa is odd
+        // update exponent, rounding bias part 1
+        f.u = f.u + ((15u - 127) << 23) + 0xfff;
+        // rounding bias part 2
+        f.u += mant_odd;
+        // take the bits!
+        o.u = (unsigned short) ( f.u >> 13 );
+      }
+    }
+
+    o.u |= sign >> 16;
+    return o;
+  }
+
+#endif
+
+
+#if defined(STBIR_FP16C)
+
+  #include <immintrin.h>
+
+  static stbir__inline void stbir__half_to_float_SIMD(float * output, stbir__FP16 const * input)
+  {
+    _mm256_storeu_ps( (float*)output, _mm256_cvtph_ps( _mm_loadu_si128( (__m128i const* )input ) ) );
+  }
+
+  static stbir__inline void stbir__float_to_half_SIMD(stbir__FP16 * output, float const * input)
+  {
+    _mm_storeu_si128( (__m128i*)output, _mm256_cvtps_ph( _mm256_loadu_ps( input ), 0 ) );
+  }
+
+  static stbir__inline float stbir__half_to_float( stbir__FP16 h )
+  {
+    return _mm_cvtss_f32( _mm_cvtph_ps( _mm_cvtsi32_si128( (int)h.u ) ) );
+  }
+
+  static stbir__inline stbir__FP16 stbir__float_to_half( float f )
+  {
+    stbir__FP16 h;
+    h.u = (unsigned short) _mm_cvtsi128_si32( _mm_cvtps_ph( _mm_set_ss( f ), 0 ) );
+    return h;
+  }
+
+#elif defined(STBIR_SSE2)
+
+  // Fabian's half float routines, see: https://gist.github.com/rygorous/2156668
+  stbir__inline static void stbir__half_to_float_SIMD(float * output, void const * input)
+  {
+    static const STBIR__SIMDI_CONST(mask_nosign,      0x7fff);
+    static const STBIR__SIMDI_CONST(smallest_normal,  0x0400);
+    static const STBIR__SIMDI_CONST(infinity,         0x7c00);
+    static const STBIR__SIMDI_CONST(expadjust_normal, (127 - 15) << 23);
+    static const STBIR__SIMDI_CONST(magic_denorm,     113 << 23);
+
+    __m128i i = _mm_loadu_si128 ( (__m128i const*)(input) );
+    __m128i h = _mm_unpacklo_epi16 ( i, _mm_setzero_si128() );
+    __m128i mnosign     = STBIR__CONSTI(mask_nosign);
+    __m128i eadjust     = STBIR__CONSTI(expadjust_normal);
+    __m128i smallest    = STBIR__CONSTI(smallest_normal);
+    __m128i infty       = STBIR__CONSTI(infinity);
+    __m128i expmant     = _mm_and_si128(mnosign, h);
+    __m128i justsign    = _mm_xor_si128(h, expmant);
+    __m128i b_notinfnan = _mm_cmpgt_epi32(infty, expmant);
+    __m128i b_isdenorm  = _mm_cmpgt_epi32(smallest, expmant);
+    __m128i shifted     = _mm_slli_epi32(expmant, 13);
+    __m128i adj_infnan  = _mm_andnot_si128(b_notinfnan, eadjust);
+    __m128i adjusted    = _mm_add_epi32(eadjust, shifted);
+    __m128i den1        = _mm_add_epi32(shifted, STBIR__CONSTI(magic_denorm));
+    __m128i adjusted2   = _mm_add_epi32(adjusted, adj_infnan);
+    __m128  den2        = _mm_sub_ps(_mm_castsi128_ps(den1), *(const __m128 *)&magic_denorm);
+    __m128  adjusted3   = _mm_and_ps(den2, _mm_castsi128_ps(b_isdenorm));
+    __m128  adjusted4   = _mm_andnot_ps(_mm_castsi128_ps(b_isdenorm), _mm_castsi128_ps(adjusted2));
+    __m128  adjusted5   = _mm_or_ps(adjusted3, adjusted4);
+    __m128i sign        = _mm_slli_epi32(justsign, 16);
+    __m128  final       = _mm_or_ps(adjusted5, _mm_castsi128_ps(sign));
+    stbir__simdf_store( output + 0,  final );
+
+    h = _mm_unpackhi_epi16 ( i, _mm_setzero_si128() );
+    expmant     = _mm_and_si128(mnosign, h);
+    justsign    = _mm_xor_si128(h, expmant);
+    b_notinfnan = _mm_cmpgt_epi32(infty, expmant);
+    b_isdenorm  = _mm_cmpgt_epi32(smallest, expmant);
+    shifted     = _mm_slli_epi32(expmant, 13);
+    adj_infnan  = _mm_andnot_si128(b_notinfnan, eadjust);
+    adjusted    = _mm_add_epi32(eadjust, shifted);
+    den1        = _mm_add_epi32(shifted, STBIR__CONSTI(magic_denorm));
+    adjusted2   = _mm_add_epi32(adjusted, adj_infnan);
+    den2        = _mm_sub_ps(_mm_castsi128_ps(den1), *(const __m128 *)&magic_denorm);
+    adjusted3   = _mm_and_ps(den2, _mm_castsi128_ps(b_isdenorm));
+    adjusted4   = _mm_andnot_ps(_mm_castsi128_ps(b_isdenorm), _mm_castsi128_ps(adjusted2));
+    adjusted5   = _mm_or_ps(adjusted3, adjusted4);
+    sign        = _mm_slli_epi32(justsign, 16);
+    final       = _mm_or_ps(adjusted5, _mm_castsi128_ps(sign));
+    stbir__simdf_store( output + 4,  final );
+
+    // ~38 SSE2 ops for 8 values
+  }
+
+  // Fabian's round-to-nearest-even float to half
+  // ~48 SSE2 ops for 8 output
+  stbir__inline static void stbir__float_to_half_SIMD(void * output, float const * input)
+  {
+    static const STBIR__SIMDI_CONST(mask_sign,      0x80000000u);
+    static const STBIR__SIMDI_CONST(c_f16max,       (127 + 16) << 23); // all FP32 values >=this round to +inf
+    static const STBIR__SIMDI_CONST(c_nanbit,        0x200);
+    static const STBIR__SIMDI_CONST(c_infty_as_fp16, 0x7c00);
+    static const STBIR__SIMDI_CONST(c_min_normal,    (127 - 14) << 23); // smallest FP32 that yields a normalized FP16
+    static const STBIR__SIMDI_CONST(c_subnorm_magic, ((127 - 15) + (23 - 10) + 1) << 23);
+    static const STBIR__SIMDI_CONST(c_normal_bias,    0xfff - ((127 - 15) << 23)); // adjust exponent and add mantissa rounding
+
+    __m128  f           =  _mm_loadu_ps(input);
+    __m128  msign       = _mm_castsi128_ps(STBIR__CONSTI(mask_sign));
+    __m128  justsign    = _mm_and_ps(msign, f);
+    __m128  absf        = _mm_xor_ps(f, justsign);
+    __m128i absf_int    = _mm_castps_si128(absf); // the cast is "free" (extra bypass latency, but no thruput hit)
+    __m128i f16max      = STBIR__CONSTI(c_f16max);
+    __m128  b_isnan     = _mm_cmpunord_ps(absf, absf); // is this a NaN?
+    __m128i b_isregular = _mm_cmpgt_epi32(f16max, absf_int); // (sub)normalized or special?
+    __m128i nanbit      = _mm_and_si128(_mm_castps_si128(b_isnan), STBIR__CONSTI(c_nanbit));
+    __m128i inf_or_nan  = _mm_or_si128(nanbit, STBIR__CONSTI(c_infty_as_fp16)); // output for specials
+
+    __m128i min_normal  = STBIR__CONSTI(c_min_normal);
+    __m128i b_issub     = _mm_cmpgt_epi32(min_normal, absf_int);
+
+    // "result is subnormal" path
+    __m128  subnorm1    = _mm_add_ps(absf, _mm_castsi128_ps(STBIR__CONSTI(c_subnorm_magic))); // magic value to round output mantissa
+    __m128i subnorm2    = _mm_sub_epi32(_mm_castps_si128(subnorm1), STBIR__CONSTI(c_subnorm_magic)); // subtract out bias
+
+    // "result is normal" path
+    __m128i mantoddbit  = _mm_slli_epi32(absf_int, 31 - 13); // shift bit 13 (mantissa LSB) to sign
+    __m128i mantodd     = _mm_srai_epi32(mantoddbit, 31); // -1 if FP16 mantissa odd, else 0
+
+    __m128i round1      = _mm_add_epi32(absf_int, STBIR__CONSTI(c_normal_bias));
+    __m128i round2      = _mm_sub_epi32(round1, mantodd); // if mantissa LSB odd, bias towards rounding up (RTNE)
+    __m128i normal      = _mm_srli_epi32(round2, 13); // rounded result
+
+    // combine the two non-specials
+    __m128i nonspecial  = _mm_or_si128(_mm_and_si128(subnorm2, b_issub), _mm_andnot_si128(b_issub, normal));
+
+    // merge in specials as well
+    __m128i joined      = _mm_or_si128(_mm_and_si128(nonspecial, b_isregular), _mm_andnot_si128(b_isregular, inf_or_nan));
+
+    __m128i sign_shift  = _mm_srai_epi32(_mm_castps_si128(justsign), 16);
+    __m128i final2, final= _mm_or_si128(joined, sign_shift);
+
+    f           =  _mm_loadu_ps(input+4);
+    justsign    = _mm_and_ps(msign, f);
+    absf        = _mm_xor_ps(f, justsign);
+    absf_int    = _mm_castps_si128(absf); // the cast is "free" (extra bypass latency, but no thruput hit)
+    b_isnan     = _mm_cmpunord_ps(absf, absf); // is this a NaN?
+    b_isregular = _mm_cmpgt_epi32(f16max, absf_int); // (sub)normalized or special?
+    nanbit      = _mm_and_si128(_mm_castps_si128(b_isnan), c_nanbit);
+    inf_or_nan  = _mm_or_si128(nanbit, STBIR__CONSTI(c_infty_as_fp16)); // output for specials
+
+    b_issub     = _mm_cmpgt_epi32(min_normal, absf_int);
+
+    // "result is subnormal" path
+    subnorm1    = _mm_add_ps(absf, _mm_castsi128_ps(STBIR__CONSTI(c_subnorm_magic))); // magic value to round output mantissa
+    subnorm2    = _mm_sub_epi32(_mm_castps_si128(subnorm1), STBIR__CONSTI(c_subnorm_magic)); // subtract out bias
+
+    // "result is normal" path
+    mantoddbit  = _mm_slli_epi32(absf_int, 31 - 13); // shift bit 13 (mantissa LSB) to sign
+    mantodd     = _mm_srai_epi32(mantoddbit, 31); // -1 if FP16 mantissa odd, else 0
+
+    round1      = _mm_add_epi32(absf_int, STBIR__CONSTI(c_normal_bias));
+    round2      = _mm_sub_epi32(round1, mantodd); // if mantissa LSB odd, bias towards rounding up (RTNE)
+    normal      = _mm_srli_epi32(round2, 13); // rounded result
+
+    // combine the two non-specials
+    nonspecial  = _mm_or_si128(_mm_and_si128(subnorm2, b_issub), _mm_andnot_si128(b_issub, normal));
+
+    // merge in specials as well
+    joined      = _mm_or_si128(_mm_and_si128(nonspecial, b_isregular), _mm_andnot_si128(b_isregular, inf_or_nan));
+
+    sign_shift  = _mm_srai_epi32(_mm_castps_si128(justsign), 16);
+    final2      = _mm_or_si128(joined, sign_shift);
+    final       = _mm_packs_epi32(final, final2);
+    stbir__simdi_store( output,final );
+  }
+
+#elif defined(STBIR_WASM) || (defined(STBIR_NEON) && defined(_MSC_VER) && defined(_M_ARM)) // WASM or 32-bit ARM on MSVC/clang
+
+  static stbir__inline void stbir__half_to_float_SIMD(float * output, stbir__FP16 const * input)
+  {
+    for (int i=0; i<8; i++)
+    {
+      output[i] = stbir__half_to_float(input[i]);
+    }
+  }
+
+  static stbir__inline void stbir__float_to_half_SIMD(stbir__FP16 * output, float const * input)
+  {
+    for (int i=0; i<8; i++)
+    {
+      output[i] = stbir__float_to_half(input[i]);
+    }
+  }
+
+#elif defined(STBIR_NEON) && defined(_MSC_VER) && defined(_M_ARM64) && !defined(__clang__) // 64-bit ARM on MSVC (not clang)
+
+  static stbir__inline void stbir__half_to_float_SIMD(float * output, stbir__FP16 const * input)
+  {
+    float16x4_t in0 = vld1_f16(input + 0);
+    float16x4_t in1 = vld1_f16(input + 4);
+    vst1q_f32(output + 0, vcvt_f32_f16(in0));
+    vst1q_f32(output + 4, vcvt_f32_f16(in1));
+  }
+
+  static stbir__inline void stbir__float_to_half_SIMD(stbir__FP16 * output, float const * input)
+  {
+    float16x4_t out0 = vcvt_f16_f32(vld1q_f32(input + 0));
+    float16x4_t out1 = vcvt_f16_f32(vld1q_f32(input + 4));
+    vst1_f16(output+0, out0);
+    vst1_f16(output+4, out1);
+  }
+
+  static stbir__inline float stbir__half_to_float( stbir__FP16 h )
+  {
+    return vgetq_lane_f32(vcvt_f32_f16(vld1_dup_f16(&h)), 0);
+  }
+
+  static stbir__inline stbir__FP16 stbir__float_to_half( float f )
+  {
+    return vget_lane_f16(vcvt_f16_f32(vdupq_n_f32(f)), 0).n16_u16[0];
+  }
+
+#elif defined(STBIR_NEON) // 64-bit ARM
+
+  static stbir__inline void stbir__half_to_float_SIMD(float * output, stbir__FP16 const * input)
+  {
+    float16x8_t in = vld1q_f16(input);
+    vst1q_f32(output + 0, vcvt_f32_f16(vget_low_f16(in)));
+    vst1q_f32(output + 4, vcvt_f32_f16(vget_high_f16(in)));
+  }
+
+  static stbir__inline void stbir__float_to_half_SIMD(stbir__FP16 * output, float const * input)
+  {
+    float16x4_t out0 = vcvt_f16_f32(vld1q_f32(input + 0));
+    float16x4_t out1 = vcvt_f16_f32(vld1q_f32(input + 4));
+    vst1q_f16(output, vcombine_f16(out0, out1));
+  }
+
+  static stbir__inline float stbir__half_to_float( stbir__FP16 h )
+  {
+    return vgetq_lane_f32(vcvt_f32_f16(vdup_n_f16(h)), 0);
+  }
+
+  static stbir__inline stbir__FP16 stbir__float_to_half( float f )
+  {
+    return vget_lane_f16(vcvt_f16_f32(vdupq_n_f32(f)), 0);
+  }
+
+#endif
+
+
+#ifdef STBIR_SIMD
+
+#define stbir__simdf_0123to3333( out, reg ) (out) = stbir__simdf_swiz( reg, 3,3,3,3 )
+#define stbir__simdf_0123to2222( out, reg ) (out) = stbir__simdf_swiz( reg, 2,2,2,2 )
+#define stbir__simdf_0123to1111( out, reg ) (out) = stbir__simdf_swiz( reg, 1,1,1,1 )
+#define stbir__simdf_0123to0000( out, reg ) (out) = stbir__simdf_swiz( reg, 0,0,0,0 )
+#define stbir__simdf_0123to0003( out, reg ) (out) = stbir__simdf_swiz( reg, 0,0,0,3 )
+#define stbir__simdf_0123to0001( out, reg ) (out) = stbir__simdf_swiz( reg, 0,0,0,1 )
+#define stbir__simdf_0123to1122( out, reg ) (out) = stbir__simdf_swiz( reg, 1,1,2,2 )
+#define stbir__simdf_0123to2333( out, reg ) (out) = stbir__simdf_swiz( reg, 2,3,3,3 )
+#define stbir__simdf_0123to0023( out, reg ) (out) = stbir__simdf_swiz( reg, 0,0,2,3 )
+#define stbir__simdf_0123to1230( out, reg ) (out) = stbir__simdf_swiz( reg, 1,2,3,0 )
+#define stbir__simdf_0123to2103( out, reg ) (out) = stbir__simdf_swiz( reg, 2,1,0,3 )
+#define stbir__simdf_0123to3210( out, reg ) (out) = stbir__simdf_swiz( reg, 3,2,1,0 )
+#define stbir__simdf_0123to2301( out, reg ) (out) = stbir__simdf_swiz( reg, 2,3,0,1 )
+#define stbir__simdf_0123to3012( out, reg ) (out) = stbir__simdf_swiz( reg, 3,0,1,2 )
+#define stbir__simdf_0123to0011( out, reg ) (out) = stbir__simdf_swiz( reg, 0,0,1,1 )
+#define stbir__simdf_0123to1100( out, reg ) (out) = stbir__simdf_swiz( reg, 1,1,0,0 )
+#define stbir__simdf_0123to2233( out, reg ) (out) = stbir__simdf_swiz( reg, 2,2,3,3 ) 
+#define stbir__simdf_0123to1133( out, reg ) (out) = stbir__simdf_swiz( reg, 1,1,3,3 ) 
+#define stbir__simdf_0123to0022( out, reg ) (out) = stbir__simdf_swiz( reg, 0,0,2,2 ) 
+#define stbir__simdf_0123to1032( out, reg ) (out) = stbir__simdf_swiz( reg, 1,0,3,2 ) 
+
+typedef union stbir__simdi_u32
+{
+  stbir_uint32 m128i_u32[4];
+  int m128i_i32[4];
+  stbir__simdi m128i_i128;
+} stbir__simdi_u32;
+
+static const int STBIR_mask[9] = { 0,0,0,-1,-1,-1,0,0,0 };
+
+static const STBIR__SIMDF_CONST(STBIR_max_uint8_as_float,           stbir__max_uint8_as_float);
+static const STBIR__SIMDF_CONST(STBIR_max_uint16_as_float,          stbir__max_uint16_as_float);
+static const STBIR__SIMDF_CONST(STBIR_max_uint8_as_float_inverted,  stbir__max_uint8_as_float_inverted);
+static const STBIR__SIMDF_CONST(STBIR_max_uint16_as_float_inverted, stbir__max_uint16_as_float_inverted);
+
+static const STBIR__SIMDF_CONST(STBIR_simd_point5,   0.5f);
+static const STBIR__SIMDF_CONST(STBIR_ones,          1.0f);
+static const STBIR__SIMDI_CONST(STBIR_almost_zero,   (127 - 13) << 23);
+static const STBIR__SIMDI_CONST(STBIR_almost_one,    0x3f7fffff);
+static const STBIR__SIMDI_CONST(STBIR_mastissa_mask, 0xff);
+static const STBIR__SIMDI_CONST(STBIR_topscale,      0x02000000);
+
+//   Basically, in simd mode, we unroll the proper amount, and we don't want
+//   the non-simd remnant loops to be unroll because they only run a few times
+//   Adding this switch saves about 5K on clang which is Captain Unroll the 3rd.
+#define STBIR_SIMD_STREAMOUT_PTR( star )  STBIR_STREAMOUT_PTR( star )
+#define STBIR_SIMD_NO_UNROLL(ptr) STBIR_NO_UNROLL(ptr)
+
+#ifdef STBIR_MEMCPY
+#undef STBIR_MEMCPY
+#define STBIR_MEMCPY stbir_simd_memcpy
+#endif
+
+// override normal use of memcpy with much simpler copy (faster and smaller with our sized copies)
+static void stbir_simd_memcpy( void * dest, void const * src, size_t bytes ) 
+{
+  char STBIR_SIMD_STREAMOUT_PTR (*) d = (char*) dest;
+  char STBIR_SIMD_STREAMOUT_PTR( * ) d_end = ((char*) dest) + bytes;
+  ptrdiff_t ofs_to_src = (char*)src - (char*)dest;
+
+  // check overlaps
+  STBIR_ASSERT( ( ( d >= ( (char*)src) + bytes ) ) || ( ( d + bytes ) <= (char*)src ) );
+
+  if ( bytes < (16*stbir__simdfX_float_count) )
+  {
+    if ( bytes < 16 )
+    {
+      if ( bytes ) 
+      {
+        do
+        {
+          STBIR_SIMD_NO_UNROLL(d);
+          d[ 0 ] = d[ ofs_to_src ];
+          ++d;
+        } while ( d < d_end );
+      }
+    }
+    else
+    {
+      stbir__simdf x;
+      // do one unaligned to get us aligned for the stream out below
+      stbir__simdf_load( x, ( d + ofs_to_src ) );
+      stbir__simdf_store( d, x );
+      d = (char*)( ( ( (ptrdiff_t)d ) + 16 ) & ~15 );
+
+      for(;;)
+      {
+        STBIR_SIMD_NO_UNROLL(d);
+
+        if ( d > ( d_end - 16 ) )
+        {
+          if ( d == d_end )
+            return;
+          d = d_end - 16;
+        }
+
+        stbir__simdf_load( x, ( d + ofs_to_src ) );
+        stbir__simdf_store( d, x );
+        d += 16;
+      }
+    }
+  }
+  else
+  {
+    stbir__simdfX x0,x1,x2,x3;
+
+    // do one unaligned to get us aligned for the stream out below
+    stbir__simdfX_load( x0, ( d + ofs_to_src ) +  0*stbir__simdfX_float_count );
+    stbir__simdfX_load( x1, ( d + ofs_to_src ) +  4*stbir__simdfX_float_count );
+    stbir__simdfX_load( x2, ( d + ofs_to_src ) +  8*stbir__simdfX_float_count );
+    stbir__simdfX_load( x3, ( d + ofs_to_src ) + 12*stbir__simdfX_float_count );
+    stbir__simdfX_store( d +  0*stbir__simdfX_float_count, x0 );
+    stbir__simdfX_store( d +  4*stbir__simdfX_float_count, x1 );
+    stbir__simdfX_store( d +  8*stbir__simdfX_float_count, x2 );
+    stbir__simdfX_store( d + 12*stbir__simdfX_float_count, x3 );
+    d = (char*)( ( ( (ptrdiff_t)d ) + (16*stbir__simdfX_float_count) ) & ~((16*stbir__simdfX_float_count)-1) );
+
+    for(;;)
+    {
+      STBIR_SIMD_NO_UNROLL(d);
+  
+      if ( d > ( d_end - (16*stbir__simdfX_float_count) ) )
+      {
+        if ( d == d_end )
+          return;
+        d = d_end - (16*stbir__simdfX_float_count);
+      }
+
+      stbir__simdfX_load( x0, ( d + ofs_to_src ) +  0*stbir__simdfX_float_count );
+      stbir__simdfX_load( x1, ( d + ofs_to_src ) +  4*stbir__simdfX_float_count );
+      stbir__simdfX_load( x2, ( d + ofs_to_src ) +  8*stbir__simdfX_float_count );
+      stbir__simdfX_load( x3, ( d + ofs_to_src ) + 12*stbir__simdfX_float_count );
+      stbir__simdfX_store( d +  0*stbir__simdfX_float_count, x0 );
+      stbir__simdfX_store( d +  4*stbir__simdfX_float_count, x1 );
+      stbir__simdfX_store( d +  8*stbir__simdfX_float_count, x2 );
+      stbir__simdfX_store( d + 12*stbir__simdfX_float_count, x3 );
+      d += (16*stbir__simdfX_float_count);
+    }
+  }
+}
+
+// memcpy that is specically intentionally overlapping (src is smaller then dest, so can be
+//   a normal forward copy, bytes is divisible by 4 and bytes is greater than or equal to
+//   the diff between dest and src)
+static void stbir_overlapping_memcpy( void * dest, void const * src, size_t bytes ) 
+{
+  char STBIR_SIMD_STREAMOUT_PTR (*) sd = (char*) src;
+  char STBIR_SIMD_STREAMOUT_PTR( * ) s_end = ((char*) src) + bytes;
+  ptrdiff_t ofs_to_dest = (char*)dest - (char*)src;
+
+  if ( ofs_to_dest >= 16 ) // is the overlap more than 16 away?
+  {
+    char STBIR_SIMD_STREAMOUT_PTR( * ) s_end16 = ((char*) src) + (bytes&~15);
+    do
+    {
+      stbir__simdf x;
+      STBIR_SIMD_NO_UNROLL(sd);
+      stbir__simdf_load( x, sd );
+      stbir__simdf_store(  ( sd + ofs_to_dest ), x );
+      sd += 16;
+    } while ( sd < s_end16 );
+
+    if ( sd == s_end )
+      return;
+  }
+
+  do
+  {
+    STBIR_SIMD_NO_UNROLL(sd);
+    *(int*)( sd + ofs_to_dest ) = *(int*) sd; 
+    sd += 4;
+  } while ( sd < s_end );
+}
+
+#else // no SSE2
+
+// when in scalar mode, we let unrolling happen, so this macro just does the __restrict
+#define STBIR_SIMD_STREAMOUT_PTR( star ) STBIR_STREAMOUT_PTR( star )
+#define STBIR_SIMD_NO_UNROLL(ptr) 
+
+#endif // SSE2
+
+
+#ifdef STBIR_PROFILE
+
+#if defined(_x86_64) || defined( __x86_64__ ) || defined( _M_X64 ) || defined(__x86_64) || defined(__SSE2__) || defined(STBIR_SSE) || defined( _M_IX86_FP ) || defined(__i386) || defined( __i386__ ) || defined( _M_IX86 ) || defined( _X86_ )
+
+#ifdef _MSC_VER
+
+  STBIRDEF stbir_uint64 __rdtsc();
+  #define STBIR_PROFILE_FUNC() __rdtsc()
+
+#else // non msvc
+
+  static stbir__inline stbir_uint64 STBIR_PROFILE_FUNC() 
+  {
+    stbir_uint32 lo, hi;
+    asm volatile ("rdtsc" : "=a" (lo), "=d" (hi) );
+    return ( ( (stbir_uint64) hi ) << 32 ) | ( (stbir_uint64) lo );
+  }
+
+#endif  // msvc
+
+#elif defined( _M_ARM64 ) || defined( __aarch64__ ) || defined( __arm64__ ) || defined(__ARM_NEON__) 
+
+#if defined( _MSC_VER ) && !defined(__clang__)
+
+  #define STBIR_PROFILE_FUNC() _ReadStatusReg(ARM64_CNTVCT)
+
+#else
+
+  static stbir__inline stbir_uint64 STBIR_PROFILE_FUNC()
+  {
+    stbir_uint64 tsc;
+    asm volatile("mrs %0, cntvct_el0" : "=r" (tsc));
+    return tsc;
+  }
+
+#endif
+
+#else // x64, arm
+
+#error Unknown platform for profiling.
+
+#endif  //x64 and   
+
+
+#define STBIR_ONLY_PROFILE_GET_SPLIT_INFO ,stbir__per_split_info * split_info
+#define STBIR_ONLY_PROFILE_SET_SPLIT_INFO ,split_info
+
+#define STBIR_ONLY_PROFILE_BUILD_GET_INFO ,stbir__info * profile_info
+#define STBIR_ONLY_PROFILE_BUILD_SET_INFO ,profile_info
+
+// super light-weight micro profiler
+#define STBIR_PROFILE_START_ll( info, wh ) { stbir_uint64 wh##thiszonetime = STBIR_PROFILE_FUNC(); stbir_uint64 * wh##save_parent_excluded_ptr = info->current_zone_excluded_ptr; stbir_uint64 wh##current_zone_excluded = 0; info->current_zone_excluded_ptr = &wh##current_zone_excluded; 
+#define STBIR_PROFILE_END_ll( info, wh ) wh##thiszonetime = STBIR_PROFILE_FUNC() - wh##thiszonetime; info->profile.named.wh += wh##thiszonetime - wh##current_zone_excluded; *wh##save_parent_excluded_ptr += wh##thiszonetime; info->current_zone_excluded_ptr = wh##save_parent_excluded_ptr; }
+#define STBIR_PROFILE_FIRST_START_ll( info, wh ) { int i; info->current_zone_excluded_ptr = &info->profile.named.total; for(i=0;i<STBIR__ARRAY_SIZE(info->profile.array);i++) info->profile.array[i]=0; } STBIR_PROFILE_START_ll( info, wh );
+#define STBIR_PROFILE_CLEAR_EXTRAS_ll( info, num ) { int extra; for(extra=1;extra<(num);extra++) { int i; for(i=0;i<STBIR__ARRAY_SIZE((info)->profile.array);i++) (info)[extra].profile.array[i]=0; } }
+
+// for thread data
+#define STBIR_PROFILE_START( wh ) STBIR_PROFILE_START_ll( split_info, wh )
+#define STBIR_PROFILE_END( wh ) STBIR_PROFILE_END_ll( split_info, wh )
+#define STBIR_PROFILE_FIRST_START( wh ) STBIR_PROFILE_FIRST_START_ll( split_info, wh )
+#define STBIR_PROFILE_CLEAR_EXTRAS() STBIR_PROFILE_CLEAR_EXTRAS_ll( split_info, split_count )
+
+// for build data
+#define STBIR_PROFILE_BUILD_START( wh ) STBIR_PROFILE_START_ll( profile_info, wh )
+#define STBIR_PROFILE_BUILD_END( wh ) STBIR_PROFILE_END_ll( profile_info, wh )
+#define STBIR_PROFILE_BUILD_FIRST_START( wh ) STBIR_PROFILE_FIRST_START_ll( profile_info, wh )
+#define STBIR_PROFILE_BUILD_CLEAR( info ) { int i; for(i=0;i<STBIR__ARRAY_SIZE(info->profile.array);i++) info->profile.array[i]=0; }
+
+#else  // no profile
+
+#define STBIR_ONLY_PROFILE_GET_SPLIT_INFO
+#define STBIR_ONLY_PROFILE_SET_SPLIT_INFO
+
+#define STBIR_ONLY_PROFILE_BUILD_GET_INFO
+#define STBIR_ONLY_PROFILE_BUILD_SET_INFO
+
+#define STBIR_PROFILE_START( wh )
+#define STBIR_PROFILE_END( wh )
+#define STBIR_PROFILE_FIRST_START( wh )
+#define STBIR_PROFILE_CLEAR_EXTRAS( )
+
+#define STBIR_PROFILE_BUILD_START( wh ) 
+#define STBIR_PROFILE_BUILD_END( wh ) 
+#define STBIR_PROFILE_BUILD_FIRST_START( wh )
+#define STBIR_PROFILE_BUILD_CLEAR( info )
+
+#endif  // stbir_profile
+
+#ifndef STBIR_CEILF
+#include <math.h>
+#if _MSC_VER <= 1200 // support VC6 for Sean
+#define STBIR_CEILF(x) ((float)ceil((float)(x)))
+#define STBIR_FLOORF(x) ((float)floor((float)(x)))
+#else
+#define STBIR_CEILF(x) ceilf(x)
+#define STBIR_FLOORF(x) floorf(x)
+#endif
+#endif
+
+#ifndef STBIR_MEMCPY
+// For memcpy
+#include <string.h>
+#define STBIR_MEMCPY( dest, src, len ) memcpy( dest, src, len )
+#endif
+
+#ifndef STBIR_SIMD
+
+// memcpy that is specically intentionally overlapping (src is smaller then dest, so can be
+//   a normal forward copy, bytes is divisible by 4 and bytes is greater than or equal to
+//   the diff between dest and src)
+static void stbir_overlapping_memcpy( void * dest, void const * src, size_t bytes ) 
+{
+  char STBIR_SIMD_STREAMOUT_PTR (*) sd = (char*) src;
+  char STBIR_SIMD_STREAMOUT_PTR( * ) s_end = ((char*) src) + bytes;
+  ptrdiff_t ofs_to_dest = (char*)dest - (char*)src;
+
+  if ( ofs_to_dest >= 8 ) // is the overlap more than 8 away?
+  {
+    char STBIR_SIMD_STREAMOUT_PTR( * ) s_end8 = ((char*) src) + (bytes&~7);
+    do
+    {
+      STBIR_NO_UNROLL(sd);
+      *(stbir_uint64*)( sd + ofs_to_dest ) = *(stbir_uint64*) sd; 
+      sd += 8;
+    } while ( sd < s_end8 );
+
+    if ( sd == s_end )
+      return;
+  }
+
+  do
+  {
+    STBIR_NO_UNROLL(sd);
+    *(int*)( sd + ofs_to_dest ) = *(int*) sd; 
+    sd += 4;
+  } while ( sd < s_end );
+}
+
+#endif
+
+static float stbir__filter_trapezoid(float x, float scale, void * user_data)
+{
+  float halfscale = scale / 2;
+  float t = 0.5f + halfscale;
+  STBIR_ASSERT(scale <= 1);
+  STBIR__UNUSED(user_data);
+
+  if ( x < 0.0f ) x = -x;
+
+  if (x >= t)
+    return 0.0f;
+  else
+  {
+    float r = 0.5f - halfscale;
+    if (x <= r)
+      return 1.0f;
+    else
+      return (t - x) / scale;
+  }
+}
+
+static float stbir__support_trapezoid(float scale, void * user_data)
+{
+  STBIR__UNUSED(user_data);
+  return 0.5f + scale / 2.0f;
+}
+
+static float stbir__filter_triangle(float x, float s, void * user_data)
+{
+  STBIR__UNUSED(s);
+  STBIR__UNUSED(user_data);
+
+  if ( x < 0.0f ) x = -x;
+
+  if (x <= 1.0f)
+    return 1.0f - x;
+  else
+    return 0.0f;
+}
+
+static float stbir__filter_point(float x, float s, void * user_data)
+{
+  STBIR__UNUSED(x);
+  STBIR__UNUSED(s);
+  STBIR__UNUSED(user_data);
+
+  return 1.0f;
+}
+
+static float stbir__filter_cubic(float x, float s, void * user_data)
+{
+  STBIR__UNUSED(s);
+  STBIR__UNUSED(user_data);
+
+  if ( x < 0.0f ) x = -x;
+
+  if (x < 1.0f)
+    return (4.0f + x*x*(3.0f*x - 6.0f))/6.0f;
+  else if (x < 2.0f)
+    return (8.0f + x*(-12.0f + x*(6.0f - x)))/6.0f;
+
+  return (0.0f);
+}
+
+static float stbir__filter_catmullrom(float x, float s, void * user_data)
+{
+  STBIR__UNUSED(s);
+  STBIR__UNUSED(user_data);
+
+  if ( x < 0.0f ) x = -x;
+
+  if (x < 1.0f)
+    return 1.0f - x*x*(2.5f - 1.5f*x);
+  else if (x < 2.0f)
+    return 2.0f - x*(4.0f + x*(0.5f*x - 2.5f));
+
+  return (0.0f);
+}
+
+static float stbir__filter_mitchell(float x, float s, void * user_data)
+{
+  STBIR__UNUSED(s);
+  STBIR__UNUSED(user_data);
+
+  if ( x < 0.0f ) x = -x;
+
+  if (x < 1.0f)
+    return (16.0f + x*x*(21.0f * x - 36.0f))/18.0f;
+  else if (x < 2.0f)
+    return (32.0f + x*(-60.0f + x*(36.0f - 7.0f*x)))/18.0f;
+
+  return (0.0f);
+}
+
+static float stbir__support_zero(float s, void * user_data)
+{
+  STBIR__UNUSED(s);
+  STBIR__UNUSED(user_data);
+  return 0;
+}
+
+static float stbir__support_zeropoint5(float s, void * user_data)
+{
+  STBIR__UNUSED(s);
+  STBIR__UNUSED(user_data);
+  return 0.5f;
+}
+
+static float stbir__support_one(float s, void * user_data)
+{
+  STBIR__UNUSED(s);
+  STBIR__UNUSED(user_data);
+  return 1;
+}
+
+static float stbir__support_two(float s, void * user_data) 
+{
+  STBIR__UNUSED(s);
+  STBIR__UNUSED(user_data);
+  return 2;
+}
+
+// This is the maximum number of input samples that can affect an output sample
+// with the given filter from the output pixel's perspective
+static int stbir__get_filter_pixel_width(stbir__support_callback * support, float scale, void * user_data)
+{
+  STBIR_ASSERT(support != 0);
+
+  if ( scale >= ( 1.0f-stbir__small_float ) ) // upscale
+    return (int)STBIR_CEILF(support(1.0f/scale,user_data) * 2.0f);
+  else
+    return (int)STBIR_CEILF(support(scale,user_data) * 2.0f / scale);
+}
+
+// this is how many coefficents per run of the filter (which is different 
+//   from the filter_pixel_width depending on if we are scattering or gathering)
+static int stbir__get_coefficient_width(stbir__sampler * samp, int is_gather, void * user_data)
+{
+  float scale = samp->scale_info.scale;
+  stbir__support_callback * support = samp->filter_support;
+
+  switch( is_gather )
+  {
+    case 1:
+      return (int)STBIR_CEILF(support(1.0f / scale, user_data) * 2.0f);
+    case 2:
+      return (int)STBIR_CEILF(support(scale, user_data) * 2.0f / scale);
+    case 0:
+      return (int)STBIR_CEILF(support(scale, user_data) * 2.0f);
+    default:
+      STBIR_ASSERT( (is_gather >= 0 ) && (is_gather <= 2 ) );
+      return 0;
+  }
+}
+
+static int stbir__get_contributors(stbir__sampler * samp, int is_gather)  
+{
+  if (is_gather)
+      return samp->scale_info.output_sub_size;
+  else
+      return (samp->scale_info.input_full_size + samp->filter_pixel_margin * 2);
+}
+
+static int stbir__edge_zero_full( int n, int max )
+{
+  STBIR__UNUSED(n);
+  STBIR__UNUSED(max);
+  return 0; // NOTREACHED
+}
+
+static int stbir__edge_clamp_full( int n, int max )
+{
+  if (n < 0)
+    return 0;
+
+  if (n >= max)
+    return max - 1;
+
+  return n; // NOTREACHED
+}
+
+static int stbir__edge_reflect_full( int n, int max )
+{
+  if (n < 0)
+  {
+    if (n > -max)    
+      return -n;
+    else
+      return max - 1;
+  }
+
+  if (n >= max)
+  {
+    int max2 = max * 2;
+    if (n >= max2)
+      return 0;
+    else
+      return max2 - n - 1;
+  }
+
+  return n; // NOTREACHED
+}
+
+static int stbir__edge_wrap_full( int n, int max )
+{
+  if (n >= 0)
+    return (n % max);
+  else
+  {
+    int m = (-n) % max;
+
+    if (m != 0)
+      m = max - m;
+
+    return (m);
+  }
+}
+
+typedef int stbir__edge_wrap_func( int n, int max );
+static stbir__edge_wrap_func * stbir__edge_wrap_slow[] =
+{
+  stbir__edge_clamp_full,    // STBIR_EDGE_CLAMP
+  stbir__edge_reflect_full,  // STBIR_EDGE_REFLECT
+  stbir__edge_wrap_full,     // STBIR_EDGE_WRAP
+  stbir__edge_zero_full,     // STBIR_EDGE_ZERO
+};
+
+stbir__inline static int stbir__edge_wrap(stbir_edge edge, int n, int max)
+{
+  // avoid per-pixel switch
+  if (n >= 0 && n < max)
+      return n;
+  return stbir__edge_wrap_slow[edge]( n, max );
+}
+
+#define STBIR__MERGE_RUNS_PIXEL_THRESHOLD 16
+
+// get information on the extents of a sampler
+static void stbir__get_extents( stbir__sampler * samp, stbir__extents * scanline_extents )
+{
+  int j, stop;
+  int left_margin, right_margin;
+  int min_n = 0x7fffffff, max_n = -0x7fffffff;
+  int min_left = 0x7fffffff, max_left = -0x7fffffff;
+  int min_right = 0x7fffffff, max_right = -0x7fffffff;
+  stbir_edge edge = samp->edge;
+  stbir__contributors* contributors = samp->contributors;
+  int output_sub_size = samp->scale_info.output_sub_size;
+  int input_full_size = samp->scale_info.input_full_size;
+  int filter_pixel_margin = samp->filter_pixel_margin;
+
+  STBIR_ASSERT( samp->is_gather );
+
+  stop = output_sub_size;
+  for (j = 0; j < stop; j++ )
+  {
+    STBIR_ASSERT( contributors[j].n1 >= contributors[j].n0 );
+    if ( contributors[j].n0 < min_n )
+    {
+      min_n = contributors[j].n0;
+      stop = j + filter_pixel_margin;  // if we find a new min, only scan another filter width
+      if ( stop > output_sub_size ) stop = output_sub_size;
+    }
+  }
+
+  stop = 0;
+  for (j = output_sub_size - 1; j >= stop; j-- )
+  {
+    STBIR_ASSERT( contributors[j].n1 >= contributors[j].n0 );
+    if ( contributors[j].n1 > max_n )
+    {
+      max_n = contributors[j].n1;
+      stop = j - filter_pixel_margin;  // if we find a new max, only scan another filter width
+      if (stop<0) stop = 0;
+    }
+  }
+
+  STBIR_ASSERT( scanline_extents->conservative.n0 <= min_n );
+  STBIR_ASSERT( scanline_extents->conservative.n1 >= max_n );
+
+  // now calculate how much into the margins we really read
+  left_margin = 0;
+  if ( min_n < 0 )
+  {
+    left_margin = -min_n;
+    min_n = 0;
+  }
+  
+  right_margin = 0;
+  if ( max_n >= input_full_size )
+  {
+    right_margin = max_n - input_full_size + 1;
+    max_n = input_full_size - 1;
+  }
+
+  // index 1 is margin pixel extents (how many pixels we hang over the edge)
+  scanline_extents->edge_sizes[0] = left_margin;
+  scanline_extents->edge_sizes[1] = right_margin;
+
+  // index 2 is pixels read from the input
+  scanline_extents->spans[0].n0 = min_n;
+  scanline_extents->spans[0].n1 = max_n;
+  scanline_extents->spans[0].pixel_offset_for_input = min_n;
+
+  // default to no other input range
+  scanline_extents->spans[1].n0 = 0;
+  scanline_extents->spans[1].n1 = -1;
+  scanline_extents->spans[1].pixel_offset_for_input = 0;
+
+  // don't have to do edge calc for zero clamp
+  if ( edge == STBIR_EDGE_ZERO )
+    return;
+  
+  // convert margin pixels to the pixels within the input (min and max)
+  for( j = -left_margin ; j < 0 ; j++ )
+  {
+      int p = stbir__edge_wrap( edge, j, input_full_size );
+      if ( p < min_left )
+        min_left = p;
+      if ( p > max_left )
+        max_left = p;
+  }
+
+  for( j = input_full_size ; j < (input_full_size + right_margin) ; j++ )
+  {
+      int p = stbir__edge_wrap( edge, j, input_full_size );
+      if ( p < min_right )
+        min_right = p;
+      if ( p > max_right )
+        max_right = p;
+  }
+
+  // merge the left margin pixel region if it connects within 4 pixels of main pixel region
+  if ( min_left != 0x7fffffff )
+  {
+    if ( ( ( min_left <= min_n ) && ( ( max_left  + STBIR__MERGE_RUNS_PIXEL_THRESHOLD ) >= min_n ) ) ||
+         ( ( min_n <= min_left ) && ( ( max_n  + STBIR__MERGE_RUNS_PIXEL_THRESHOLD ) >= max_left ) ) )
+    {
+      scanline_extents->spans[0].n0 = min_n = stbir__min( min_n, min_left );
+      scanline_extents->spans[0].n1 = max_n = stbir__max( max_n, max_left );
+      scanline_extents->spans[0].pixel_offset_for_input = min_n;
+      left_margin = 0;
+    }
+  }
+
+  // merge the right margin pixel region if it connects within 4 pixels of main pixel region
+  if ( min_right != 0x7fffffff )
+  {
+    if ( ( ( min_right <= min_n ) && ( ( max_right  + STBIR__MERGE_RUNS_PIXEL_THRESHOLD ) >= min_n ) ) ||
+         ( ( min_n <= min_right ) && ( ( max_n  + STBIR__MERGE_RUNS_PIXEL_THRESHOLD ) >= max_right ) ) )
+    {
+      scanline_extents->spans[0].n0 = min_n = stbir__min( min_n, min_right );
+      scanline_extents->spans[0].n1 = max_n = stbir__max( max_n, max_right );
+      scanline_extents->spans[0].pixel_offset_for_input = min_n;
+      right_margin = 0;
+    }
+  }
+
+  STBIR_ASSERT( scanline_extents->conservative.n0 <= min_n );
+  STBIR_ASSERT( scanline_extents->conservative.n1 >= max_n );
+
+  // you get two ranges when you have the WRAP edge mode and you are doing just the a piece of the resize
+  //   so you need to get a second run of pixels from the opposite side of the scanline (which you
+  //   wouldn't need except for WRAP)
+
+
+  // if we can't merge the min_left range, add it as a second range
+  if ( ( left_margin ) && ( min_left != 0x7fffffff ) )
+  {
+    stbir__span * newspan = scanline_extents->spans + 1;
+    STBIR_ASSERT( right_margin == 0 );
+    if ( min_left < scanline_extents->spans[0].n0 )
+    {
+      scanline_extents->spans[1].pixel_offset_for_input = scanline_extents->spans[0].n0;
+      scanline_extents->spans[1].n0 = scanline_extents->spans[0].n0;
+      scanline_extents->spans[1].n1 = scanline_extents->spans[0].n1;
+      --newspan;
+    }
+    newspan->pixel_offset_for_input = min_left;
+    newspan->n0 = -left_margin;
+    newspan->n1 = ( max_left - min_left ) - left_margin;
+    scanline_extents->edge_sizes[0] = 0;  // don't need to copy the left margin, since we are directly decoding into the margin
+    return;
+  }
+
+  // if we can't merge the min_left range, add it as a second range
+  if ( ( right_margin ) && ( min_right != 0x7fffffff ) )
+  {
+    stbir__span * newspan = scanline_extents->spans + 1;
+    if ( min_right < scanline_extents->spans[0].n0 )
+    {
+      scanline_extents->spans[1].pixel_offset_for_input = scanline_extents->spans[0].n0;
+      scanline_extents->spans[1].n0 = scanline_extents->spans[0].n0;
+      scanline_extents->spans[1].n1 = scanline_extents->spans[0].n1;
+      --newspan;
+    }
+    newspan->pixel_offset_for_input = min_right;
+    newspan->n0 = scanline_extents->spans[1].n1 + 1;
+    newspan->n1 = scanline_extents->spans[1].n1 + 1 + ( max_right - min_right );
+    scanline_extents->edge_sizes[1] = 0;  // don't need to copy the right margin, since we are directly decoding into the margin
+    return;
+  }
+}
+
+static void stbir__calculate_in_pixel_range( int * first_pixel, int * last_pixel, float out_pixel_center, float out_filter_radius, float inv_scale, float out_shift, int input_size, stbir_edge edge )
+{
+  int first, last;
+  float out_pixel_influence_lowerbound = out_pixel_center - out_filter_radius;
+  float out_pixel_influence_upperbound = out_pixel_center + out_filter_radius;
+
+  float in_pixel_influence_lowerbound = (out_pixel_influence_lowerbound + out_shift) * inv_scale; 
+  float in_pixel_influence_upperbound = (out_pixel_influence_upperbound + out_shift) * inv_scale; 
+
+  first = (int)(STBIR_FLOORF(in_pixel_influence_lowerbound + 0.5f));
+  last = (int)(STBIR_FLOORF(in_pixel_influence_upperbound - 0.5f));
+
+  if ( edge == STBIR_EDGE_WRAP )
+  {
+    if ( first <= -input_size )
+      first = -(input_size-1);
+    if ( last >= (input_size*2))
+      last = (input_size*2) - 1;
+  }
+  
+  *first_pixel = first;
+  *last_pixel = last;
+}
+
+static void stbir__calculate_coefficients_for_gather_upsample( float out_filter_radius, stbir__kernel_callback * kernel, stbir__scale_info * scale_info, int num_contributors, stbir__contributors* contributors, float* coefficient_group, int coefficient_width, stbir_edge edge, void * user_data )
+{
+  int n, end;
+  float inv_scale = scale_info->inv_scale;
+  float out_shift = scale_info->pixel_shift;
+  int input_size  = scale_info->input_full_size;
+  int numerator = scale_info->scale_numerator;
+  int polyphase = ( ( scale_info->scale_is_rational ) && ( numerator < num_contributors ) );
+
+  // Looping through out pixels
+  end = num_contributors; if ( polyphase ) end = numerator;
+  for (n = 0; n < end; n++)
+  {
+    int i;
+    int last_non_zero;
+    float out_pixel_center = (float)n + 0.5f;
+    float in_center_of_out = (out_pixel_center + out_shift) * inv_scale;  
+
+    int in_first_pixel, in_last_pixel;
+    
+    stbir__calculate_in_pixel_range( &in_first_pixel, &in_last_pixel, out_pixel_center, out_filter_radius, inv_scale, out_shift, input_size, edge );
+
+    last_non_zero = -1;
+    for (i = 0; i <= in_last_pixel - in_first_pixel; i++)
+    {
+      float in_pixel_center = (float)(i + in_first_pixel) + 0.5f;
+      float coeff = kernel(in_center_of_out - in_pixel_center, inv_scale, user_data);
+
+      // kill denormals
+      if ( ( ( coeff < stbir__small_float ) && ( coeff > -stbir__small_float ) ) )
+      {
+        if ( i == 0 )  // if we're at the front, just eat zero contributors
+        { 
+          STBIR_ASSERT ( ( in_last_pixel - in_first_pixel ) != 0 ); // there should be at least one contrib
+          ++in_first_pixel;
+          i--;
+          continue;
+        }
+        coeff = 0;  // make sure is fully zero (should keep denormals away)
+      }
+      else
+        last_non_zero = i;
+      
+      coefficient_group[i] = coeff;
+    }
+    
+    in_last_pixel = last_non_zero+in_first_pixel; // kills trailing zeros
+    contributors->n0 = in_first_pixel;
+    contributors->n1 = in_last_pixel;
+
+    STBIR_ASSERT(contributors->n1 >= contributors->n0);
+
+    ++contributors;
+    coefficient_group += coefficient_width;
+  }
+}
+
+static void stbir__insert_coeff( stbir__contributors * contribs, float * coeffs, int new_pixel, float new_coeff )
+{
+  if ( new_pixel <= contribs->n1 )  // before the end
+  {
+    if ( new_pixel < contribs->n0 ) // before the front?
+    {
+      int j, o = contribs->n0 - new_pixel;
+      for ( j = contribs->n1 - contribs->n0 ; j <= 0 ; j-- )
+        coeffs[ j + o ] = coeffs[ j ];
+      for ( j = 1 ; j < o ; j-- )
+        coeffs[ j ] = coeffs[ 0 ];
+      coeffs[ 0 ] = new_coeff;
+      contribs->n0 = new_pixel;
+    }
+    else
+    {
+      coeffs[ new_pixel - contribs->n0 ] += new_coeff;
+    }
+  }
+  else
+  {
+    int j, e = new_pixel - contribs->n0;
+    for( j = ( contribs->n1 - contribs->n0 ) + 1 ; j < e ; j++ ) // clear in-betweens coeffs if there are any
+      coeffs[j] = 0;
+
+    coeffs[ e ] = new_coeff;
+    contribs->n1 = new_pixel;
+  }
+}
+
+static void stbir__calculate_out_pixel_range( int * first_pixel, int * last_pixel, float in_pixel_center, float in_pixels_radius, float scale, float out_shift, int out_size )
+{
+  float in_pixel_influence_lowerbound = in_pixel_center - in_pixels_radius;
+  float in_pixel_influence_upperbound = in_pixel_center + in_pixels_radius;
+  float out_pixel_influence_lowerbound = in_pixel_influence_lowerbound * scale - out_shift;
+  float out_pixel_influence_upperbound = in_pixel_influence_upperbound * scale - out_shift;
+  int out_first_pixel = (int)(STBIR_FLOORF(out_pixel_influence_lowerbound + 0.5f));
+  int out_last_pixel = (int)(STBIR_FLOORF(out_pixel_influence_upperbound - 0.5f));
+
+  if ( out_first_pixel < 0 )
+    out_first_pixel = 0;
+  if ( out_last_pixel >= out_size )
+    out_last_pixel = out_size - 1;
+  *first_pixel = out_first_pixel;
+  *last_pixel = out_last_pixel;
+}
+
+static void stbir__calculate_coefficients_for_gather_downsample( int start, int end, float in_pixels_radius, stbir__kernel_callback * kernel, stbir__scale_info * scale_info, int coefficient_width, int num_contributors, stbir__contributors * contributors, float * coefficient_group, void * user_data )
+{
+  int in_pixel;
+  int i;
+  int first_out_inited = -1;
+  float scale = scale_info->scale;
+  float out_shift = scale_info->pixel_shift;
+  int out_size = scale_info->output_sub_size;
+  int numerator = scale_info->scale_numerator;
+  int polyphase = ( ( scale_info->scale_is_rational ) && ( numerator < out_size ) );
+
+  STBIR__UNUSED(num_contributors);
+
+  // Loop through the input pixels
+  for (in_pixel = start; in_pixel < end; in_pixel++)
+  {
+    float in_pixel_center = (float)in_pixel + 0.5f;
+    float out_center_of_in = in_pixel_center * scale - out_shift;
+    int out_first_pixel, out_last_pixel;
+
+    stbir__calculate_out_pixel_range( &out_first_pixel, &out_last_pixel, in_pixel_center, in_pixels_radius, scale, out_shift, out_size );
+
+    if ( out_first_pixel > out_last_pixel )
+      continue;
+
+    // clamp or exit if we are using polyphase filtering, and the limit is up
+    if ( polyphase )
+    {
+      // when polyphase, you only have to do coeffs up to the numerator count
+      if ( out_first_pixel == numerator )
+        break;
+
+      // don't do any extra work, clamp last pixel at numerator too
+      if ( out_last_pixel >= numerator )
+        out_last_pixel = numerator - 1;
+    }
+
+    for (i = 0; i <= out_last_pixel - out_first_pixel; i++)
+    {
+      float out_pixel_center = (float)(i + out_first_pixel) + 0.5f;
+      float x = out_pixel_center - out_center_of_in;
+      float coeff = kernel(x, scale, user_data) * scale;
+
+      // kill the coeff if it's too small (avoid denormals)
+      if ( ( ( coeff < stbir__small_float ) && ( coeff > -stbir__small_float ) ) )
+        coeff = 0.0f;
+
+      {
+        int out = i + out_first_pixel;
+        float * coeffs = coefficient_group + out * coefficient_width;
+        stbir__contributors * contribs = contributors + out;
+
+        // is this the first time this output pixel has been seen?  Init it.
+        if ( out > first_out_inited ) 
+        {
+          STBIR_ASSERT( out == ( first_out_inited + 1 ) ); // ensure we have only advanced one at time
+          first_out_inited = out;
+          contribs->n0 = in_pixel;
+          contribs->n1 = in_pixel;
+          coeffs[0]  = coeff;
+        }
+        else 
+        {
+          // insert on end (always in order)
+          if ( coeffs[0] == 0.0f )  // if the first coefficent is zero, then zap it for this coeffs
+          {
+            STBIR_ASSERT( ( in_pixel - contribs->n0 ) == 1 ); // ensure that when we zap, we're at the 2nd pos
+            contribs->n0 = in_pixel;
+          }
+          contribs->n1 = in_pixel;
+          STBIR_ASSERT( ( in_pixel - contribs->n0 ) < coefficient_width );
+          coeffs[in_pixel - contribs->n0]  = coeff;
+        }
+      }
+    }
+  }
+}
+
+static void stbir__cleanup_gathered_coefficients( stbir_edge edge, stbir__filter_extent_info* filter_info, stbir__scale_info * scale_info, int num_contributors, stbir__contributors* contributors, float * coefficient_group, int coefficient_width )
+{
+  int input_size = scale_info->input_full_size;
+  int input_last_n1 = input_size - 1; 
+  int n, end;
+  int lowest = 0x7fffffff;
+  int highest = -0x7fffffff;
+  int widest = -1;
+  int numerator = scale_info->scale_numerator;
+  int denominator = scale_info->scale_denominator;
+  int polyphase = ( ( scale_info->scale_is_rational ) && ( numerator < num_contributors ) );
+  float * coeffs;
+  stbir__contributors * contribs;
+
+  // weight all the coeffs for each sample
+  coeffs = coefficient_group;
+  contribs = contributors;
+  end = num_contributors; if ( polyphase ) end = numerator;
+  for (n = 0; n < end; n++)
+  {
+    int i;
+    float filter_scale, total_filter = 0;
+    int e;
+
+    // add all contribs
+    e = contribs->n1 - contribs->n0;
+    for( i = 0 ; i <= e ; i++ )
+    {
+      total_filter += coeffs[i];
+      STBIR_ASSERT( ( coeffs[i] >= -2.0f ) && ( coeffs[i] <= 2.0f )  ); // check for wonky weights
+    }
+
+    // rescale
+    if ( ( total_filter < stbir__small_float ) && ( total_filter > -stbir__small_float ) )
+    {
+      // all coeffs are extremely small, just zero it
+      contribs->n1 = contribs->n0;
+      coeffs[0] = 0.0f;
+    }
+    else
+    {
+      // if the total isn't 1.0, rescale everything
+      if ( ( total_filter < (1.0f-stbir__small_float) ) || ( total_filter > (1.0f+stbir__small_float) ) )
+      {
+        filter_scale = 1.0f / total_filter;
+        // scale them all
+        for (i = 0; i <= e; i++)
+          coeffs[i] *= filter_scale;
+      }
+    }
+    ++contribs;
+    coeffs += coefficient_width;
+  }
+
+  // if we have a rational for the scale, we can exploit the polyphaseness to not calculate
+  //   most of the coefficients, so we copy them here
+  if ( polyphase )
+  {
+    stbir__contributors * prev_contribs = contributors;
+    stbir__contributors * cur_contribs = contributors + numerator;
+
+    for( n = numerator ; n < num_contributors ; n++ )
+    {
+      cur_contribs->n0 = prev_contribs->n0 + denominator;
+      cur_contribs->n1 = prev_contribs->n1 + denominator;
+      ++cur_contribs;
+      ++prev_contribs;
+    }
+    stbir_overlapping_memcpy( coefficient_group + numerator * coefficient_width, coefficient_group, ( num_contributors - numerator ) * coefficient_width * sizeof( coeffs[ 0 ] ) );
+  }
+
+  coeffs = coefficient_group;
+  contribs = contributors;
+  for (n = 0; n < num_contributors; n++)
+  {
+    int i;
+
+    // in zero edge mode, just remove out of bounds contribs completely (since their weights are accounted for now)
+    if ( edge == STBIR_EDGE_ZERO )
+    {
+      // shrink the right side if necessary
+      if ( contribs->n1 > input_last_n1 )
+        contribs->n1 = input_last_n1;
+
+      // shrink the left side
+      if ( contribs->n0 < 0 )
+      {
+        int j, left, skips = 0;
+
+        skips = -contribs->n0;
+        contribs->n0 = 0;
+
+        // now move down the weights
+        left = contribs->n1 - contribs->n0 + 1;
+        if ( left > 0 )
+        {
+          for( j = 0 ; j < left ; j++ )
+            coeffs[ j ] = coeffs[ j + skips ];
+        }
+      }
+    }
+    else if ( ( edge == STBIR_EDGE_CLAMP ) || ( edge == STBIR_EDGE_REFLECT ) )
+    {
+      // for clamp and reflect, calculate the true inbounds position (based on edge type) and just add that to the existing weight
+      
+      // right hand side first
+      if ( contribs->n1 > input_last_n1 )
+      {
+        int start = contribs->n0;
+        int endi = contribs->n1;
+        contribs->n1 = input_last_n1;  
+        for( i = input_size; i <= endi; i++ )
+          stbir__insert_coeff( contribs, coeffs, stbir__edge_wrap_slow[edge]( i, input_size ), coeffs[i-start] );
+      }
+
+      // now check left hand edge
+      if ( contribs->n0 < 0 )
+      {
+        int save_n0;
+        float save_n0_coeff;
+        float * c = coeffs - ( contribs->n0 + 1 );
+        
+        // reinsert the coeffs with it reflected or clamped (insert accumulates, if the coeffs exist)
+        for( i = -1 ; i > contribs->n0 ; i-- ) 
+          stbir__insert_coeff( contribs, coeffs, stbir__edge_wrap_slow[edge]( i, input_size ), *c-- );
+        save_n0 = contribs->n0;
+        save_n0_coeff = c[0]; // save it, since we didn't do the final one (i==n0), because there might be too many coeffs to hold (before we resize)!
+
+        // now slide all the coeffs down (since we have accumulated them in the positive contribs) and reset the first contrib
+        contribs->n0 = 0;  
+        for(i = 0 ; i <= contribs->n1 ; i++ )
+          coeffs[i] = coeffs[i-save_n0];
+        
+        // now that we have shrunk down the contribs, we insert the first one safely
+        stbir__insert_coeff( contribs, coeffs, stbir__edge_wrap_slow[edge]( save_n0, input_size ), save_n0_coeff );
+      }
+    }
+
+    if ( contribs->n0 <= contribs->n1 )
+    {
+      int diff = contribs->n1 - contribs->n0 + 1;
+      while ( diff && ( coeffs[ diff-1 ] == 0.0f ) )
+        --diff;
+      contribs->n1 = contribs->n0 + diff - 1;
+
+      if ( contribs->n0 <= contribs->n1 )
+      {
+        if ( contribs->n0 < lowest )
+          lowest = contribs->n0;
+        if ( contribs->n1 > highest )
+          highest = contribs->n1;
+        if ( diff > widest )
+          widest = diff;
+      }
+
+      // re-zero out unused coefficients (if any)
+      for( i = diff ; i < coefficient_width ; i++ )
+        coeffs[i] = 0.0f;
+    }
+
+    ++contribs;
+    coeffs += coefficient_width;
+  }
+  filter_info->lowest = lowest;
+  filter_info->highest = highest;
+  filter_info->widest = widest;
+}
+
+static int stbir__pack_coefficients( int num_contributors, stbir__contributors* contributors, float * coefficents, int coefficient_width, int widest, int row_width )
+{
+  #define STBIR_MOVE_1( dest, src ) { STBIR_NO_UNROLL(dest); ((stbir_uint32*)(dest))[0] = ((stbir_uint32*)(src))[0]; }
+  #define STBIR_MOVE_2( dest, src ) { STBIR_NO_UNROLL(dest); ((stbir_uint64*)(dest))[0] = ((stbir_uint64*)(src))[0]; }
+  #ifdef STBIR_SIMD
+  #define STBIR_MOVE_4( dest, src ) { stbir__simdf t; STBIR_NO_UNROLL(dest); stbir__simdf_load( t, src ); stbir__simdf_store( dest, t ); }
+  #else
+  #define STBIR_MOVE_4( dest, src ) { STBIR_NO_UNROLL(dest); ((stbir_uint64*)(dest))[0] = ((stbir_uint64*)(src))[0]; ((stbir_uint64*)(dest))[1] = ((stbir_uint64*)(src))[1]; }
+  #endif
+  if ( coefficient_width != widest )
+  {
+    float * pc = coefficents;
+    float * coeffs = coefficents;
+    float * pc_end = coefficents + num_contributors * widest;
+    switch( widest )
+    {
+      case 1:
+        do {
+          STBIR_MOVE_1( pc, coeffs );
+          ++pc;
+          coeffs += coefficient_width;
+        } while ( pc < pc_end );
+        break;
+      case 2:
+        do {
+          STBIR_MOVE_2( pc, coeffs );
+          pc += 2;
+          coeffs += coefficient_width;
+        } while ( pc < pc_end );
+        break;
+      case 3:
+        do {
+          STBIR_MOVE_2( pc, coeffs );
+          STBIR_MOVE_1( pc+2, coeffs+2 );
+          pc += 3;
+          coeffs += coefficient_width;
+        } while ( pc < pc_end );
+        break;
+      case 4:
+        do {
+          STBIR_MOVE_4( pc, coeffs );
+          pc += 4;
+          coeffs += coefficient_width;
+        } while ( pc < pc_end );
+        break;
+      case 5:
+        do {
+          STBIR_MOVE_4( pc, coeffs );
+          STBIR_MOVE_1( pc+4, coeffs+4 );
+          pc += 5;
+          coeffs += coefficient_width;
+        } while ( pc < pc_end );
+        break;
+      case 6:
+        do {
+          STBIR_MOVE_4( pc, coeffs );
+          STBIR_MOVE_2( pc+4, coeffs+4 );
+          pc += 6;
+          coeffs += coefficient_width;
+        } while ( pc < pc_end );
+        break;
+      case 7:
+        do {
+          STBIR_MOVE_4( pc, coeffs );
+          STBIR_MOVE_2( pc+4, coeffs+4 );
+          STBIR_MOVE_1( pc+6, coeffs+6 );
+          pc += 7;
+          coeffs += coefficient_width;
+        } while ( pc < pc_end );
+        break;
+      case 8:
+        do {
+          STBIR_MOVE_4( pc, coeffs );
+          STBIR_MOVE_4( pc+4, coeffs+4 );
+          pc += 8;
+          coeffs += coefficient_width;
+        } while ( pc < pc_end );
+        break;
+      case 9:
+        do {
+          STBIR_MOVE_4( pc, coeffs );
+          STBIR_MOVE_4( pc+4, coeffs+4 );
+          STBIR_MOVE_1( pc+8, coeffs+8 );
+          pc += 9;
+          coeffs += coefficient_width;
+        } while ( pc < pc_end );
+        break;
+      case 10:
+        do {
+          STBIR_MOVE_4( pc, coeffs );
+          STBIR_MOVE_4( pc+4, coeffs+4 );
+          STBIR_MOVE_2( pc+8, coeffs+8 );
+          pc += 10;
+          coeffs += coefficient_width;
+        } while ( pc < pc_end );
+        break;
+      case 11:
+        do {
+          STBIR_MOVE_4( pc, coeffs );
+          STBIR_MOVE_4( pc+4, coeffs+4 );
+          STBIR_MOVE_2( pc+8, coeffs+8 );
+          STBIR_MOVE_1( pc+10, coeffs+10 );
+          pc += 11;
+          coeffs += coefficient_width;
+        } while ( pc < pc_end );
+        break;
+      case 12:
+        do {
+          STBIR_MOVE_4( pc, coeffs );
+          STBIR_MOVE_4( pc+4, coeffs+4 );
+          STBIR_MOVE_4( pc+8, coeffs+8 );
+          pc += 12;
+          coeffs += coefficient_width;
+        } while ( pc < pc_end );
+        break;
+      default:
+        do {
+          float * copy_end = pc + widest - 4;
+          float * c = coeffs;
+          do {
+            STBIR_NO_UNROLL( pc );
+            STBIR_MOVE_4( pc, c );
+            pc += 4;
+            c += 4;
+          } while ( pc <= copy_end );
+          copy_end += 4;
+          while ( pc < copy_end )
+          {
+            STBIR_MOVE_1( pc, c );
+            ++pc; ++c;
+          }
+          coeffs += coefficient_width;
+        } while ( pc < pc_end );
+        break;
+    }
+  }
+
+  // some horizontal routines read one float off the end (which is then masked off), so put in a sentinal so we don't read an snan or denormal
+  coefficents[ widest * num_contributors ] = 8888.0f;
+
+  // the minimum we might read for unrolled filters widths is 12. So, we need to
+  //   make sure we never read outside the decode buffer, by possibly moving 
+  //   the sample area back into the scanline, and putting zeros weights first.
+  // we start on the right edge and check until we're well past the possible
+  //   clip area (2*widest).
+  {
+    stbir__contributors * contribs = contributors + num_contributors - 1;
+    float * coeffs = coefficents + widest * ( num_contributors - 1 );
+
+    // go until no chance of clipping (this is usually less than 8 lops)
+    while ( ( ( contribs->n0 + widest*2 ) >= row_width ) && ( contribs >= contributors ) )
+    {
+      // might we clip??
+      if ( ( contribs->n0 + widest ) > row_width )
+      {
+        int stop_range = widest;
+      
+        // if range is larger than 12, it will be handled by generic loops that can terminate on the exact length
+        //   of this contrib n1, instead of a fixed widest amount - so calculate this
+        if ( widest > 12 )
+        {
+          int mod;
+
+          // how far will be read in the n_coeff loop (which depends on the widest count mod4);
+          mod = widest & 3;
+          stop_range = ( ( ( contribs->n1 - contribs->n0 + 1 ) - mod + 3 ) & ~3 ) + mod; 
+
+          // the n_coeff loops do a minimum amount of coeffs, so factor that in!
+          if ( stop_range < ( 8 + mod ) ) stop_range = 8 + mod;
+        }
+
+        // now see if we still clip with the refined range
+        if ( ( contribs->n0 + stop_range ) > row_width )
+        {
+          int new_n0 = row_width - stop_range;
+          int num = contribs->n1 - contribs->n0 + 1;
+          int backup = contribs->n0 - new_n0;
+          float * from_co = coeffs + num - 1;
+          float * to_co = from_co + backup;
+
+          STBIR_ASSERT( ( new_n0 >= 0 ) && ( new_n0 < contribs->n0 ) );
+
+          // move the coeffs over
+          while( num )
+          {
+            *to_co-- = *from_co--;
+            --num;
+          }
+          // zero new positions
+          while ( to_co >= coeffs )
+            *to_co-- = 0;
+          // set new start point
+          contribs->n0 = new_n0;
+          if ( widest > 12 )
+          {
+            int mod;
+
+            // how far will be read in the n_coeff loop (which depends on the widest count mod4);
+            mod = widest & 3;
+            stop_range = ( ( ( contribs->n1 - contribs->n0 + 1 ) - mod + 3 ) & ~3 ) + mod; 
+
+            // the n_coeff loops do a minimum amount of coeffs, so factor that in!
+            if ( stop_range < ( 8 + mod ) ) stop_range = 8 + mod;
+          }
+        }
+      }
+      --contribs;
+      coeffs -= widest;
+    }
+  }
+
+  return widest;
+  #undef STBIR_MOVE_1
+  #undef STBIR_MOVE_2
+  #undef STBIR_MOVE_4
+}
+
+static void stbir__calculate_filters( stbir__sampler * samp, stbir__sampler * other_axis_for_pivot, void * user_data STBIR_ONLY_PROFILE_BUILD_GET_INFO )
+{
+  int n;
+  float scale = samp->scale_info.scale;
+  stbir__kernel_callback * kernel = samp->filter_kernel;
+  stbir__support_callback * support = samp->filter_support;
+  float inv_scale = samp->scale_info.inv_scale;
+  int input_full_size = samp->scale_info.input_full_size;
+  int gather_num_contributors = samp->num_contributors;
+  stbir__contributors* gather_contributors = samp->contributors;
+  float * gather_coeffs = samp->coefficients; 
+  int gather_coefficient_width = samp->coefficient_width;
+
+  switch ( samp->is_gather )
+  {
+    case 1: // gather upsample
+    {
+      float out_pixels_radius = support(inv_scale,user_data) * scale;
+
+      stbir__calculate_coefficients_for_gather_upsample( out_pixels_radius, kernel, &samp->scale_info, gather_num_contributors, gather_contributors, gather_coeffs, gather_coefficient_width, samp->edge, user_data );
+
+      STBIR_PROFILE_BUILD_START( cleanup );
+      stbir__cleanup_gathered_coefficients( samp->edge, &samp->extent_info, &samp->scale_info, gather_num_contributors, gather_contributors, gather_coeffs, gather_coefficient_width );
+      STBIR_PROFILE_BUILD_END( cleanup );
+    }
+    break;
+
+    case 0: // scatter downsample (only on vertical)
+    case 2: // gather downsample  
+    {
+      float in_pixels_radius = support(scale,user_data) * inv_scale;
+      int filter_pixel_margin = samp->filter_pixel_margin;
+      int input_end = input_full_size + filter_pixel_margin;
+      
+      // if this is a scatter, we do a downsample gather to get the coeffs, and then pivot after
+      if ( !samp->is_gather )
+      {
+        // check if we are using the same gather downsample on the horizontal as this vertical, 
+        //   if so, then we don't have to generate them, we can just pivot from the horizontal.
+        if ( other_axis_for_pivot )
+        {
+          gather_contributors = other_axis_for_pivot->contributors;
+          gather_coeffs = other_axis_for_pivot->coefficients;
+          gather_coefficient_width = other_axis_for_pivot->coefficient_width;
+          gather_num_contributors = other_axis_for_pivot->num_contributors;
+          samp->extent_info.lowest = other_axis_for_pivot->extent_info.lowest;
+          samp->extent_info.highest = other_axis_for_pivot->extent_info.highest;
+          samp->extent_info.widest = other_axis_for_pivot->extent_info.widest;
+          goto jump_right_to_pivot;
+        }
+
+        gather_contributors = samp->gather_prescatter_contributors;
+        gather_coeffs = samp->gather_prescatter_coefficients;
+        gather_coefficient_width = samp->gather_prescatter_coefficient_width;
+        gather_num_contributors = samp->gather_prescatter_num_contributors;
+      }
+
+      stbir__calculate_coefficients_for_gather_downsample( -filter_pixel_margin, input_end, in_pixels_radius, kernel, &samp->scale_info, gather_coefficient_width, gather_num_contributors, gather_contributors, gather_coeffs, user_data );
+
+      STBIR_PROFILE_BUILD_START( cleanup );
+      stbir__cleanup_gathered_coefficients( samp->edge, &samp->extent_info, &samp->scale_info, gather_num_contributors, gather_contributors, gather_coeffs, gather_coefficient_width );
+      STBIR_PROFILE_BUILD_END( cleanup );
+
+      if ( !samp->is_gather )
+      {
+        // if this is a scatter (vertical only), then we need to pivot the coeffs
+        stbir__contributors * scatter_contributors;
+        int highest_set;
+
+        jump_right_to_pivot:
+
+        STBIR_PROFILE_BUILD_START( pivot );
+
+        highest_set = (-filter_pixel_margin) - 1;
+        for (n = 0; n < gather_num_contributors; n++)
+        {
+          int k;
+          int gn0 = gather_contributors->n0, gn1 = gather_contributors->n1;
+          int scatter_coefficient_width = samp->coefficient_width;
+          float * scatter_coeffs = samp->coefficients + ( gn0 + filter_pixel_margin ) * scatter_coefficient_width;
+          float * g_coeffs = gather_coeffs;
+          scatter_contributors = samp->contributors + ( gn0 + filter_pixel_margin );
+          
+          for (k = gn0 ; k <= gn1 ; k++ )
+          {
+            float gc = *g_coeffs++;
+            if ( ( k > highest_set ) || ( scatter_contributors->n0 > scatter_contributors->n1 ) )
+            {
+              {
+                // if we are skipping over several contributors, we need to clear the skipped ones
+                stbir__contributors * clear_contributors = samp->contributors + ( highest_set + filter_pixel_margin + 1);
+                while ( clear_contributors < scatter_contributors )
+                {
+                  clear_contributors->n0 = 0; 
+                  clear_contributors->n1 = -1;
+                  ++clear_contributors;
+                }
+              }
+              scatter_contributors->n0 = n;
+              scatter_contributors->n1 = n;
+              scatter_coeffs[0]  = gc;
+              highest_set = k;
+            }
+            else
+            {
+              stbir__insert_coeff( scatter_contributors, scatter_coeffs, n, gc );
+            }
+            ++scatter_contributors;
+            scatter_coeffs += scatter_coefficient_width;
+          }
+
+          ++gather_contributors;
+          gather_coeffs += gather_coefficient_width;
+        }
+
+        // now clear any unset contribs
+        {
+          stbir__contributors * clear_contributors = samp->contributors + ( highest_set + filter_pixel_margin + 1);
+          stbir__contributors * end_contributors = samp->contributors + samp->num_contributors;
+          while ( clear_contributors < end_contributors )
+          {
+            clear_contributors->n0 = 0;
+            clear_contributors->n1 = -1;
+            ++clear_contributors;
+          }
+        }
+
+        STBIR_PROFILE_BUILD_END( pivot );
+      }
+    }
+    break;
+  }
+}
+
+
+//========================================================================================================
+// scanline decoders and encoders
+
+#define stbir__coder_min_num 1
+#define STB_IMAGE_RESIZE_DO_CODERS
+#include STBIR__HEADER_FILENAME
+
+#define stbir__decode_suffix BGRA
+#define stbir__decode_swizzle
+#define stbir__decode_order0  2 
+#define stbir__decode_order1  1
+#define stbir__decode_order2  0
+#define stbir__decode_order3  3
+#define stbir__encode_order0  2 
+#define stbir__encode_order1  1
+#define stbir__encode_order2  0
+#define stbir__encode_order3  3
+#define stbir__coder_min_num 4
+#define STB_IMAGE_RESIZE_DO_CODERS
+#include STBIR__HEADER_FILENAME
+
+#define stbir__decode_suffix ARGB
+#define stbir__decode_swizzle
+#define stbir__decode_order0  1 
+#define stbir__decode_order1  2
+#define stbir__decode_order2  3
+#define stbir__decode_order3  0
+#define stbir__encode_order0  3 
+#define stbir__encode_order1  0
+#define stbir__encode_order2  1
+#define stbir__encode_order3  2
+#define stbir__coder_min_num 4
+#define STB_IMAGE_RESIZE_DO_CODERS
+#include STBIR__HEADER_FILENAME
+
+#define stbir__decode_suffix ABGR
+#define stbir__decode_swizzle
+#define stbir__decode_order0  3 
+#define stbir__decode_order1  2
+#define stbir__decode_order2  1
+#define stbir__decode_order3  0
+#define stbir__encode_order0  3 
+#define stbir__encode_order1  2
+#define stbir__encode_order2  1
+#define stbir__encode_order3  0
+#define stbir__coder_min_num 4
+#define STB_IMAGE_RESIZE_DO_CODERS
+#include STBIR__HEADER_FILENAME
+
+#define stbir__decode_suffix AR
+#define stbir__decode_swizzle
+#define stbir__decode_order0  1 
+#define stbir__decode_order1  0 
+#define stbir__decode_order2  3
+#define stbir__decode_order3  2
+#define stbir__encode_order0  1 
+#define stbir__encode_order1  0 
+#define stbir__encode_order2  3
+#define stbir__encode_order3  2
+#define stbir__coder_min_num 2
+#define STB_IMAGE_RESIZE_DO_CODERS
+#include STBIR__HEADER_FILENAME
+
+
+// fancy alpha means we expand to keep both premultipied and non-premultiplied color channels
+static void stbir__fancy_alpha_weight_4ch( float * out_buffer, int width_times_channels )
+{
+  float STBIR_STREAMOUT_PTR(*) out = out_buffer;
+  float const * end_decode = out_buffer + ( width_times_channels / 4 ) * 7;  // decode buffer aligned to end of out_buffer
+  float STBIR_STREAMOUT_PTR(*) decode = (float*)end_decode - width_times_channels;
+
+  // fancy alpha is stored internally as R G B A Rpm Gpm Bpm
+
+  #ifdef STBIR_SIMD
+  
+  #ifdef STBIR_SIMD8
+  decode += 16;
+  while ( decode <= end_decode )
+  {
+    stbir__simdf8 d0,d1,a0,a1,p0,p1;
+    STBIR_NO_UNROLL(decode);
+    stbir__simdf8_load( d0, decode-16 );
+    stbir__simdf8_load( d1, decode-16+8 );
+    stbir__simdf8_0123to33333333( a0, d0 );
+    stbir__simdf8_0123to33333333( a1, d1 );
+    stbir__simdf8_mult( p0, a0, d0 );
+    stbir__simdf8_mult( p1, a1, d1 );
+    stbir__simdf8_bot4s( a0, d0, p0 );
+    stbir__simdf8_bot4s( a1, d1, p1 );
+    stbir__simdf8_top4s( d0, d0, p0 );
+    stbir__simdf8_top4s( d1, d1, p1 );
+    stbir__simdf8_store ( out, a0 );
+    stbir__simdf8_store ( out+7, d0 );
+    stbir__simdf8_store ( out+14, a1 );
+    stbir__simdf8_store ( out+21, d1 );
+    decode += 16;
+    out += 28;
+  }
+  decode -= 16;
+  #else  
+  decode += 8;
+  while ( decode <= end_decode )
+  {
+    stbir__simdf d0,a0,d1,a1,p0,p1;
+    STBIR_NO_UNROLL(decode);
+    stbir__simdf_load( d0, decode-8 );
+    stbir__simdf_load( d1, decode-8+4 );
+    stbir__simdf_0123to3333( a0, d0 );
+    stbir__simdf_0123to3333( a1, d1 );
+    stbir__simdf_mult( p0, a0, d0 );
+    stbir__simdf_mult( p1, a1, d1 );
+    stbir__simdf_store ( out, d0 );
+    stbir__simdf_store ( out+4, p0 );
+    stbir__simdf_store ( out+7, d1 );
+    stbir__simdf_store ( out+7+4, p1 );
+    decode += 8;
+    out += 14;
+  }
+  decode -= 8;
+  #endif
+
+  // might be one last odd pixel
+  #ifdef STBIR_SIMD8
+  while ( decode < end_decode )
+  #else
+  if ( decode < end_decode )
+  #endif
+  {
+    stbir__simdf d,a,p;
+    stbir__simdf_load( d, decode );
+    stbir__simdf_0123to3333( a, d );
+    stbir__simdf_mult( p, a, d );
+    stbir__simdf_store ( out, d );
+    stbir__simdf_store ( out+4, p );
+    decode += 4;
+    out += 7;
+  }
+
+  #else
+
+  while( decode < end_decode )
+  {
+    float r = decode[0], g = decode[1], b = decode[2], alpha = decode[3];
+    out[0] = r;
+    out[1] = g;
+    out[2] = b;
+    out[3] = alpha;
+    out[4] = r * alpha;
+    out[5] = g * alpha;
+    out[6] = b * alpha;
+    out += 7;
+    decode += 4;
+  }
+
+  #endif
+}
+
+static void stbir__fancy_alpha_weight_2ch( float * out_buffer, int width_times_channels )
+{
+  float STBIR_STREAMOUT_PTR(*) out = out_buffer;
+  float const * end_decode = out_buffer + ( width_times_channels / 2 ) * 3;
+  float STBIR_STREAMOUT_PTR(*) decode = (float*)end_decode - width_times_channels;
+
+  //  for fancy alpha, turns into: [X A Xpm][X A Xpm],etc
+
+  #ifdef STBIR_SIMD
+
+  decode += 8;
+  if ( decode <= end_decode )
+  {
+    do {
+      #ifdef STBIR_SIMD8
+      stbir__simdf8 d0,a0,p0;
+      STBIR_NO_UNROLL(decode);
+      stbir__simdf8_load( d0, decode-8 );
+      stbir__simdf8_0123to11331133( p0, d0 );
+      stbir__simdf8_0123to00220022( a0, d0 );
+      stbir__simdf8_mult( p0, p0, a0 );
+ 
+      stbir__simdf_store2( out, stbir__if_simdf8_cast_to_simdf4( d0 ) );
+      stbir__simdf_store( out+2, stbir__if_simdf8_cast_to_simdf4( p0 ) );
+      stbir__simdf_store2h( out+3, stbir__if_simdf8_cast_to_simdf4( d0 ) );
+      
+      stbir__simdf_store2( out+6, stbir__simdf8_gettop4( d0 ) );
+      stbir__simdf_store( out+8, stbir__simdf8_gettop4( p0 ) );
+      stbir__simdf_store2h( out+9, stbir__simdf8_gettop4( d0 ) );
+      #else
+      stbir__simdf d0,a0,d1,a1,p0,p1;
+      STBIR_NO_UNROLL(decode);
+      stbir__simdf_load( d0, decode-8 );
+      stbir__simdf_load( d1, decode-8+4 );
+      stbir__simdf_0123to1133( p0, d0 );
+      stbir__simdf_0123to1133( p1, d1 );
+      stbir__simdf_0123to0022( a0, d0 );
+      stbir__simdf_0123to0022( a1, d1 );
+      stbir__simdf_mult( p0, p0, a0 );
+      stbir__simdf_mult( p1, p1, a1 );
+
+      stbir__simdf_store2( out, d0 );
+      stbir__simdf_store( out+2, p0 );
+      stbir__simdf_store2h( out+3, d0 );
+
+      stbir__simdf_store2( out+6, d1 );
+      stbir__simdf_store( out+8, p1 );
+      stbir__simdf_store2h( out+9, d1 );
+      #endif
+      decode += 8;
+      out += 12;
+    } while ( decode <= end_decode );
+  }
+  decode -= 8;
+  #endif
+
+  while( decode < end_decode )
+  {
+    float x = decode[0], y = decode[1];
+    STBIR_SIMD_NO_UNROLL(decode);
+    out[0] = x;
+    out[1] = y;
+    out[2] = x * y;
+    out += 3;
+    decode += 2;
+  }
+}
+
+static void stbir__fancy_alpha_unweight_4ch( float * encode_buffer, int width_times_channels )
+{
+  float STBIR_SIMD_STREAMOUT_PTR(*) encode = encode_buffer;
+  float STBIR_SIMD_STREAMOUT_PTR(*) input = encode_buffer;
+  float const * end_output = encode_buffer + width_times_channels;
+
+  // fancy RGBA is stored internally as R G B A Rpm Gpm Bpm
+
+  do {
+    float alpha = input[3];
+#ifdef STBIR_SIMD
+    stbir__simdf i,ia;
+    STBIR_SIMD_NO_UNROLL(encode);
+    if ( alpha < stbir__small_float )
+    {
+      stbir__simdf_load( i, input );
+      stbir__simdf_store( encode, i );
+    }
+    else
+    {
+      stbir__simdf_load1frep4( ia, 1.0f / alpha );
+      stbir__simdf_load( i, input+4 );
+      stbir__simdf_mult( i, i, ia );
+      stbir__simdf_store( encode, i );
+      encode[3] = alpha;
+    }
+#else
+    if ( alpha < stbir__small_float )
+    {
+      encode[0] = input[0];
+      encode[1] = input[1];
+      encode[2] = input[2];
+    }
+    else
+    {
+      float ialpha = 1.0f / alpha;
+      encode[0] = input[4] * ialpha;
+      encode[1] = input[5] * ialpha;
+      encode[2] = input[6] * ialpha;
+    }
+    encode[3] = alpha;
+#endif
+
+    input += 7;
+    encode += 4;
+  } while ( encode < end_output );
+}
+
+//  format: [X A Xpm][X A Xpm] etc
+static void stbir__fancy_alpha_unweight_2ch( float * encode_buffer, int width_times_channels )
+{
+  float STBIR_SIMD_STREAMOUT_PTR(*) encode = encode_buffer;
+  float STBIR_SIMD_STREAMOUT_PTR(*) input = encode_buffer;
+  float const * end_output = encode_buffer + width_times_channels;
+
+  do {
+    float alpha = input[1];
+    encode[0] = input[0];
+    if ( alpha >= stbir__small_float )
+      encode[0] = input[2] / alpha;
+    encode[1] = alpha;
+
+    input += 3;
+    encode += 2;
+  } while ( encode < end_output );
+}
+
+static void stbir__simple_alpha_weight_4ch( float * decode_buffer, int width_times_channels )
+{
+  float STBIR_STREAMOUT_PTR(*) decode = decode_buffer;
+  float const * end_decode = decode_buffer + width_times_channels;
+
+  #ifdef STBIR_SIMD
+  {
+    decode += 2 * stbir__simdfX_float_count;
+    while ( decode <= end_decode )
+    {
+      stbir__simdfX d0,a0,d1,a1;
+      STBIR_NO_UNROLL(decode);
+      stbir__simdfX_load( d0, decode-2*stbir__simdfX_float_count );
+      stbir__simdfX_load( d1, decode-2*stbir__simdfX_float_count+stbir__simdfX_float_count );
+      stbir__simdfX_aaa1( a0, d0, STBIR_onesX );
+      stbir__simdfX_aaa1( a1, d1, STBIR_onesX );
+      stbir__simdfX_mult( d0, d0, a0 );
+      stbir__simdfX_mult( d1, d1, a1 );
+      stbir__simdfX_store ( decode-2*stbir__simdfX_float_count, d0 );
+      stbir__simdfX_store ( decode-2*stbir__simdfX_float_count+stbir__simdfX_float_count, d1 );
+      decode += 2 * stbir__simdfX_float_count;
+    }
+    decode -= 2 * stbir__simdfX_float_count;
+
+    // few last pixels remnants
+    #ifdef STBIR_SIMD8
+    while ( decode < end_decode )
+    #else
+    if ( decode < end_decode )
+    #endif
+    {
+      stbir__simdf d,a;
+      stbir__simdf_load( d, decode );
+      stbir__simdf_aaa1( a, d, STBIR__CONSTF(STBIR_ones) );
+      stbir__simdf_mult( d, d, a );
+      stbir__simdf_store ( decode, d );
+      decode += 4;
+    }
+  }
+
+  #else
+
+  while( decode < end_decode )
+  {
+    float alpha = decode[3];
+    decode[0] *= alpha;
+    decode[1] *= alpha;
+    decode[2] *= alpha;
+    decode += 4;
+  }
+
+  #endif
+}
+
+static void stbir__simple_alpha_weight_2ch( float * decode_buffer, int width_times_channels )
+{
+  float STBIR_STREAMOUT_PTR(*) decode = decode_buffer;
+  float const * end_decode = decode_buffer + width_times_channels;
+
+  #ifdef STBIR_SIMD
+  decode += 2 * stbir__simdfX_float_count;
+  while ( decode <= end_decode )
+  {
+    stbir__simdfX d0,a0,d1,a1;
+    STBIR_NO_UNROLL(decode);
+    stbir__simdfX_load( d0, decode-2*stbir__simdfX_float_count );
+    stbir__simdfX_load( d1, decode-2*stbir__simdfX_float_count+stbir__simdfX_float_count );
+    stbir__simdfX_a1a1( a0, d0, STBIR_onesX );
+    stbir__simdfX_a1a1( a1, d1, STBIR_onesX );
+    stbir__simdfX_mult( d0, d0, a0 );
+    stbir__simdfX_mult( d1, d1, a1 );
+    stbir__simdfX_store ( decode-2*stbir__simdfX_float_count, d0 );
+    stbir__simdfX_store ( decode-2*stbir__simdfX_float_count+stbir__simdfX_float_count, d1 );
+    decode += 2 * stbir__simdfX_float_count;
+  }
+  decode -= 2 * stbir__simdfX_float_count;
+  #endif
+
+  while( decode < end_decode )
+  {
+    float alpha = decode[1];
+    STBIR_SIMD_NO_UNROLL(decode);
+    decode[0] *= alpha;
+    decode += 2;
+  }
+}
+
+static void stbir__simple_alpha_unweight_4ch( float * encode_buffer, int width_times_channels )
+{
+  float STBIR_SIMD_STREAMOUT_PTR(*) encode = encode_buffer;
+  float const * end_output = encode_buffer + width_times_channels;
+
+  do {
+    float alpha = encode[3];
+
+#ifdef STBIR_SIMD
+    stbir__simdf i,ia;
+    STBIR_SIMD_NO_UNROLL(encode);
+    if ( alpha >= stbir__small_float )
+    {
+      stbir__simdf_load1frep4( ia, 1.0f / alpha );
+      stbir__simdf_load( i, encode );
+      stbir__simdf_mult( i, i, ia );
+      stbir__simdf_store( encode, i );
+      encode[3] = alpha;
+    }
+#else
+    if ( alpha >= stbir__small_float )
+    {
+      float ialpha = 1.0f / alpha;
+      encode[0] *= ialpha;
+      encode[1] *= ialpha;
+      encode[2] *= ialpha;
+    }
+#endif
+    encode += 4;
+  } while ( encode < end_output );
+}
+
+static void stbir__simple_alpha_unweight_2ch( float * encode_buffer, int width_times_channels )
+{
+  float STBIR_SIMD_STREAMOUT_PTR(*) encode = encode_buffer;
+  float const * end_output = encode_buffer + width_times_channels;
+
+  do {
+    float alpha = encode[1];
+    if ( alpha >= stbir__small_float )
+      encode[0] /= alpha;
+    encode += 2;
+  } while ( encode < end_output );
+}
+
+
+// only used in RGB->BGR or BGR->RGB
+static void stbir__simple_flip_3ch( float * decode_buffer, int width_times_channels )
+{
+  float STBIR_STREAMOUT_PTR(*) decode = decode_buffer;
+  float const * end_decode = decode_buffer + width_times_channels;
+
+  decode += 12;
+  while( decode <= end_decode )
+  {
+    float t0,t1,t2,t3;
+    STBIR_NO_UNROLL(decode);
+    t0 = decode[0]; t1 = decode[3]; t2 = decode[6]; t3 = decode[9];
+    decode[0] = decode[2]; decode[3] = decode[5]; decode[6] = decode[8]; decode[9] = decode[11];
+    decode[2] = t0; decode[5] = t1; decode[8] = t2; decode[11] = t3;
+    decode += 12;
+  }
+  decode -= 12;
+
+  while( decode < end_decode )
+  {
+    float t = decode[0];
+    STBIR_NO_UNROLL(decode);
+    decode[0] = decode[2];
+    decode[2] = t;
+    decode += 3;
+  }
+}
+
+
+
+static void stbir__decode_scanline(stbir__info const * stbir_info, int n, float * output_buffer STBIR_ONLY_PROFILE_GET_SPLIT_INFO )
+{
+  int channels = stbir_info->channels;
+  int effective_channels = stbir_info->effective_channels;
+  int input_sample_in_bytes = stbir__type_size[stbir_info->input_type] * channels;
+  stbir_edge edge_horizontal = stbir_info->horizontal.edge;
+  stbir_edge edge_vertical = stbir_info->vertical.edge;
+  int row = stbir__edge_wrap(edge_vertical, n, stbir_info->vertical.scale_info.input_full_size);
+  const void* input_plane_data = ( (char *) stbir_info->input_data ) + (ptrdiff_t)row * (ptrdiff_t) stbir_info->input_stride_bytes;
+  stbir__span const * spans = stbir_info->scanline_extents.spans;
+  float* full_decode_buffer = output_buffer - stbir_info->scanline_extents.conservative.n0 * effective_channels;
+
+  // if we are on edge_zero, and we get in here with an out of bounds n, then the calculate filters has failed
+  STBIR_ASSERT( !(edge_vertical == STBIR_EDGE_ZERO && (n < 0 || n >= stbir_info->vertical.scale_info.input_full_size)) );
+
+  do 
+  {
+    float * decode_buffer;
+    void const * input_data;
+    float * end_decode;
+    int width_times_channels;
+    int width;
+
+    if ( spans->n1 < spans->n0 )    
+      break;
+
+    width = spans->n1 + 1 - spans->n0;
+    decode_buffer = full_decode_buffer + spans->n0 * effective_channels;
+    end_decode = full_decode_buffer + ( spans->n1 + 1 ) * effective_channels;
+    width_times_channels = width * channels;
+
+    // read directly out of input plane by default
+    input_data = ( (char*)input_plane_data ) + spans->pixel_offset_for_input * input_sample_in_bytes;
+
+    // if we have an input callback, call it to get the input data
+    if ( stbir_info->in_pixels_cb )
+    {
+      // call the callback with a temp buffer (that they can choose to use or not).  the temp is just right aligned memory in the decode_buffer itself
+      input_data = stbir_info->in_pixels_cb( ( (char*) end_decode ) - ( width * input_sample_in_bytes ), input_plane_data, width, spans->pixel_offset_for_input, row, stbir_info->user_data );
+    }
+    
+    STBIR_PROFILE_START( decode );
+    // convert the pixels info the float decode_buffer, (we index from end_decode, so that when channels<effective_channels, we are right justified in the buffer)
+    stbir_info->decode_pixels( (float*)end_decode - width_times_channels, width_times_channels, input_data );
+    STBIR_PROFILE_END( decode );
+
+    if (stbir_info->alpha_weight)
+    {
+      STBIR_PROFILE_START( alpha );
+      stbir_info->alpha_weight( decode_buffer, width_times_channels );
+      STBIR_PROFILE_END( alpha );
+    }
+
+    ++spans;
+  } while ( spans <= ( &stbir_info->scanline_extents.spans[1] ) );
+
+  // handle the edge_wrap filter (all other types are handled back out at the calculate_filter stage)
+  // basically the idea here is that if we have the whole scanline in memory, we don't redecode the
+  //   wrapped edge pixels, and instead just memcpy them from the scanline into the edge positions
+  if ( ( edge_horizontal == STBIR_EDGE_WRAP ) && ( stbir_info->scanline_extents.edge_sizes[0] | stbir_info->scanline_extents.edge_sizes[1] ) )
+  {
+    // this code only runs if we're in edge_wrap, and we're doing the entire scanline
+    int e, start_x[2];
+    int input_full_size = stbir_info->horizontal.scale_info.input_full_size;
+    
+    start_x[0] = -stbir_info->scanline_extents.edge_sizes[0];  // left edge start x
+    start_x[1] =  input_full_size;                             // right edge
+
+    for( e = 0; e < 2 ; e++ )
+    {
+      // do each margin
+      int margin = stbir_info->scanline_extents.edge_sizes[e];
+      if ( margin )
+      {
+        int x = start_x[e];
+        float * marg = full_decode_buffer + x * effective_channels;
+        float const * src = full_decode_buffer + stbir__edge_wrap(edge_horizontal, x, input_full_size) * effective_channels;
+        STBIR_MEMCPY( marg, src, margin * effective_channels * sizeof(float) );
+      }
+    }
+  }
+}
+
+
+//=================
+// Do 1 channel horizontal routines
+
+#ifdef STBIR_SIMD
+
+#define stbir__1_coeff_only()          \
+    stbir__simdf tot,c;                \
+    STBIR_SIMD_NO_UNROLL(decode);      \
+    stbir__simdf_load1( c, hc );       \
+    stbir__simdf_mult1_mem( tot, c, decode ); 
+
+#define stbir__2_coeff_only()          \
+    stbir__simdf tot,c,d;              \
+    STBIR_SIMD_NO_UNROLL(decode);      \
+    stbir__simdf_load2z( c, hc );      \
+    stbir__simdf_load2( d, decode );   \
+    stbir__simdf_mult( tot, c, d );    \
+    stbir__simdf_0123to1230( c, tot ); \
+    stbir__simdf_add1( tot, tot, c );          
+
+#define stbir__3_coeff_only()                  \
+    stbir__simdf tot,c,t;                      \
+    STBIR_SIMD_NO_UNROLL(decode);              \
+    stbir__simdf_load( c, hc );                \
+    stbir__simdf_mult_mem( tot, c, decode );   \
+    stbir__simdf_0123to1230( c, tot );         \
+    stbir__simdf_0123to2301( t, tot );         \
+    stbir__simdf_add1( tot, tot, c );          \
+    stbir__simdf_add1( tot, tot, t );    
+
+#define stbir__store_output_tiny()                \
+    stbir__simdf_store1( output, tot );           \
+    horizontal_coefficients += coefficient_width; \
+    ++horizontal_contributors;                    \
+    output += 1;
+
+#define stbir__4_coeff_start()                 \
+    stbir__simdf tot,c;                        \
+    STBIR_SIMD_NO_UNROLL(decode);              \
+    stbir__simdf_load( c, hc );                \
+    stbir__simdf_mult_mem( tot, c, decode );   \
+
+#define stbir__4_coeff_continue_from_4( ofs )  \
+    STBIR_SIMD_NO_UNROLL(decode);              \
+    stbir__simdf_load( c, hc + (ofs) );        \
+    stbir__simdf_madd_mem( tot, tot, c, decode+(ofs) ); 
+
+#define stbir__1_coeff_remnant( ofs )          \
+    { stbir__simdf d;                          \
+    stbir__simdf_load1z( c, hc + (ofs) );      \
+    stbir__simdf_load1( d, decode + (ofs) );   \
+    stbir__simdf_madd( tot, tot, d, c ); }
+
+#define stbir__2_coeff_remnant( ofs )          \
+    { stbir__simdf d;                          \
+    stbir__simdf_load2z( c, hc+(ofs) );        \
+    stbir__simdf_load2( d, decode+(ofs) );     \
+    stbir__simdf_madd( tot, tot, d, c ); }   
+
+#define stbir__3_coeff_setup()                 \
+    stbir__simdf mask;                         \
+    stbir__simdf_load( mask, STBIR_mask + 3 );
+
+#define stbir__3_coeff_remnant( ofs )                  \
+    stbir__simdf_load( c, hc+(ofs) );                  \
+    stbir__simdf_and( c, c, mask );                    \
+    stbir__simdf_madd_mem( tot, tot, c, decode+(ofs) );
+
+#define stbir__store_output()                     \
+    stbir__simdf_0123to2301( c, tot );            \
+    stbir__simdf_add( tot, tot, c );              \
+    stbir__simdf_0123to1230( c, tot );            \
+    stbir__simdf_add1( tot, tot, c );             \
+    stbir__simdf_store1( output, tot );           \
+    horizontal_coefficients += coefficient_width; \
+    ++horizontal_contributors;                    \
+    output += 1;
+
+#else
+
+#define stbir__1_coeff_only()  \
+    float tot;                 \
+    tot = decode[0]*hc[0];     
+
+#define stbir__2_coeff_only()  \
+    float tot;                 \
+    tot = decode[0] * hc[0];   \
+    tot += decode[1] * hc[1];    
+
+#define stbir__3_coeff_only()  \
+    float tot;                 \
+    tot = decode[0] * hc[0];   \
+    tot += decode[1] * hc[1];  \
+    tot += decode[2] * hc[2];    
+
+#define stbir__store_output_tiny()                \
+    output[0] = tot;                              \
+    horizontal_coefficients += coefficient_width; \
+    ++horizontal_contributors;                    \
+    output += 1;
+
+#define stbir__4_coeff_start()  \
+    float tot0,tot1,tot2,tot3;  \
+    tot0 = decode[0] * hc[0];   \
+    tot1 = decode[1] * hc[1];   \
+    tot2 = decode[2] * hc[2];   \
+    tot3 = decode[3] * hc[3];     
+
+#define stbir__4_coeff_continue_from_4( ofs )  \
+    tot0 += decode[0+(ofs)] * hc[0+(ofs)];     \
+    tot1 += decode[1+(ofs)] * hc[1+(ofs)];     \
+    tot2 += decode[2+(ofs)] * hc[2+(ofs)];     \
+    tot3 += decode[3+(ofs)] * hc[3+(ofs)];     
+
+#define stbir__1_coeff_remnant( ofs )        \
+    tot0 += decode[0+(ofs)] * hc[0+(ofs)];   
+
+#define stbir__2_coeff_remnant( ofs )        \
+    tot0 += decode[0+(ofs)] * hc[0+(ofs)];   \
+    tot1 += decode[1+(ofs)] * hc[1+(ofs)];   \
+
+#define stbir__3_coeff_remnant( ofs )        \
+    tot0 += decode[0+(ofs)] * hc[0+(ofs)];   \
+    tot1 += decode[1+(ofs)] * hc[1+(ofs)];   \
+    tot2 += decode[2+(ofs)] * hc[2+(ofs)];   
+
+#define stbir__store_output()                     \
+    output[0] = (tot0+tot2)+(tot1+tot3);          \
+    horizontal_coefficients += coefficient_width; \
+    ++horizontal_contributors;                    \
+    output += 1;
+
+#endif  
+
+#define STBIR__horizontal_channels 1
+#define STB_IMAGE_RESIZE_DO_HORIZONTALS
+#include STBIR__HEADER_FILENAME
+
+
+//=================
+// Do 2 channel horizontal routines
+
+#ifdef STBIR_SIMD
+
+#define stbir__1_coeff_only()         \
+    stbir__simdf tot,c,d;             \
+    STBIR_SIMD_NO_UNROLL(decode);     \
+    stbir__simdf_load1z( c, hc );     \
+    stbir__simdf_0123to0011( c, c );  \
+    stbir__simdf_load2( d, decode );  \
+    stbir__simdf_mult( tot, d, c ); 
+
+#define stbir__2_coeff_only()         \
+    stbir__simdf tot,c;               \
+    STBIR_SIMD_NO_UNROLL(decode);     \
+    stbir__simdf_load2( c, hc );      \
+    stbir__simdf_0123to0011( c, c );  \
+    stbir__simdf_mult_mem( tot, c, decode ); 
+
+#define stbir__3_coeff_only()                \
+    stbir__simdf tot,c,cs,d;                 \
+    STBIR_SIMD_NO_UNROLL(decode);            \
+    stbir__simdf_load( cs, hc );             \
+    stbir__simdf_0123to0011( c, cs );        \
+    stbir__simdf_mult_mem( tot, c, decode ); \
+    stbir__simdf_0123to2222( c, cs );        \
+    stbir__simdf_load2z( d, decode+4 );      \
+    stbir__simdf_madd( tot, tot, d, c );   
+
+#define stbir__store_output_tiny()                \
+    stbir__simdf_0123to2301( c, tot );            \
+    stbir__simdf_add( tot, tot, c );              \
+    stbir__simdf_store2( output, tot );           \
+    horizontal_coefficients += coefficient_width; \
+    ++horizontal_contributors;                    \
+    output += 2;
+
+#ifdef STBIR_SIMD8
+
+#define stbir__4_coeff_start()                    \
+    stbir__simdf8 tot0,c,cs;                      \
+    STBIR_SIMD_NO_UNROLL(decode);                 \
+    stbir__simdf8_load4b( cs, hc );               \
+    stbir__simdf8_0123to00112233( c, cs );        \
+    stbir__simdf8_mult_mem( tot0, c, decode );
+
+#define stbir__4_coeff_continue_from_4( ofs )        \
+    STBIR_SIMD_NO_UNROLL(decode);                    \
+    stbir__simdf8_load4b( cs, hc + (ofs) );          \
+    stbir__simdf8_0123to00112233( c, cs );           \
+    stbir__simdf8_madd_mem( tot0, tot0, c, decode+(ofs)*2 ); 
+
+#define stbir__1_coeff_remnant( ofs )                \
+    { stbir__simdf t;                                \
+    stbir__simdf_load1z( t, hc + (ofs) );            \
+    stbir__simdf_0123to0011( t, t );                 \
+    stbir__simdf_mult_mem( t, t, decode+(ofs)*2 );   \
+    stbir__simdf8_add4( tot0, tot0, t ); }
+
+#define stbir__2_coeff_remnant( ofs )                \
+    { stbir__simdf t;                                \
+    stbir__simdf_load2( t, hc + (ofs) );             \
+    stbir__simdf_0123to0011( t, t );                 \
+    stbir__simdf_mult_mem( t, t, decode+(ofs)*2 );   \
+    stbir__simdf8_add4( tot0, tot0, t ); }
+
+#define stbir__3_coeff_remnant( ofs )                \
+    { stbir__simdf8 d;                               \
+    stbir__simdf8_load4b( cs, hc + (ofs) );          \
+    stbir__simdf8_0123to00112233( c, cs );           \
+    stbir__simdf8_load6z( d, decode+(ofs)*2 );       \
+    stbir__simdf8_madd( tot0, tot0, c, d ); }               
+
+#define stbir__store_output()                     \
+    { stbir__simdf t,c;                           \
+    stbir__simdf8_add4halves( t, stbir__if_simdf8_cast_to_simdf4(tot0), tot0 );    \
+    stbir__simdf_0123to2301( c, t );              \
+    stbir__simdf_add( t, t, c );                  \
+    stbir__simdf_store2( output, t );             \
+    horizontal_coefficients += coefficient_width; \
+    ++horizontal_contributors;                    \
+    output += 2; }
+
+#else
+
+#define stbir__4_coeff_start()                   \
+    stbir__simdf tot0,tot1,c,cs;                 \
+    STBIR_SIMD_NO_UNROLL(decode);                \
+    stbir__simdf_load( cs, hc );                 \
+    stbir__simdf_0123to0011( c, cs );            \
+    stbir__simdf_mult_mem( tot0, c, decode );    \
+    stbir__simdf_0123to2233( c, cs );            \
+    stbir__simdf_mult_mem( tot1, c, decode+4 );   
+
+#define stbir__4_coeff_continue_from_4( ofs )                \
+    STBIR_SIMD_NO_UNROLL(decode);                            \
+    stbir__simdf_load( cs, hc + (ofs) );                     \
+    stbir__simdf_0123to0011( c, cs );                        \
+    stbir__simdf_madd_mem( tot0, tot0, c, decode+(ofs)*2 );  \
+    stbir__simdf_0123to2233( c, cs );                        \
+    stbir__simdf_madd_mem( tot1, tot1, c, decode+(ofs)*2+4 );   
+
+#define stbir__1_coeff_remnant( ofs )            \
+    { stbir__simdf d;                            \
+    stbir__simdf_load1z( cs, hc + (ofs) );       \
+    stbir__simdf_0123to0011( c, cs );            \
+    stbir__simdf_load2( d, decode + (ofs) * 2 ); \
+    stbir__simdf_madd( tot0, tot0, d, c ); }
+
+#define stbir__2_coeff_remnant( ofs )                      \
+    stbir__simdf_load2( cs, hc + (ofs) );                  \
+    stbir__simdf_0123to0011( c, cs );                      \
+    stbir__simdf_madd_mem( tot0, tot0, c, decode+(ofs)*2 );       
+
+#define stbir__3_coeff_remnant( ofs )                       \
+    { stbir__simdf d;                                       \
+    stbir__simdf_load( cs, hc + (ofs) );                    \
+    stbir__simdf_0123to0011( c, cs );                       \
+    stbir__simdf_madd_mem( tot0, tot0, c, decode+(ofs)*2 ); \
+    stbir__simdf_0123to2222( c, cs );                       \
+    stbir__simdf_load2z( d, decode + (ofs) * 2 + 4 );       \
+    stbir__simdf_madd( tot1, tot1, d, c ); }  
+
+#define stbir__store_output()                     \
+    stbir__simdf_add( tot0, tot0, tot1 );         \
+    stbir__simdf_0123to2301( c, tot0 );           \
+    stbir__simdf_add( tot0, tot0, c );            \
+    stbir__simdf_store2( output, tot0 );          \
+    horizontal_coefficients += coefficient_width; \
+    ++horizontal_contributors;                    \
+    output += 2;
+
+#endif
+
+#else
+
+#define stbir__1_coeff_only()  \
+    float tota,totb,c;         \
+    c = hc[0];                 \
+    tota = decode[0]*c;        \
+    totb = decode[1]*c;     
+
+#define stbir__2_coeff_only()  \
+    float tota,totb,c;         \
+    c = hc[0];                 \
+    tota = decode[0]*c;        \
+    totb = decode[1]*c;        \
+    c = hc[1];                 \
+    tota += decode[2]*c;       \
+    totb += decode[3]*c;     
+
+// this weird order of add matches the simd
+#define stbir__3_coeff_only()  \
+    float tota,totb,c;         \
+    c = hc[0];                 \
+    tota = decode[0]*c;        \
+    totb = decode[1]*c;        \
+    c = hc[2];                 \
+    tota += decode[4]*c;       \
+    totb += decode[5]*c;       \
+    c = hc[1];                 \
+    tota += decode[2]*c;       \
+    totb += decode[3]*c;     
+
+#define stbir__store_output_tiny()                \
+    output[0] = tota;                             \
+    output[1] = totb;                             \
+    horizontal_coefficients += coefficient_width; \
+    ++horizontal_contributors;                    \
+    output += 2;
+
+#define stbir__4_coeff_start()      \
+    float tota0,tota1,tota2,tota3,totb0,totb1,totb2,totb3,c;  \
+    c = hc[0];                      \
+    tota0 = decode[0]*c;            \
+    totb0 = decode[1]*c;            \
+    c = hc[1];                      \
+    tota1 = decode[2]*c;            \
+    totb1 = decode[3]*c;            \
+    c = hc[2];                      \
+    tota2 = decode[4]*c;            \
+    totb2 = decode[5]*c;            \
+    c = hc[3];                      \
+    tota3 = decode[6]*c;            \
+    totb3 = decode[7]*c;     
+
+#define stbir__4_coeff_continue_from_4( ofs )  \
+    c = hc[0+(ofs)];                           \
+    tota0 += decode[0+(ofs)*2]*c;              \
+    totb0 += decode[1+(ofs)*2]*c;              \
+    c = hc[1+(ofs)];                           \
+    tota1 += decode[2+(ofs)*2]*c;              \
+    totb1 += decode[3+(ofs)*2]*c;              \
+    c = hc[2+(ofs)];                           \
+    tota2 += decode[4+(ofs)*2]*c;              \
+    totb2 += decode[5+(ofs)*2]*c;              \
+    c = hc[3+(ofs)];                           \
+    tota3 += decode[6+(ofs)*2]*c;              \
+    totb3 += decode[7+(ofs)*2]*c;     
+
+#define stbir__1_coeff_remnant( ofs )  \
+    c = hc[0+(ofs)];                   \
+    tota0 += decode[0+(ofs)*2] * c;    \
+    totb0 += decode[1+(ofs)*2] * c;   
+
+#define stbir__2_coeff_remnant( ofs )  \
+    c = hc[0+(ofs)];                   \
+    tota0 += decode[0+(ofs)*2] * c;    \
+    totb0 += decode[1+(ofs)*2] * c;    \
+    c = hc[1+(ofs)];                   \
+    tota1 += decode[2+(ofs)*2] * c;    \
+    totb1 += decode[3+(ofs)*2] * c;   
+
+#define stbir__3_coeff_remnant( ofs )  \
+    c = hc[0+(ofs)];                   \
+    tota0 += decode[0+(ofs)*2] * c;    \
+    totb0 += decode[1+(ofs)*2] * c;    \
+    c = hc[1+(ofs)];                   \
+    tota1 += decode[2+(ofs)*2] * c;    \
+    totb1 += decode[3+(ofs)*2] * c;    \
+    c = hc[2+(ofs)];                   \
+    tota2 += decode[4+(ofs)*2] * c;    \
+    totb2 += decode[5+(ofs)*2] * c;    
+
+#define stbir__store_output()                     \
+    output[0] = (tota0+tota2)+(tota1+tota3);      \
+    output[1] = (totb0+totb2)+(totb1+totb3);      \
+    horizontal_coefficients += coefficient_width; \
+    ++horizontal_contributors;                    \
+    output += 2;
+
+#endif  
+
+#define STBIR__horizontal_channels 2
+#define STB_IMAGE_RESIZE_DO_HORIZONTALS
+#include STBIR__HEADER_FILENAME
+
+
+//=================
+// Do 3 channel horizontal routines
+
+#ifdef STBIR_SIMD
+
+#define stbir__1_coeff_only()         \
+    stbir__simdf tot,c,d;             \
+    STBIR_SIMD_NO_UNROLL(decode);     \
+    stbir__simdf_load1z( c, hc );     \
+    stbir__simdf_0123to0001( c, c );  \
+    stbir__simdf_load( d, decode );   \
+    stbir__simdf_mult( tot, d, c ); 
+
+#define stbir__2_coeff_only()         \
+    stbir__simdf tot,c,cs,d;          \
+    STBIR_SIMD_NO_UNROLL(decode);     \
+    stbir__simdf_load2( cs, hc );     \
+    stbir__simdf_0123to0000( c, cs ); \
+    stbir__simdf_load( d, decode );   \
+    stbir__simdf_mult( tot, d, c );   \
+    stbir__simdf_0123to1111( c, cs ); \
+    stbir__simdf_load( d, decode+3 ); \
+    stbir__simdf_madd( tot, tot, d, c ); 
+
+#define stbir__3_coeff_only()            \
+    stbir__simdf tot,c,d,cs;             \
+    STBIR_SIMD_NO_UNROLL(decode);        \
+    stbir__simdf_load( cs, hc );         \
+    stbir__simdf_0123to0000( c, cs );    \
+    stbir__simdf_load( d, decode );      \
+    stbir__simdf_mult( tot, d, c );      \
+    stbir__simdf_0123to1111( c, cs );    \
+    stbir__simdf_load( d, decode+3 );    \
+    stbir__simdf_madd( tot, tot, d, c ); \
+    stbir__simdf_0123to2222( c, cs );    \
+    stbir__simdf_load( d, decode+6 );    \
+    stbir__simdf_madd( tot, tot, d, c ); 
+
+#define stbir__store_output_tiny()                \
+    stbir__simdf_store2( output, tot );           \
+    stbir__simdf_0123to2301( tot, tot );          \
+    stbir__simdf_store1( output+2, tot );         \
+    horizontal_coefficients += coefficient_width; \
+    ++horizontal_contributors;                    \
+    output += 3;
+
+#ifdef STBIR_SIMD8
+
+// we're loading from the XXXYYY decode by -1 to get the XXXYYY into different halves of the AVX reg fyi
+#define stbir__4_coeff_start()                     \
+    stbir__simdf8 tot0,tot1,c,cs; stbir__simdf t;  \
+    STBIR_SIMD_NO_UNROLL(decode);                  \
+    stbir__simdf8_load4b( cs, hc );                \
+    stbir__simdf8_0123to00001111( c, cs );         \
+    stbir__simdf8_mult_mem( tot0, c, decode - 1 ); \
+    stbir__simdf8_0123to22223333( c, cs );         \
+    stbir__simdf8_mult_mem( tot1, c, decode+6 - 1 );    
+
+#define stbir__4_coeff_continue_from_4( ofs )      \
+    STBIR_SIMD_NO_UNROLL(decode);                  \
+    stbir__simdf8_load4b( cs, hc + (ofs) );        \
+    stbir__simdf8_0123to00001111( c, cs );         \
+    stbir__simdf8_madd_mem( tot0, tot0, c, decode+(ofs)*3 - 1 ); \
+    stbir__simdf8_0123to22223333( c, cs );         \
+    stbir__simdf8_madd_mem( tot1, tot1, c, decode+(ofs)*3 + 6 - 1 );    
+
+#define stbir__1_coeff_remnant( ofs )                          \
+    STBIR_SIMD_NO_UNROLL(decode);                              \
+    stbir__simdf_load1rep4( t, hc + (ofs) );                   \
+    stbir__simdf8_madd_mem4( tot0, tot0, t, decode+(ofs)*3 - 1 ); 
+
+#define stbir__2_coeff_remnant( ofs )                          \
+    STBIR_SIMD_NO_UNROLL(decode);                              \
+    stbir__simdf8_load4b( cs, hc + (ofs) - 2 );                \
+    stbir__simdf8_0123to22223333( c, cs );                     \
+    stbir__simdf8_madd_mem( tot0, tot0, c, decode+(ofs)*3 - 1 );   
+ 
+ #define stbir__3_coeff_remnant( ofs )                           \
+    STBIR_SIMD_NO_UNROLL(decode);                                \
+    stbir__simdf8_load4b( cs, hc + (ofs) );                      \
+    stbir__simdf8_0123to00001111( c, cs );                       \
+    stbir__simdf8_madd_mem( tot0, tot0, c, decode+(ofs)*3 - 1 ); \
+    stbir__simdf8_0123to2222( t, cs );                           \
+    stbir__simdf8_madd_mem4( tot1, tot1, t, decode+(ofs)*3 + 6 - 1 ); 
+
+#define stbir__store_output()                       \
+    stbir__simdf8_add( tot0, tot0, tot1 );          \
+    stbir__simdf_0123to1230( t, stbir__if_simdf8_cast_to_simdf4( tot0 ) ); \
+    stbir__simdf8_add4halves( t, t, tot0 );         \
+    horizontal_coefficients += coefficient_width;   \
+    ++horizontal_contributors;                      \
+    output += 3;                                    \
+    if ( output < output_end )                      \
+    {                                               \
+      stbir__simdf_store( output-3, t );            \
+      continue;                                     \
+    }                                               \
+    { stbir__simdf tt; stbir__simdf_0123to2301( tt, t ); \
+    stbir__simdf_store2( output-3, t );             \
+    stbir__simdf_store1( output+2-3, tt ); }        \
+    break;
+
+
+#else
+
+#define stbir__4_coeff_start()                  \
+    stbir__simdf tot0,tot1,tot2,c,cs;           \
+    STBIR_SIMD_NO_UNROLL(decode);               \
+    stbir__simdf_load( cs, hc );                \
+    stbir__simdf_0123to0001( c, cs );           \
+    stbir__simdf_mult_mem( tot0, c, decode );   \
+    stbir__simdf_0123to1122( c, cs );           \
+    stbir__simdf_mult_mem( tot1, c, decode+4 ); \
+    stbir__simdf_0123to2333( c, cs );           \
+    stbir__simdf_mult_mem( tot2, c, decode+8 ); 
+
+#define stbir__4_coeff_continue_from_4( ofs )                 \
+    STBIR_SIMD_NO_UNROLL(decode);                             \
+    stbir__simdf_load( cs, hc + (ofs) );                      \
+    stbir__simdf_0123to0001( c, cs );                         \
+    stbir__simdf_madd_mem( tot0, tot0, c, decode+(ofs)*3 );   \
+    stbir__simdf_0123to1122( c, cs );                         \
+    stbir__simdf_madd_mem( tot1, tot1, c, decode+(ofs)*3+4 ); \
+    stbir__simdf_0123to2333( c, cs );                         \
+    stbir__simdf_madd_mem( tot2, tot2, c, decode+(ofs)*3+8 );   
+
+#define stbir__1_coeff_remnant( ofs )         \
+    STBIR_SIMD_NO_UNROLL(decode);             \
+    stbir__simdf_load1z( c, hc + (ofs) );     \
+    stbir__simdf_0123to0001( c, c );          \
+    stbir__simdf_madd_mem( tot0, tot0, c, decode+(ofs)*3 );   
+
+#define stbir__2_coeff_remnant( ofs )                       \
+    { stbir__simdf d;                                       \
+    STBIR_SIMD_NO_UNROLL(decode);                           \
+    stbir__simdf_load2z( cs, hc + (ofs) );                  \
+    stbir__simdf_0123to0001( c, cs );                       \
+    stbir__simdf_madd_mem( tot0, tot0, c, decode+(ofs)*3 ); \
+    stbir__simdf_0123to1122( c, cs );                       \
+    stbir__simdf_load2z( d, decode+(ofs)*3+4 );             \
+    stbir__simdf_madd( tot1, tot1, c, d ); }                 
+
+#define stbir__3_coeff_remnant( ofs )                         \
+    { stbir__simdf d;                                         \
+    STBIR_SIMD_NO_UNROLL(decode);                             \
+    stbir__simdf_load( cs, hc + (ofs) );                      \
+    stbir__simdf_0123to0001( c, cs );                         \
+    stbir__simdf_madd_mem( tot0, tot0, c, decode+(ofs)*3 );   \
+    stbir__simdf_0123to1122( c, cs );                         \
+    stbir__simdf_madd_mem( tot1, tot1, c, decode+(ofs)*3+4 ); \
+    stbir__simdf_0123to2222( c, cs );                         \
+    stbir__simdf_load1z( d, decode+(ofs)*3+8 );               \
+    stbir__simdf_madd( tot2, tot2, c, d );  }                
+
+#define stbir__store_output()                       \
+    stbir__simdf_0123ABCDto3ABx( c, tot0, tot1 );   \
+    stbir__simdf_0123ABCDto23Ax( cs, tot1, tot2 );  \
+    stbir__simdf_0123to1230( tot2, tot2 );          \
+    stbir__simdf_add( tot0, tot0, cs );             \
+    stbir__simdf_add( c, c, tot2 );                 \
+    stbir__simdf_add( tot0, tot0, c );              \
+    horizontal_coefficients += coefficient_width;   \
+    ++horizontal_contributors;                      \
+    output += 3;                                    \
+    if ( output < output_end )                      \
+    {                                               \
+      stbir__simdf_store( output-3, tot0 );         \
+      continue;                                     \
+    }                                               \
+    stbir__simdf_0123to2301( tot1, tot0 );          \
+    stbir__simdf_store2( output-3, tot0 );          \
+    stbir__simdf_store1( output+2-3, tot1 );        \
+    break;
+
+#endif
+
+#else
+
+#define stbir__1_coeff_only()  \
+    float tot0, tot1, tot2, c; \
+    c = hc[0];                 \
+    tot0 = decode[0]*c;        \
+    tot1 = decode[1]*c;        \
+    tot2 = decode[2]*c;              
+
+#define stbir__2_coeff_only()  \
+    float tot0, tot1, tot2, c; \
+    c = hc[0];                 \
+    tot0 = decode[0]*c;        \
+    tot1 = decode[1]*c;        \
+    tot2 = decode[2]*c;        \
+    c = hc[1];                 \
+    tot0 += decode[3]*c;       \
+    tot1 += decode[4]*c;       \
+    tot2 += decode[5]*c;              
+
+#define stbir__3_coeff_only()  \
+    float tot0, tot1, tot2, c; \
+    c = hc[0];                 \
+    tot0 = decode[0]*c;        \
+    tot1 = decode[1]*c;        \
+    tot2 = decode[2]*c;        \
+    c = hc[1];                 \
+    tot0 += decode[3]*c;       \
+    tot1 += decode[4]*c;       \
+    tot2 += decode[5]*c;       \
+    c = hc[2];                 \
+    tot0 += decode[6]*c;       \
+    tot1 += decode[7]*c;       \
+    tot2 += decode[8]*c;              
+
+#define stbir__store_output_tiny()                \
+    output[0] = tot0;                             \
+    output[1] = tot1;                             \
+    output[2] = tot2;                             \
+    horizontal_coefficients += coefficient_width; \
+    ++horizontal_contributors;                    \
+    output += 3;
+
+#define stbir__4_coeff_start()      \
+    float tota0,tota1,tota2,totb0,totb1,totb2,totc0,totc1,totc2,totd0,totd1,totd2,c;  \
+    c = hc[0];                      \
+    tota0 = decode[0]*c;            \
+    tota1 = decode[1]*c;            \
+    tota2 = decode[2]*c;            \
+    c = hc[1];                      \
+    totb0 = decode[3]*c;            \
+    totb1 = decode[4]*c;            \
+    totb2 = decode[5]*c;            \
+    c = hc[2];                      \
+    totc0 = decode[6]*c;            \
+    totc1 = decode[7]*c;            \
+    totc2 = decode[8]*c;            \
+    c = hc[3];                      \
+    totd0 = decode[9]*c;            \
+    totd1 = decode[10]*c;           \
+    totd2 = decode[11]*c;            
+
+#define stbir__4_coeff_continue_from_4( ofs )  \
+    c = hc[0+(ofs)];                           \
+    tota0 += decode[0+(ofs)*3]*c;              \
+    tota1 += decode[1+(ofs)*3]*c;              \
+    tota2 += decode[2+(ofs)*3]*c;              \
+    c = hc[1+(ofs)];                           \
+    totb0 += decode[3+(ofs)*3]*c;              \
+    totb1 += decode[4+(ofs)*3]*c;              \
+    totb2 += decode[5+(ofs)*3]*c;              \
+    c = hc[2+(ofs)];                           \
+    totc0 += decode[6+(ofs)*3]*c;              \
+    totc1 += decode[7+(ofs)*3]*c;              \
+    totc2 += decode[8+(ofs)*3]*c;              \
+    c = hc[3+(ofs)];                           \
+    totd0 += decode[9+(ofs)*3]*c;              \
+    totd1 += decode[10+(ofs)*3]*c;             \
+    totd2 += decode[11+(ofs)*3]*c;              
+
+#define stbir__1_coeff_remnant( ofs )  \
+    c = hc[0+(ofs)];                   \
+    tota0 += decode[0+(ofs)*3]*c;      \
+    tota1 += decode[1+(ofs)*3]*c;      \
+    tota2 += decode[2+(ofs)*3]*c;
+
+#define stbir__2_coeff_remnant( ofs )  \
+    c = hc[0+(ofs)];                   \
+    tota0 += decode[0+(ofs)*3]*c;      \
+    tota1 += decode[1+(ofs)*3]*c;      \
+    tota2 += decode[2+(ofs)*3]*c;      \
+    c = hc[1+(ofs)];                   \
+    totb0 += decode[3+(ofs)*3]*c;      \
+    totb1 += decode[4+(ofs)*3]*c;      \
+    totb2 += decode[5+(ofs)*3]*c;      \
+
+#define stbir__3_coeff_remnant( ofs )  \
+    c = hc[0+(ofs)];                   \
+    tota0 += decode[0+(ofs)*3]*c;      \
+    tota1 += decode[1+(ofs)*3]*c;      \
+    tota2 += decode[2+(ofs)*3]*c;      \
+    c = hc[1+(ofs)];                   \
+    totb0 += decode[3+(ofs)*3]*c;      \
+    totb1 += decode[4+(ofs)*3]*c;      \
+    totb2 += decode[5+(ofs)*3]*c;      \
+    c = hc[2+(ofs)];                   \
+    totc0 += decode[6+(ofs)*3]*c;      \
+    totc1 += decode[7+(ofs)*3]*c;      \
+    totc2 += decode[8+(ofs)*3]*c;              
+
+#define stbir__store_output()                     \
+    output[0] = (tota0+totc0)+(totb0+totd0);      \
+    output[1] = (tota1+totc1)+(totb1+totd1);      \
+    output[2] = (tota2+totc2)+(totb2+totd2);      \
+    horizontal_coefficients += coefficient_width; \
+    ++horizontal_contributors;                    \
+    output += 3;
+
+#endif  
+
+#define STBIR__horizontal_channels 3
+#define STB_IMAGE_RESIZE_DO_HORIZONTALS
+#include STBIR__HEADER_FILENAME
+
+//=================
+// Do 4 channel horizontal routines
+
+#ifdef STBIR_SIMD
+
+#define stbir__1_coeff_only()             \
+    stbir__simdf tot,c;                   \
+    STBIR_SIMD_NO_UNROLL(decode);         \
+    stbir__simdf_load1( c, hc );          \
+    stbir__simdf_0123to0000( c, c );      \
+    stbir__simdf_mult_mem( tot, c, decode ); 
+
+#define stbir__2_coeff_only()                       \
+    stbir__simdf tot,c,cs;                          \
+    STBIR_SIMD_NO_UNROLL(decode);                   \
+    stbir__simdf_load2( cs, hc );                   \
+    stbir__simdf_0123to0000( c, cs );               \
+    stbir__simdf_mult_mem( tot, c, decode );        \
+    stbir__simdf_0123to1111( c, cs );               \
+    stbir__simdf_madd_mem( tot, tot, c, decode+4 ); 
+
+#define stbir__3_coeff_only()                       \
+    stbir__simdf tot,c,cs;                          \
+    STBIR_SIMD_NO_UNROLL(decode);                   \
+    stbir__simdf_load( cs, hc );                    \
+    stbir__simdf_0123to0000( c, cs );               \
+    stbir__simdf_mult_mem( tot, c, decode );        \
+    stbir__simdf_0123to1111( c, cs );               \
+    stbir__simdf_madd_mem( tot, tot, c, decode+4 ); \
+    stbir__simdf_0123to2222( c, cs );               \
+    stbir__simdf_madd_mem( tot, tot, c, decode+8 ); 
+
+#define stbir__store_output_tiny()                \
+    stbir__simdf_store( output, tot );            \
+    horizontal_coefficients += coefficient_width; \
+    ++horizontal_contributors;                    \
+    output += 4;
+
+#ifdef STBIR_SIMD8
+
+#define stbir__4_coeff_start()                     \
+    stbir__simdf8 tot0,c,cs; stbir__simdf t;  \
+    STBIR_SIMD_NO_UNROLL(decode);                  \
+    stbir__simdf8_load4b( cs, hc );                \
+    stbir__simdf8_0123to00001111( c, cs );         \
+    stbir__simdf8_mult_mem( tot0, c, decode );     \
+    stbir__simdf8_0123to22223333( c, cs );         \
+    stbir__simdf8_madd_mem( tot0, tot0, c, decode+8 );    
+
+#define stbir__4_coeff_continue_from_4( ofs )                  \
+    STBIR_SIMD_NO_UNROLL(decode);                              \
+    stbir__simdf8_load4b( cs, hc + (ofs) );                    \
+    stbir__simdf8_0123to00001111( c, cs );                     \
+    stbir__simdf8_madd_mem( tot0, tot0, c, decode+(ofs)*4 );   \
+    stbir__simdf8_0123to22223333( c, cs );                     \
+    stbir__simdf8_madd_mem( tot0, tot0, c, decode+(ofs)*4+8 );    
+
+#define stbir__1_coeff_remnant( ofs )                          \
+    STBIR_SIMD_NO_UNROLL(decode);                              \
+    stbir__simdf_load1rep4( t, hc + (ofs) );                   \
+    stbir__simdf8_madd_mem4( tot0, tot0, t, decode+(ofs)*4 ); 
+
+#define stbir__2_coeff_remnant( ofs )                          \
+    STBIR_SIMD_NO_UNROLL(decode);                              \
+    stbir__simdf8_load4b( cs, hc + (ofs) - 2 );                \
+    stbir__simdf8_0123to22223333( c, cs );                     \
+    stbir__simdf8_madd_mem( tot0, tot0, c, decode+(ofs)*4 );   
+ 
+ #define stbir__3_coeff_remnant( ofs )                         \
+    STBIR_SIMD_NO_UNROLL(decode);                              \
+    stbir__simdf8_load4b( cs, hc + (ofs) );                    \
+    stbir__simdf8_0123to00001111( c, cs );                     \
+    stbir__simdf8_madd_mem( tot0, tot0, c, decode+(ofs)*4 );   \
+    stbir__simdf8_0123to2222( t, cs );                         \
+    stbir__simdf8_madd_mem4( tot0, tot0, t, decode+(ofs)*4+8 ); 
+
+#define stbir__store_output()                      \
+    stbir__simdf8_add4halves( t, stbir__if_simdf8_cast_to_simdf4(tot0), tot0 );     \
+    stbir__simdf_store( output, t );               \
+    horizontal_coefficients += coefficient_width;  \
+    ++horizontal_contributors;                     \
+    output += 4;
+
+#else    
+
+#define stbir__4_coeff_start()                        \
+    stbir__simdf tot0,tot1,c,cs;                      \
+    STBIR_SIMD_NO_UNROLL(decode);                     \
+    stbir__simdf_load( cs, hc );                      \
+    stbir__simdf_0123to0000( c, cs );                 \
+    stbir__simdf_mult_mem( tot0, c, decode );         \
+    stbir__simdf_0123to1111( c, cs );                 \
+    stbir__simdf_mult_mem( tot1, c, decode+4 );       \
+    stbir__simdf_0123to2222( c, cs );                 \
+    stbir__simdf_madd_mem( tot0, tot0, c, decode+8 ); \
+    stbir__simdf_0123to3333( c, cs );                 \
+    stbir__simdf_madd_mem( tot1, tot1, c, decode+12 ); 
+
+#define stbir__4_coeff_continue_from_4( ofs )                  \
+    STBIR_SIMD_NO_UNROLL(decode);                              \
+    stbir__simdf_load( cs, hc + (ofs) );                       \
+    stbir__simdf_0123to0000( c, cs );                          \
+    stbir__simdf_madd_mem( tot0, tot0, c, decode+(ofs)*4 );    \
+    stbir__simdf_0123to1111( c, cs );                          \
+    stbir__simdf_madd_mem( tot1, tot1, c, decode+(ofs)*4+4 );  \
+    stbir__simdf_0123to2222( c, cs );                          \
+    stbir__simdf_madd_mem( tot0, tot0, c, decode+(ofs)*4+8 );  \
+    stbir__simdf_0123to3333( c, cs );                          \
+    stbir__simdf_madd_mem( tot1, tot1, c, decode+(ofs)*4+12 ); 
+
+#define stbir__1_coeff_remnant( ofs )                       \
+    STBIR_SIMD_NO_UNROLL(decode);                           \
+    stbir__simdf_load1( c, hc + (ofs) );                    \
+    stbir__simdf_0123to0000( c, c );                        \
+    stbir__simdf_madd_mem( tot0, tot0, c, decode+(ofs)*4 ); 
+
+#define stbir__2_coeff_remnant( ofs )                         \
+    STBIR_SIMD_NO_UNROLL(decode);                             \
+    stbir__simdf_load2( cs, hc + (ofs) );                     \
+    stbir__simdf_0123to0000( c, cs );                         \
+    stbir__simdf_madd_mem( tot0, tot0, c, decode+(ofs)*4 );   \
+    stbir__simdf_0123to1111( c, cs );                         \
+    stbir__simdf_madd_mem( tot1, tot1, c, decode+(ofs)*4+4 ); 
+  
+#define stbir__3_coeff_remnant( ofs )                          \
+    STBIR_SIMD_NO_UNROLL(decode);                              \
+    stbir__simdf_load( cs, hc + (ofs) );                       \
+    stbir__simdf_0123to0000( c, cs );                          \
+    stbir__simdf_madd_mem( tot0, tot0, c, decode+(ofs)*4 );    \
+    stbir__simdf_0123to1111( c, cs );                          \
+    stbir__simdf_madd_mem( tot1, tot1, c, decode+(ofs)*4+4 );  \
+    stbir__simdf_0123to2222( c, cs );                          \
+    stbir__simdf_madd_mem( tot0, tot0, c, decode+(ofs)*4+8 );
+
+#define stbir__store_output()                     \
+    stbir__simdf_add( tot0, tot0, tot1 );         \
+    stbir__simdf_store( output, tot0 );           \
+    horizontal_coefficients += coefficient_width; \
+    ++horizontal_contributors;                    \
+    output += 4;
+
+#endif
+
+#else
+
+#define stbir__1_coeff_only()         \
+    float p0,p1,p2,p3,c;              \
+    STBIR_SIMD_NO_UNROLL(decode);     \
+    c = hc[0];                        \
+    p0 = decode[0] * c;               \
+    p1 = decode[1] * c;               \
+    p2 = decode[2] * c;               \
+    p3 = decode[3] * c;
+
+#define stbir__2_coeff_only()         \
+    float p0,p1,p2,p3,c;              \
+    STBIR_SIMD_NO_UNROLL(decode);     \
+    c = hc[0];                        \
+    p0 = decode[0] * c;               \
+    p1 = decode[1] * c;               \
+    p2 = decode[2] * c;               \
+    p3 = decode[3] * c;               \
+    c = hc[1];                        \
+    p0 += decode[4] * c;              \
+    p1 += decode[5] * c;              \
+    p2 += decode[6] * c;              \
+    p3 += decode[7] * c;
+
+#define stbir__3_coeff_only()         \
+    float p0,p1,p2,p3,c;              \
+    STBIR_SIMD_NO_UNROLL(decode);     \
+    c = hc[0];                        \
+    p0 = decode[0] * c;               \
+    p1 = decode[1] * c;               \
+    p2 = decode[2] * c;               \
+    p3 = decode[3] * c;               \
+    c = hc[1];                        \
+    p0 += decode[4] * c;              \
+    p1 += decode[5] * c;              \
+    p2 += decode[6] * c;              \
+    p3 += decode[7] * c;              \
+    c = hc[2];                        \
+    p0 += decode[8] * c;              \
+    p1 += decode[9] * c;              \
+    p2 += decode[10] * c;             \
+    p3 += decode[11] * c;
+
+#define stbir__store_output_tiny()                \
+    output[0] = p0;                               \
+    output[1] = p1;                               \
+    output[2] = p2;                               \
+    output[3] = p3;                               \
+    horizontal_coefficients += coefficient_width; \
+    ++horizontal_contributors;                    \
+    output += 4;
+
+#define stbir__4_coeff_start()        \
+    float x0,x1,x2,x3,y0,y1,y2,y3,c;  \
+    STBIR_SIMD_NO_UNROLL(decode);     \
+    c = hc[0];                        \
+    x0 = decode[0] * c;               \
+    x1 = decode[1] * c;               \
+    x2 = decode[2] * c;               \
+    x3 = decode[3] * c;               \
+    c = hc[1];                        \
+    y0 = decode[4] * c;               \
+    y1 = decode[5] * c;               \
+    y2 = decode[6] * c;               \
+    y3 = decode[7] * c;               \
+    c = hc[2];                        \
+    x0 += decode[8] * c;              \
+    x1 += decode[9] * c;              \
+    x2 += decode[10] * c;             \
+    x3 += decode[11] * c;             \
+    c = hc[3];                        \
+    y0 += decode[12] * c;             \
+    y1 += decode[13] * c;             \
+    y2 += decode[14] * c;             \
+    y3 += decode[15] * c;
+
+#define stbir__4_coeff_continue_from_4( ofs ) \
+    STBIR_SIMD_NO_UNROLL(decode);     \
+    c = hc[0+(ofs)];                  \
+    x0 += decode[0+(ofs)*4] * c;      \
+    x1 += decode[1+(ofs)*4] * c;      \
+    x2 += decode[2+(ofs)*4] * c;      \
+    x3 += decode[3+(ofs)*4] * c;      \
+    c = hc[1+(ofs)];                  \
+    y0 += decode[4+(ofs)*4] * c;      \
+    y1 += decode[5+(ofs)*4] * c;      \
+    y2 += decode[6+(ofs)*4] * c;      \
+    y3 += decode[7+(ofs)*4] * c;      \
+    c = hc[2+(ofs)];                  \
+    x0 += decode[8+(ofs)*4] * c;      \
+    x1 += decode[9+(ofs)*4] * c;      \
+    x2 += decode[10+(ofs)*4] * c;     \
+    x3 += decode[11+(ofs)*4] * c;     \
+    c = hc[3+(ofs)];                  \
+    y0 += decode[12+(ofs)*4] * c;     \
+    y1 += decode[13+(ofs)*4] * c;     \
+    y2 += decode[14+(ofs)*4] * c;     \
+    y3 += decode[15+(ofs)*4] * c;
+
+#define stbir__1_coeff_remnant( ofs ) \
+    STBIR_SIMD_NO_UNROLL(decode);     \
+    c = hc[0+(ofs)];                  \
+    x0 += decode[0+(ofs)*4] * c;      \
+    x1 += decode[1+(ofs)*4] * c;      \
+    x2 += decode[2+(ofs)*4] * c;      \
+    x3 += decode[3+(ofs)*4] * c;      
+
+#define stbir__2_coeff_remnant( ofs ) \
+    STBIR_SIMD_NO_UNROLL(decode);     \
+    c = hc[0+(ofs)];                  \
+    x0 += decode[0+(ofs)*4] * c;      \
+    x1 += decode[1+(ofs)*4] * c;      \
+    x2 += decode[2+(ofs)*4] * c;      \
+    x3 += decode[3+(ofs)*4] * c;      \
+    c = hc[1+(ofs)];                  \
+    y0 += decode[4+(ofs)*4] * c;      \
+    y1 += decode[5+(ofs)*4] * c;      \
+    y2 += decode[6+(ofs)*4] * c;      \
+    y3 += decode[7+(ofs)*4] * c;    
+  
+#define stbir__3_coeff_remnant( ofs ) \
+    STBIR_SIMD_NO_UNROLL(decode);     \
+    c = hc[0+(ofs)];                  \
+    x0 += decode[0+(ofs)*4] * c;      \
+    x1 += decode[1+(ofs)*4] * c;      \
+    x2 += decode[2+(ofs)*4] * c;      \
+    x3 += decode[3+(ofs)*4] * c;      \
+    c = hc[1+(ofs)];                  \
+    y0 += decode[4+(ofs)*4] * c;      \
+    y1 += decode[5+(ofs)*4] * c;      \
+    y2 += decode[6+(ofs)*4] * c;      \
+    y3 += decode[7+(ofs)*4] * c;      \
+    c = hc[2+(ofs)];                  \
+    x0 += decode[8+(ofs)*4] * c;      \
+    x1 += decode[9+(ofs)*4] * c;      \
+    x2 += decode[10+(ofs)*4] * c;     \
+    x3 += decode[11+(ofs)*4] * c;     
+
+#define stbir__store_output()                     \
+    output[0] = x0 + y0;                          \
+    output[1] = x1 + y1;                          \
+    output[2] = x2 + y2;                          \
+    output[3] = x3 + y3;                          \
+    horizontal_coefficients += coefficient_width; \
+    ++horizontal_contributors;                    \
+    output += 4;
+
+#endif  
+
+#define STBIR__horizontal_channels 4
+#define STB_IMAGE_RESIZE_DO_HORIZONTALS
+#include STBIR__HEADER_FILENAME
+
+
+
+//=================
+// Do 7 channel horizontal routines
+
+#ifdef STBIR_SIMD
+
+#define stbir__1_coeff_only()                   \
+    stbir__simdf tot0,tot1,c;                   \
+    STBIR_SIMD_NO_UNROLL(decode);               \
+    stbir__simdf_load1( c, hc );                \
+    stbir__simdf_0123to0000( c, c );            \
+    stbir__simdf_mult_mem( tot0, c, decode );   \
+    stbir__simdf_mult_mem( tot1, c, decode+3 ); 
+
+#define stbir__2_coeff_only()                         \
+    stbir__simdf tot0,tot1,c,cs;                      \
+    STBIR_SIMD_NO_UNROLL(decode);                     \
+    stbir__simdf_load2( cs, hc );                     \
+    stbir__simdf_0123to0000( c, cs );                 \
+    stbir__simdf_mult_mem( tot0, c, decode );         \
+    stbir__simdf_mult_mem( tot1, c, decode+3 );       \
+    stbir__simdf_0123to1111( c, cs );                 \
+    stbir__simdf_madd_mem( tot0, tot0, c, decode+7 ); \
+    stbir__simdf_madd_mem( tot1, tot1, c,decode+10 ); 
+
+#define stbir__3_coeff_only()                           \
+    stbir__simdf tot0,tot1,c,cs;                        \
+    STBIR_SIMD_NO_UNROLL(decode);                       \
+    stbir__simdf_load( cs, hc );                        \
+    stbir__simdf_0123to0000( c, cs );                   \
+    stbir__simdf_mult_mem( tot0, c, decode );           \
+    stbir__simdf_mult_mem( tot1, c, decode+3 );         \
+    stbir__simdf_0123to1111( c, cs );                   \
+    stbir__simdf_madd_mem( tot0, tot0, c, decode+7 );   \
+    stbir__simdf_madd_mem( tot1, tot1, c, decode+10 );  \
+    stbir__simdf_0123to2222( c, cs );                   \
+    stbir__simdf_madd_mem( tot0, tot0, c, decode+14 );  \
+    stbir__simdf_madd_mem( tot1, tot1, c, decode+17 );  
+
+#define stbir__store_output_tiny()                \
+    stbir__simdf_store( output+3, tot1 );         \
+    stbir__simdf_store( output, tot0 );           \
+    horizontal_coefficients += coefficient_width; \
+    ++horizontal_contributors;                    \
+    output += 7;
+
+#ifdef STBIR_SIMD8
+
+#define stbir__4_coeff_start()                     \
+    stbir__simdf8 tot0,tot1,c,cs;                  \
+    STBIR_SIMD_NO_UNROLL(decode);                  \
+    stbir__simdf8_load4b( cs, hc );                \
+    stbir__simdf8_0123to00000000( c, cs );         \
+    stbir__simdf8_mult_mem( tot0, c, decode );     \
+    stbir__simdf8_0123to11111111( c, cs );         \
+    stbir__simdf8_mult_mem( tot1, c, decode+7 );   \
+    stbir__simdf8_0123to22222222( c, cs );         \
+    stbir__simdf8_madd_mem( tot0, tot0, c, decode+14 );  \
+    stbir__simdf8_0123to33333333( c, cs );         \
+    stbir__simdf8_madd_mem( tot1, tot1, c, decode+21 );  
+
+#define stbir__4_coeff_continue_from_4( ofs )                   \
+    STBIR_SIMD_NO_UNROLL(decode);                               \
+    stbir__simdf8_load4b( cs, hc + (ofs) );                     \
+    stbir__simdf8_0123to00000000( c, cs );                      \
+    stbir__simdf8_madd_mem( tot0, tot0, c, decode+(ofs)*7 );    \
+    stbir__simdf8_0123to11111111( c, cs );                      \
+    stbir__simdf8_madd_mem( tot1, tot1, c, decode+(ofs)*7+7 );  \
+    stbir__simdf8_0123to22222222( c, cs );                      \
+    stbir__simdf8_madd_mem( tot0, tot0, c, decode+(ofs)*7+14 ); \
+    stbir__simdf8_0123to33333333( c, cs );                      \
+    stbir__simdf8_madd_mem( tot1, tot1, c, decode+(ofs)*7+21 ); 
+
+#define stbir__1_coeff_remnant( ofs )                           \
+    STBIR_SIMD_NO_UNROLL(decode);                               \
+    stbir__simdf8_load1b( c, hc + (ofs) );                      \
+    stbir__simdf8_madd_mem( tot0, tot0, c, decode+(ofs)*7 );    
+
+#define stbir__2_coeff_remnant( ofs )                           \
+    STBIR_SIMD_NO_UNROLL(decode);                               \
+    stbir__simdf8_load1b( c, hc + (ofs) );                      \
+    stbir__simdf8_madd_mem( tot0, tot0, c, decode+(ofs)*7 );    \
+    stbir__simdf8_load1b( c, hc + (ofs)+1 );                    \
+    stbir__simdf8_madd_mem( tot1, tot1, c, decode+(ofs)*7+7 );   
+
+#define stbir__3_coeff_remnant( ofs )                           \
+    STBIR_SIMD_NO_UNROLL(decode);                               \
+    stbir__simdf8_load4b( cs, hc + (ofs) );                     \
+    stbir__simdf8_0123to00000000( c, cs );                      \
+    stbir__simdf8_madd_mem( tot0, tot0, c, decode+(ofs)*7 );    \
+    stbir__simdf8_0123to11111111( c, cs );                      \
+    stbir__simdf8_madd_mem( tot1, tot1, c, decode+(ofs)*7+7 );  \
+    stbir__simdf8_0123to22222222( c, cs );                      \
+    stbir__simdf8_madd_mem( tot0, tot0, c, decode+(ofs)*7+14 ); 
+
+#define stbir__store_output()                     \
+    stbir__simdf8_add( tot0, tot0, tot1 );        \
+    horizontal_coefficients += coefficient_width; \
+    ++horizontal_contributors;                    \
+    output += 7;                                  \
+    if ( output < output_end )                    \
+    {                                             \
+      stbir__simdf8_store( output-7, tot0 );      \
+      continue;                                   \
+    }                                             \
+    stbir__simdf_store( output-7+3, stbir__simdf_swiz(stbir__simdf8_gettop4(tot0),0,0,1,2) ); \
+    stbir__simdf_store( output-7, stbir__if_simdf8_cast_to_simdf4(tot0) );           \
+    break;
+
+#else
+
+#define stbir__4_coeff_start()                    \
+    stbir__simdf tot0,tot1,tot2,tot3,c,cs;        \
+    STBIR_SIMD_NO_UNROLL(decode);                 \
+    stbir__simdf_load( cs, hc );                  \
+    stbir__simdf_0123to0000( c, cs );             \
+    stbir__simdf_mult_mem( tot0, c, decode );     \
+    stbir__simdf_mult_mem( tot1, c, decode+3 );   \
+    stbir__simdf_0123to1111( c, cs );             \
+    stbir__simdf_mult_mem( tot2, c, decode+7 );   \
+    stbir__simdf_mult_mem( tot3, c, decode+10 );  \
+    stbir__simdf_0123to2222( c, cs );             \
+    stbir__simdf_madd_mem( tot0, tot0, c, decode+14 );  \
+    stbir__simdf_madd_mem( tot1, tot1, c, decode+17 );  \
+    stbir__simdf_0123to3333( c, cs );                   \
+    stbir__simdf_madd_mem( tot2, tot2, c, decode+21 );  \
+    stbir__simdf_madd_mem( tot3, tot3, c, decode+24 );         
+
+#define stbir__4_coeff_continue_from_4( ofs )                   \
+    STBIR_SIMD_NO_UNROLL(decode);                               \
+    stbir__simdf_load( cs, hc + (ofs) );                        \
+    stbir__simdf_0123to0000( c, cs );                           \
+    stbir__simdf_madd_mem( tot0, tot0, c, decode+(ofs)*7 );     \
+    stbir__simdf_madd_mem( tot1, tot1, c, decode+(ofs)*7+3 );   \
+    stbir__simdf_0123to1111( c, cs );                           \
+    stbir__simdf_madd_mem( tot2, tot2, c, decode+(ofs)*7+7 );   \
+    stbir__simdf_madd_mem( tot3, tot3, c, decode+(ofs)*7+10 );  \
+    stbir__simdf_0123to2222( c, cs );                           \
+    stbir__simdf_madd_mem( tot0, tot0, c, decode+(ofs)*7+14 );  \
+    stbir__simdf_madd_mem( tot1, tot1, c, decode+(ofs)*7+17 );  \
+    stbir__simdf_0123to3333( c, cs );                           \
+    stbir__simdf_madd_mem( tot2, tot2, c, decode+(ofs)*7+21 );  \
+    stbir__simdf_madd_mem( tot3, tot3, c, decode+(ofs)*7+24 );   
+
+#define stbir__1_coeff_remnant( ofs )                           \
+    STBIR_SIMD_NO_UNROLL(decode);                               \
+    stbir__simdf_load1( c, hc + (ofs) );                        \
+    stbir__simdf_0123to0000( c, c );                            \
+    stbir__simdf_madd_mem( tot0, tot0, c, decode+(ofs)*7 );     \
+    stbir__simdf_madd_mem( tot1, tot1, c, decode+(ofs)*7+3 );   \
+
+#define stbir__2_coeff_remnant( ofs )                           \
+    STBIR_SIMD_NO_UNROLL(decode);                               \
+    stbir__simdf_load2( cs, hc + (ofs) );                       \
+    stbir__simdf_0123to0000( c, cs );                           \
+    stbir__simdf_madd_mem( tot0, tot0, c, decode+(ofs)*7 );     \
+    stbir__simdf_madd_mem( tot1, tot1, c, decode+(ofs)*7+3 );   \
+    stbir__simdf_0123to1111( c, cs );                           \
+    stbir__simdf_madd_mem( tot2, tot2, c, decode+(ofs)*7+7 );   \
+    stbir__simdf_madd_mem( tot3, tot3, c, decode+(ofs)*7+10 );  
+  
+#define stbir__3_coeff_remnant( ofs )                           \
+    STBIR_SIMD_NO_UNROLL(decode);                               \
+    stbir__simdf_load( cs, hc + (ofs) );                        \
+    stbir__simdf_0123to0000( c, cs );                           \
+    stbir__simdf_madd_mem( tot0, tot0, c, decode+(ofs)*7 );     \
+    stbir__simdf_madd_mem( tot1, tot1, c, decode+(ofs)*7+3 );   \
+    stbir__simdf_0123to1111( c, cs );                           \
+    stbir__simdf_madd_mem( tot2, tot2, c, decode+(ofs)*7+7 );   \
+    stbir__simdf_madd_mem( tot3, tot3, c, decode+(ofs)*7+10 );  \
+    stbir__simdf_0123to2222( c, cs );                           \
+    stbir__simdf_madd_mem( tot0, tot0, c, decode+(ofs)*7+14 );  \
+    stbir__simdf_madd_mem( tot1, tot1, c, decode+(ofs)*7+17 );  
+
+#define stbir__store_output()                     \
+    stbir__simdf_add( tot0, tot0, tot2 );         \
+    stbir__simdf_add( tot1, tot1, tot3 );         \
+    stbir__simdf_store( output+3, tot1 );         \
+    stbir__simdf_store( output, tot0 );           \
+    horizontal_coefficients += coefficient_width; \
+    ++horizontal_contributors;                    \
+    output += 7;
+
+#endif
+
+#else
+
+#define stbir__1_coeff_only()        \
+    float tot0, tot1, tot2, tot3, tot4, tot5, tot6, c; \
+    c = hc[0];                       \
+    tot0 = decode[0]*c;              \
+    tot1 = decode[1]*c;              \
+    tot2 = decode[2]*c;              \
+    tot3 = decode[3]*c;              \
+    tot4 = decode[4]*c;              \
+    tot5 = decode[5]*c;              \
+    tot6 = decode[6]*c;              
+
+#define stbir__2_coeff_only()        \
+    float tot0, tot1, tot2, tot3, tot4, tot5, tot6, c; \
+    c = hc[0];                       \
+    tot0 = decode[0]*c;              \
+    tot1 = decode[1]*c;              \
+    tot2 = decode[2]*c;              \
+    tot3 = decode[3]*c;              \
+    tot4 = decode[4]*c;              \
+    tot5 = decode[5]*c;              \
+    tot6 = decode[6]*c;              \
+    c = hc[1];                       \
+    tot0 += decode[7]*c;             \
+    tot1 += decode[8]*c;             \
+    tot2 += decode[9]*c;             \
+    tot3 += decode[10]*c;            \
+    tot4 += decode[11]*c;            \
+    tot5 += decode[12]*c;            \
+    tot6 += decode[13]*c;            \
+
+#define stbir__3_coeff_only()        \
+    float tot0, tot1, tot2, tot3, tot4, tot5, tot6, c; \
+    c = hc[0];                       \
+    tot0 = decode[0]*c;              \
+    tot1 = decode[1]*c;              \
+    tot2 = decode[2]*c;              \
+    tot3 = decode[3]*c;              \
+    tot4 = decode[4]*c;              \
+    tot5 = decode[5]*c;              \
+    tot6 = decode[6]*c;              \
+    c = hc[1];                       \
+    tot0 += decode[7]*c;             \
+    tot1 += decode[8]*c;             \
+    tot2 += decode[9]*c;             \
+    tot3 += decode[10]*c;            \
+    tot4 += decode[11]*c;            \
+    tot5 += decode[12]*c;            \
+    tot6 += decode[13]*c;            \
+    c = hc[2];                       \
+    tot0 += decode[14]*c;            \
+    tot1 += decode[15]*c;            \
+    tot2 += decode[16]*c;            \
+    tot3 += decode[17]*c;            \
+    tot4 += decode[18]*c;            \
+    tot5 += decode[19]*c;            \
+    tot6 += decode[20]*c;            \
+
+#define stbir__store_output_tiny()                \
+    output[0] = tot0;                             \
+    output[1] = tot1;                             \
+    output[2] = tot2;                             \
+    output[3] = tot3;                             \
+    output[4] = tot4;                             \
+    output[5] = tot5;                             \
+    output[6] = tot6;                             \
+    horizontal_coefficients += coefficient_width; \
+    ++horizontal_contributors;                    \
+    output += 7;
+
+#define stbir__4_coeff_start()    \
+    float x0,x1,x2,x3,x4,x5,x6,y0,y1,y2,y3,y4,y5,y6,c; \
+    STBIR_SIMD_NO_UNROLL(decode); \
+    c = hc[0];                    \
+    x0 = decode[0] * c;           \
+    x1 = decode[1] * c;           \
+    x2 = decode[2] * c;           \
+    x3 = decode[3] * c;           \
+    x4 = decode[4] * c;           \
+    x5 = decode[5] * c;           \
+    x6 = decode[6] * c;           \
+    c = hc[1];                    \
+    y0 = decode[7] * c;           \
+    y1 = decode[8] * c;           \
+    y2 = decode[9] * c;           \
+    y3 = decode[10] * c;          \
+    y4 = decode[11] * c;          \
+    y5 = decode[12] * c;          \
+    y6 = decode[13] * c;          \
+    c = hc[2];                    \
+    x0 += decode[14] * c;         \
+    x1 += decode[15] * c;         \
+    x2 += decode[16] * c;         \
+    x3 += decode[17] * c;         \
+    x4 += decode[18] * c;         \
+    x5 += decode[19] * c;         \
+    x6 += decode[20] * c;         \
+    c = hc[3];                    \
+    y0 += decode[21] * c;         \
+    y1 += decode[22] * c;         \
+    y2 += decode[23] * c;         \
+    y3 += decode[24] * c;         \
+    y4 += decode[25] * c;         \
+    y5 += decode[26] * c;         \
+    y6 += decode[27] * c; 
+
+#define stbir__4_coeff_continue_from_4( ofs ) \
+    STBIR_SIMD_NO_UNROLL(decode);  \
+    c = hc[0+(ofs)];               \
+    x0 += decode[0+(ofs)*7] * c;   \
+    x1 += decode[1+(ofs)*7] * c;   \
+    x2 += decode[2+(ofs)*7] * c;   \
+    x3 += decode[3+(ofs)*7] * c;   \
+    x4 += decode[4+(ofs)*7] * c;   \
+    x5 += decode[5+(ofs)*7] * c;   \
+    x6 += decode[6+(ofs)*7] * c;   \
+    c = hc[1+(ofs)];               \
+    y0 += decode[7+(ofs)*7] * c;   \
+    y1 += decode[8+(ofs)*7] * c;   \
+    y2 += decode[9+(ofs)*7] * c;   \
+    y3 += decode[10+(ofs)*7] * c;  \
+    y4 += decode[11+(ofs)*7] * c;  \
+    y5 += decode[12+(ofs)*7] * c;  \
+    y6 += decode[13+(ofs)*7] * c;  \
+    c = hc[2+(ofs)];               \
+    x0 += decode[14+(ofs)*7] * c;  \
+    x1 += decode[15+(ofs)*7] * c;  \
+    x2 += decode[16+(ofs)*7] * c;  \
+    x3 += decode[17+(ofs)*7] * c;  \
+    x4 += decode[18+(ofs)*7] * c;  \
+    x5 += decode[19+(ofs)*7] * c;  \
+    x6 += decode[20+(ofs)*7] * c;  \
+    c = hc[3+(ofs)];               \
+    y0 += decode[21+(ofs)*7] * c;  \
+    y1 += decode[22+(ofs)*7] * c;  \
+    y2 += decode[23+(ofs)*7] * c;  \
+    y3 += decode[24+(ofs)*7] * c;  \
+    y4 += decode[25+(ofs)*7] * c;  \
+    y5 += decode[26+(ofs)*7] * c;  \
+    y6 += decode[27+(ofs)*7] * c; 
+
+#define stbir__1_coeff_remnant( ofs ) \
+    STBIR_SIMD_NO_UNROLL(decode);  \
+    c = hc[0+(ofs)];               \
+    x0 += decode[0+(ofs)*7] * c;   \
+    x1 += decode[1+(ofs)*7] * c;   \
+    x2 += decode[2+(ofs)*7] * c;   \
+    x3 += decode[3+(ofs)*7] * c;   \
+    x4 += decode[4+(ofs)*7] * c;   \
+    x5 += decode[5+(ofs)*7] * c;   \
+    x6 += decode[6+(ofs)*7] * c;   \
+
+#define stbir__2_coeff_remnant( ofs ) \
+    STBIR_SIMD_NO_UNROLL(decode);  \
+    c = hc[0+(ofs)];               \
+    x0 += decode[0+(ofs)*7] * c;   \
+    x1 += decode[1+(ofs)*7] * c;   \
+    x2 += decode[2+(ofs)*7] * c;   \
+    x3 += decode[3+(ofs)*7] * c;   \
+    x4 += decode[4+(ofs)*7] * c;   \
+    x5 += decode[5+(ofs)*7] * c;   \
+    x6 += decode[6+(ofs)*7] * c;   \
+    c = hc[1+(ofs)];               \
+    y0 += decode[7+(ofs)*7] * c;   \
+    y1 += decode[8+(ofs)*7] * c;   \
+    y2 += decode[9+(ofs)*7] * c;   \
+    y3 += decode[10+(ofs)*7] * c;  \
+    y4 += decode[11+(ofs)*7] * c;  \
+    y5 += decode[12+(ofs)*7] * c;  \
+    y6 += decode[13+(ofs)*7] * c;  \
+  
+#define stbir__3_coeff_remnant( ofs ) \
+    STBIR_SIMD_NO_UNROLL(decode);  \
+    c = hc[0+(ofs)];               \
+    x0 += decode[0+(ofs)*7] * c;   \
+    x1 += decode[1+(ofs)*7] * c;   \
+    x2 += decode[2+(ofs)*7] * c;   \
+    x3 += decode[3+(ofs)*7] * c;   \
+    x4 += decode[4+(ofs)*7] * c;   \
+    x5 += decode[5+(ofs)*7] * c;   \
+    x6 += decode[6+(ofs)*7] * c;   \
+    c = hc[1+(ofs)];               \
+    y0 += decode[7+(ofs)*7] * c;   \
+    y1 += decode[8+(ofs)*7] * c;   \
+    y2 += decode[9+(ofs)*7] * c;   \
+    y3 += decode[10+(ofs)*7] * c;  \
+    y4 += decode[11+(ofs)*7] * c;  \
+    y5 += decode[12+(ofs)*7] * c;  \
+    y6 += decode[13+(ofs)*7] * c;  \
+    c = hc[2+(ofs)];               \
+    x0 += decode[14+(ofs)*7] * c;  \
+    x1 += decode[15+(ofs)*7] * c;  \
+    x2 += decode[16+(ofs)*7] * c;  \
+    x3 += decode[17+(ofs)*7] * c;  \
+    x4 += decode[18+(ofs)*7] * c;  \
+    x5 += decode[19+(ofs)*7] * c;  \
+    x6 += decode[20+(ofs)*7] * c;  \
+
+#define stbir__store_output()                     \
+    output[0] = x0 + y0;                          \
+    output[1] = x1 + y1;                          \
+    output[2] = x2 + y2;                          \
+    output[3] = x3 + y3;                          \
+    output[4] = x4 + y4;                          \
+    output[5] = x5 + y5;                          \
+    output[6] = x6 + y6;                          \
+    horizontal_coefficients += coefficient_width; \
+    ++horizontal_contributors;                    \
+    output += 7;
+
+#endif  
+
+#define STBIR__horizontal_channels 7
+#define STB_IMAGE_RESIZE_DO_HORIZONTALS
+#include STBIR__HEADER_FILENAME
+
+
+// include all of the vertical resamplers (both scatter and gather versions)
+
+#define STBIR__vertical_channels 1
+#define STB_IMAGE_RESIZE_DO_VERTICALS
+#include STBIR__HEADER_FILENAME
+
+#define STBIR__vertical_channels 1
+#define STB_IMAGE_RESIZE_DO_VERTICALS
+#define STB_IMAGE_RESIZE_VERTICAL_CONTINUE
+#include STBIR__HEADER_FILENAME
+
+#define STBIR__vertical_channels 2
+#define STB_IMAGE_RESIZE_DO_VERTICALS
+#include STBIR__HEADER_FILENAME
+
+#define STBIR__vertical_channels 2
+#define STB_IMAGE_RESIZE_DO_VERTICALS
+#define STB_IMAGE_RESIZE_VERTICAL_CONTINUE
+#include STBIR__HEADER_FILENAME
+
+#define STBIR__vertical_channels 3
+#define STB_IMAGE_RESIZE_DO_VERTICALS
+#include STBIR__HEADER_FILENAME
+
+#define STBIR__vertical_channels 3
+#define STB_IMAGE_RESIZE_DO_VERTICALS
+#define STB_IMAGE_RESIZE_VERTICAL_CONTINUE
+#include STBIR__HEADER_FILENAME
+
+#define STBIR__vertical_channels 4
+#define STB_IMAGE_RESIZE_DO_VERTICALS
+#include STBIR__HEADER_FILENAME
+
+#define STBIR__vertical_channels 4
+#define STB_IMAGE_RESIZE_DO_VERTICALS
+#define STB_IMAGE_RESIZE_VERTICAL_CONTINUE
+#include STBIR__HEADER_FILENAME
+
+#define STBIR__vertical_channels 5
+#define STB_IMAGE_RESIZE_DO_VERTICALS
+#include STBIR__HEADER_FILENAME
+
+#define STBIR__vertical_channels 5
+#define STB_IMAGE_RESIZE_DO_VERTICALS
+#define STB_IMAGE_RESIZE_VERTICAL_CONTINUE
+#include STBIR__HEADER_FILENAME
+
+#define STBIR__vertical_channels 6
+#define STB_IMAGE_RESIZE_DO_VERTICALS
+#include STBIR__HEADER_FILENAME
+
+#define STBIR__vertical_channels 6
+#define STB_IMAGE_RESIZE_DO_VERTICALS
+#define STB_IMAGE_RESIZE_VERTICAL_CONTINUE
+#include STBIR__HEADER_FILENAME
+
+#define STBIR__vertical_channels 7
+#define STB_IMAGE_RESIZE_DO_VERTICALS
+#include STBIR__HEADER_FILENAME
+
+#define STBIR__vertical_channels 7
+#define STB_IMAGE_RESIZE_DO_VERTICALS
+#define STB_IMAGE_RESIZE_VERTICAL_CONTINUE
+#include STBIR__HEADER_FILENAME
+
+#define STBIR__vertical_channels 8
+#define STB_IMAGE_RESIZE_DO_VERTICALS
+#include STBIR__HEADER_FILENAME
+
+#define STBIR__vertical_channels 8
+#define STB_IMAGE_RESIZE_DO_VERTICALS
+#define STB_IMAGE_RESIZE_VERTICAL_CONTINUE
+#include STBIR__HEADER_FILENAME
+
+typedef void STBIR_VERTICAL_GATHERFUNC( float * output, float const * coeffs, float const ** inputs, float const * input0_end );
+
+static STBIR_VERTICAL_GATHERFUNC * stbir__vertical_gathers[ 8 ] =
+{
+  stbir__vertical_gather_with_1_coeffs,stbir__vertical_gather_with_2_coeffs,stbir__vertical_gather_with_3_coeffs,stbir__vertical_gather_with_4_coeffs,stbir__vertical_gather_with_5_coeffs,stbir__vertical_gather_with_6_coeffs,stbir__vertical_gather_with_7_coeffs,stbir__vertical_gather_with_8_coeffs
+};
+
+static STBIR_VERTICAL_GATHERFUNC * stbir__vertical_gathers_continues[ 8 ] =
+{
+  stbir__vertical_gather_with_1_coeffs_cont,stbir__vertical_gather_with_2_coeffs_cont,stbir__vertical_gather_with_3_coeffs_cont,stbir__vertical_gather_with_4_coeffs_cont,stbir__vertical_gather_with_5_coeffs_cont,stbir__vertical_gather_with_6_coeffs_cont,stbir__vertical_gather_with_7_coeffs_cont,stbir__vertical_gather_with_8_coeffs_cont
+};
+
+typedef void STBIR_VERTICAL_SCATTERFUNC( float ** outputs, float const * coeffs, float const * input, float const * input_end );
+
+static STBIR_VERTICAL_SCATTERFUNC * stbir__vertical_scatter_sets[ 8 ] =
+{
+  stbir__vertical_scatter_with_1_coeffs,stbir__vertical_scatter_with_2_coeffs,stbir__vertical_scatter_with_3_coeffs,stbir__vertical_scatter_with_4_coeffs,stbir__vertical_scatter_with_5_coeffs,stbir__vertical_scatter_with_6_coeffs,stbir__vertical_scatter_with_7_coeffs,stbir__vertical_scatter_with_8_coeffs
+};
+
+static STBIR_VERTICAL_SCATTERFUNC * stbir__vertical_scatter_blends[ 8 ] =
+{
+  stbir__vertical_scatter_with_1_coeffs_cont,stbir__vertical_scatter_with_2_coeffs_cont,stbir__vertical_scatter_with_3_coeffs_cont,stbir__vertical_scatter_with_4_coeffs_cont,stbir__vertical_scatter_with_5_coeffs_cont,stbir__vertical_scatter_with_6_coeffs_cont,stbir__vertical_scatter_with_7_coeffs_cont,stbir__vertical_scatter_with_8_coeffs_cont
+};
+
+
+static void stbir__encode_scanline( stbir__info const * stbir_info, void *output_buffer_data, float * encode_buffer, int row  STBIR_ONLY_PROFILE_GET_SPLIT_INFO )
+{
+  int num_pixels = stbir_info->horizontal.scale_info.output_sub_size;
+  int channels = stbir_info->channels;
+  int width_times_channels = num_pixels * channels;
+  void * output_buffer;
+
+  // un-alpha weight if we need to
+  if ( stbir_info->alpha_unweight )
+  {
+    STBIR_PROFILE_START( unalpha );
+    stbir_info->alpha_unweight( encode_buffer, width_times_channels );
+    STBIR_PROFILE_END( unalpha );
+  }
+
+  // write directly into output by default
+  output_buffer = output_buffer_data;
+
+  // if we have an output callback, we first convert the decode buffer in place (and then hand that to the callback)
+  if ( stbir_info->out_pixels_cb )
+    output_buffer = encode_buffer;
+  
+  STBIR_PROFILE_START( encode );
+  // convert into the output buffer
+  stbir_info->encode_pixels( output_buffer, width_times_channels, encode_buffer );
+  STBIR_PROFILE_END( encode );
+
+  // if we have an output callback, call it to send the data
+  if ( stbir_info->out_pixels_cb )
+    stbir_info->out_pixels_cb( output_buffer_data, num_pixels, row, stbir_info->user_data );
+}
+
+
+// Get the ring buffer pointer for an index
+static float* stbir__get_ring_buffer_entry(stbir__info const * stbir_info, stbir__per_split_info const * split_info, int index )
+{
+  STBIR_ASSERT( index < stbir_info->ring_buffer_num_entries );
+
+  #ifdef STBIR__SEPARATE_ALLOCATIONS
+    return split_info->ring_buffers[ index ];
+  #else
+    return (float*) ( ( (char*) split_info->ring_buffer ) + ( index * stbir_info->ring_buffer_length_bytes ) );
+  #endif
+}
+
+// Get the specified scan line from the ring buffer
+static float* stbir__get_ring_buffer_scanline(stbir__info const * stbir_info, stbir__per_split_info const * split_info, int get_scanline)
+{
+  int ring_buffer_index = (split_info->ring_buffer_begin_index + (get_scanline - split_info->ring_buffer_first_scanline)) % stbir_info->ring_buffer_num_entries;
+  return stbir__get_ring_buffer_entry( stbir_info, split_info, ring_buffer_index );
+}
+
+static void stbir__resample_horizontal_gather(stbir__info const * stbir_info, float* output_buffer, float const * input_buffer STBIR_ONLY_PROFILE_GET_SPLIT_INFO )
+{
+  float const * decode_buffer = input_buffer - ( stbir_info->scanline_extents.conservative.n0 * stbir_info->effective_channels );
+
+  STBIR_PROFILE_START( horizontal );
+  if ( ( stbir_info->horizontal.filter_enum == STBIR_FILTER_POINT_SAMPLE ) && ( stbir_info->horizontal.scale_info.scale == 1.0f ) )
+    STBIR_MEMCPY( output_buffer, input_buffer, stbir_info->horizontal.scale_info.output_sub_size * sizeof( float ) * stbir_info->effective_channels );
+  else
+    stbir_info->horizontal_gather_channels( output_buffer, stbir_info->horizontal.scale_info.output_sub_size, decode_buffer, stbir_info->horizontal.contributors, stbir_info->horizontal.coefficients, stbir_info->horizontal.coefficient_width );
+  STBIR_PROFILE_END( horizontal );
+}
+
+static void stbir__resample_vertical_gather(stbir__info const * stbir_info, stbir__per_split_info* split_info, int n, int contrib_n0, int contrib_n1, float const * vertical_coefficients )
+{
+  float* encode_buffer = split_info->vertical_buffer;
+  float* decode_buffer = split_info->decode_buffer;
+  int vertical_first = stbir_info->vertical_first;
+  int width = (vertical_first) ? ( stbir_info->scanline_extents.conservative.n1-stbir_info->scanline_extents.conservative.n0+1 ) : stbir_info->horizontal.scale_info.output_sub_size;
+  int width_times_channels = stbir_info->effective_channels * width;
+
+  STBIR_ASSERT( stbir_info->vertical.is_gather );
+
+  // loop over the contributing scanlines and scale into the buffer
+  STBIR_PROFILE_START( vertical );
+  {
+    int k = 0, total = contrib_n1 - contrib_n0 + 1;
+    STBIR_ASSERT( total > 0 );
+    do {
+      float const * inputs[8];
+      int i, cnt = total; if ( cnt > 8 ) cnt = 8;
+      for( i = 0 ; i < cnt ; i++ )
+        inputs[ i ] = stbir__get_ring_buffer_scanline(stbir_info, split_info, k+i+contrib_n0 );
+
+      // call the N scanlines at a time function (up to 8 scanlines of blending at once)
+      ((k==0)?stbir__vertical_gathers:stbir__vertical_gathers_continues)[cnt-1]( (vertical_first) ? decode_buffer : encode_buffer, vertical_coefficients + k, inputs, inputs[0] + width_times_channels );
+      k += cnt;
+      total -= cnt;
+    } while ( total );
+  }
+  STBIR_PROFILE_END( vertical );
+
+  if ( vertical_first )
+  {
+    // Now resample the gathered vertical data in the horizontal axis into the encode buffer
+    stbir__resample_horizontal_gather(stbir_info, encode_buffer, decode_buffer  STBIR_ONLY_PROFILE_SET_SPLIT_INFO );
+  }
+
+  stbir__encode_scanline( stbir_info, ( (char *) stbir_info->output_data ) + ((ptrdiff_t)n * (ptrdiff_t)stbir_info->output_stride_bytes), 
+                          encode_buffer, n  STBIR_ONLY_PROFILE_SET_SPLIT_INFO );
+}
+
+static void stbir__decode_and_resample_for_vertical_gather_loop(stbir__info const * stbir_info, stbir__per_split_info* split_info, int n)
+{
+  int ring_buffer_index;
+  float* ring_buffer;
+
+  // Decode the nth scanline from the source image into the decode buffer.
+  stbir__decode_scanline( stbir_info, n, split_info->decode_buffer  STBIR_ONLY_PROFILE_SET_SPLIT_INFO );
+
+  // update new end scanline
+  split_info->ring_buffer_last_scanline = n;
+
+  // get ring buffer 
+  ring_buffer_index = (split_info->ring_buffer_begin_index + (split_info->ring_buffer_last_scanline - split_info->ring_buffer_first_scanline)) % stbir_info->ring_buffer_num_entries;
+  ring_buffer = stbir__get_ring_buffer_entry(stbir_info, split_info, ring_buffer_index);
+
+  // Now resample it into the ring buffer.
+  stbir__resample_horizontal_gather( stbir_info, ring_buffer, split_info->decode_buffer  STBIR_ONLY_PROFILE_SET_SPLIT_INFO );
+
+  // Now it's sitting in the ring buffer ready to be used as source for the vertical sampling.
+}
+
+static void stbir__vertical_gather_loop( stbir__info const * stbir_info, stbir__per_split_info* split_info, int split_count )
+{
+  int y, start_output_y, end_output_y;
+  stbir__contributors* vertical_contributors = stbir_info->vertical.contributors;
+  float const * vertical_coefficients = stbir_info->vertical.coefficients;
+
+  STBIR_ASSERT( stbir_info->vertical.is_gather );
+
+  start_output_y = split_info->start_output_y;
+  end_output_y = split_info[split_count-1].end_output_y;
+
+  vertical_contributors += start_output_y;
+  vertical_coefficients += start_output_y * stbir_info->vertical.coefficient_width;
+
+  // initialize the ring buffer for gathering
+  split_info->ring_buffer_begin_index = 0;
+  split_info->ring_buffer_first_scanline = stbir_info->vertical.extent_info.lowest;  
+  split_info->ring_buffer_last_scanline = split_info->ring_buffer_first_scanline - 1; // means "empty"
+
+  for (y = start_output_y; y < end_output_y; y++)
+  {
+    int in_first_scanline, in_last_scanline;
+
+    in_first_scanline = vertical_contributors->n0;
+    in_last_scanline = vertical_contributors->n1;
+
+    // make sure the indexing hasn't broken
+    STBIR_ASSERT( in_first_scanline >= split_info->ring_buffer_first_scanline );
+
+    // Load in new scanlines
+    while (in_last_scanline > split_info->ring_buffer_last_scanline)
+    {
+      STBIR_ASSERT( ( split_info->ring_buffer_last_scanline - split_info->ring_buffer_first_scanline + 1 ) <= stbir_info->ring_buffer_num_entries );
+
+      // make sure there was room in the ring buffer when we add new scanlines
+      if ( ( split_info->ring_buffer_last_scanline - split_info->ring_buffer_first_scanline + 1 ) == stbir_info->ring_buffer_num_entries )
+      {
+        split_info->ring_buffer_first_scanline++;
+        split_info->ring_buffer_begin_index++;
+      }
+      
+      if ( stbir_info->vertical_first )
+      {
+        float * ring_buffer = stbir__get_ring_buffer_scanline( stbir_info, split_info, ++split_info->ring_buffer_last_scanline );
+        // Decode the nth scanline from the source image into the decode buffer.
+        stbir__decode_scanline( stbir_info, split_info->ring_buffer_last_scanline, ring_buffer  STBIR_ONLY_PROFILE_SET_SPLIT_INFO ); 
+      }
+      else
+      {
+        stbir__decode_and_resample_for_vertical_gather_loop(stbir_info, split_info, split_info->ring_buffer_last_scanline + 1);
+      }
+    }
+
+    // Now all buffers should be ready to write a row of vertical sampling, so do it.
+    stbir__resample_vertical_gather(stbir_info, split_info, y, in_first_scanline, in_last_scanline, vertical_coefficients );
+
+    ++vertical_contributors;
+    vertical_coefficients += stbir_info->vertical.coefficient_width;
+  }
+}
+
+#define STBIR__FLOAT_EMPTY_MARKER 3.0e+38F
+#define STBIR__FLOAT_BUFFER_IS_EMPTY(ptr) ((ptr)[0]==STBIR__FLOAT_EMPTY_MARKER)
+
+static void stbir__encode_first_scanline_from_scatter(stbir__info const * stbir_info, stbir__per_split_info* split_info)
+{
+  // evict a scanline out into the output buffer
+  float* ring_buffer_entry = stbir__get_ring_buffer_entry(stbir_info, split_info, split_info->ring_buffer_begin_index );
+  
+  // dump the scanline out
+  stbir__encode_scanline( stbir_info, ( (char *)stbir_info->output_data ) + ( (ptrdiff_t)split_info->ring_buffer_first_scanline * (ptrdiff_t)stbir_info->output_stride_bytes ), ring_buffer_entry, split_info->ring_buffer_first_scanline  STBIR_ONLY_PROFILE_SET_SPLIT_INFO );
+  
+  // mark it as empty
+  ring_buffer_entry[ 0 ] = STBIR__FLOAT_EMPTY_MARKER;
+
+  // advance the first scanline
+  split_info->ring_buffer_first_scanline++;
+  if ( ++split_info->ring_buffer_begin_index == stbir_info->ring_buffer_num_entries )
+    split_info->ring_buffer_begin_index = 0;
+}
+
+static void stbir__horizontal_resample_and_encode_first_scanline_from_scatter(stbir__info const * stbir_info, stbir__per_split_info* split_info)
+{
+  // evict a scanline out into the output buffer
+
+  float* ring_buffer_entry = stbir__get_ring_buffer_entry(stbir_info, split_info, split_info->ring_buffer_begin_index );
+
+  // Now resample it into the buffer.
+  stbir__resample_horizontal_gather( stbir_info, split_info->vertical_buffer, ring_buffer_entry  STBIR_ONLY_PROFILE_SET_SPLIT_INFO );
+  
+  // dump the scanline out
+  stbir__encode_scanline( stbir_info, ( (char *)stbir_info->output_data ) + ( (ptrdiff_t)split_info->ring_buffer_first_scanline * (ptrdiff_t)stbir_info->output_stride_bytes ), split_info->vertical_buffer, split_info->ring_buffer_first_scanline  STBIR_ONLY_PROFILE_SET_SPLIT_INFO );
+  
+  // mark it as empty
+  ring_buffer_entry[ 0 ] = STBIR__FLOAT_EMPTY_MARKER;
+
+  // advance the first scanline
+  split_info->ring_buffer_first_scanline++;
+  if ( ++split_info->ring_buffer_begin_index == stbir_info->ring_buffer_num_entries )
+    split_info->ring_buffer_begin_index = 0;
+}
+
+static void stbir__resample_vertical_scatter(stbir__info const * stbir_info, stbir__per_split_info* split_info, int n0, int n1, float const * vertical_coefficients, float const * vertical_buffer, float const * vertical_buffer_end )
+{
+  STBIR_ASSERT( !stbir_info->vertical.is_gather );
+
+  STBIR_PROFILE_START( vertical );
+  {
+    int k = 0, total = n1 - n0 + 1;
+    STBIR_ASSERT( total > 0 );
+    do {
+      float * outputs[8];
+      int i, n = total; if ( n > 8 ) n = 8;
+      for( i = 0 ; i < n ; i++ )
+      {
+        outputs[ i ] = stbir__get_ring_buffer_scanline(stbir_info, split_info, k+i+n0 );
+        if ( ( i ) && ( STBIR__FLOAT_BUFFER_IS_EMPTY( outputs[i] ) != STBIR__FLOAT_BUFFER_IS_EMPTY( outputs[0] ) ) ) // make sure runs are of the same type
+        {
+          n = i;
+          break;
+        }
+      }
+      // call the scatter to N scanlines at a time function (up to 8 scanlines of scattering at once)
+      ((STBIR__FLOAT_BUFFER_IS_EMPTY( outputs[0] ))?stbir__vertical_scatter_sets:stbir__vertical_scatter_blends)[n-1]( outputs, vertical_coefficients + k, vertical_buffer, vertical_buffer_end );
+      k += n;
+      total -= n;
+    } while ( total );
+  }
+
+  STBIR_PROFILE_END( vertical );
+}
+
+typedef void stbir__handle_scanline_for_scatter_func(stbir__info const * stbir_info, stbir__per_split_info* split_info); 
+
+static void stbir__vertical_scatter_loop( stbir__info const * stbir_info, stbir__per_split_info* split_info, int split_count )
+{
+  int y, start_output_y, end_output_y, start_input_y, end_input_y;
+  stbir__contributors* vertical_contributors = stbir_info->vertical.contributors;
+  float const * vertical_coefficients = stbir_info->vertical.coefficients;
+  stbir__handle_scanline_for_scatter_func * handle_scanline_for_scatter;
+  void * scanline_scatter_buffer;
+  void * scanline_scatter_buffer_end;
+  int on_first_input_y, last_input_y;
+
+  STBIR_ASSERT( !stbir_info->vertical.is_gather );
+
+  start_output_y = split_info->start_output_y;
+  end_output_y = split_info[split_count-1].end_output_y;  // may do multiple split counts
+
+  start_input_y = split_info->start_input_y;
+  end_input_y = split_info[split_count-1].end_input_y;
+
+  // adjust for starting offset start_input_y
+  y = start_input_y + stbir_info->vertical.filter_pixel_margin; 
+  vertical_contributors += y ;
+  vertical_coefficients += stbir_info->vertical.coefficient_width * y;
+
+  if ( stbir_info->vertical_first )
+  {
+    handle_scanline_for_scatter = stbir__horizontal_resample_and_encode_first_scanline_from_scatter;
+    scanline_scatter_buffer = split_info->decode_buffer;
+    scanline_scatter_buffer_end = ( (char*) scanline_scatter_buffer ) + sizeof( float ) * stbir_info->effective_channels * (stbir_info->scanline_extents.conservative.n1-stbir_info->scanline_extents.conservative.n0+1);
+  }
+  else
+  {
+    handle_scanline_for_scatter = stbir__encode_first_scanline_from_scatter;
+    scanline_scatter_buffer = split_info->vertical_buffer;
+    scanline_scatter_buffer_end = ( (char*) scanline_scatter_buffer ) + sizeof( float ) * stbir_info->effective_channels * stbir_info->horizontal.scale_info.output_sub_size;
+  }
+
+  // initialize the ring buffer for scattering
+  split_info->ring_buffer_first_scanline = start_output_y;
+  split_info->ring_buffer_last_scanline = -1;
+  split_info->ring_buffer_begin_index = -1;
+
+  // mark all the buffers as empty to start
+  for( y = 0 ; y < stbir_info->ring_buffer_num_entries ; y++ )
+    stbir__get_ring_buffer_entry( stbir_info, split_info, y )[0] = STBIR__FLOAT_EMPTY_MARKER; // only used on scatter
+
+  // do the loop in input space
+  on_first_input_y = 1; last_input_y = start_input_y;
+  for (y = start_input_y ; y < end_input_y; y++)
+  {
+    int out_first_scanline, out_last_scanline;
+
+    out_first_scanline = vertical_contributors->n0;
+    out_last_scanline = vertical_contributors->n1;
+
+    STBIR_ASSERT(out_last_scanline - out_first_scanline + 1 <= stbir_info->ring_buffer_num_entries);
+
+    if ( ( out_last_scanline >= out_first_scanline ) && ( ( ( out_first_scanline >= start_output_y ) && ( out_first_scanline < end_output_y ) ) || ( ( out_last_scanline >= start_output_y ) && ( out_last_scanline < end_output_y ) ) ) )
+    {
+      float const * vc = vertical_coefficients;
+
+      // keep track of the range actually seen for the next resize
+      last_input_y = y;
+      if ( ( on_first_input_y ) && ( y > start_input_y ) )
+        split_info->start_input_y = y;
+      on_first_input_y = 0;
+
+      // clip the region 
+      if ( out_first_scanline < start_output_y )
+      {
+        vc += start_output_y - out_first_scanline;
+        out_first_scanline = start_output_y;
+      }
+
+      if ( out_last_scanline >= end_output_y )
+        out_last_scanline = end_output_y - 1;
+
+      // if very first scanline, init the index
+      if (split_info->ring_buffer_begin_index < 0)
+        split_info->ring_buffer_begin_index = out_first_scanline - start_output_y;
+      
+      STBIR_ASSERT( split_info->ring_buffer_begin_index <= out_first_scanline );
+
+      // Decode the nth scanline from the source image into the decode buffer.
+      stbir__decode_scanline( stbir_info, y, split_info->decode_buffer  STBIR_ONLY_PROFILE_SET_SPLIT_INFO ); 
+
+      // When horizontal first, we resample horizontally into the vertical buffer before we scatter it out
+      if ( !stbir_info->vertical_first )
+        stbir__resample_horizontal_gather( stbir_info, split_info->vertical_buffer, split_info->decode_buffer  STBIR_ONLY_PROFILE_SET_SPLIT_INFO );
+
+      // Now it's sitting in the buffer ready to be distributed into the ring buffers.
+
+      // evict from the ringbuffer, if we need are full
+      if ( ( ( split_info->ring_buffer_last_scanline - split_info->ring_buffer_first_scanline + 1 ) == stbir_info->ring_buffer_num_entries ) &&
+           ( out_last_scanline > split_info->ring_buffer_last_scanline ) )
+        handle_scanline_for_scatter( stbir_info, split_info );
+    
+      // Now the horizontal buffer is ready to write to all ring buffer rows, so do it.
+      stbir__resample_vertical_scatter(stbir_info, split_info, out_first_scanline, out_last_scanline, vc, (float*)scanline_scatter_buffer, (float*)scanline_scatter_buffer_end );
+
+      // update the end of the buffer
+      if ( out_last_scanline > split_info->ring_buffer_last_scanline )
+        split_info->ring_buffer_last_scanline = out_last_scanline;
+    }
+    ++vertical_contributors;
+    vertical_coefficients += stbir_info->vertical.coefficient_width;
+  }
+
+  // now evict the scanlines that are left over in the ring buffer
+  while ( split_info->ring_buffer_first_scanline < end_output_y )
+    handle_scanline_for_scatter(stbir_info, split_info);
+
+  // update the end_input_y if we do multiple resizes with the same data
+  ++last_input_y;
+  for( y = 0 ; y < split_count; y++ )
+    if ( split_info[y].end_input_y > last_input_y )
+      split_info[y].end_input_y = last_input_y;
+}
+
+
+static stbir__kernel_callback * stbir__builtin_kernels[] =   { 0, stbir__filter_trapezoid,  stbir__filter_triangle, stbir__filter_cubic, stbir__filter_catmullrom, stbir__filter_mitchell, stbir__filter_point };
+static stbir__support_callback * stbir__builtin_supports[] = { 0, stbir__support_trapezoid, stbir__support_one,     stbir__support_two,  stbir__support_two,       stbir__support_two,     stbir__support_zeropoint5 };
+
+static void stbir__set_sampler(stbir__sampler * samp, stbir_filter filter, stbir__kernel_callback * kernel, stbir__support_callback * support, stbir_edge edge, stbir__scale_info * scale_info, int always_gather, void * user_data )
+{
+  // set filter
+  if (filter == 0)
+  {
+    filter = STBIR_DEFAULT_FILTER_DOWNSAMPLE; // default to downsample
+    if (scale_info->scale >= ( 1.0f - stbir__small_float ) )
+    {
+      if ( (scale_info->scale <= ( 1.0f + stbir__small_float ) ) && ( STBIR_CEILF(scale_info->pixel_shift) == scale_info->pixel_shift ) )
+        filter = STBIR_FILTER_POINT_SAMPLE;  
+      else
+        filter = STBIR_DEFAULT_FILTER_UPSAMPLE;
+    }
+  }
+  samp->filter_enum = filter;
+
+  STBIR_ASSERT(samp->filter_enum != 0);
+  STBIR_ASSERT((unsigned)samp->filter_enum < STBIR_FILTER_OTHER); 
+  samp->filter_kernel = stbir__builtin_kernels[ filter ];
+  samp->filter_support = stbir__builtin_supports[ filter ];
+
+  if ( kernel && support )
+  {
+    samp->filter_kernel = kernel;
+    samp->filter_support = support;
+    samp->filter_enum = STBIR_FILTER_OTHER;
+  }
+
+  samp->edge = edge;
+  samp->filter_pixel_width  = stbir__get_filter_pixel_width (samp->filter_support, scale_info->scale, user_data );
+  // Gather is always better, but in extreme downsamples, you have to most or all of the data in memory
+  //    For horizontal, we always have all the pixels, so we always use gather here (always_gather==1).
+  //    For vertical, we use gather if scaling up (which means we will have samp->filter_pixel_width
+  //    scanlines in memory at once).
+  samp->is_gather = 0;
+  if ( scale_info->scale >= ( 1.0f - stbir__small_float ) )
+    samp->is_gather = 1;
+  else if ( ( always_gather ) || ( samp->filter_pixel_width <= STBIR_FORCE_GATHER_FILTER_SCANLINES_AMOUNT ) )
+    samp->is_gather = 2;
+
+  // pre calculate stuff based on the above
+  samp->coefficient_width = stbir__get_coefficient_width(samp, samp->is_gather, user_data);
+
+  if ( edge == STBIR_EDGE_WRAP )
+    if ( samp->filter_pixel_width > ( scale_info->input_full_size * 2 ) )  // this can only happen when shrinking to a single pixel
+      samp->filter_pixel_width = scale_info->input_full_size * 2;
+
+  // This is how much to expand buffers to account for filters seeking outside
+  // the image boundaries.
+  samp->filter_pixel_margin = samp->filter_pixel_width / 2;
+
+  samp->num_contributors = stbir__get_contributors(samp, samp->is_gather);
+  samp->contributors_size = samp->num_contributors * sizeof(stbir__contributors);
+  samp->coefficients_size = samp->num_contributors * samp->coefficient_width * sizeof(float) + sizeof(float); // extra sizeof(float) is padding
+
+  samp->gather_prescatter_contributors = 0;
+  samp->gather_prescatter_coefficients = 0;
+  if ( samp->is_gather == 0 )
+  {
+    samp->gather_prescatter_coefficient_width = samp->filter_pixel_width;
+    samp->gather_prescatter_num_contributors  = stbir__get_contributors(samp, 2);
+    samp->gather_prescatter_contributors_size = samp->gather_prescatter_num_contributors * sizeof(stbir__contributors);
+    samp->gather_prescatter_coefficients_size = samp->gather_prescatter_num_contributors * samp->gather_prescatter_coefficient_width * sizeof(float);
+  }
+}
+
+static void stbir__get_conservative_extents( stbir__sampler * samp, stbir__contributors * range, void * user_data )
+{
+  float scale = samp->scale_info.scale;
+  float out_shift = samp->scale_info.pixel_shift;
+  stbir__support_callback * support = samp->filter_support;
+  int input_full_size = samp->scale_info.input_full_size;
+  stbir_edge edge = samp->edge;
+  float inv_scale = samp->scale_info.inv_scale;
+
+  STBIR_ASSERT( samp->is_gather != 0 );
+
+  if ( samp->is_gather == 1 )
+  {
+    int in_first_pixel, in_last_pixel;
+    float out_filter_radius = support(inv_scale, user_data) * scale;
+
+    stbir__calculate_in_pixel_range( &in_first_pixel, &in_last_pixel, 0.5, out_filter_radius, inv_scale, out_shift, input_full_size, edge );
+    range->n0 = in_first_pixel;
+    stbir__calculate_in_pixel_range( &in_first_pixel, &in_last_pixel, ( (float)(samp->scale_info.output_sub_size-1) ) + 0.5f, out_filter_radius, inv_scale, out_shift, input_full_size, edge );
+    range->n1 = in_last_pixel;
+  }
+  else if ( samp->is_gather == 2 ) // downsample gather, refine
+  {
+    float in_pixels_radius = support(scale, user_data) * inv_scale;
+    int filter_pixel_margin = samp->filter_pixel_margin;
+    int output_sub_size = samp->scale_info.output_sub_size;
+    int input_end;
+    int n;
+    int in_first_pixel, in_last_pixel;
+
+    // get a conservative area of the input range
+    stbir__calculate_in_pixel_range( &in_first_pixel, &in_last_pixel, 0, 0, inv_scale, out_shift, input_full_size, edge );
+    range->n0 = in_first_pixel;
+    stbir__calculate_in_pixel_range( &in_first_pixel, &in_last_pixel, (float)output_sub_size, 0, inv_scale, out_shift, input_full_size, edge );
+    range->n1 = in_last_pixel;
+     
+    // now go through the margin to the start of area to find bottom 
+    n = range->n0 + 1;
+    input_end = -filter_pixel_margin;
+    while( n >= input_end )
+    {
+      int out_first_pixel, out_last_pixel;
+      stbir__calculate_out_pixel_range( &out_first_pixel, &out_last_pixel, ((float)n)+0.5f, in_pixels_radius, scale, out_shift, output_sub_size );
+      if ( out_first_pixel > out_last_pixel )
+        break;
+
+      if ( ( out_first_pixel < output_sub_size ) || ( out_last_pixel >= 0 ) )
+        range->n0 = n;
+      --n;
+    }
+
+    // now go through the end of the area through the margin to find top 
+    n = range->n1 - 1;
+    input_end = n + 1 + filter_pixel_margin;
+    while( n <= input_end )
+    {
+      int out_first_pixel, out_last_pixel;
+      stbir__calculate_out_pixel_range( &out_first_pixel, &out_last_pixel, ((float)n)+0.5f, in_pixels_radius, scale, out_shift, output_sub_size );
+      if ( out_first_pixel > out_last_pixel )
+        break;
+      if ( ( out_first_pixel < output_sub_size ) || ( out_last_pixel >= 0 ) )
+        range->n1 = n;
+      ++n;
+    }
+  }
+
+  if ( samp->edge == STBIR_EDGE_WRAP )
+  {
+    // if we are wrapping, and we are very close to the image size (so the edges might merge), just use the scanline up to the edge
+    if ( ( range->n0 > 0 ) && ( range->n1 >= input_full_size ) )
+    {
+      int marg = range->n1 - input_full_size + 1;
+      if ( ( marg + STBIR__MERGE_RUNS_PIXEL_THRESHOLD ) >= range->n0 )
+        range->n0 = 0;
+    }
+    if ( ( range->n0 < 0 ) && ( range->n1 < (input_full_size-1) ) )
+    {
+      int marg = -range->n0;
+      if ( ( input_full_size - marg - STBIR__MERGE_RUNS_PIXEL_THRESHOLD - 1 ) <= range->n1 )
+        range->n1 = input_full_size - 1;
+    }
+  }
+  else
+  {
+    // for non-edge-wrap modes, we never read over the edge, so clamp
+    if ( range->n0 < 0 )
+      range->n0 = 0;
+    if ( range->n1 >= input_full_size )
+      range->n1 = input_full_size - 1;
+  }
+}
+
+static void stbir__get_split_info( stbir__per_split_info* split_info, int splits, int output_height, int vertical_pixel_margin, int input_full_height )
+{
+  int i, cur;
+  int left = output_height;
+
+  cur = 0;
+  for( i = 0 ; i < splits ; i++ )
+  {
+    int each; 
+    split_info[i].start_output_y = cur;
+    each = left / ( splits - i );
+    split_info[i].end_output_y = cur + each;
+    cur += each;
+    left -= each;
+
+    // scatter range (updated to minimum as you run it)
+    split_info[i].start_input_y = -vertical_pixel_margin;
+    split_info[i].end_input_y = input_full_height + vertical_pixel_margin;
+  }
+}
+
+static void stbir__free_internal_mem( stbir__info *info )
+{
+  #define STBIR__FREE_AND_CLEAR( ptr ) { if ( ptr ) { void * p = (ptr); (ptr) = 0; STBIR_FREE( p, info->user_data); } }
+  
+  if ( info )
+  {
+  #ifndef STBIR__SEPARATE_ALLOCATIONS
+    STBIR__FREE_AND_CLEAR( info->alloced_mem );
+  #else
+    int i,j;
+
+    if ( ( info->vertical.gather_prescatter_contributors ) && ( (void*)info->vertical.gather_prescatter_contributors != (void*)info->split_info[0].decode_buffer ) )
+    {
+      STBIR__FREE_AND_CLEAR( info->vertical.gather_prescatter_coefficients );
+      STBIR__FREE_AND_CLEAR( info->vertical.gather_prescatter_contributors );
+    }
+    for( i = 0 ; i < info->splits ; i++ )
+    {
+      for( j = 0 ; j < info->alloc_ring_buffer_num_entries ; j++ )
+      {
+        #ifdef STBIR_SIMD8
+        if ( info->effective_channels == 3 ) 
+          --info->split_info[i].ring_buffers[j]; // avx in 3 channel mode needs one float at the start of the buffer
+        #endif  
+        STBIR__FREE_AND_CLEAR( info->split_info[i].ring_buffers[j] );
+      }
+
+      #ifdef STBIR_SIMD8
+      if ( info->effective_channels == 3 ) 
+        --info->split_info[i].decode_buffer; // avx in 3 channel mode needs one float at the start of the buffer
+      #endif  
+      STBIR__FREE_AND_CLEAR( info->split_info[i].decode_buffer );
+      STBIR__FREE_AND_CLEAR( info->split_info[i].ring_buffers );
+      STBIR__FREE_AND_CLEAR( info->split_info[i].vertical_buffer );
+    }
+    STBIR__FREE_AND_CLEAR( info->split_info );
+    if ( info->vertical.coefficients != info->horizontal.coefficients )
+    {
+      STBIR__FREE_AND_CLEAR( info->vertical.coefficients );
+      STBIR__FREE_AND_CLEAR( info->vertical.contributors );
+    }
+    STBIR__FREE_AND_CLEAR( info->horizontal.coefficients );
+    STBIR__FREE_AND_CLEAR( info->horizontal.contributors );
+    STBIR__FREE_AND_CLEAR( info->alloced_mem );
+    STBIR__FREE_AND_CLEAR( info );
+  #endif
+  }
+  
+  #undef STBIR__FREE_AND_CLEAR
+}
+
+static int stbir__get_max_split( int splits, int height )
+{
+  int i;
+  int max = 0;
+
+  for( i = 0 ; i < splits ; i++ )
+  {
+    int each = height / ( splits - i );
+    if ( each > max ) 
+      max = each;
+    height -= each;
+  }
+  return max;
+}
+
+static stbir__horizontal_gather_channels_func ** stbir__horizontal_gather_n_coeffs_funcs[8] = 
+{ 
+  0, stbir__horizontal_gather_1_channels_with_n_coeffs_funcs, stbir__horizontal_gather_2_channels_with_n_coeffs_funcs, stbir__horizontal_gather_3_channels_with_n_coeffs_funcs, stbir__horizontal_gather_4_channels_with_n_coeffs_funcs, 0,0, stbir__horizontal_gather_7_channels_with_n_coeffs_funcs
+};
+
+static stbir__horizontal_gather_channels_func ** stbir__horizontal_gather_channels_funcs[8] = 
+{ 
+  0, stbir__horizontal_gather_1_channels_funcs, stbir__horizontal_gather_2_channels_funcs, stbir__horizontal_gather_3_channels_funcs, stbir__horizontal_gather_4_channels_funcs, 0,0, stbir__horizontal_gather_7_channels_funcs
+};
+
+// there are six resize classifications: 0 == vertical scatter, 1 == vertical gather < 1x scale, 2 == vertical gather 1x-2x scale, 4 == vertical gather < 3x scale, 4 == vertical gather > 3x scale, 5 == <=4 pixel height, 6 == <=4 pixel wide column
+#define STBIR_RESIZE_CLASSIFICATIONS 8
+
+static float stbir__compute_weights[5][STBIR_RESIZE_CLASSIFICATIONS][4]=  // 5 = 0=1chan, 1=2chan, 2=3chan, 3=4chan, 4=7chan
+{
+  {
+    { 1.00000f, 1.00000f, 0.31250f, 1.00000f },
+    { 0.56250f, 0.59375f, 0.00000f, 0.96875f },
+    { 1.00000f, 0.06250f, 0.00000f, 1.00000f },
+    { 0.00000f, 0.09375f, 1.00000f, 1.00000f },
+    { 1.00000f, 1.00000f, 1.00000f, 1.00000f },
+    { 0.03125f, 0.12500f, 1.00000f, 1.00000f },
+    { 0.06250f, 0.12500f, 0.00000f, 1.00000f },
+    { 0.00000f, 1.00000f, 0.00000f, 0.03125f },
+  }, {
+    { 0.00000f, 0.84375f, 0.00000f, 0.03125f },
+    { 0.09375f, 0.93750f, 0.00000f, 0.78125f },
+    { 0.87500f, 0.21875f, 0.00000f, 0.96875f },
+    { 0.09375f, 0.09375f, 1.00000f, 1.00000f },
+    { 1.00000f, 1.00000f, 1.00000f, 1.00000f },
+    { 0.03125f, 0.12500f, 1.00000f, 1.00000f },
+    { 0.06250f, 0.12500f, 0.00000f, 1.00000f },
+    { 0.00000f, 1.00000f, 0.00000f, 0.53125f },
+  }, {
+    { 0.00000f, 0.53125f, 0.00000f, 0.03125f },
+    { 0.06250f, 0.96875f, 0.00000f, 0.53125f },
+    { 0.87500f, 0.18750f, 0.00000f, 0.93750f },
+    { 0.00000f, 0.09375f, 1.00000f, 1.00000f },
+    { 1.00000f, 1.00000f, 1.00000f, 1.00000f },
+    { 0.03125f, 0.12500f, 1.00000f, 1.00000f },
+    { 0.06250f, 0.12500f, 0.00000f, 1.00000f },
+    { 0.00000f, 1.00000f, 0.00000f, 0.56250f },
+  }, {
+    { 0.00000f, 0.50000f, 0.00000f, 0.71875f },
+    { 0.06250f, 0.84375f, 0.00000f, 0.87500f },
+    { 1.00000f, 0.50000f, 0.50000f, 0.96875f },
+    { 1.00000f, 0.09375f, 0.31250f, 0.50000f },
+    { 1.00000f, 1.00000f, 1.00000f, 1.00000f },
+    { 1.00000f, 0.03125f, 0.03125f, 0.53125f },
+    { 0.18750f, 0.12500f, 0.00000f, 1.00000f },
+    { 0.00000f, 1.00000f, 0.03125f, 0.18750f },
+  }, {
+    { 0.00000f, 0.59375f, 0.00000f, 0.96875f },
+    { 0.06250f, 0.81250f, 0.06250f, 0.59375f },
+    { 0.75000f, 0.43750f, 0.12500f, 0.96875f },
+    { 0.87500f, 0.06250f, 0.18750f, 0.43750f },
+    { 1.00000f, 1.00000f, 1.00000f, 1.00000f },
+    { 0.15625f, 0.12500f, 1.00000f, 1.00000f },
+    { 0.06250f, 0.12500f, 0.00000f, 1.00000f },
+    { 0.00000f, 1.00000f, 0.03125f, 0.34375f },
+  }
+};
+
+// structure that allow us to query and override info for training the costs
+typedef struct STBIR__V_FIRST_INFO
+{
+  double v_cost, h_cost;
+  int control_v_first; // 0 = no control, 1 = force hori, 2 = force vert
+  int v_first;
+  int v_resize_classification;
+  int is_gather;
+} STBIR__V_FIRST_INFO;
+
+#ifdef STBIR__V_FIRST_INFO_BUFFER
+static STBIR__V_FIRST_INFO STBIR__V_FIRST_INFO_BUFFER = {0};
+#define STBIR__V_FIRST_INFO_POINTER &STBIR__V_FIRST_INFO_BUFFER
+#else
+#define STBIR__V_FIRST_INFO_POINTER 0
+#endif
+
+// Figure out whether to scale along the horizontal or vertical first.
+//   This only *super* important when you are scaling by a massively 
+//   different amount in the vertical vs the horizontal (for example, if 
+//   you are scaling by 2x in the width, and 0.5x in the height, then you 
+//   want to do the vertical scale first, because it's around 3x faster 
+//   in that order.
+//
+//   In more normal circumstances, this makes a 20-40% differences, so 
+//     it's good to get right, but not critical. The normal way that you
+//     decide which direction goes first is just figuring out which 
+//     direction does more multiplies. But with modern CPUs with their 
+//     fancy caches and SIMD and high IPC abilities, so there's just a lot
+//     more that goes into it. 
+//
+//   My handwavy sort of solution is to have an app that does a whole 
+//     bunch of timing for both vertical and horizontal first modes,
+//     and then another app that can read lots of these timing files
+//     and try to search for the best weights to use. Dotimings.c
+//     is the app that does a bunch of timings, and vf_train.c is the
+//     app that solves for the best weights (and shows how well it 
+//     does currently).
+
+static int stbir__should_do_vertical_first( float weights_table[STBIR_RESIZE_CLASSIFICATIONS][4], int horizontal_filter_pixel_width, float horizontal_scale, int horizontal_output_size, int vertical_filter_pixel_width, float vertical_scale, int vertical_output_size, int is_gather, STBIR__V_FIRST_INFO * info )    
+{
+  double v_cost, h_cost;
+  float * weights;
+  int vertical_first;
+  int v_classification;
+
+  // categorize the resize into buckets
+  if ( ( vertical_output_size <= 4 ) || ( horizontal_output_size <= 4 ) )
+    v_classification = ( vertical_output_size < horizontal_output_size ) ? 6 : 7;
+  else if ( vertical_scale <= 1.0f )
+    v_classification = ( is_gather ) ? 1 : 0;
+  else if ( vertical_scale <= 2.0f) 
+    v_classification = 2;
+  else if ( vertical_scale <= 3.0f) 
+    v_classification = 3;
+  else if ( vertical_scale <= 4.0f) 
+    v_classification = 5;
+  else 
+    v_classification = 6;
+  
+  // use the right weights
+  weights = weights_table[ v_classification ];
+
+  // this is the costs when you don't take into account modern CPUs with high ipc and simd and caches - wish we had a better estimate
+  h_cost = (float)horizontal_filter_pixel_width * weights[0] + horizontal_scale * (float)vertical_filter_pixel_width * weights[1];
+  v_cost = (float)vertical_filter_pixel_width  * weights[2] + vertical_scale * (float)horizontal_filter_pixel_width * weights[3];
+
+  // use computation estimate to decide vertical first or not
+  vertical_first = ( v_cost <= h_cost ) ? 1 : 0;
+
+  // save these, if requested
+  if ( info )
+  {
+    info->h_cost = h_cost;
+    info->v_cost = v_cost;
+    info->v_resize_classification = v_classification;
+    info->v_first = vertical_first;
+    info->is_gather = is_gather;
+  }
+
+  // and this allows us to override everything for testing (see dotiming.c) 
+  if ( ( info ) && ( info->control_v_first ) ) 
+    vertical_first = ( info->control_v_first == 2 ) ? 1 : 0;
+  
+  return vertical_first;
+}
+
+// layout lookups - must match stbir_internal_pixel_layout
+static unsigned char stbir__pixel_channels[] = {
+  1,2,3,3,4,   // 1ch, 2ch, rgb, bgr, 4ch
+  4,4,4,4,2,2, // RGBA,BGRA,ARGB,ABGR,RA,AR
+  4,4,4,4,2,2, // RGBA_PM,BGRA_PM,ARGB_PM,ABGR_PM,RA_PM,AR_PM
+};
+
+// the internal pixel layout enums are in a different order, so we can easily do range comparisons of types
+//   the public pixel layout is ordered in a way that if you cast num_channels (1-4) to the enum, you get something sensible 
+static stbir_internal_pixel_layout stbir__pixel_layout_convert_public_to_internal[] = {
+  STBIRI_BGR, STBIRI_1CHANNEL, STBIRI_2CHANNEL, STBIRI_RGB, STBIRI_RGBA, 
+  STBIRI_4CHANNEL, STBIRI_BGRA, STBIRI_ARGB, STBIRI_ABGR, STBIRI_RA, STBIRI_AR,
+  STBIRI_RGBA_PM, STBIRI_BGRA_PM, STBIRI_ARGB_PM, STBIRI_ABGR_PM, STBIRI_RA_PM, STBIRI_AR_PM,
+};
+
+static stbir__info * stbir__alloc_internal_mem_and_build_samplers( stbir__sampler * horizontal, stbir__sampler * vertical, stbir__contributors * conservative, stbir_pixel_layout input_pixel_layout_public, stbir_pixel_layout output_pixel_layout_public, int splits, int new_x, int new_y, int fast_alpha, void * user_data STBIR_ONLY_PROFILE_BUILD_GET_INFO )
+{
+  static char stbir_channel_count_index[8]={ 9,0,1,2, 3,9,9,4 };
+
+  stbir__info * info = 0;
+  void * alloced = 0;
+  int alloced_total = 0;
+  int vertical_first;
+  int decode_buffer_size, ring_buffer_length_bytes, ring_buffer_size, vertical_buffer_size, alloc_ring_buffer_num_entries;
+
+  int alpha_weighting_type = 0; // 0=none, 1=simple, 2=fancy
+  int conservative_split_output_size = stbir__get_max_split( splits, vertical->scale_info.output_sub_size ); 
+  stbir_internal_pixel_layout input_pixel_layout = stbir__pixel_layout_convert_public_to_internal[ input_pixel_layout_public ];  
+  stbir_internal_pixel_layout output_pixel_layout = stbir__pixel_layout_convert_public_to_internal[ output_pixel_layout_public ];
+  int channels = stbir__pixel_channels[ input_pixel_layout ];      
+  int effective_channels = channels;
+  
+  // first figure out what type of alpha weighting to use (if any)
+  if ( ( horizontal->filter_enum != STBIR_FILTER_POINT_SAMPLE ) || ( vertical->filter_enum != STBIR_FILTER_POINT_SAMPLE ) ) // no alpha weighting on point sampling
+  {
+    if ( ( input_pixel_layout >= STBIRI_RGBA ) && ( input_pixel_layout <= STBIRI_AR ) && ( output_pixel_layout >= STBIRI_RGBA ) && ( output_pixel_layout <= STBIRI_AR ) )
+    {
+      if ( fast_alpha )
+      {
+        alpha_weighting_type = 4;
+      }
+      else
+      {
+        static int fancy_alpha_effective_cnts[6] = { 7, 7, 7, 7, 3, 3 };
+        alpha_weighting_type = 2;
+        effective_channels = fancy_alpha_effective_cnts[ input_pixel_layout - STBIRI_RGBA ];
+      }
+    }
+    else if ( ( input_pixel_layout >= STBIRI_RGBA_PM ) && ( input_pixel_layout <= STBIRI_AR_PM ) && ( output_pixel_layout >= STBIRI_RGBA ) && ( output_pixel_layout <= STBIRI_AR ) )
+    {
+      // input premult, output non-premult
+      alpha_weighting_type = 3;
+    }
+    else if ( ( input_pixel_layout >= STBIRI_RGBA ) && ( input_pixel_layout <= STBIRI_AR ) && ( output_pixel_layout >= STBIRI_RGBA_PM ) && ( output_pixel_layout <= STBIRI_AR_PM ) )
+    {
+      // input non-premult, output premult
+      alpha_weighting_type = 1;
+    }
+  }
+
+  // channel in and out count must match currently
+  if ( channels != stbir__pixel_channels[ output_pixel_layout ] )
+    return 0;
+
+  // get vertical first
+  vertical_first = stbir__should_do_vertical_first( stbir__compute_weights[ (int)stbir_channel_count_index[ effective_channels ] ], horizontal->filter_pixel_width, horizontal->scale_info.scale, horizontal->scale_info.output_sub_size, vertical->filter_pixel_width, vertical->scale_info.scale, vertical->scale_info.output_sub_size, vertical->is_gather, STBIR__V_FIRST_INFO_POINTER );
+
+  // sometimes read one float off in some of the unrolled loops (with a weight of zero coeff, so it doesn't have an effect)
+  decode_buffer_size = ( conservative->n1 - conservative->n0 + 1 ) * effective_channels * sizeof(float) + sizeof(float); // extra float for padding
+  
+#if defined( STBIR__SEPARATE_ALLOCATIONS ) && defined(STBIR_SIMD8)
+  if ( effective_channels == 3 )
+    decode_buffer_size += sizeof(float); // avx in 3 channel mode needs one float at the start of the buffer (only with separate allocations)
+#endif  
+
+  ring_buffer_length_bytes = horizontal->scale_info.output_sub_size * effective_channels * sizeof(float) + sizeof(float); // extra float for padding
+
+  // if we do vertical first, the ring buffer holds a whole decoded line
+  if ( vertical_first )
+    ring_buffer_length_bytes = ( decode_buffer_size + 15 ) & ~15;
+
+  if ( ( ring_buffer_length_bytes & 4095 ) == 0 ) ring_buffer_length_bytes += 64*3; // avoid 4k alias
+
+  // One extra entry because floating point precision problems sometimes cause an extra to be necessary.
+  alloc_ring_buffer_num_entries = vertical->filter_pixel_width + 1;
+
+  // we never need more ring buffer entries than the scanlines we're outputting when in scatter mode
+  if ( ( !vertical->is_gather ) && ( alloc_ring_buffer_num_entries > conservative_split_output_size ) )
+    alloc_ring_buffer_num_entries = conservative_split_output_size;
+
+  ring_buffer_size = alloc_ring_buffer_num_entries * ring_buffer_length_bytes;
+
+  // The vertical buffer is used differently, depending on whether we are scattering
+  //   the vertical scanlines, or gathering them.
+  //   If scattering, it's used at the temp buffer to accumulate each output.
+  //   If gathering, it's just the output buffer.
+  vertical_buffer_size = horizontal->scale_info.output_sub_size * effective_channels * sizeof(float) + sizeof(float);  // extra float for padding
+
+  // we make two passes through this loop, 1st to add everything up, 2nd to allocate and init
+  for(;;)
+  {
+    int i;
+    void * advance_mem = alloced;
+    int copy_horizontal = 0;
+    stbir__sampler * possibly_use_horizontal_for_pivot = 0;
+
+#ifdef STBIR__SEPARATE_ALLOCATIONS
+    #define STBIR__NEXT_PTR( ptr, size, ntype ) if ( alloced ) { void * p = STBIR_MALLOC( size, user_data); if ( p == 0 ) { stbir__free_internal_mem( info ); return 0; } (ptr) = (ntype*)p; }
+#else
+    #define STBIR__NEXT_PTR( ptr, size, ntype ) advance_mem = (void*) ( ( ((size_t)advance_mem) + 15 ) & ~15 ); if ( alloced ) ptr = (ntype*)advance_mem; advance_mem = ((char*)advance_mem) + (size);
+#endif
+
+    STBIR__NEXT_PTR( info, sizeof( stbir__info ), stbir__info );      
+
+    STBIR__NEXT_PTR( info->split_info, sizeof( stbir__per_split_info ) * splits, stbir__per_split_info );      
+
+    if ( info )
+    {
+      static stbir__alpha_weight_func * fancy_alpha_weights[6]  =    { stbir__fancy_alpha_weight_4ch,   stbir__fancy_alpha_weight_4ch,   stbir__fancy_alpha_weight_4ch,   stbir__fancy_alpha_weight_4ch,   stbir__fancy_alpha_weight_2ch,   stbir__fancy_alpha_weight_2ch };
+      static stbir__alpha_unweight_func * fancy_alpha_unweights[6] = { stbir__fancy_alpha_unweight_4ch, stbir__fancy_alpha_unweight_4ch, stbir__fancy_alpha_unweight_4ch, stbir__fancy_alpha_unweight_4ch, stbir__fancy_alpha_unweight_2ch, stbir__fancy_alpha_unweight_2ch };
+      static stbir__alpha_weight_func * simple_alpha_weights[6] = { stbir__simple_alpha_weight_4ch, stbir__simple_alpha_weight_4ch, stbir__simple_alpha_weight_4ch, stbir__simple_alpha_weight_4ch, stbir__simple_alpha_weight_2ch, stbir__simple_alpha_weight_2ch };
+      static stbir__alpha_unweight_func * simple_alpha_unweights[6] = { stbir__simple_alpha_unweight_4ch, stbir__simple_alpha_unweight_4ch, stbir__simple_alpha_unweight_4ch, stbir__simple_alpha_unweight_4ch, stbir__simple_alpha_unweight_2ch, stbir__simple_alpha_unweight_2ch };
+
+      // initialize info fields
+      info->alloced_mem = alloced;
+      info->alloced_total = alloced_total;
+
+      info->channels = channels;
+      info->effective_channels = effective_channels;
+  
+      info->offset_x = new_x;
+      info->offset_y = new_y;
+      info->alloc_ring_buffer_num_entries = alloc_ring_buffer_num_entries;
+      info->ring_buffer_num_entries = 0;  
+      info->ring_buffer_length_bytes = ring_buffer_length_bytes;
+      info->splits = splits;
+      info->vertical_first = vertical_first;
+
+      info->input_pixel_layout_internal = input_pixel_layout;  
+      info->output_pixel_layout_internal = output_pixel_layout;
+
+      // setup alpha weight functions
+      info->alpha_weight = 0;
+      info->alpha_unweight = 0;
+    
+      // handle alpha weighting functions and overrides
+      if ( alpha_weighting_type == 2 )
+      {
+        // high quality alpha multiplying on the way in, dividing on the way out
+        info->alpha_weight = fancy_alpha_weights[ input_pixel_layout - STBIRI_RGBA ];  
+        info->alpha_unweight = fancy_alpha_unweights[ output_pixel_layout - STBIRI_RGBA ];
+      }
+      else if ( alpha_weighting_type == 4 )
+      {
+        // fast alpha multiplying on the way in, dividing on the way out
+        info->alpha_weight = simple_alpha_weights[ input_pixel_layout - STBIRI_RGBA ];  
+        info->alpha_unweight = simple_alpha_unweights[ output_pixel_layout - STBIRI_RGBA ];
+      }
+      else if ( alpha_weighting_type == 1 )
+      {
+        // fast alpha on the way in, leave in premultiplied form on way out
+        info->alpha_weight = simple_alpha_weights[ input_pixel_layout - STBIRI_RGBA ]; 
+      }
+      else if ( alpha_weighting_type == 3 )
+      {
+        // incoming is premultiplied, fast alpha dividing on the way out - non-premultiplied output
+        info->alpha_unweight = simple_alpha_unweights[ output_pixel_layout - STBIRI_RGBA ];
+      }
+
+      // handle 3-chan color flipping, using the alpha weight path
+      if ( ( ( input_pixel_layout == STBIRI_RGB ) && ( output_pixel_layout == STBIRI_BGR ) ) ||
+           ( ( input_pixel_layout == STBIRI_BGR ) && ( output_pixel_layout == STBIRI_RGB ) ) )
+      {
+        // do the flipping on the smaller of the two ends
+        if ( horizontal->scale_info.scale < 1.0f )
+          info->alpha_unweight = stbir__simple_flip_3ch;
+        else
+          info->alpha_weight = stbir__simple_flip_3ch;
+      }
+
+    }        
+
+    // get all the per-split buffers
+    for( i = 0 ; i < splits ; i++ )
+    {
+      STBIR__NEXT_PTR( info->split_info[i].decode_buffer, decode_buffer_size, float );
+
+#ifdef STBIR__SEPARATE_ALLOCATIONS
+
+      #ifdef STBIR_SIMD8
+      if ( ( info ) && ( effective_channels == 3 ) )
+        ++info->split_info[i].decode_buffer; // avx in 3 channel mode needs one float at the start of the buffer
+      #endif  
+
+      STBIR__NEXT_PTR( info->split_info[i].ring_buffers, alloc_ring_buffer_num_entries * sizeof(float*), float* );
+      {
+        int j;
+        for( j = 0 ; j < alloc_ring_buffer_num_entries ; j++ )
+        {
+          STBIR__NEXT_PTR( info->split_info[i].ring_buffers[j], ring_buffer_length_bytes, float );
+          #ifdef STBIR_SIMD8
+          if ( ( info ) && ( effective_channels == 3 ) )
+            ++info->split_info[i].ring_buffers[j]; // avx in 3 channel mode needs one float at the start of the buffer
+          #endif  
+        }
+      }
+#else
+      STBIR__NEXT_PTR( info->split_info[i].ring_buffer, ring_buffer_size, float );
+#endif
+      STBIR__NEXT_PTR( info->split_info[i].vertical_buffer, vertical_buffer_size, float );
+    }
+
+    // alloc memory for to-be-pivoted coeffs (if necessary)
+    if ( vertical->is_gather == 0 )
+    {
+      int both;
+      int temp_mem_amt;
+
+      // when in vertical scatter mode, we first build the coefficients in gather mode, and then pivot after,
+      //   that means we need two buffers, so we try to use the decode buffer and ring buffer for this. if that
+      //   is too small, we just allocate extra memory to use as this temp.
+
+      both = vertical->gather_prescatter_contributors_size + vertical->gather_prescatter_coefficients_size;
+
+#ifdef STBIR__SEPARATE_ALLOCATIONS
+      temp_mem_amt = decode_buffer_size;
+#else
+      temp_mem_amt = ( decode_buffer_size + ring_buffer_size + vertical_buffer_size ) * splits;
+#endif
+      if ( temp_mem_amt >= both )
+      {
+        if ( info ) 
+        { 
+          vertical->gather_prescatter_contributors = (stbir__contributors*)info->split_info[0].decode_buffer; 
+          vertical->gather_prescatter_coefficients = (float*) ( ( (char*)info->split_info[0].decode_buffer ) + vertical->gather_prescatter_contributors_size ); 
+        }
+      }
+      else
+      {
+        // ring+decode memory is too small, so allocate temp memory
+        STBIR__NEXT_PTR( vertical->gather_prescatter_contributors, vertical->gather_prescatter_contributors_size, stbir__contributors );
+        STBIR__NEXT_PTR( vertical->gather_prescatter_coefficients, vertical->gather_prescatter_coefficients_size, float );
+      }
+    }
+
+    STBIR__NEXT_PTR( horizontal->contributors, horizontal->contributors_size, stbir__contributors );
+    STBIR__NEXT_PTR( horizontal->coefficients, horizontal->coefficients_size, float );
+
+    // are the two filters identical?? (happens a lot with mipmap generation)
+    if ( ( horizontal->filter_kernel == vertical->filter_kernel ) && ( horizontal->filter_support == vertical->filter_support ) && ( horizontal->edge == vertical->edge ) && ( horizontal->scale_info.output_sub_size == vertical->scale_info.output_sub_size ) )
+    {
+      float diff_scale = horizontal->scale_info.scale - vertical->scale_info.scale;
+      float diff_shift = horizontal->scale_info.pixel_shift - vertical->scale_info.pixel_shift;
+      if ( diff_scale < 0.0f ) diff_scale = -diff_scale;
+      if ( diff_shift < 0.0f ) diff_shift = -diff_shift;
+      if ( ( diff_scale <= stbir__small_float ) && ( diff_shift <= stbir__small_float ) )
+      {
+        if ( horizontal->is_gather == vertical->is_gather ) 
+        {
+          copy_horizontal = 1;
+          goto no_vert_alloc;
+        }
+        // everything matches, but vertical is scatter, horizontal is gather, use horizontal coeffs for vertical pivot coeffs
+        possibly_use_horizontal_for_pivot = horizontal;
+      }
+    }
+
+    STBIR__NEXT_PTR( vertical->contributors, vertical->contributors_size, stbir__contributors );
+    STBIR__NEXT_PTR( vertical->coefficients, vertical->coefficients_size, float );
+
+   no_vert_alloc:
+
+    if ( info )
+    {
+      STBIR_PROFILE_BUILD_START( horizontal );
+
+      stbir__calculate_filters( horizontal, 0, user_data STBIR_ONLY_PROFILE_BUILD_SET_INFO );
+
+      // setup the horizontal gather functions
+      // start with defaulting to the n_coeffs functions (specialized on channels and remnant leftover)
+      info->horizontal_gather_channels = stbir__horizontal_gather_n_coeffs_funcs[ effective_channels ][ horizontal->extent_info.widest & 3 ];
+      // but if the number of coeffs <= 12, use another set of special cases. <=12 coeffs is any enlarging resize, or shrinking resize down to about 1/3 size
+      if ( horizontal->extent_info.widest <= 12 )
+        info->horizontal_gather_channels = stbir__horizontal_gather_channels_funcs[ effective_channels ][ horizontal->extent_info.widest - 1 ];
+      
+      info->scanline_extents.conservative.n0 = conservative->n0;
+      info->scanline_extents.conservative.n1 = conservative->n1;
+      
+      // get exact extents
+      stbir__get_extents( horizontal, &info->scanline_extents );
+
+      // pack the horizontal coeffs
+      horizontal->coefficient_width = stbir__pack_coefficients(horizontal->num_contributors, horizontal->contributors, horizontal->coefficients, horizontal->coefficient_width, horizontal->extent_info.widest, info->scanline_extents.conservative.n1 + 1 );
+      
+      STBIR_MEMCPY( &info->horizontal, horizontal, sizeof( stbir__sampler ) );
+
+      STBIR_PROFILE_BUILD_END( horizontal );
+
+      if ( copy_horizontal )
+      {
+        STBIR_MEMCPY( &info->vertical, horizontal, sizeof( stbir__sampler ) );
+      }
+      else
+      {
+        STBIR_PROFILE_BUILD_START( vertical );
+
+        stbir__calculate_filters( vertical, possibly_use_horizontal_for_pivot, user_data STBIR_ONLY_PROFILE_BUILD_SET_INFO );
+        STBIR_MEMCPY( &info->vertical, vertical, sizeof( stbir__sampler ) );
+
+        STBIR_PROFILE_BUILD_END( vertical );
+      }
+
+      // setup the vertical split ranges
+      stbir__get_split_info( info->split_info, info->splits, info->vertical.scale_info.output_sub_size, info->vertical.filter_pixel_margin, info->vertical.scale_info.input_full_size );
+
+      // now we know precisely how many entries we need
+      info->ring_buffer_num_entries = info->vertical.extent_info.widest;
+
+      // we never need more ring buffer entries than the scanlines we're outputting
+      if ( ( !info->vertical.is_gather ) && ( info->ring_buffer_num_entries > conservative_split_output_size ) )
+        info->ring_buffer_num_entries = conservative_split_output_size;
+      STBIR_ASSERT( info->ring_buffer_num_entries <= info->alloc_ring_buffer_num_entries );
+
+      // a few of the horizontal gather functions read one dword past the end (but mask it out), so put in a normal value so no snans or denormals accidentally sneak in
+      for( i = 0 ; i < splits ; i++ )
+      {
+        int width, ofs;
+        
+        // find the right most span
+        if ( info->scanline_extents.spans[0].n1 > info->scanline_extents.spans[1].n1 )
+          width = info->scanline_extents.spans[0].n1 - info->scanline_extents.spans[0].n0;
+        else
+          width = info->scanline_extents.spans[1].n1 - info->scanline_extents.spans[1].n0;
+        
+        // this calc finds the exact end of the decoded scanline for all filter modes.
+        //   usually this is just the width * effective channels.  But we have to account 
+        //   for the area to the left of the scanline for wrap filtering and alignment, this 
+        //   is stored as a negative value in info->scanline_extents.conservative.n0. Next,
+        //   we need to skip the exact size of the right hand size filter area (again for
+        //   wrap mode), this is in info->scanline_extents.edge_sizes[1]).
+        ofs = ( width + 1 - info->scanline_extents.conservative.n0 + info->scanline_extents.edge_sizes[1] ) * effective_channels;
+        
+        // place a known, but numerically valid value in the decode buffer
+        info->split_info[i].decode_buffer[ ofs ] = 9999.0f;
+
+        // if vertical filtering first, place a known, but numerically valid value in the all
+        //   of the ring buffer accumulators
+        if ( vertical_first )
+        {
+          int j;  
+          for( j = 0; j < info->ring_buffer_num_entries ; j++ )
+          {
+            stbir__get_ring_buffer_entry( info, info->split_info + i, j )[ ofs ] = 9999.0f;
+          }
+        }
+      }
+    }
+
+    #undef STBIR__NEXT_PTR
+
+
+    // is this the first time through loop?
+    if ( info == 0 )
+    {
+      alloced_total = (int) ( 15 + (size_t)advance_mem );
+      alloced = STBIR_MALLOC( alloced_total, user_data );
+      if ( alloced == 0 )
+        return 0;
+    }
+    else
+      return info;  // success
+  }
+}
+
+static int stbir__perform_resize( stbir__info const * info, int split_start, int split_count ) 
+{
+  stbir__per_split_info * split_info = info->split_info + split_start;
+
+  STBIR_PROFILE_CLEAR_EXTRAS();
+
+  STBIR_PROFILE_FIRST_START( looping );
+  if (info->vertical.is_gather)
+    stbir__vertical_gather_loop( info, split_info, split_count );
+  else
+    stbir__vertical_scatter_loop( info, split_info, split_count );
+  STBIR_PROFILE_END( looping );
+
+  return 1;
+}
+
+static void stbir__update_info_from_resize( stbir__info * info, STBIR_RESIZE * resize )
+{
+  static stbir__decode_pixels_func * decode_simple[STBIR_TYPE_HALF_FLOAT-STBIR_TYPE_UINT8_SRGB+1]=
+  {
+    /* 1ch-4ch */ stbir__decode_uint8_srgb, stbir__decode_uint8_srgb, 0, stbir__decode_float_linear, stbir__decode_half_float_linear, 
+  };
+
+  static stbir__decode_pixels_func * decode_alphas[STBIRI_AR-STBIRI_RGBA+1][STBIR_TYPE_HALF_FLOAT-STBIR_TYPE_UINT8_SRGB+1]=
+  {
+    { /* RGBA */ stbir__decode_uint8_srgb4_linearalpha,      stbir__decode_uint8_srgb,      0, stbir__decode_float_linear,      stbir__decode_half_float_linear },
+    { /* BGRA */ stbir__decode_uint8_srgb4_linearalpha_BGRA, stbir__decode_uint8_srgb_BGRA, 0, stbir__decode_float_linear_BGRA, stbir__decode_half_float_linear_BGRA },
+    { /* ARGB */ stbir__decode_uint8_srgb4_linearalpha_ARGB, stbir__decode_uint8_srgb_ARGB, 0, stbir__decode_float_linear_ARGB, stbir__decode_half_float_linear_ARGB },
+    { /* ABGR */ stbir__decode_uint8_srgb4_linearalpha_ABGR, stbir__decode_uint8_srgb_ABGR, 0, stbir__decode_float_linear_ABGR, stbir__decode_half_float_linear_ABGR },
+    { /* RA   */ stbir__decode_uint8_srgb2_linearalpha,      stbir__decode_uint8_srgb,      0, stbir__decode_float_linear,      stbir__decode_half_float_linear },
+    { /* AR   */ stbir__decode_uint8_srgb2_linearalpha_AR,   stbir__decode_uint8_srgb_AR,   0, stbir__decode_float_linear_AR,   stbir__decode_half_float_linear_AR },
+  };
+
+  static stbir__decode_pixels_func * decode_simple_scaled_or_not[2][2]=
+  {
+    { stbir__decode_uint8_linear_scaled,  stbir__decode_uint8_linear }, { stbir__decode_uint16_linear_scaled, stbir__decode_uint16_linear },
+  };
+
+  static stbir__decode_pixels_func * decode_alphas_scaled_or_not[STBIRI_AR-STBIRI_RGBA+1][2][2]=
+  {
+    { /* RGBA */ { stbir__decode_uint8_linear_scaled,       stbir__decode_uint8_linear },      { stbir__decode_uint16_linear_scaled,      stbir__decode_uint16_linear } },
+    { /* BGRA */ { stbir__decode_uint8_linear_scaled_BGRA,  stbir__decode_uint8_linear_BGRA }, { stbir__decode_uint16_linear_scaled_BGRA, stbir__decode_uint16_linear_BGRA } },
+    { /* ARGB */ { stbir__decode_uint8_linear_scaled_ARGB,  stbir__decode_uint8_linear_ARGB }, { stbir__decode_uint16_linear_scaled_ARGB, stbir__decode_uint16_linear_ARGB } },
+    { /* ABGR */ { stbir__decode_uint8_linear_scaled_ABGR,  stbir__decode_uint8_linear_ABGR }, { stbir__decode_uint16_linear_scaled_ABGR, stbir__decode_uint16_linear_ABGR } },
+    { /* RA   */ { stbir__decode_uint8_linear_scaled,       stbir__decode_uint8_linear },      { stbir__decode_uint16_linear_scaled,      stbir__decode_uint16_linear } },
+    { /* AR   */ { stbir__decode_uint8_linear_scaled_AR,    stbir__decode_uint8_linear_AR },   { stbir__decode_uint16_linear_scaled_AR,   stbir__decode_uint16_linear_AR } }
+  };
+
+  static stbir__encode_pixels_func * encode_simple[STBIR_TYPE_HALF_FLOAT-STBIR_TYPE_UINT8_SRGB+1]=
+  {
+    /* 1ch-4ch */ stbir__encode_uint8_srgb, stbir__encode_uint8_srgb, 0, stbir__encode_float_linear, stbir__encode_half_float_linear,
+  };
+
+  static stbir__encode_pixels_func * encode_alphas[STBIRI_AR-STBIRI_RGBA+1][STBIR_TYPE_HALF_FLOAT-STBIR_TYPE_UINT8_SRGB+1]=
+  {
+    { /* RGBA */ stbir__encode_uint8_srgb4_linearalpha,      stbir__encode_uint8_srgb,      0, stbir__encode_float_linear,      stbir__encode_half_float_linear },
+    { /* BGRA */ stbir__encode_uint8_srgb4_linearalpha_BGRA, stbir__encode_uint8_srgb_BGRA, 0, stbir__encode_float_linear_BGRA, stbir__encode_half_float_linear_BGRA },
+    { /* ARGB */ stbir__encode_uint8_srgb4_linearalpha_ARGB, stbir__encode_uint8_srgb_ARGB, 0, stbir__encode_float_linear_ARGB, stbir__encode_half_float_linear_ARGB },
+    { /* ABGR */ stbir__encode_uint8_srgb4_linearalpha_ABGR, stbir__encode_uint8_srgb_ABGR, 0, stbir__encode_float_linear_ABGR, stbir__encode_half_float_linear_ABGR },
+    { /* RA   */ stbir__encode_uint8_srgb2_linearalpha,      stbir__encode_uint8_srgb,      0, stbir__encode_float_linear,      stbir__encode_half_float_linear },
+    { /* AR   */ stbir__encode_uint8_srgb2_linearalpha_AR,   stbir__encode_uint8_srgb_AR,   0, stbir__encode_float_linear_AR,   stbir__encode_half_float_linear_AR }
+  };
+
+  static stbir__encode_pixels_func * encode_simple_scaled_or_not[2][2]=
+  {
+    { stbir__encode_uint8_linear_scaled,  stbir__encode_uint8_linear }, { stbir__encode_uint16_linear_scaled, stbir__encode_uint16_linear },
+  };
+
+  static stbir__encode_pixels_func * encode_alphas_scaled_or_not[STBIRI_AR-STBIRI_RGBA+1][2][2]=
+  {
+    { /* RGBA */ { stbir__encode_uint8_linear_scaled,       stbir__encode_uint8_linear },       { stbir__encode_uint16_linear_scaled,      stbir__encode_uint16_linear } },
+    { /* BGRA */ { stbir__encode_uint8_linear_scaled_BGRA,  stbir__encode_uint8_linear_BGRA },  { stbir__encode_uint16_linear_scaled_BGRA, stbir__encode_uint16_linear_BGRA } },
+    { /* ARGB */ { stbir__encode_uint8_linear_scaled_ARGB,  stbir__encode_uint8_linear_ARGB },  { stbir__encode_uint16_linear_scaled_ARGB, stbir__encode_uint16_linear_ARGB } },
+    { /* ABGR */ { stbir__encode_uint8_linear_scaled_ABGR,  stbir__encode_uint8_linear_ABGR },  { stbir__encode_uint16_linear_scaled_ABGR, stbir__encode_uint16_linear_ABGR } },
+    { /* RA   */ { stbir__encode_uint8_linear_scaled,       stbir__encode_uint8_linear },       { stbir__encode_uint16_linear_scaled,      stbir__encode_uint16_linear } },
+    { /* AR   */ { stbir__encode_uint8_linear_scaled_AR,    stbir__encode_uint8_linear_AR },    { stbir__encode_uint16_linear_scaled_AR,   stbir__encode_uint16_linear_AR } }
+  };
+
+  stbir__decode_pixels_func * decode_pixels = 0;
+  stbir__encode_pixels_func * encode_pixels = 0;
+  stbir_datatype input_type, output_type;
+
+  input_type = resize->input_data_type;
+  output_type = resize->output_data_type; 
+  info->input_data = resize->input_pixels;
+  info->input_stride_bytes = resize->input_stride_in_bytes;
+  info->output_stride_bytes = resize->output_stride_in_bytes;
+
+  // if we're completely point sampling, then we can turn off SRGB
+  if ( ( info->horizontal.filter_enum == STBIR_FILTER_POINT_SAMPLE ) && ( info->vertical.filter_enum == STBIR_FILTER_POINT_SAMPLE ) )
+  {
+    if ( ( ( input_type  == STBIR_TYPE_UINT8_SRGB ) || ( input_type  == STBIR_TYPE_UINT8_SRGB_ALPHA ) ) && 
+         ( ( output_type == STBIR_TYPE_UINT8_SRGB ) || ( output_type == STBIR_TYPE_UINT8_SRGB_ALPHA ) ) )
+    {
+      input_type = STBIR_TYPE_UINT8;
+      output_type = STBIR_TYPE_UINT8;
+    }
+  }
+
+  // recalc the output and input strides  
+  if ( info->input_stride_bytes == 0 )
+    info->input_stride_bytes = info->channels * info->horizontal.scale_info.input_full_size * stbir__type_size[input_type];
+
+  if ( info->output_stride_bytes == 0 )
+    info->output_stride_bytes = info->channels * info->horizontal.scale_info.output_sub_size * stbir__type_size[output_type];
+
+  // calc offset
+  info->output_data = ( (char*) resize->output_pixels ) + ( (ptrdiff_t) info->offset_y * (ptrdiff_t) resize->output_stride_in_bytes ) + ( info->offset_x * info->channels * stbir__type_size[output_type] );
+
+  info->in_pixels_cb = resize->input_cb;
+  info->user_data = resize->user_data;
+  info->out_pixels_cb = resize->output_cb;
+
+  // setup the input format converters
+  if ( ( input_type == STBIR_TYPE_UINT8 ) || ( input_type == STBIR_TYPE_UINT16 ) )
+  {
+    int non_scaled = 0;
+
+    // check if we can run unscaled - 0-255.0/0-65535.0 instead of 0-1.0 (which is a tiny bit faster when doing linear 8->8 or 16->16)
+    if ( ( !info->alpha_weight ) && ( !info->alpha_unweight )  ) // don't short circuit when alpha weighting (get everything to 0-1.0 as usual)
+      if ( ( ( input_type == STBIR_TYPE_UINT8 ) && ( output_type == STBIR_TYPE_UINT8 ) ) || ( ( input_type == STBIR_TYPE_UINT16 ) && ( output_type == STBIR_TYPE_UINT16 ) ) )
+        non_scaled = 1;
+
+    if ( info->input_pixel_layout_internal <= STBIRI_4CHANNEL )
+      decode_pixels = decode_simple_scaled_or_not[ input_type == STBIR_TYPE_UINT16 ][ non_scaled ];
+    else
+      decode_pixels = decode_alphas_scaled_or_not[ ( info->input_pixel_layout_internal - STBIRI_RGBA ) % ( STBIRI_AR-STBIRI_RGBA+1 ) ][ input_type == STBIR_TYPE_UINT16 ][ non_scaled ];
+  }
+  else
+  {
+    if ( info->input_pixel_layout_internal <= STBIRI_4CHANNEL )
+      decode_pixels = decode_simple[ input_type - STBIR_TYPE_UINT8_SRGB ];
+    else
+      decode_pixels = decode_alphas[ ( info->input_pixel_layout_internal - STBIRI_RGBA ) % ( STBIRI_AR-STBIRI_RGBA+1 ) ][ input_type - STBIR_TYPE_UINT8_SRGB ];
+  }
+
+  // setup the output format converters
+  if ( ( output_type == STBIR_TYPE_UINT8 ) || ( output_type == STBIR_TYPE_UINT16 ) )
+  {
+    int non_scaled = 0;
+    
+    // check if we can run unscaled - 0-255.0/0-65535.0 instead of 0-1.0 (which is a tiny bit faster when doing linear 8->8 or 16->16)
+    if ( ( !info->alpha_weight ) && ( !info->alpha_unweight ) ) // don't short circuit when alpha weighting (get everything to 0-1.0 as usual)
+      if ( ( ( input_type == STBIR_TYPE_UINT8 ) && ( output_type == STBIR_TYPE_UINT8 ) ) || ( ( input_type == STBIR_TYPE_UINT16 ) && ( output_type == STBIR_TYPE_UINT16 ) ) )
+        non_scaled = 1;
+
+    if ( info->output_pixel_layout_internal <= STBIRI_4CHANNEL )
+      encode_pixels = encode_simple_scaled_or_not[ output_type == STBIR_TYPE_UINT16 ][ non_scaled ];
+    else
+      encode_pixels = encode_alphas_scaled_or_not[ ( info->output_pixel_layout_internal - STBIRI_RGBA ) % ( STBIRI_AR-STBIRI_RGBA+1 ) ][ output_type == STBIR_TYPE_UINT16 ][ non_scaled ];
+  }
+  else
+  {
+    if ( info->output_pixel_layout_internal <= STBIRI_4CHANNEL )
+      encode_pixels = encode_simple[ output_type - STBIR_TYPE_UINT8_SRGB ];
+    else
+      encode_pixels = encode_alphas[ ( info->output_pixel_layout_internal - STBIRI_RGBA ) % ( STBIRI_AR-STBIRI_RGBA+1 ) ][ output_type - STBIR_TYPE_UINT8_SRGB ];
+  }
+
+  info->input_type = input_type;
+  info->output_type = output_type; 
+  info->decode_pixels = decode_pixels;
+  info->encode_pixels = encode_pixels; 
+}
+
+static void stbir__clip( int * outx, int * outsubw, int outw, double * u0, double * u1 )
+{
+  double per, adj;
+  int over;
+  
+  // do left/top edge
+  if ( *outx < 0 )
+  {
+    per = ( (double)*outx ) / ( (double)*outsubw ); // is negative
+    adj = per * ( *u1 - *u0 );
+    *u0 -= adj; // increases u0
+    *outx = 0;
+  }
+
+  // do right/bot edge
+  over = outw - ( *outx + *outsubw );
+  if ( over < 0 )
+  {
+    per = ( (double)over ) / ( (double)*outsubw ); // is negative
+    adj = per * ( *u1 - *u0 );
+    *u1 += adj; // decrease u1
+    *outsubw = outw - *outx;
+  }
+}    
+
+// converts a double to a rational that has less than one float bit of error (returns 0 if unable to do so)
+static int stbir__double_to_rational(double f, stbir_uint32 limit, stbir_uint32 *numer, stbir_uint32 *denom, int limit_denom ) // limit_denom (1) or limit numer (0)
+{
+  double err;
+  stbir_uint64 top, bot;
+  stbir_uint64 numer_last = 0;
+  stbir_uint64 denom_last = 1;
+  stbir_uint64 numer_estimate = 1;
+  stbir_uint64 denom_estimate = 0;
+
+  // scale to past float error range
+  top = (stbir_uint64)( f * (double)(1 << 25) );
+  bot = 1 << 25;
+
+  // keep refining, but usually stops in a few loops - usually 5 for bad cases
+  for(;;)  
+  {
+    stbir_uint64 est, temp;
+
+    // hit limit, break out and do best full range estimate
+    if ( ( ( limit_denom ) ? denom_estimate : numer_estimate ) >= limit )
+      break;
+
+    // is the current error less than 1 bit of a float? if so, we're done
+    if ( denom_estimate )
+    {
+      err = ( (double)numer_estimate / (double)denom_estimate ) - f;
+      if ( err < 0.0 ) err = -err;
+      if ( err < ( 1.0 / (double)(1<<24) ) )
+      {
+        // yup, found it
+        *numer = (stbir_uint32) numer_estimate;
+        *denom = (stbir_uint32) denom_estimate;
+        return 1;
+      }
+    }
+
+    // no more refinement bits left? break out and do full range estimate
+    if ( bot == 0 )
+      break;
+
+    // gcd the estimate bits
+    est = top / bot;
+    temp = top % bot;
+    top = bot;
+    bot = temp;
+
+    // move remainders
+    temp = est * denom_estimate + denom_last; 
+    denom_last = denom_estimate; 
+    denom_estimate = temp;
+
+    // move remainders
+    temp = est * numer_estimate + numer_last; 
+    numer_last = numer_estimate; 
+    numer_estimate = temp;
+  }
+
+  // we didn't fine anything good enough for float, use a full range estimate
+  if ( limit_denom )
+  {
+    numer_estimate= (stbir_uint64)( f * (double)limit + 0.5 );
+    denom_estimate = limit;
+  }
+  else
+  {
+    numer_estimate = limit;
+    denom_estimate = (stbir_uint64)( ( (double)limit / f ) + 0.5 );
+  }
+
+  *numer = (stbir_uint32) numer_estimate;
+  *denom = (stbir_uint32) denom_estimate;
+
+  err = ( denom_estimate ) ? ( ( (double)(stbir_uint32)numer_estimate / (double)(stbir_uint32)denom_estimate ) - f ) : 1.0;
+  if ( err < 0.0 ) err = -err;
+  return ( err < ( 1.0 / (double)(1<<24) ) ) ? 1 : 0;
+}
+
+static int stbir__calculate_region_transform( stbir__scale_info * scale_info, int output_full_range, int * output_offset, int output_sub_range, int input_full_range, double input_s0, double input_s1 )
+{
+  double output_range, input_range, output_s, input_s, ratio, scale;
+
+  input_s = input_s1 - input_s0;
+
+  // null area
+  if ( ( output_full_range == 0 ) || ( input_full_range == 0 ) ||
+       ( output_sub_range == 0 ) || ( input_s <= stbir__small_float ) )
+    return 0;
+
+  // are either of the ranges completely out of bounds?
+  if ( ( *output_offset >= output_full_range ) || ( ( *output_offset + output_sub_range ) <= 0 ) || ( input_s0 >= (1.0f-stbir__small_float) ) || ( input_s1 <= stbir__small_float ) )
+    return 0;
+
+  output_range = (double)output_full_range;
+  input_range = (double)input_full_range;
+
+  output_s = ( (double)output_sub_range) / output_range;
+
+  // figure out the scaling to use 
+  ratio = output_s / input_s; 
+
+  // save scale before clipping
+  scale = ( output_range / input_range ) * ratio; 
+  scale_info->scale = (float)scale;
+  scale_info->inv_scale = (float)( 1.0 / scale );
+
+  // clip output area to left/right output edges (and adjust input area)
+  stbir__clip( output_offset, &output_sub_range, output_full_range, &input_s0, &input_s1 );
+
+  // recalc input area
+  input_s = input_s1 - input_s0;
+
+  // after clipping do we have zero input area?
+  if ( input_s <= stbir__small_float ) 
+    return 0;
+
+  // calculate and store the starting source offsets in output pixel space 
+  scale_info->pixel_shift = (float) ( input_s0 * ratio * output_range ); 
+
+  scale_info->scale_is_rational = stbir__double_to_rational( scale, ( scale <= 1.0 ) ? output_full_range : input_full_range, &scale_info->scale_numerator, &scale_info->scale_denominator, ( scale >= 1.0 ) );
+
+  scale_info->input_full_size = input_full_range;
+  scale_info->output_sub_size = output_sub_range;
+
+  return 1;
+}
+
+
+static void stbir__init_and_set_layout( STBIR_RESIZE * resize, stbir_pixel_layout pixel_layout, stbir_datatype data_type )
+{
+  resize->input_cb = 0;
+  resize->output_cb = 0;
+  resize->user_data = resize;
+  resize->samplers = 0;
+  resize->needs_rebuild = 1;
+  resize->called_alloc = 0;
+  resize->horizontal_filter = STBIR_FILTER_DEFAULT;
+  resize->horizontal_filter_kernel = 0; resize->horizontal_filter_support = 0;
+  resize->vertical_filter = STBIR_FILTER_DEFAULT;
+  resize->vertical_filter_kernel = 0; resize->vertical_filter_support = 0;
+  resize->horizontal_edge = STBIR_EDGE_CLAMP;
+  resize->vertical_edge = STBIR_EDGE_CLAMP;
+  resize->input_s0 = 0; resize->input_t0 = 0; resize->input_s1 = 1; resize->input_t1 = 1;
+  resize->output_subx = 0; resize->output_suby = 0; resize->output_subw = resize->output_w; resize->output_subh = resize->output_h;
+  resize->input_data_type = data_type;
+  resize->output_data_type = data_type;
+  resize->input_pixel_layout_public = pixel_layout;
+  resize->output_pixel_layout_public = pixel_layout;
+}
+
+STBIRDEF void stbir_resize_init( STBIR_RESIZE * resize, 
+                                 const void *input_pixels,  int input_w,  int input_h, int input_stride_in_bytes, // stride can be zero
+                                       void *output_pixels, int output_w, int output_h, int output_stride_in_bytes, // stride can be zero
+                                 stbir_pixel_layout pixel_layout, stbir_datatype data_type )
+{
+  resize->input_pixels = input_pixels;
+  resize->input_w = input_w;
+  resize->input_h = input_h;
+  resize->input_stride_in_bytes = input_stride_in_bytes;
+  resize->output_pixels = output_pixels;
+  resize->output_w = output_w;
+  resize->output_h = output_h;
+  resize->output_stride_in_bytes = output_stride_in_bytes;
+  resize->fast_alpha = 0;
+
+  stbir__init_and_set_layout( resize, pixel_layout, data_type );
+}
+
+// You can update parameters any time after resize_init
+STBIRDEF void stbir_set_datatypes( STBIR_RESIZE * resize, stbir_datatype input_type, stbir_datatype output_type )  // by default, datatype from resize_init
+{
+  resize->input_data_type = input_type;
+  resize->output_data_type = output_type;
+}
+
+STBIRDEF void stbir_set_pixel_callbacks( STBIR_RESIZE * resize, stbir_input_callback * input_cb, stbir_output_callback * output_cb )   // no callbacks by default
+{
+  resize->input_cb = input_cb;
+  resize->output_cb = output_cb;
+}
+
+STBIRDEF void stbir_set_user_data( STBIR_RESIZE * resize, void * user_data )                                     // pass back STBIR_RESIZE* by default
+{
+  resize->user_data = user_data;
+}
+
+STBIRDEF void stbir_set_buffer_ptrs( STBIR_RESIZE * resize, const void * input_pixels, int input_stride_in_bytes, void * output_pixels, int output_stride_in_bytes )
+{
+  resize->input_pixels = input_pixels;
+  resize->input_stride_in_bytes = input_stride_in_bytes;
+  resize->output_pixels = output_pixels;
+  resize->output_stride_in_bytes = output_stride_in_bytes;
+}
+
+
+STBIRDEF int stbir_set_edgemodes( STBIR_RESIZE * resize, stbir_edge horizontal_edge, stbir_edge vertical_edge )       // CLAMP by default
+{
+  resize->horizontal_edge = horizontal_edge;
+  resize->vertical_edge = vertical_edge;
+  resize->needs_rebuild = 1;
+  return 1;
+}
+
+STBIRDEF int stbir_set_filters( STBIR_RESIZE * resize, stbir_filter horizontal_filter, stbir_filter vertical_filter ) // STBIR_DEFAULT_FILTER_UPSAMPLE/DOWNSAMPLE by default
+{
+  resize->horizontal_filter = horizontal_filter;
+  resize->vertical_filter = vertical_filter;
+  resize->needs_rebuild = 1;
+  return 1;
+}
+
+STBIRDEF int stbir_set_filter_callbacks( STBIR_RESIZE * resize, stbir__kernel_callback * horizontal_filter, stbir__support_callback * horizontal_support, stbir__kernel_callback * vertical_filter, stbir__support_callback * vertical_support )
+{
+  resize->horizontal_filter_kernel = horizontal_filter; resize->horizontal_filter_support = horizontal_support;
+  resize->vertical_filter_kernel = vertical_filter; resize->vertical_filter_support = vertical_support;
+  resize->needs_rebuild = 1;
+  return 1;
+}
+
+STBIRDEF int stbir_set_pixel_layouts( STBIR_RESIZE * resize, stbir_pixel_layout input_pixel_layout, stbir_pixel_layout output_pixel_layout )   // sets new pixel layouts
+{
+  resize->input_pixel_layout_public = input_pixel_layout;
+  resize->output_pixel_layout_public = output_pixel_layout;
+  resize->needs_rebuild = 1;
+  return 1;
+}
+
+
+STBIRDEF int stbir_set_non_pm_alpha_speed_over_quality( STBIR_RESIZE * resize, int non_pma_alpha_speed_over_quality )   // sets alpha speed
+{
+  resize->fast_alpha = non_pma_alpha_speed_over_quality;
+  resize->needs_rebuild = 1;
+  return 1;
+}
+
+STBIRDEF int stbir_set_input_subrect( STBIR_RESIZE * resize, double s0, double t0, double s1, double t1 )                 // sets input region (full region by default)
+{
+  resize->input_s0 = s0;
+  resize->input_t0 = t0;
+  resize->input_s1 = s1;
+  resize->input_t1 = t1;
+  resize->needs_rebuild = 1;
+
+  // are we inbounds?
+  if ( ( s1 < stbir__small_float ) || ( (s1-s0) < stbir__small_float ) ||
+       ( t1 < stbir__small_float ) || ( (t1-t0) < stbir__small_float ) ||
+       ( s0 > (1.0f-stbir__small_float) ) ||
+       ( t0 > (1.0f-stbir__small_float) ) )
+    return 0;
+
+  return 1;
+}
+
+STBIRDEF int stbir_set_output_pixel_subrect( STBIR_RESIZE * resize, int subx, int suby, int subw, int subh )          // sets input region (full region by default)
+{
+  resize->output_subx = subx;
+  resize->output_suby = suby;
+  resize->output_subw = subw;
+  resize->output_subh = subh;
+  resize->needs_rebuild = 1;
+
+  // are we inbounds?
+  if ( ( subx >= resize->output_w ) || ( ( subx + subw ) <= 0 ) || ( suby >= resize->output_h ) || ( ( suby + subh ) <= 0 ) || ( subw == 0 ) || ( subh == 0 ) )
+    return 0;
+
+  return 1;
+}
+
+STBIRDEF int stbir_set_pixel_subrect( STBIR_RESIZE * resize, int subx, int suby, int subw, int subh )                 // sets both regions (full regions by default)
+{
+  double s0, t0, s1, t1;
+
+  s0 = ( (double)subx ) / ( (double)resize->output_w );
+  t0 = ( (double)suby ) / ( (double)resize->output_h );
+  s1 = ( (double)(subx+subw) ) / ( (double)resize->output_w );
+  t1 = ( (double)(suby+subh) ) / ( (double)resize->output_h );
+
+  resize->input_s0 = s0;
+  resize->input_t0 = t0;
+  resize->input_s1 = s1;
+  resize->input_t1 = t1;
+  resize->output_subx = subx;
+  resize->output_suby = suby;
+  resize->output_subw = subw;
+  resize->output_subh = subh;
+  resize->needs_rebuild = 1;
+
+  // are we inbounds?
+  if ( ( subx >= resize->output_w ) || ( ( subx + subw ) <= 0 ) || ( suby >= resize->output_h ) || ( ( suby + subh ) <= 0 ) || ( subw == 0 ) || ( subh == 0 ) )
+    return 0;
+
+  return 1;
+}
+
+static int stbir__perform_build( STBIR_RESIZE * resize, int splits ) 
+{
+  stbir__contributors conservative = { 0, 0 };
+  stbir__sampler horizontal, vertical;
+  int new_output_subx, new_output_suby;
+  stbir__info * out_info;
+  #ifdef STBIR_PROFILE
+  stbir__info profile_infod;  // used to contain building profile info before everything is allocated
+  stbir__info * profile_info = &profile_infod;
+  #endif
+
+  // have we already built the samplers?
+  if ( resize->samplers )
+    return 0;
+  
+  #define STBIR_RETURN_ERROR_AND_ASSERT( exp )  STBIR_ASSERT( !(exp) ); if (exp) return 0;
+  STBIR_RETURN_ERROR_AND_ASSERT( (unsigned)resize->horizontal_filter >= STBIR_FILTER_OTHER)
+  STBIR_RETURN_ERROR_AND_ASSERT( (unsigned)resize->vertical_filter >= STBIR_FILTER_OTHER)
+  #undef STBIR_RETURN_ERROR_AND_ASSERT
+
+  if ( splits <= 0 ) 
+    return 0;
+
+  STBIR_PROFILE_BUILD_FIRST_START( build );
+
+  new_output_subx = resize->output_subx;
+  new_output_suby = resize->output_suby;
+
+  // do horizontal clip and scale calcs
+  if ( !stbir__calculate_region_transform( &horizontal.scale_info, resize->output_w, &new_output_subx, resize->output_subw, resize->input_w, resize->input_s0, resize->input_s1 ) )
+    return 0;
+
+  // do vertical clip and scale calcs
+  if ( !stbir__calculate_region_transform( &vertical.scale_info, resize->output_h, &new_output_suby, resize->output_subh, resize->input_h, resize->input_t0, resize->input_t1 ) )
+    return 0;
+
+  // if nothing to do, just return
+  if ( ( horizontal.scale_info.output_sub_size == 0 ) || ( vertical.scale_info.output_sub_size == 0 ) )
+    return 0;
+
+  stbir__set_sampler(&horizontal, resize->horizontal_filter, resize->horizontal_filter_kernel, resize->horizontal_filter_support, resize->horizontal_edge, &horizontal.scale_info, 1, resize->user_data );
+  stbir__get_conservative_extents( &horizontal, &conservative, resize->user_data );
+  stbir__set_sampler(&vertical, resize->vertical_filter, resize->horizontal_filter_kernel, resize->vertical_filter_support, resize->vertical_edge, &vertical.scale_info, 0, resize->user_data );
+
+  if ( ( vertical.scale_info.output_sub_size / splits ) < 4 ) // each split should be a minimum of 4 scanlines (handwavey choice)
+  {
+    splits = vertical.scale_info.output_sub_size / 4;
+    if ( splits == 0 ) splits = 1;
+  }
+
+  STBIR_PROFILE_BUILD_START( alloc );
+  out_info = stbir__alloc_internal_mem_and_build_samplers( &horizontal, &vertical, &conservative, resize->input_pixel_layout_public, resize->output_pixel_layout_public, splits, new_output_subx, new_output_suby, resize->fast_alpha, resize->user_data STBIR_ONLY_PROFILE_BUILD_SET_INFO );
+  STBIR_PROFILE_BUILD_END( alloc );
+  STBIR_PROFILE_BUILD_END( build );
+ 
+  if ( out_info )
+  {
+    resize->splits = splits;
+    resize->samplers = out_info;
+    resize->needs_rebuild = 0;
+    #ifdef STBIR_PROFILE
+      STBIR_MEMCPY( &out_info->profile, &profile_infod.profile, sizeof( out_info->profile ) );
+    #endif
+    return splits;
+  }
+
+  return 0;
+}
+
+void stbir_free_samplers( STBIR_RESIZE * resize )
+{
+  if ( resize->samplers )
+  {
+    stbir__free_internal_mem( resize->samplers );
+    resize->samplers = 0;
+    resize->called_alloc = 0;
+  }
+}
+
+STBIRDEF int stbir_build_samplers_with_splits( STBIR_RESIZE * resize, int splits )
+{
+  if ( ( resize->samplers == 0 ) || ( resize->needs_rebuild ) )
+  {
+    if ( resize->samplers )
+      stbir_free_samplers( resize );
+
+    resize->called_alloc = 1;
+    return stbir__perform_build( resize, splits );
+  }
+
+  STBIR_PROFILE_BUILD_CLEAR( resize->samplers );
+  
+  return 1;
+}
+
+STBIRDEF int stbir_build_samplers( STBIR_RESIZE * resize )
+{
+  return stbir_build_samplers_with_splits( resize, 1 );
+}
+
+STBIRDEF int stbir_resize_extended( STBIR_RESIZE * resize )
+{
+  int result;
+  
+  if ( ( resize->samplers == 0 ) || ( resize->needs_rebuild ) )
+  {
+    int alloc_state = resize->called_alloc;  // remember allocated state
+
+    if ( resize->samplers )
+    {
+      stbir__free_internal_mem( resize->samplers );
+      resize->samplers = 0;
+    }
+
+    if ( !stbir_build_samplers( resize ) )
+      return 0;
+    
+    resize->called_alloc = alloc_state;
+
+    // if build_samplers succeeded (above), but there are no samplers set, then 
+    //   the area to stretch into was zero pixels, so don't do anything and return
+    //   success
+    if ( resize->samplers == 0 )
+      return 1;
+  }
+  else
+  {
+    // didn't build anything - clear it
+    STBIR_PROFILE_BUILD_CLEAR( resize->samplers );
+  }
+
+
+  // update anything that can be changed without recalcing samplers
+  stbir__update_info_from_resize( resize->samplers, resize );
+
+  // do resize
+  result = stbir__perform_resize( resize->samplers, 0, resize->splits );
+
+  // if we alloced, then free
+  if ( !resize->called_alloc )
+  {
+    stbir_free_samplers( resize );
+    resize->samplers = 0;
+  } 
+
+  return result;
+}
+
+STBIRDEF int stbir_resize_extended_split( STBIR_RESIZE * resize, int split_start, int split_count )
+{
+  STBIR_ASSERT( resize->samplers );
+
+  // if we're just doing the whole thing, call full
+  if ( ( split_start == -1 ) || ( ( split_start == 0 ) && ( split_count == resize->splits ) ) )
+    return stbir_resize_extended( resize );
+
+  // you **must** build samplers first when using split resize
+  if ( ( resize->samplers == 0 ) || ( resize->needs_rebuild ) )
+    return 0; 
+    
+  if ( ( split_start >= resize->splits ) || ( split_start < 0 ) || ( ( split_start + split_count ) > resize->splits ) || ( split_count <= 0 ) )
+    return 0;
+  
+  // update anything that can be changed without recalcing samplers
+  stbir__update_info_from_resize( resize->samplers, resize );
+ 
+  // do resize
+  return stbir__perform_resize( resize->samplers, split_start, split_count );
+}
+
+static int stbir__check_output_stuff( void ** ret_ptr, int * ret_pitch, void * output_pixels, int type_size, int output_w, int output_h, int output_stride_in_bytes, stbir_internal_pixel_layout pixel_layout )
+{
+  size_t size;
+  int pitch;
+  void * ptr;
+
+  pitch = output_w * type_size * stbir__pixel_channels[ pixel_layout ];
+  if ( pitch == 0 )
+    return 0;
+
+  if ( output_stride_in_bytes == 0 )
+    output_stride_in_bytes = pitch;
+
+  if ( output_stride_in_bytes < pitch )
+    return 0;
+
+  size = output_stride_in_bytes * output_h;
+  if ( size == 0 )
+    return 0;
+
+  *ret_ptr = 0;
+  *ret_pitch = output_stride_in_bytes;
+
+  if ( output_pixels == 0 )
+  {
+    ptr = STBIR_MALLOC( size, 0 );
+    if ( ptr == 0 )
+      return 0;
+
+    *ret_ptr = ptr;
+    *ret_pitch = pitch;
+  }
+
+  return 1;  
+}
+
+
+STBIRDEF unsigned char * stbir_resize_uint8_linear( const unsigned char *input_pixels , int input_w , int input_h, int input_stride_in_bytes,
+                                                          unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+                                                          stbir_pixel_layout pixel_layout )
+{
+  STBIR_RESIZE resize;
+  unsigned char * optr;
+  int opitch;
+
+  if ( !stbir__check_output_stuff( (void**)&optr, &opitch, output_pixels, sizeof( unsigned char ), output_w, output_h, output_stride_in_bytes, stbir__pixel_layout_convert_public_to_internal[ pixel_layout ] ) )
+    return 0;
+
+  stbir_resize_init( &resize, 
+                     input_pixels,  input_w,  input_h,  input_stride_in_bytes, 
+                     (optr) ? optr : output_pixels, output_w, output_h, opitch, 
+                     pixel_layout, STBIR_TYPE_UINT8 );
+
+  if ( !stbir_resize_extended( &resize ) )
+  {
+    if ( optr )
+      STBIR_FREE( optr, 0 );
+    return 0;
+  }
+
+  return (optr) ? optr : output_pixels;
+}
+
+STBIRDEF unsigned char * stbir_resize_uint8_srgb( const unsigned char *input_pixels , int input_w , int input_h, int input_stride_in_bytes,
+                                                        unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+                                                        stbir_pixel_layout pixel_layout )
+{
+  STBIR_RESIZE resize;
+  unsigned char * optr;
+  int opitch;
+
+  if ( !stbir__check_output_stuff( (void**)&optr, &opitch, output_pixels, sizeof( unsigned char ), output_w, output_h, output_stride_in_bytes, stbir__pixel_layout_convert_public_to_internal[ pixel_layout ] ) )
+    return 0;
+
+  stbir_resize_init( &resize, 
+                     input_pixels,  input_w,  input_h,  input_stride_in_bytes, 
+                     (optr) ? optr : output_pixels, output_w, output_h, opitch, 
+                     pixel_layout, STBIR_TYPE_UINT8_SRGB );
+
+  if ( !stbir_resize_extended( &resize ) )
+  {
+    if ( optr )
+      STBIR_FREE( optr, 0 );
+    return 0;
+  }
+
+  return (optr) ? optr : output_pixels;
+}
+
+
+STBIRDEF float * stbir_resize_float_linear( const float *input_pixels , int input_w , int input_h, int input_stride_in_bytes,
+                                                  float *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+                                                  stbir_pixel_layout pixel_layout )
+{
+  STBIR_RESIZE resize;
+  float * optr;
+  int opitch;
+
+  if ( !stbir__check_output_stuff( (void**)&optr, &opitch, output_pixels, sizeof( float ), output_w, output_h, output_stride_in_bytes, stbir__pixel_layout_convert_public_to_internal[ pixel_layout ] ) )
+    return 0;
+
+  stbir_resize_init( &resize, 
+                     input_pixels,  input_w,  input_h,  input_stride_in_bytes, 
+                     (optr) ? optr : output_pixels, output_w, output_h, opitch, 
+                     pixel_layout, STBIR_TYPE_FLOAT );
+
+  if ( !stbir_resize_extended( &resize ) )
+  {
+    if ( optr )
+      STBIR_FREE( optr, 0 );
+    return 0;
+  }
+
+  return (optr) ? optr : output_pixels;
+}
+
+
+STBIRDEF void * stbir_resize( const void *input_pixels , int input_w , int input_h, int input_stride_in_bytes,
+                                    void *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+                              stbir_pixel_layout pixel_layout, stbir_datatype data_type, 
+                              stbir_edge edge, stbir_filter filter )
+{
+  STBIR_RESIZE resize;
+  float * optr;
+  int opitch;
+
+  if ( !stbir__check_output_stuff( (void**)&optr, &opitch, output_pixels, stbir__type_size[data_type], output_w, output_h, output_stride_in_bytes, stbir__pixel_layout_convert_public_to_internal[ pixel_layout ] ) )
+    return 0;
+
+  stbir_resize_init( &resize, 
+                     input_pixels,  input_w,  input_h,  input_stride_in_bytes, 
+                     (optr) ? optr : output_pixels, output_w, output_h, output_stride_in_bytes, 
+                     pixel_layout, data_type );
+
+  resize.horizontal_edge = edge;
+  resize.vertical_edge = edge;
+  resize.horizontal_filter = filter;
+  resize.vertical_filter = filter;
+
+  if ( !stbir_resize_extended( &resize ) )
+  {
+    if ( optr )
+      STBIR_FREE( optr, 0 );
+    return 0;
+  }
+
+  return (optr) ? optr : output_pixels;
+}
+
+#ifdef STBIR_PROFILE
+
+STBIRDEF void stbir_resize_build_profile_info( STBIR_PROFILE_INFO * info, STBIR_RESIZE const * resize )
+{
+  static char const * bdescriptions[6] = { "Building", "Allocating", "Horizontal sampler", "Vertical sampler", "Coefficient cleanup", "Coefficient piovot" } ;
+  stbir__info* samp = resize->samplers;
+  int i;
+
+  typedef int testa[ (STBIR__ARRAY_SIZE( bdescriptions ) == (STBIR__ARRAY_SIZE( samp->profile.array )-1) )?1:-1];
+  typedef int testb[ (sizeof( samp->profile.array ) == (sizeof(samp->profile.named)) )?1:-1];
+  typedef int testc[ (sizeof( info->clocks ) >= (sizeof(samp->profile.named)) )?1:-1];
+
+  for( i = 0 ; i < STBIR__ARRAY_SIZE( bdescriptions ) ; i++)
+    info->clocks[i] = samp->profile.array[i+1];
+
+  info->total_clocks = samp->profile.named.total;
+  info->descriptions = bdescriptions;
+  info->count = STBIR__ARRAY_SIZE( bdescriptions );
+}
+
+STBIRDEF void stbir_resize_split_profile_info( STBIR_PROFILE_INFO * info, STBIR_RESIZE const * resize, int split_start, int split_count )
+{
+  static char const * descriptions[7] = { "Looping", "Vertical sampling", "Horizontal sampling", "Scanline input", "Scanline output", "Alpha weighting", "Alpha unweighting" };
+  stbir__per_split_info * split_info;
+  int s, i;
+
+  typedef int testa[ (STBIR__ARRAY_SIZE( descriptions ) == (STBIR__ARRAY_SIZE( split_info->profile.array )-1) )?1:-1];
+  typedef int testb[ (sizeof( split_info->profile.array ) == (sizeof(split_info->profile.named)) )?1:-1];
+  typedef int testc[ (sizeof( info->clocks ) >= (sizeof(split_info->profile.named)) )?1:-1];
+
+  if ( split_start == -1 )
+  {
+    split_start = 0;
+    split_count = resize->samplers->splits;
+  }
+
+  if ( ( split_start >= resize->splits ) || ( split_start < 0 ) || ( ( split_start + split_count ) > resize->splits ) || ( split_count <= 0 ) )
+  {
+    info->total_clocks = 0;
+    info->descriptions = 0;
+    info->count = 0;
+    return;
+  }
+
+  split_info = resize->samplers->split_info + split_start;
+
+  // sum up the profile from all the splits
+  for( i = 0 ; i < STBIR__ARRAY_SIZE( descriptions ) ; i++ )
+  {
+    stbir_uint64 sum = 0;
+    for( s = 0 ; s < split_count ; s++ )
+      sum += split_info[s].profile.array[i+1];
+    info->clocks[i] = sum;
+  }
+
+  info->total_clocks = split_info->profile.named.total;
+  info->descriptions = descriptions;
+  info->count = STBIR__ARRAY_SIZE( descriptions );
+}
+
+STBIRDEF void stbir_resize_extended_profile_info( STBIR_PROFILE_INFO * info, STBIR_RESIZE const * resize )
+{
+  stbir_resize_split_profile_info( info, resize, -1, 0 );
+}
+
+#endif // STBIR_PROFILE
+
+#undef STBIR_BGR
+#undef STBIR_1CHANNEL
+#undef STBIR_2CHANNEL
+#undef STBIR_RGB
+#undef STBIR_RGBA
+#undef STBIR_4CHANNEL
+#undef STBIR_BGRA
+#undef STBIR_ARGB
+#undef STBIR_ABGR
+#undef STBIR_RA
+#undef STBIR_AR
+#undef STBIR_RGBA_PM
+#undef STBIR_BGRA_PM
+#undef STBIR_ARGB_PM
+#undef STBIR_ABGR_PM
+#undef STBIR_RA_PM
+#undef STBIR_AR_PM
+
+#endif // STB_IMAGE_RESIZE_IMPLEMENTATION
+
+#else  // STB_IMAGE_RESIZE_HORIZONTALS&STB_IMAGE_RESIZE_DO_VERTICALS
+
+// we reinclude the header file to define all the horizontal functions
+//   specializing each function for the number of coeffs is 20-40% faster *OVERALL* 
+
+// by including the header file again this way, we can still debug the functions
+
+#define STBIR_strs_join2( start, mid, end ) start##mid##end
+#define STBIR_strs_join1( start, mid, end ) STBIR_strs_join2( start, mid, end )
+
+#define STBIR_strs_join24( start, mid1, mid2, end ) start##mid1##mid2##end
+#define STBIR_strs_join14( start, mid1, mid2, end ) STBIR_strs_join24( start, mid1, mid2, end )
+
+#ifdef STB_IMAGE_RESIZE_DO_CODERS
+
+#ifdef stbir__decode_suffix
+#define STBIR__CODER_NAME( name ) STBIR_strs_join1( name, _, stbir__decode_suffix )
+#else
+#define STBIR__CODER_NAME( name ) name
+#endif
+
+#ifdef stbir__decode_swizzle
+#define stbir__decode_simdf8_flip(reg) STBIR_strs_join1( STBIR_strs_join1( STBIR_strs_join1( STBIR_strs_join1( stbir__simdf8_0123to,stbir__decode_order0,stbir__decode_order1),stbir__decode_order2,stbir__decode_order3),stbir__decode_order0,stbir__decode_order1),stbir__decode_order2,stbir__decode_order3)(reg, reg)
+#define stbir__decode_simdf4_flip(reg) STBIR_strs_join1( STBIR_strs_join1( stbir__simdf_0123to,stbir__decode_order0,stbir__decode_order1),stbir__decode_order2,stbir__decode_order3)(reg, reg)
+#define stbir__encode_simdf8_unflip(reg) STBIR_strs_join1( STBIR_strs_join1( STBIR_strs_join1( STBIR_strs_join1( stbir__simdf8_0123to,stbir__encode_order0,stbir__encode_order1),stbir__encode_order2,stbir__encode_order3),stbir__encode_order0,stbir__encode_order1),stbir__encode_order2,stbir__encode_order3)(reg, reg)
+#define stbir__encode_simdf4_unflip(reg) STBIR_strs_join1( STBIR_strs_join1( stbir__simdf_0123to,stbir__encode_order0,stbir__encode_order1),stbir__encode_order2,stbir__encode_order3)(reg, reg)
+#else
+#define stbir__decode_order0 0
+#define stbir__decode_order1 1
+#define stbir__decode_order2 2
+#define stbir__decode_order3 3
+#define stbir__encode_order0 0
+#define stbir__encode_order1 1
+#define stbir__encode_order2 2
+#define stbir__encode_order3 3
+#define stbir__decode_simdf8_flip(reg)
+#define stbir__decode_simdf4_flip(reg) 
+#define stbir__encode_simdf8_unflip(reg)
+#define stbir__encode_simdf4_unflip(reg) 
+#endif
+
+#ifdef STBIR_SIMD8
+#define stbir__encode_simdfX_unflip  stbir__encode_simdf8_unflip
+#else
+#define stbir__encode_simdfX_unflip  stbir__encode_simdf4_unflip
+#endif 
+
+static void STBIR__CODER_NAME( stbir__decode_uint8_linear_scaled )( float * decodep, int width_times_channels, void const * inputp )
+{
+  float STBIR_STREAMOUT_PTR( * ) decode = decodep;
+  float * decode_end = (float*) decode + width_times_channels;
+  unsigned char const * input = (unsigned char const*)inputp;
+
+  #ifdef STBIR_SIMD
+  unsigned char const * end_input_m16 = input + width_times_channels - 16;
+  if ( width_times_channels >= 16 )
+  {
+    decode_end -= 16;
+    for(;;)
+    {
+      #ifdef STBIR_SIMD8
+      stbir__simdi i; stbir__simdi8 o0,o1;
+      stbir__simdf8 of0, of1;
+      STBIR_NO_UNROLL(decode);
+      stbir__simdi_load( i, input );
+      stbir__simdi8_expand_u8_to_u32( o0, o1, i );
+      stbir__simdi8_convert_i32_to_float( of0, o0 );
+      stbir__simdi8_convert_i32_to_float( of1, o1 );
+      stbir__simdf8_mult( of0, of0, STBIR_max_uint8_as_float_inverted8);
+      stbir__simdf8_mult( of1, of1, STBIR_max_uint8_as_float_inverted8);
+      stbir__decode_simdf8_flip( of0 );
+      stbir__decode_simdf8_flip( of1 );
+      stbir__simdf8_store( decode + 0, of0 );
+      stbir__simdf8_store( decode + 8, of1 );
+      #else
+      stbir__simdi i, o0, o1, o2, o3;
+      stbir__simdf of0, of1, of2, of3;
+      STBIR_NO_UNROLL(decode);
+      stbir__simdi_load( i, input );
+      stbir__simdi_expand_u8_to_u32( o0,o1,o2,o3,i);
+      stbir__simdi_convert_i32_to_float( of0, o0 );
+      stbir__simdi_convert_i32_to_float( of1, o1 );
+      stbir__simdi_convert_i32_to_float( of2, o2 );
+      stbir__simdi_convert_i32_to_float( of3, o3 );
+      stbir__simdf_mult( of0, of0, STBIR__CONSTF(STBIR_max_uint8_as_float_inverted) );
+      stbir__simdf_mult( of1, of1, STBIR__CONSTF(STBIR_max_uint8_as_float_inverted) );
+      stbir__simdf_mult( of2, of2, STBIR__CONSTF(STBIR_max_uint8_as_float_inverted) );
+      stbir__simdf_mult( of3, of3, STBIR__CONSTF(STBIR_max_uint8_as_float_inverted) );
+      stbir__decode_simdf4_flip( of0 );
+      stbir__decode_simdf4_flip( of1 );
+      stbir__decode_simdf4_flip( of2 );
+      stbir__decode_simdf4_flip( of3 );
+      stbir__simdf_store( decode + 0,  of0 );
+      stbir__simdf_store( decode + 4,  of1 );
+      stbir__simdf_store( decode + 8,  of2 );
+      stbir__simdf_store( decode + 12, of3 );
+      #endif
+      decode += 16;
+      input += 16;
+      if ( decode <= decode_end ) 
+        continue;
+      if ( decode == ( decode_end + 16 ) )
+        break;
+      decode = decode_end; // backup and do last couple
+      input = end_input_m16;
+    }
+    return;
+  }
+  #endif
+
+  // try to do blocks of 4 when you can
+  #if stbir__coder_min_num != 3 // doesn't divide cleanly by four
+  decode += 4;
+  while( decode <= decode_end )
+  {
+    STBIR_SIMD_NO_UNROLL(decode);
+    decode[0-4] = ((float)(input[stbir__decode_order0])) * stbir__max_uint8_as_float_inverted;
+    decode[1-4] = ((float)(input[stbir__decode_order1])) * stbir__max_uint8_as_float_inverted;
+    decode[2-4] = ((float)(input[stbir__decode_order2])) * stbir__max_uint8_as_float_inverted;
+    decode[3-4] = ((float)(input[stbir__decode_order3])) * stbir__max_uint8_as_float_inverted;
+    decode += 4;
+    input += 4;
+  }
+  decode -= 4;
+  #endif
+
+  // do the remnants
+  #if stbir__coder_min_num < 4
+  while( decode < decode_end )
+  {
+    STBIR_NO_UNROLL(decode);
+    decode[0] = ((float)(input[stbir__decode_order0])) * stbir__max_uint8_as_float_inverted;
+    #if stbir__coder_min_num >= 2
+    decode[1] = ((float)(input[stbir__decode_order1])) * stbir__max_uint8_as_float_inverted;
+    #endif
+    #if stbir__coder_min_num >= 3
+    decode[2] = ((float)(input[stbir__decode_order2])) * stbir__max_uint8_as_float_inverted;
+    #endif
+    decode += stbir__coder_min_num;
+    input += stbir__coder_min_num;
+  }
+  #endif
+}
+
+static void STBIR__CODER_NAME( stbir__encode_uint8_linear_scaled )( void * outputp, int width_times_channels, float const * encode )
+{
+  unsigned char STBIR_SIMD_STREAMOUT_PTR( * ) output = (unsigned char *) outputp;
+  unsigned char * end_output = ( (unsigned char *) output ) + width_times_channels;
+
+  #ifdef STBIR_SIMD
+  if ( width_times_channels >= stbir__simdfX_float_count*2 )
+  {
+    float const * end_encode_m8 = encode + width_times_channels - stbir__simdfX_float_count*2;
+    end_output -= stbir__simdfX_float_count*2;
+    for(;;)
+    {
+      stbir__simdfX e0, e1;
+      stbir__simdi i;
+      STBIR_SIMD_NO_UNROLL(encode);
+      stbir__simdfX_madd_mem( e0, STBIR_simd_point5X, STBIR_max_uint8_as_floatX, encode );
+      stbir__simdfX_madd_mem( e1, STBIR_simd_point5X, STBIR_max_uint8_as_floatX, encode+stbir__simdfX_float_count );
+      stbir__encode_simdfX_unflip( e0 );
+      stbir__encode_simdfX_unflip( e1 );
+      #ifdef STBIR_SIMD8
+      stbir__simdf8_pack_to_16bytes( i, e0, e1 ); 
+      stbir__simdi_store( output, i );
+      #else
+      stbir__simdf_pack_to_8bytes( i, e0, e1 ); 
+      stbir__simdi_store2( output, i );
+      #endif
+      encode += stbir__simdfX_float_count*2;
+      output += stbir__simdfX_float_count*2;
+      if ( output <= end_output ) 
+        continue;
+      if ( output == ( end_output + stbir__simdfX_float_count*2 ) )
+        break;
+      output = end_output; // backup and do last couple
+      encode = end_encode_m8;
+    }
+    return;
+  }
+
+  // try to do blocks of 4 when you can
+  #if stbir__coder_min_num != 3 // doesn't divide cleanly by four
+  output += 4;
+  while( output <= end_output )
+  {
+    stbir__simdf e0;
+    stbir__simdi i0;
+    STBIR_NO_UNROLL(encode);
+    stbir__simdf_load( e0, encode );
+    stbir__simdf_madd( e0, STBIR__CONSTF(STBIR_simd_point5), STBIR__CONSTF(STBIR_max_uint8_as_float), e0 );
+    stbir__encode_simdf4_unflip( e0 );
+    stbir__simdf_pack_to_8bytes( i0, e0, e0 );  // only use first 4
+    *(int*)(output-4) = stbir__simdi_to_int( i0 );
+    output += 4;
+    encode += 4;
+  }
+  output -= 4;
+  #endif
+
+  // do the remnants
+  #if stbir__coder_min_num < 4
+  while( output < end_output )
+  {
+    stbir__simdf e0; 
+    STBIR_NO_UNROLL(encode);
+    stbir__simdf_madd1_mem( e0, STBIR__CONSTF(STBIR_simd_point5), STBIR__CONSTF(STBIR_max_uint8_as_float), encode+stbir__encode_order0 ); output[0] = stbir__simdf_convert_float_to_uint8( e0 );
+    #if stbir__coder_min_num >= 2
+    stbir__simdf_madd1_mem( e0, STBIR__CONSTF(STBIR_simd_point5), STBIR__CONSTF(STBIR_max_uint8_as_float), encode+stbir__encode_order1 ); output[1] = stbir__simdf_convert_float_to_uint8( e0 );
+    #endif
+    #if stbir__coder_min_num >= 3
+    stbir__simdf_madd1_mem( e0, STBIR__CONSTF(STBIR_simd_point5), STBIR__CONSTF(STBIR_max_uint8_as_float), encode+stbir__encode_order2 ); output[2] = stbir__simdf_convert_float_to_uint8( e0 );
+    #endif
+    output += stbir__coder_min_num;
+    encode += stbir__coder_min_num;
+  }
+  #endif
+  
+  #else
+
+  // try to do blocks of 4 when you can
+  #if stbir__coder_min_num != 3 // doesn't divide cleanly by four
+  output += 4;
+  while( output <= end_output )
+  {
+    float f;
+    f = encode[stbir__encode_order0] * stbir__max_uint8_as_float + 0.5f; STBIR_CLAMP(f, 0, 255); output[0-4] = (unsigned char)f;
+    f = encode[stbir__encode_order1] * stbir__max_uint8_as_float + 0.5f; STBIR_CLAMP(f, 0, 255); output[1-4] = (unsigned char)f;
+    f = encode[stbir__encode_order2] * stbir__max_uint8_as_float + 0.5f; STBIR_CLAMP(f, 0, 255); output[2-4] = (unsigned char)f;
+    f = encode[stbir__encode_order3] * stbir__max_uint8_as_float + 0.5f; STBIR_CLAMP(f, 0, 255); output[3-4] = (unsigned char)f;
+    output += 4;
+    encode += 4;
+  }
+  output -= 4;
+  #endif
+
+  // do the remnants
+  #if stbir__coder_min_num < 4
+  while( output < end_output )
+  {
+    float f;
+    STBIR_NO_UNROLL(encode);
+    f = encode[stbir__encode_order0] * stbir__max_uint8_as_float + 0.5f; STBIR_CLAMP(f, 0, 255); output[0] = (unsigned char)f;
+    #if stbir__coder_min_num >= 2
+    f = encode[stbir__encode_order1] * stbir__max_uint8_as_float + 0.5f; STBIR_CLAMP(f, 0, 255); output[1] = (unsigned char)f;
+    #endif
+    #if stbir__coder_min_num >= 3
+    f = encode[stbir__encode_order2] * stbir__max_uint8_as_float + 0.5f; STBIR_CLAMP(f, 0, 255); output[2] = (unsigned char)f;
+    #endif
+    output += stbir__coder_min_num;
+    encode += stbir__coder_min_num;
+  }
+  #endif
+  #endif
+}
+
+static void STBIR__CODER_NAME(stbir__decode_uint8_linear)( float * decodep, int width_times_channels, void const * inputp )
+{
+  float STBIR_STREAMOUT_PTR( * ) decode = decodep;
+  float * decode_end = (float*) decode + width_times_channels;
+  unsigned char const * input = (unsigned char const*)inputp;
+
+  #ifdef STBIR_SIMD
+  unsigned char const * end_input_m16 = input + width_times_channels - 16;
+  if ( width_times_channels >= 16 )
+  {
+    decode_end -= 16;
+    for(;;)
+    {
+      #ifdef STBIR_SIMD8
+      stbir__simdi i; stbir__simdi8 o0,o1;
+      stbir__simdf8 of0, of1;
+      STBIR_NO_UNROLL(decode);
+      stbir__simdi_load( i, input );
+      stbir__simdi8_expand_u8_to_u32( o0, o1, i );
+      stbir__simdi8_convert_i32_to_float( of0, o0 );
+      stbir__simdi8_convert_i32_to_float( of1, o1 );
+      stbir__decode_simdf8_flip( of0 );
+      stbir__decode_simdf8_flip( of1 );
+      stbir__simdf8_store( decode + 0, of0 );
+      stbir__simdf8_store( decode + 8, of1 );
+      #else
+      stbir__simdi i, o0, o1, o2, o3;
+      stbir__simdf of0, of1, of2, of3;
+      STBIR_NO_UNROLL(decode);
+      stbir__simdi_load( i, input );
+      stbir__simdi_expand_u8_to_u32( o0,o1,o2,o3,i);
+      stbir__simdi_convert_i32_to_float( of0, o0 );
+      stbir__simdi_convert_i32_to_float( of1, o1 );
+      stbir__simdi_convert_i32_to_float( of2, o2 );
+      stbir__simdi_convert_i32_to_float( of3, o3 );
+      stbir__decode_simdf4_flip( of0 );
+      stbir__decode_simdf4_flip( of1 );
+      stbir__decode_simdf4_flip( of2 );
+      stbir__decode_simdf4_flip( of3 );
+      stbir__simdf_store( decode + 0,  of0 );
+      stbir__simdf_store( decode + 4,  of1 );
+      stbir__simdf_store( decode + 8,  of2 );
+      stbir__simdf_store( decode + 12, of3 );
+#endif
+      decode += 16;
+      input += 16;
+      if ( decode <= decode_end ) 
+        continue;
+      if ( decode == ( decode_end + 16 ) )
+        break;
+      decode = decode_end; // backup and do last couple
+      input = end_input_m16;
+    }
+    return;
+  }
+  #endif
+
+  // try to do blocks of 4 when you can
+  #if stbir__coder_min_num != 3 // doesn't divide cleanly by four
+  decode += 4;
+  while( decode <= decode_end )
+  {
+    STBIR_SIMD_NO_UNROLL(decode);
+    decode[0-4] = ((float)(input[stbir__decode_order0]));
+    decode[1-4] = ((float)(input[stbir__decode_order1]));
+    decode[2-4] = ((float)(input[stbir__decode_order2]));
+    decode[3-4] = ((float)(input[stbir__decode_order3]));
+    decode += 4;
+    input += 4;
+  }
+  decode -= 4;
+  #endif
+
+  // do the remnants
+  #if stbir__coder_min_num < 4
+  while( decode < decode_end )
+  {
+    STBIR_NO_UNROLL(decode);
+    decode[0] = ((float)(input[stbir__decode_order0]));
+    #if stbir__coder_min_num >= 2
+    decode[1] = ((float)(input[stbir__decode_order1]));
+    #endif
+    #if stbir__coder_min_num >= 3
+    decode[2] = ((float)(input[stbir__decode_order2]));
+    #endif
+    decode += stbir__coder_min_num;
+    input += stbir__coder_min_num;
+  }
+  #endif
+}
+
+static void STBIR__CODER_NAME( stbir__encode_uint8_linear )( void * outputp, int width_times_channels, float const * encode )
+{
+  unsigned char STBIR_SIMD_STREAMOUT_PTR( * ) output = (unsigned char *) outputp;
+  unsigned char * end_output = ( (unsigned char *) output ) + width_times_channels;
+
+  #ifdef STBIR_SIMD
+  if ( width_times_channels >= stbir__simdfX_float_count*2 )
+  {
+    float const * end_encode_m8 = encode + width_times_channels - stbir__simdfX_float_count*2;
+    end_output -= stbir__simdfX_float_count*2;
+    for(;;)
+    {
+      stbir__simdfX e0, e1;
+      stbir__simdi i;
+      STBIR_SIMD_NO_UNROLL(encode);
+      stbir__simdfX_add_mem( e0, STBIR_simd_point5X, encode );
+      stbir__simdfX_add_mem( e1, STBIR_simd_point5X, encode+stbir__simdfX_float_count );
+      stbir__encode_simdfX_unflip( e0 );
+      stbir__encode_simdfX_unflip( e1 );
+      #ifdef STBIR_SIMD8
+      stbir__simdf8_pack_to_16bytes( i, e0, e1 ); 
+      stbir__simdi_store( output, i );
+      #else
+      stbir__simdf_pack_to_8bytes( i, e0, e1 ); 
+      stbir__simdi_store2( output, i );
+      #endif
+      encode += stbir__simdfX_float_count*2;
+      output += stbir__simdfX_float_count*2;
+      if ( output <= end_output ) 
+        continue;
+      if ( output == ( end_output + stbir__simdfX_float_count*2 ) )
+        break;
+      output = end_output; // backup and do last couple
+      encode = end_encode_m8;
+    }
+    return;
+  }
+
+  // try to do blocks of 4 when you can
+  #if stbir__coder_min_num != 3 // doesn't divide cleanly by four
+  output += 4;
+  while( output <= end_output )
+  {
+    stbir__simdf e0;
+    stbir__simdi i0;
+    STBIR_NO_UNROLL(encode);
+    stbir__simdf_load( e0, encode );
+    stbir__simdf_add( e0, STBIR__CONSTF(STBIR_simd_point5), e0 );
+    stbir__encode_simdf4_unflip( e0 );
+    stbir__simdf_pack_to_8bytes( i0, e0, e0 );  // only use first 4
+    *(int*)(output-4) = stbir__simdi_to_int( i0 );
+    output += 4;
+    encode += 4;
+  }
+  output -= 4;
+  #endif
+
+  #else
+
+  // try to do blocks of 4 when you can
+  #if stbir__coder_min_num != 3 // doesn't divide cleanly by four
+  output += 4;
+  while( output <= end_output )
+  {
+    float f;
+    f = encode[stbir__encode_order0] + 0.5f; STBIR_CLAMP(f, 0, 255); output[0-4] = (unsigned char)f;
+    f = encode[stbir__encode_order1] + 0.5f; STBIR_CLAMP(f, 0, 255); output[1-4] = (unsigned char)f;
+    f = encode[stbir__encode_order2] + 0.5f; STBIR_CLAMP(f, 0, 255); output[2-4] = (unsigned char)f;
+    f = encode[stbir__encode_order3] + 0.5f; STBIR_CLAMP(f, 0, 255); output[3-4] = (unsigned char)f;
+    output += 4;
+    encode += 4;
+  }
+  output -= 4;
+  #endif
+
+  #endif
+
+  // do the remnants
+  #if stbir__coder_min_num < 4
+  while( output < end_output )
+  {
+    float f;
+    STBIR_NO_UNROLL(encode);
+    f = encode[stbir__encode_order0] + 0.5f; STBIR_CLAMP(f, 0, 255); output[0] = (unsigned char)f;
+    #if stbir__coder_min_num >= 2
+    f = encode[stbir__encode_order1] + 0.5f; STBIR_CLAMP(f, 0, 255); output[1] = (unsigned char)f;
+    #endif
+    #if stbir__coder_min_num >= 3
+    f = encode[stbir__encode_order2] + 0.5f; STBIR_CLAMP(f, 0, 255); output[2] = (unsigned char)f;
+    #endif
+    output += stbir__coder_min_num;
+    encode += stbir__coder_min_num;
+  }
+  #endif
+}
+
+static void STBIR__CODER_NAME(stbir__decode_uint8_srgb)( float * decodep, int width_times_channels, void const * inputp )
+{
+  float STBIR_STREAMOUT_PTR( * ) decode = decodep;
+  float const * decode_end = (float*) decode + width_times_channels;
+  unsigned char const * input = (unsigned char const *)inputp;
+
+  // try to do blocks of 4 when you can
+  #if stbir__coder_min_num != 3 // doesn't divide cleanly by four
+  decode += 4;
+  while( decode <= decode_end )
+  {
+    decode[0-4] = stbir__srgb_uchar_to_linear_float[ input[ stbir__decode_order0 ] ];
+    decode[1-4] = stbir__srgb_uchar_to_linear_float[ input[ stbir__decode_order1 ] ];
+    decode[2-4] = stbir__srgb_uchar_to_linear_float[ input[ stbir__decode_order2 ] ];
+    decode[3-4] = stbir__srgb_uchar_to_linear_float[ input[ stbir__decode_order3 ] ];
+    decode += 4;
+    input += 4;
+  }
+  decode -= 4;
+  #endif
+
+  // do the remnants
+  #if stbir__coder_min_num < 4
+  while( decode < decode_end )
+  {
+    STBIR_NO_UNROLL(decode);
+    decode[0] = stbir__srgb_uchar_to_linear_float[ input[ stbir__decode_order0 ] ];
+    #if stbir__coder_min_num >= 2
+    decode[1] = stbir__srgb_uchar_to_linear_float[ input[ stbir__decode_order1 ] ];
+    #endif
+    #if stbir__coder_min_num >= 3
+    decode[2] = stbir__srgb_uchar_to_linear_float[ input[ stbir__decode_order2 ] ];
+    #endif
+    decode += stbir__coder_min_num;
+    input += stbir__coder_min_num;
+  }
+  #endif
+}
+
+#define stbir__min_max_shift20( i, f ) \
+    stbir__simdf_max( f, f, stbir_simdf_casti(STBIR__CONSTI( STBIR_almost_zero )) ); \
+    stbir__simdf_min( f, f, stbir_simdf_casti(STBIR__CONSTI( STBIR_almost_one  )) ); \
+    stbir__simdi_32shr( i, stbir_simdi_castf( f ), 20 ); 
+
+#define stbir__scale_and_convert( i, f ) \
+    stbir__simdf_madd( f, STBIR__CONSTF( STBIR_simd_point5 ), STBIR__CONSTF( STBIR_max_uint8_as_float ), f ); \
+    stbir__simdf_max( f, f, stbir__simdf_zeroP() ); \
+    stbir__simdf_min( f, f, STBIR__CONSTF( STBIR_max_uint8_as_float ) ); \
+    stbir__simdf_convert_float_to_i32( i, f );
+
+#define stbir__linear_to_srgb_finish( i, f ) \
+{ \
+    stbir__simdi temp;  \
+    stbir__simdi_32shr( temp, stbir_simdi_castf( f ), 12 ) ; \
+    stbir__simdi_and( temp, temp, STBIR__CONSTI(STBIR_mastissa_mask) ); \
+    stbir__simdi_or( temp, temp, STBIR__CONSTI(STBIR_topscale) ); \
+    stbir__simdi_16madd( i, i, temp ); \
+    stbir__simdi_32shr( i, i, 16 ); \
+}
+
+#define stbir__simdi_table_lookup2( v0,v1, table ) \
+{ \
+  stbir__simdi_u32 temp0,temp1; \
+  temp0.m128i_i128 = v0; \
+  temp1.m128i_i128 = v1; \
+  temp0.m128i_u32[0] = table[temp0.m128i_i32[0]]; temp0.m128i_u32[1] = table[temp0.m128i_i32[1]]; temp0.m128i_u32[2] = table[temp0.m128i_i32[2]]; temp0.m128i_u32[3] = table[temp0.m128i_i32[3]]; \
+  temp1.m128i_u32[0] = table[temp1.m128i_i32[0]]; temp1.m128i_u32[1] = table[temp1.m128i_i32[1]]; temp1.m128i_u32[2] = table[temp1.m128i_i32[2]]; temp1.m128i_u32[3] = table[temp1.m128i_i32[3]]; \
+  v0 = temp0.m128i_i128; \
+  v1 = temp1.m128i_i128; \
+} 
+
+#define stbir__simdi_table_lookup3( v0,v1,v2, table ) \
+{ \
+  stbir__simdi_u32 temp0,temp1,temp2; \
+  temp0.m128i_i128 = v0; \
+  temp1.m128i_i128 = v1; \
+  temp2.m128i_i128 = v2; \
+  temp0.m128i_u32[0] = table[temp0.m128i_i32[0]]; temp0.m128i_u32[1] = table[temp0.m128i_i32[1]]; temp0.m128i_u32[2] = table[temp0.m128i_i32[2]]; temp0.m128i_u32[3] = table[temp0.m128i_i32[3]]; \
+  temp1.m128i_u32[0] = table[temp1.m128i_i32[0]]; temp1.m128i_u32[1] = table[temp1.m128i_i32[1]]; temp1.m128i_u32[2] = table[temp1.m128i_i32[2]]; temp1.m128i_u32[3] = table[temp1.m128i_i32[3]]; \
+  temp2.m128i_u32[0] = table[temp2.m128i_i32[0]]; temp2.m128i_u32[1] = table[temp2.m128i_i32[1]]; temp2.m128i_u32[2] = table[temp2.m128i_i32[2]]; temp2.m128i_u32[3] = table[temp2.m128i_i32[3]]; \
+  v0 = temp0.m128i_i128; \
+  v1 = temp1.m128i_i128; \
+  v2 = temp2.m128i_i128; \
+}
+
+#define stbir__simdi_table_lookup4( v0,v1,v2,v3, table ) \
+{ \
+  stbir__simdi_u32 temp0,temp1,temp2,temp3; \
+  temp0.m128i_i128 = v0; \
+  temp1.m128i_i128 = v1; \
+  temp2.m128i_i128 = v2; \
+  temp3.m128i_i128 = v3; \
+  temp0.m128i_u32[0] = table[temp0.m128i_i32[0]]; temp0.m128i_u32[1] = table[temp0.m128i_i32[1]]; temp0.m128i_u32[2] = table[temp0.m128i_i32[2]]; temp0.m128i_u32[3] = table[temp0.m128i_i32[3]]; \
+  temp1.m128i_u32[0] = table[temp1.m128i_i32[0]]; temp1.m128i_u32[1] = table[temp1.m128i_i32[1]]; temp1.m128i_u32[2] = table[temp1.m128i_i32[2]]; temp1.m128i_u32[3] = table[temp1.m128i_i32[3]]; \
+  temp2.m128i_u32[0] = table[temp2.m128i_i32[0]]; temp2.m128i_u32[1] = table[temp2.m128i_i32[1]]; temp2.m128i_u32[2] = table[temp2.m128i_i32[2]]; temp2.m128i_u32[3] = table[temp2.m128i_i32[3]]; \
+  temp3.m128i_u32[0] = table[temp3.m128i_i32[0]]; temp3.m128i_u32[1] = table[temp3.m128i_i32[1]]; temp3.m128i_u32[2] = table[temp3.m128i_i32[2]]; temp3.m128i_u32[3] = table[temp3.m128i_i32[3]]; \
+  v0 = temp0.m128i_i128; \
+  v1 = temp1.m128i_i128; \
+  v2 = temp2.m128i_i128; \
+  v3 = temp3.m128i_i128; \
+} 
+
+static void STBIR__CODER_NAME( stbir__encode_uint8_srgb )( void * outputp, int width_times_channels, float const * encode )
+{
+  unsigned char STBIR_SIMD_STREAMOUT_PTR( * ) output = (unsigned char*) outputp;
+  unsigned char * end_output = ( (unsigned char*) output ) + width_times_channels;
+
+  #ifdef STBIR_SIMD
+  stbir_uint32 const * to_srgb = fp32_to_srgb8_tab4 - (127-13)*8;
+
+  if ( width_times_channels >= 16 )
+  {
+    float const * end_encode_m16 = encode + width_times_channels - 16;
+    end_output -= 16;
+    for(;;)
+    {
+      stbir__simdf f0, f1, f2, f3;
+      stbir__simdi i0, i1, i2, i3; 
+      STBIR_SIMD_NO_UNROLL(encode);
+
+      stbir__simdf_load4_transposed( f0, f1, f2, f3, encode );
+
+      stbir__min_max_shift20( i0, f0 );
+      stbir__min_max_shift20( i1, f1 );
+      stbir__min_max_shift20( i2, f2 );
+      stbir__min_max_shift20( i3, f3 );
+      
+      stbir__simdi_table_lookup4( i0, i1, i2, i3, to_srgb );
+     
+      stbir__linear_to_srgb_finish( i0, f0 );
+      stbir__linear_to_srgb_finish( i1, f1 );
+      stbir__linear_to_srgb_finish( i2, f2 );
+      stbir__linear_to_srgb_finish( i3, f3 );
+
+      stbir__interleave_pack_and_store_16_u8( output,  STBIR_strs_join1(i, ,stbir__encode_order0), STBIR_strs_join1(i, ,stbir__encode_order1), STBIR_strs_join1(i, ,stbir__encode_order2), STBIR_strs_join1(i, ,stbir__encode_order3) );
+
+      encode += 16;
+      output += 16;
+      if ( output <= end_output ) 
+        continue;
+      if ( output == ( end_output + 16 ) )
+        break;
+      output = end_output; // backup and do last couple
+      encode = end_encode_m16;
+    }
+    return;
+  }
+  #endif
+
+  // try to do blocks of 4 when you can
+  #if stbir__coder_min_num != 3 // doesn't divide cleanly by four
+  output += 4;
+  while ( output <= end_output )
+  {
+    STBIR_SIMD_NO_UNROLL(encode);
+
+    output[0-4] = stbir__linear_to_srgb_uchar( encode[stbir__encode_order0] );
+    output[1-4] = stbir__linear_to_srgb_uchar( encode[stbir__encode_order1] );
+    output[2-4] = stbir__linear_to_srgb_uchar( encode[stbir__encode_order2] );
+    output[3-4] = stbir__linear_to_srgb_uchar( encode[stbir__encode_order3] );
+
+    output += 4;
+    encode += 4;
+  }
+  output -= 4;
+  #endif
+
+  // do the remnants
+  #if stbir__coder_min_num < 4
+  while( output < end_output ) 
+  {
+    STBIR_NO_UNROLL(encode);
+    output[0] = stbir__linear_to_srgb_uchar( encode[stbir__encode_order0] );
+    #if stbir__coder_min_num >= 2
+    output[1] = stbir__linear_to_srgb_uchar( encode[stbir__encode_order1] );
+    #endif
+    #if stbir__coder_min_num >= 3
+    output[2] = stbir__linear_to_srgb_uchar( encode[stbir__encode_order2] );
+    #endif
+    output += stbir__coder_min_num;
+    encode += stbir__coder_min_num;
+  }
+  #endif
+}
+
+#if ( stbir__coder_min_num == 4 ) || ( ( stbir__coder_min_num == 1 ) && ( !defined(stbir__decode_swizzle) ) )
+
+static void STBIR__CODER_NAME(stbir__decode_uint8_srgb4_linearalpha)( float * decodep, int width_times_channels, void const * inputp )
+{
+  float STBIR_STREAMOUT_PTR( * ) decode = decodep;
+  float const * decode_end = (float*) decode + width_times_channels;
+  unsigned char const * input = (unsigned char const *)inputp;
+  do {
+    decode[0] = stbir__srgb_uchar_to_linear_float[ input[stbir__decode_order0] ];
+    decode[1] = stbir__srgb_uchar_to_linear_float[ input[stbir__decode_order1] ];
+    decode[2] = stbir__srgb_uchar_to_linear_float[ input[stbir__decode_order2] ];
+    decode[3] = ( (float) input[stbir__decode_order3] ) * stbir__max_uint8_as_float_inverted;
+    input += 4;
+    decode += 4;
+  } while( decode < decode_end );
+}
+
+
+static void STBIR__CODER_NAME( stbir__encode_uint8_srgb4_linearalpha )( void * outputp, int width_times_channels, float const * encode )
+{
+  unsigned char STBIR_SIMD_STREAMOUT_PTR( * ) output = (unsigned char*) outputp;
+  unsigned char * end_output = ( (unsigned char*) output ) + width_times_channels;
+
+  #ifdef STBIR_SIMD
+  stbir_uint32 const * to_srgb = fp32_to_srgb8_tab4 - (127-13)*8;
+
+  if ( width_times_channels >= 16 )
+  {
+    float const * end_encode_m16 = encode + width_times_channels - 16;
+    end_output -= 16;
+    for(;;)
+    {
+      stbir__simdf f0, f1, f2, f3;
+      stbir__simdi i0, i1, i2, i3;
+
+      STBIR_SIMD_NO_UNROLL(encode);
+      stbir__simdf_load4_transposed( f0, f1, f2, f3, encode );
+
+      stbir__min_max_shift20( i0, f0 );
+      stbir__min_max_shift20( i1, f1 );
+      stbir__min_max_shift20( i2, f2 );
+      stbir__scale_and_convert( i3, f3 ); 
+      
+      stbir__simdi_table_lookup3( i0, i1, i2, to_srgb );
+     
+      stbir__linear_to_srgb_finish( i0, f0 );
+      stbir__linear_to_srgb_finish( i1, f1 );
+      stbir__linear_to_srgb_finish( i2, f2 );
+
+      stbir__interleave_pack_and_store_16_u8( output,  STBIR_strs_join1(i, ,stbir__encode_order0), STBIR_strs_join1(i, ,stbir__encode_order1), STBIR_strs_join1(i, ,stbir__encode_order2), STBIR_strs_join1(i, ,stbir__encode_order3) );
+
+      output += 16;
+      encode += 16;
+
+      if ( output <= end_output ) 
+        continue;
+      if ( output == ( end_output + 16 ) )
+        break;
+      output = end_output; // backup and do last couple
+      encode = end_encode_m16;
+    }
+    return;
+  }
+  #endif
+
+  do {
+    float f;
+    STBIR_SIMD_NO_UNROLL(encode);                                        
+
+    output[stbir__decode_order0] = stbir__linear_to_srgb_uchar( encode[0] );
+    output[stbir__decode_order1] = stbir__linear_to_srgb_uchar( encode[1] );
+    output[stbir__decode_order2] = stbir__linear_to_srgb_uchar( encode[2] );
+
+    f = encode[3] * stbir__max_uint8_as_float + 0.5f;
+    STBIR_CLAMP(f, 0, 255);
+    output[stbir__decode_order3] = (unsigned char) f;
+
+    output += 4;
+    encode += 4;
+  } while( output < end_output );
+}
+
+#endif
+
+#if ( stbir__coder_min_num == 2 ) || ( ( stbir__coder_min_num == 1 ) && ( !defined(stbir__decode_swizzle) ) )
+
+static void STBIR__CODER_NAME(stbir__decode_uint8_srgb2_linearalpha)( float * decodep, int width_times_channels, void const * inputp )
+{
+  float STBIR_STREAMOUT_PTR( * ) decode = decodep;
+  float const * decode_end = (float*) decode + width_times_channels;
+  unsigned char const * input = (unsigned char const *)inputp;
+  decode += 4;
+  while( decode <= decode_end )
+  {
+    decode[0-4] = stbir__srgb_uchar_to_linear_float[ input[stbir__decode_order0] ];
+    decode[1-4] = ( (float) input[stbir__decode_order1] ) * stbir__max_uint8_as_float_inverted;
+    decode[2-4] = stbir__srgb_uchar_to_linear_float[ input[stbir__decode_order0+2] ];
+    decode[3-4] = ( (float) input[stbir__decode_order1+2] ) * stbir__max_uint8_as_float_inverted;
+    input += 4;
+    decode += 4;
+  }
+  decode -= 4;
+  if( decode < decode_end ) 
+  {
+    decode[0] = stbir__srgb_uchar_to_linear_float[ stbir__decode_order0 ];
+    decode[1] = ( (float) input[stbir__decode_order1] ) * stbir__max_uint8_as_float_inverted;
+  }
+}
+
+static void STBIR__CODER_NAME( stbir__encode_uint8_srgb2_linearalpha )( void * outputp, int width_times_channels, float const * encode )
+{
+  unsigned char STBIR_SIMD_STREAMOUT_PTR( * ) output = (unsigned char*) outputp;
+  unsigned char * end_output = ( (unsigned char*) output ) + width_times_channels;
+
+  #ifdef STBIR_SIMD
+  stbir_uint32 const * to_srgb = fp32_to_srgb8_tab4 - (127-13)*8;
+
+  if ( width_times_channels >= 16 )
+  {
+    float const * end_encode_m16 = encode + width_times_channels - 16;
+    end_output -= 16;
+    for(;;)
+    {
+      stbir__simdf f0, f1, f2, f3;
+      stbir__simdi i0, i1, i2, i3; 
+
+      STBIR_SIMD_NO_UNROLL(encode);
+      stbir__simdf_load4_transposed( f0, f1, f2, f3, encode );
+
+      stbir__min_max_shift20( i0, f0 );
+      stbir__scale_and_convert( i1, f1 );
+      stbir__min_max_shift20( i2, f2 );
+      stbir__scale_and_convert( i3, f3 );
+      
+      stbir__simdi_table_lookup2( i0, i2, to_srgb );
+     
+      stbir__linear_to_srgb_finish( i0, f0 );
+      stbir__linear_to_srgb_finish( i2, f2 );
+
+      stbir__interleave_pack_and_store_16_u8( output,  STBIR_strs_join1(i, ,stbir__encode_order0), STBIR_strs_join1(i, ,stbir__encode_order1), STBIR_strs_join1(i, ,stbir__encode_order2), STBIR_strs_join1(i, ,stbir__encode_order3) );
+
+      output += 16;
+      encode += 16;
+      if ( output <= end_output ) 
+        continue;
+      if ( output == ( end_output + 16 ) )
+        break;
+      output = end_output; // backup and do last couple
+      encode = end_encode_m16;
+    }
+    return;
+  }
+  #endif
+
+  do {
+    float f;
+    STBIR_SIMD_NO_UNROLL(encode);
+
+    output[stbir__decode_order0] = stbir__linear_to_srgb_uchar( encode[0] );
+
+    f = encode[1] * stbir__max_uint8_as_float + 0.5f;
+    STBIR_CLAMP(f, 0, 255);
+    output[stbir__decode_order1] = (unsigned char) f;
+
+    output += 2;
+    encode += 2;
+  } while( output < end_output );
+}
+
+#endif
+
+static void STBIR__CODER_NAME(stbir__decode_uint16_linear_scaled)( float * decodep, int width_times_channels, void const * inputp )
+{
+  float STBIR_STREAMOUT_PTR( * ) decode = decodep;
+  float * decode_end = (float*) decode + width_times_channels;
+  unsigned short const * input = (unsigned short const *)inputp;
+
+  #ifdef STBIR_SIMD
+  unsigned short const * end_input_m8 = input + width_times_channels - 8;
+  if ( width_times_channels >= 8 )
+  {
+    decode_end -= 8;
+    for(;;)
+    {
+      #ifdef STBIR_SIMD8
+      stbir__simdi i; stbir__simdi8 o;
+      stbir__simdf8 of;
+      STBIR_NO_UNROLL(decode);
+      stbir__simdi_load( i, input );
+      stbir__simdi8_expand_u16_to_u32( o, i );
+      stbir__simdi8_convert_i32_to_float( of, o );
+      stbir__simdf8_mult( of, of, STBIR_max_uint16_as_float_inverted8);
+      stbir__decode_simdf8_flip( of );
+      stbir__simdf8_store( decode + 0, of );
+      #else
+      stbir__simdi i, o0, o1;
+      stbir__simdf of0, of1;
+      STBIR_NO_UNROLL(decode);
+      stbir__simdi_load( i, input );
+      stbir__simdi_expand_u16_to_u32( o0,o1,i );
+      stbir__simdi_convert_i32_to_float( of0, o0 );
+      stbir__simdi_convert_i32_to_float( of1, o1 );
+      stbir__simdf_mult( of0, of0, STBIR__CONSTF(STBIR_max_uint16_as_float_inverted) );
+      stbir__simdf_mult( of1, of1, STBIR__CONSTF(STBIR_max_uint16_as_float_inverted));
+      stbir__decode_simdf4_flip( of0 );
+      stbir__decode_simdf4_flip( of1 );
+      stbir__simdf_store( decode + 0,  of0 );
+      stbir__simdf_store( decode + 4,  of1 );
+      #endif
+      decode += 8;  
+      input += 8;
+      if ( decode <= decode_end ) 
+        continue;
+      if ( decode == ( decode_end + 8 ) )
+        break;
+      decode = decode_end; // backup and do last couple
+      input = end_input_m8;
+    }
+    return;
+  }
+  #endif
+
+  // try to do blocks of 4 when you can
+  #if stbir__coder_min_num != 3 // doesn't divide cleanly by four
+  decode += 4;
+  while( decode <= decode_end )
+  {
+    STBIR_SIMD_NO_UNROLL(decode);
+    decode[0-4] = ((float)(input[stbir__decode_order0])) * stbir__max_uint16_as_float_inverted;
+    decode[1-4] = ((float)(input[stbir__decode_order1])) * stbir__max_uint16_as_float_inverted;
+    decode[2-4] = ((float)(input[stbir__decode_order2])) * stbir__max_uint16_as_float_inverted;
+    decode[3-4] = ((float)(input[stbir__decode_order3])) * stbir__max_uint16_as_float_inverted;
+    decode += 4;
+    input += 4;
+  }
+  decode -= 4;
+  #endif
+
+  // do the remnants
+  #if stbir__coder_min_num < 4
+  while( decode < decode_end )
+  {
+    STBIR_NO_UNROLL(decode);
+    decode[0] = ((float)(input[stbir__decode_order0])) * stbir__max_uint16_as_float_inverted;
+    #if stbir__coder_min_num >= 2
+    decode[1] = ((float)(input[stbir__decode_order1])) * stbir__max_uint16_as_float_inverted;
+    #endif
+    #if stbir__coder_min_num >= 3
+    decode[2] = ((float)(input[stbir__decode_order2])) * stbir__max_uint16_as_float_inverted;
+    #endif
+    decode += stbir__coder_min_num;
+    input += stbir__coder_min_num;
+  }
+  #endif
+}
+
+
+static void STBIR__CODER_NAME(stbir__encode_uint16_linear_scaled)( void * outputp, int width_times_channels, float const * encode )
+{
+  unsigned short STBIR_SIMD_STREAMOUT_PTR( * ) output = (unsigned short*) outputp;
+  unsigned short * end_output = ( (unsigned short*) output ) + width_times_channels;
+
+  #ifdef STBIR_SIMD
+  {
+    if ( width_times_channels >= stbir__simdfX_float_count*2 )
+    {
+      float const * end_encode_m8 = encode + width_times_channels - stbir__simdfX_float_count*2;
+      end_output -= stbir__simdfX_float_count*2;
+      for(;;)
+      {
+        stbir__simdfX e0, e1;
+        stbir__simdiX i;
+        STBIR_SIMD_NO_UNROLL(encode);
+        stbir__simdfX_madd_mem( e0, STBIR_simd_point5X, STBIR_max_uint16_as_floatX, encode );
+        stbir__simdfX_madd_mem( e1, STBIR_simd_point5X, STBIR_max_uint16_as_floatX, encode+stbir__simdfX_float_count );
+        stbir__encode_simdfX_unflip( e0 );
+        stbir__encode_simdfX_unflip( e1 );
+        stbir__simdfX_pack_to_words( i, e0, e1 );
+        stbir__simdiX_store( output, i );
+        encode += stbir__simdfX_float_count*2;
+        output += stbir__simdfX_float_count*2;
+        if ( output <= end_output ) 
+          continue;
+        if ( output == ( end_output + stbir__simdfX_float_count*2 ) )
+          break;
+        output = end_output;     // backup and do last couple
+        encode = end_encode_m8;
+      }
+      return;
+    }
+  }
+
+  // try to do blocks of 4 when you can
+  #if stbir__coder_min_num != 3 // doesn't divide cleanly by four
+  output += 4;
+  while( output <= end_output )
+  {
+    stbir__simdf e;
+    stbir__simdi i;
+    STBIR_NO_UNROLL(encode);
+    stbir__simdf_load( e, encode );
+    stbir__simdf_madd( e, STBIR__CONSTF(STBIR_simd_point5), STBIR__CONSTF(STBIR_max_uint16_as_float), e );
+    stbir__encode_simdf4_unflip( e );
+    stbir__simdf_pack_to_8words( i, e, e );  // only use first 4
+    stbir__simdi_store2( output-4, i );
+    output += 4;
+    encode += 4;
+  }
+  output -= 4;
+  #endif
+
+  // do the remnants
+  #if stbir__coder_min_num < 4
+  while( output < end_output )
+  {
+    stbir__simdf e;
+    STBIR_NO_UNROLL(encode);
+    stbir__simdf_madd1_mem( e, STBIR__CONSTF(STBIR_simd_point5), STBIR__CONSTF(STBIR_max_uint16_as_float), encode+stbir__encode_order0 ); output[0] = stbir__simdf_convert_float_to_short( e );
+    #if stbir__coder_min_num >= 2
+    stbir__simdf_madd1_mem( e, STBIR__CONSTF(STBIR_simd_point5), STBIR__CONSTF(STBIR_max_uint16_as_float), encode+stbir__encode_order1 ); output[1] = stbir__simdf_convert_float_to_short( e );
+    #endif
+    #if stbir__coder_min_num >= 3
+    stbir__simdf_madd1_mem( e, STBIR__CONSTF(STBIR_simd_point5), STBIR__CONSTF(STBIR_max_uint16_as_float), encode+stbir__encode_order2 ); output[2] = stbir__simdf_convert_float_to_short( e );
+    #endif
+    output += stbir__coder_min_num;
+    encode += stbir__coder_min_num;
+  }
+  #endif
+  
+  #else
+
+  // try to do blocks of 4 when you can
+  #if stbir__coder_min_num != 3 // doesn't divide cleanly by four
+  output += 4;
+  while( output <= end_output )
+  {
+    float f;
+    STBIR_SIMD_NO_UNROLL(encode);
+    f = encode[stbir__encode_order0] * stbir__max_uint16_as_float + 0.5f; STBIR_CLAMP(f, 0, 65535); output[0-4] = (unsigned short)f;
+    f = encode[stbir__encode_order1] * stbir__max_uint16_as_float + 0.5f; STBIR_CLAMP(f, 0, 65535); output[1-4] = (unsigned short)f;
+    f = encode[stbir__encode_order2] * stbir__max_uint16_as_float + 0.5f; STBIR_CLAMP(f, 0, 65535); output[2-4] = (unsigned short)f;
+    f = encode[stbir__encode_order3] * stbir__max_uint16_as_float + 0.5f; STBIR_CLAMP(f, 0, 65535); output[3-4] = (unsigned short)f;
+    output += 4;
+    encode += 4;
+  }
+  output -= 4;
+  #endif
+
+  // do the remnants
+  #if stbir__coder_min_num < 4
+  while( output < end_output )
+  {
+    float f;
+    STBIR_NO_UNROLL(encode);
+    f = encode[stbir__encode_order0] * stbir__max_uint16_as_float + 0.5f; STBIR_CLAMP(f, 0, 65535); output[0] = (unsigned short)f;
+    #if stbir__coder_min_num >= 2
+    f = encode[stbir__encode_order1] * stbir__max_uint16_as_float + 0.5f; STBIR_CLAMP(f, 0, 65535); output[1] = (unsigned short)f;
+    #endif
+    #if stbir__coder_min_num >= 3
+    f = encode[stbir__encode_order2] * stbir__max_uint16_as_float + 0.5f; STBIR_CLAMP(f, 0, 65535); output[2] = (unsigned short)f;
+    #endif
+    output += stbir__coder_min_num;
+    encode += stbir__coder_min_num;
+  }
+  #endif
+  #endif
+}
+
+static void STBIR__CODER_NAME(stbir__decode_uint16_linear)( float * decodep, int width_times_channels, void const * inputp )
+{
+  float STBIR_STREAMOUT_PTR( * ) decode = decodep;
+  float * decode_end = (float*) decode + width_times_channels;
+  unsigned short const * input = (unsigned short const *)inputp;
+
+  #ifdef STBIR_SIMD
+  unsigned short const * end_input_m8 = input + width_times_channels - 8;
+  if ( width_times_channels >= 8 )
+  {
+    decode_end -= 8;
+    for(;;)
+    {
+      #ifdef STBIR_SIMD8
+      stbir__simdi i; stbir__simdi8 o;
+      stbir__simdf8 of;
+      STBIR_NO_UNROLL(decode);
+      stbir__simdi_load( i, input );
+      stbir__simdi8_expand_u16_to_u32( o, i );
+      stbir__simdi8_convert_i32_to_float( of, o );
+      stbir__decode_simdf8_flip( of );
+      stbir__simdf8_store( decode + 0, of );
+      #else
+      stbir__simdi i, o0, o1;
+      stbir__simdf of0, of1;
+      STBIR_NO_UNROLL(decode);
+      stbir__simdi_load( i, input );
+      stbir__simdi_expand_u16_to_u32( o0, o1, i );
+      stbir__simdi_convert_i32_to_float( of0, o0 );
+      stbir__simdi_convert_i32_to_float( of1, o1 );
+      stbir__decode_simdf4_flip( of0 );
+      stbir__decode_simdf4_flip( of1 );
+      stbir__simdf_store( decode + 0,  of0 );
+      stbir__simdf_store( decode + 4,  of1 );
+      #endif
+      decode += 8;
+      input += 8;
+      if ( decode <= decode_end ) 
+        continue;
+      if ( decode == ( decode_end + 8 ) )
+        break;
+      decode = decode_end; // backup and do last couple
+      input = end_input_m8;
+    }
+    return;
+  }
+  #endif
+
+  // try to do blocks of 4 when you can
+  #if stbir__coder_min_num != 3 // doesn't divide cleanly by four
+  decode += 4;
+  while( decode <= decode_end )
+  {
+    STBIR_SIMD_NO_UNROLL(decode);
+    decode[0-4] = ((float)(input[stbir__decode_order0]));
+    decode[1-4] = ((float)(input[stbir__decode_order1]));
+    decode[2-4] = ((float)(input[stbir__decode_order2]));
+    decode[3-4] = ((float)(input[stbir__decode_order3]));
+    decode += 4;
+    input += 4;
+  }
+  decode -= 4;
+  #endif
+
+  // do the remnants
+  #if stbir__coder_min_num < 4
+  while( decode < decode_end )
+  {
+    STBIR_NO_UNROLL(decode);
+    decode[0] = ((float)(input[stbir__decode_order0]));
+    #if stbir__coder_min_num >= 2
+    decode[1] = ((float)(input[stbir__decode_order1]));
+    #endif
+    #if stbir__coder_min_num >= 3
+    decode[2] = ((float)(input[stbir__decode_order2]));
+    #endif
+    decode += stbir__coder_min_num;
+    input += stbir__coder_min_num;
+  }
+  #endif
+}
+
+static void STBIR__CODER_NAME(stbir__encode_uint16_linear)( void * outputp, int width_times_channels, float const * encode )
+{
+  unsigned short STBIR_SIMD_STREAMOUT_PTR( * ) output = (unsigned short*) outputp;
+  unsigned short * end_output = ( (unsigned short*) output ) + width_times_channels;
+
+  #ifdef STBIR_SIMD
+  {
+    if ( width_times_channels >= stbir__simdfX_float_count*2 )
+    {
+      float const * end_encode_m8 = encode + width_times_channels - stbir__simdfX_float_count*2;
+      end_output -= stbir__simdfX_float_count*2;
+      for(;;)
+      {
+        stbir__simdfX e0, e1;
+        stbir__simdiX i;
+        STBIR_SIMD_NO_UNROLL(encode);
+        stbir__simdfX_add_mem( e0, STBIR_simd_point5X, encode );
+        stbir__simdfX_add_mem( e1, STBIR_simd_point5X, encode+stbir__simdfX_float_count );
+        stbir__encode_simdfX_unflip( e0 );
+        stbir__encode_simdfX_unflip( e1 );
+        stbir__simdfX_pack_to_words( i, e0, e1 );
+        stbir__simdiX_store( output, i );
+        encode += stbir__simdfX_float_count*2;
+        output += stbir__simdfX_float_count*2;
+        if ( output <= end_output ) 
+          continue;
+        if ( output == ( end_output + stbir__simdfX_float_count*2 ) )
+          break;
+        output = end_output; // backup and do last couple
+        encode = end_encode_m8;
+      }
+      return;
+    }
+  }
+
+  // try to do blocks of 4 when you can
+  #if stbir__coder_min_num != 3 // doesn't divide cleanly by four
+  output += 4;
+  while( output <= end_output )
+  {
+    stbir__simdf e;
+    stbir__simdi i;
+    STBIR_NO_UNROLL(encode);
+    stbir__simdf_load( e, encode );
+    stbir__simdf_add( e, STBIR__CONSTF(STBIR_simd_point5), e );
+    stbir__encode_simdf4_unflip( e );
+    stbir__simdf_pack_to_8words( i, e, e );  // only use first 4
+    stbir__simdi_store2( output-4, i );
+    output += 4;
+    encode += 4;
+  }
+  output -= 4;
+  #endif
+
+  #else
+
+  // try to do blocks of 4 when you can
+  #if  stbir__coder_min_num != 3 // doesn't divide cleanly by four
+  output += 4;
+  while( output <= end_output )
+  {
+    float f;
+    STBIR_SIMD_NO_UNROLL(encode);
+    f = encode[stbir__encode_order0] + 0.5f; STBIR_CLAMP(f, 0, 65535); output[0-4] = (unsigned short)f;
+    f = encode[stbir__encode_order1] + 0.5f; STBIR_CLAMP(f, 0, 65535); output[1-4] = (unsigned short)f;
+    f = encode[stbir__encode_order2] + 0.5f; STBIR_CLAMP(f, 0, 65535); output[2-4] = (unsigned short)f;
+    f = encode[stbir__encode_order3] + 0.5f; STBIR_CLAMP(f, 0, 65535); output[3-4] = (unsigned short)f;
+    output += 4;
+    encode += 4;
+  }
+  output -= 4;
+  #endif
+
+  #endif
+
+  // do the remnants
+  #if stbir__coder_min_num < 4
+  while( output < end_output )
+  {
+    float f;
+    STBIR_NO_UNROLL(encode);
+    f = encode[stbir__encode_order0] + 0.5f; STBIR_CLAMP(f, 0, 65535); output[0] = (unsigned short)f;
+    #if stbir__coder_min_num >= 2
+    f = encode[stbir__encode_order1] + 0.5f; STBIR_CLAMP(f, 0, 65535); output[1] = (unsigned short)f;
+    #endif
+    #if stbir__coder_min_num >= 3
+    f = encode[stbir__encode_order2] + 0.5f; STBIR_CLAMP(f, 0, 65535); output[2] = (unsigned short)f;
+    #endif
+    output += stbir__coder_min_num;
+    encode += stbir__coder_min_num;
+  }
+  #endif
+}
+
+static void STBIR__CODER_NAME(stbir__decode_half_float_linear)( float * decodep, int width_times_channels, void const * inputp )
+{
+  float STBIR_STREAMOUT_PTR( * ) decode = decodep;
+  float * decode_end = (float*) decode + width_times_channels;
+  stbir__FP16 const * input = (stbir__FP16 const *)inputp;
+
+  #ifdef STBIR_SIMD
+  if ( width_times_channels >= 8 )
+  {
+    stbir__FP16 const * end_input_m8 = input + width_times_channels - 8;
+    decode_end -= 8;
+    for(;;)
+    {
+      STBIR_NO_UNROLL(decode);
+
+      stbir__half_to_float_SIMD( decode, input );
+      #ifdef stbir__decode_swizzle
+      #ifdef STBIR_SIMD8
+      {
+        stbir__simdf8 of;
+        stbir__simdf8_load( of, decode );
+        stbir__decode_simdf8_flip( of );
+        stbir__simdf8_store( decode, of );
+      }
+      #else
+      {
+        stbir__simdf of0,of1;
+        stbir__simdf_load( of0, decode );
+        stbir__simdf_load( of1, decode+4 );
+        stbir__decode_simdf4_flip( of0 );
+        stbir__decode_simdf4_flip( of1 );
+        stbir__simdf_store( decode, of0 );
+        stbir__simdf_store( decode+4, of1 );
+      }
+      #endif
+      #endif
+      decode += 8;
+      input += 8;
+      if ( decode <= decode_end ) 
+        continue;
+      if ( decode == ( decode_end + 8 ) )
+        break;
+      decode = decode_end; // backup and do last couple
+      input = end_input_m8;
+    }
+    return;
+  }
+  #endif
+
+  // try to do blocks of 4 when you can
+  #if stbir__coder_min_num != 3 // doesn't divide cleanly by four
+  decode += 4;
+  while( decode <= decode_end )
+  {
+    STBIR_SIMD_NO_UNROLL(decode);
+    decode[0-4] = stbir__half_to_float(input[stbir__decode_order0]);
+    decode[1-4] = stbir__half_to_float(input[stbir__decode_order1]);
+    decode[2-4] = stbir__half_to_float(input[stbir__decode_order2]);
+    decode[3-4] = stbir__half_to_float(input[stbir__decode_order3]);
+    decode += 4;
+    input += 4;
+  }
+  decode -= 4;
+  #endif
+
+  // do the remnants
+  #if stbir__coder_min_num < 4
+  while( decode < decode_end )
+  {
+    STBIR_NO_UNROLL(decode);
+    decode[0] = stbir__half_to_float(input[stbir__decode_order0]);
+    #if stbir__coder_min_num >= 2
+    decode[1] = stbir__half_to_float(input[stbir__decode_order1]);
+    #endif
+    #if stbir__coder_min_num >= 3
+    decode[2] = stbir__half_to_float(input[stbir__decode_order2]);
+    #endif
+    decode += stbir__coder_min_num;
+    input += stbir__coder_min_num;
+  }
+  #endif
+}
+
+static void STBIR__CODER_NAME( stbir__encode_half_float_linear )( void * outputp, int width_times_channels, float const * encode )
+{
+  stbir__FP16 STBIR_SIMD_STREAMOUT_PTR( * ) output = (stbir__FP16*) outputp;
+  stbir__FP16 * end_output = ( (stbir__FP16*) output ) + width_times_channels;
+
+  #ifdef STBIR_SIMD
+  if ( width_times_channels >= 8 )
+  {
+    float const * end_encode_m8 = encode + width_times_channels - 8;
+    end_output -= 8;
+    for(;;)
+    {
+      STBIR_SIMD_NO_UNROLL(encode);
+      #ifdef stbir__decode_swizzle
+      #ifdef STBIR_SIMD8
+      {
+        stbir__simdf8 of;
+        stbir__simdf8_load( of, encode );
+        stbir__encode_simdf8_unflip( of );
+        stbir__float_to_half_SIMD( output, (float*)&of );
+      }
+      #else
+      {
+        stbir__simdf of[2];
+        stbir__simdf_load( of[0], encode );
+        stbir__simdf_load( of[1], encode+4 );
+        stbir__encode_simdf4_unflip( of[0] );
+        stbir__encode_simdf4_unflip( of[1] );
+        stbir__float_to_half_SIMD( output, (float*)of );
+      }
+      #endif
+      #else
+      stbir__float_to_half_SIMD( output, encode );
+      #endif
+      encode += 8;
+      output += 8;
+      if ( output <= end_output ) 
+        continue;
+      if ( output == ( end_output + 8 ) )
+        break;
+      output = end_output; // backup and do last couple
+      encode = end_encode_m8;
+    }
+    return;
+  }
+  #endif
+
+  // try to do blocks of 4 when you can
+  #if stbir__coder_min_num != 3 // doesn't divide cleanly by four
+  output += 4;
+  while( output <= end_output )
+  {
+    STBIR_SIMD_NO_UNROLL(output);
+    output[0-4] = stbir__float_to_half(encode[stbir__encode_order0]);
+    output[1-4] = stbir__float_to_half(encode[stbir__encode_order1]);
+    output[2-4] = stbir__float_to_half(encode[stbir__encode_order2]);
+    output[3-4] = stbir__float_to_half(encode[stbir__encode_order3]);
+    output += 4;
+    encode += 4;
+  }
+  output -= 4;
+  #endif
+
+  // do the remnants
+  #if stbir__coder_min_num < 4
+  while( output < end_output )
+  {
+    STBIR_NO_UNROLL(output);
+    output[0] = stbir__float_to_half(encode[stbir__encode_order0]);
+    #if stbir__coder_min_num >= 2
+    output[1] = stbir__float_to_half(encode[stbir__encode_order1]);
+    #endif
+    #if stbir__coder_min_num >= 3
+    output[2] = stbir__float_to_half(encode[stbir__encode_order2]);
+    #endif
+    output += stbir__coder_min_num;
+    encode += stbir__coder_min_num;
+  }
+  #endif
+}
+
+static void STBIR__CODER_NAME(stbir__decode_float_linear)( float * decodep, int width_times_channels, void const * inputp )
+{
+  #ifdef stbir__decode_swizzle
+  float STBIR_STREAMOUT_PTR( * ) decode = decodep;
+  float * decode_end = (float*) decode + width_times_channels;
+  float const * input = (float const *)inputp;
+
+  #ifdef STBIR_SIMD
+  if ( width_times_channels >= 16 )
+  {
+    float const * end_input_m16 = input + width_times_channels - 16;
+    decode_end -= 16;
+    for(;;)
+    {
+      STBIR_NO_UNROLL(decode);
+      #ifdef stbir__decode_swizzle
+      #ifdef STBIR_SIMD8
+      {
+        stbir__simdf8 of0,of1;
+        stbir__simdf8_load( of0, input );
+        stbir__simdf8_load( of1, input+8 );
+        stbir__decode_simdf8_flip( of0 );
+        stbir__decode_simdf8_flip( of1 );
+        stbir__simdf8_store( decode, of0 );
+        stbir__simdf8_store( decode+8, of1 );
+      }
+      #else
+      {
+        stbir__simdf of0,of1,of2,of3;
+        stbir__simdf_load( of0, input );
+        stbir__simdf_load( of1, input+4 );
+        stbir__simdf_load( of2, input+8 );
+        stbir__simdf_load( of3, input+12 );
+        stbir__decode_simdf4_flip( of0 );
+        stbir__decode_simdf4_flip( of1 );
+        stbir__decode_simdf4_flip( of2 );
+        stbir__decode_simdf4_flip( of3 );
+        stbir__simdf_store( decode, of0 );
+        stbir__simdf_store( decode+4, of1 );
+        stbir__simdf_store( decode+8, of2 );
+        stbir__simdf_store( decode+12, of3 );
+      }
+      #endif
+      #endif
+      decode += 16;
+      input += 16;
+      if ( decode <= decode_end ) 
+        continue;
+      if ( decode == ( decode_end + 16 ) )
+        break;
+      decode = decode_end; // backup and do last couple
+      input = end_input_m16;
+    }
+    return;
+  }
+  #endif
+
+  // try to do blocks of 4 when you can
+  #if stbir__coder_min_num != 3 // doesn't divide cleanly by four
+  decode += 4;
+  while( decode <= decode_end )
+  {
+    STBIR_SIMD_NO_UNROLL(decode);
+    decode[0-4] = input[stbir__decode_order0];
+    decode[1-4] = input[stbir__decode_order1];
+    decode[2-4] = input[stbir__decode_order2];
+    decode[3-4] = input[stbir__decode_order3];
+    decode += 4;
+    input += 4;
+  }
+  decode -= 4;
+  #endif
+
+  // do the remnants
+  #if stbir__coder_min_num < 4
+  while( decode < decode_end )
+  {
+    STBIR_NO_UNROLL(decode);
+    decode[0] = input[stbir__decode_order0];
+    #if stbir__coder_min_num >= 2
+    decode[1] = input[stbir__decode_order1];
+    #endif
+    #if stbir__coder_min_num >= 3
+    decode[2] = input[stbir__decode_order2];
+    #endif
+    decode += stbir__coder_min_num;
+    input += stbir__coder_min_num;
+  }
+  #endif
+
+  #else
+  
+  if ( (void*)decodep != inputp )
+    STBIR_MEMCPY( decodep, inputp, width_times_channels * sizeof( float ) );
+  
+  #endif
+}
+
+static void STBIR__CODER_NAME( stbir__encode_float_linear )( void * outputp, int width_times_channels, float const * encode )
+{
+  #if !defined( STBIR_FLOAT_HIGH_CLAMP ) && !defined(STBIR_FLOAT_LO_CLAMP) && !defined(stbir__decode_swizzle)
+
+  if ( (void*)outputp != (void*) encode )
+    STBIR_MEMCPY( outputp, encode, width_times_channels * sizeof( float ) );
+
+  #else
+
+  float STBIR_SIMD_STREAMOUT_PTR( * ) output = (float*) outputp;
+  float * end_output = ( (float*) output ) + width_times_channels;
+
+  #ifdef STBIR_FLOAT_HIGH_CLAMP
+  #define stbir_scalar_hi_clamp( v ) if ( v > STBIR_FLOAT_HIGH_CLAMP ) v = STBIR_FLOAT_HIGH_CLAMP;
+  #else
+  #define stbir_scalar_hi_clamp( v )
+  #endif
+  #ifdef STBIR_FLOAT_LOW_CLAMP
+  #define stbir_scalar_lo_clamp( v ) if ( v < STBIR_FLOAT_LOW_CLAMP ) v = STBIR_FLOAT_LOW_CLAMP;
+  #else
+  #define stbir_scalar_lo_clamp( v )
+  #endif
+
+  #ifdef STBIR_SIMD
+
+  #ifdef STBIR_FLOAT_HIGH_CLAMP
+  const stbir__simdfX high_clamp = stbir__simdf_frepX(STBIR_FLOAT_HIGH_CLAMP);
+  #endif
+  #ifdef STBIR_FLOAT_LOW_CLAMP
+  const stbir__simdfX low_clamp = stbir__simdf_frepX(STBIR_FLOAT_LOW_CLAMP);
+  #endif
+
+  if ( width_times_channels >= ( stbir__simdfX_float_count * 2 ) )
+  {
+    float const * end_encode_m8 = encode + width_times_channels - ( stbir__simdfX_float_count * 2 );
+    end_output -= ( stbir__simdfX_float_count * 2 );
+    for(;;)
+    {
+      stbir__simdfX e0, e1;
+      STBIR_SIMD_NO_UNROLL(encode);
+      stbir__simdfX_load( e0, encode );
+      stbir__simdfX_load( e1, encode+stbir__simdfX_float_count );
+#ifdef STBIR_FLOAT_HIGH_CLAMP
+      stbir__simdfX_min( e0, e0, high_clamp );
+      stbir__simdfX_min( e1, e1, high_clamp );
+#endif      
+#ifdef STBIR_FLOAT_LOW_CLAMP
+      stbir__simdfX_max( e0, e0, low_clamp );
+      stbir__simdfX_max( e1, e1, low_clamp );
+#endif      
+      stbir__encode_simdfX_unflip( e0 );
+      stbir__encode_simdfX_unflip( e1 );
+      stbir__simdfX_store( output, e0 );
+      stbir__simdfX_store( output+stbir__simdfX_float_count, e1 );
+      encode += stbir__simdfX_float_count * 2;
+      output += stbir__simdfX_float_count * 2;
+      if ( output < end_output ) 
+        continue;
+      if ( output == ( end_output + ( stbir__simdfX_float_count * 2 ) ) )
+        break;
+      output = end_output; // backup and do last couple
+      encode = end_encode_m8;
+    }
+    return;
+  }
+
+  // try to do blocks of 4 when you can
+  #if stbir__coder_min_num != 3 // doesn't divide cleanly by four
+  output += 4;
+  while( output <= end_output )
+  {
+    stbir__simdf e0;
+    STBIR_NO_UNROLL(encode);
+    stbir__simdf_load( e0, encode );
+#ifdef STBIR_FLOAT_HIGH_CLAMP
+    stbir__simdf_min( e0, e0, high_clamp );
+#endif      
+#ifdef STBIR_FLOAT_LOW_CLAMP
+    stbir__simdf_max( e0, e0, low_clamp );
+#endif      
+    stbir__encode_simdf4_unflip( e0 );
+    stbir__simdf_store( output-4, e0 );
+    output += 4;
+    encode += 4;
+  }
+  output -= 4;
+  #endif
+
+  #else
+
+  // try to do blocks of 4 when you can
+  #if stbir__coder_min_num != 3 // doesn't divide cleanly by four
+  output += 4;
+  while( output <= end_output )
+  {
+    float e;
+    STBIR_SIMD_NO_UNROLL(encode);
+    e = encode[ stbir__encode_order0 ]; stbir_scalar_hi_clamp( e ); stbir_scalar_lo_clamp( e ); output[0-4] = e;
+    e = encode[ stbir__encode_order1 ]; stbir_scalar_hi_clamp( e ); stbir_scalar_lo_clamp( e ); output[1-4] = e;
+    e = encode[ stbir__encode_order2 ]; stbir_scalar_hi_clamp( e ); stbir_scalar_lo_clamp( e ); output[2-4] = e;
+    e = encode[ stbir__encode_order3 ]; stbir_scalar_hi_clamp( e ); stbir_scalar_lo_clamp( e ); output[3-4] = e;
+    output += 4;
+    encode += 4;
+  }
+  output -= 4;
+
+  #endif
+
+  #endif
+
+  // do the remnants
+  #if stbir__coder_min_num < 4
+  while( output < end_output )
+  {
+    float e;
+    STBIR_NO_UNROLL(encode);
+    e = encode[ stbir__encode_order0 ]; stbir_scalar_hi_clamp( e ); stbir_scalar_lo_clamp( e ); output[0] = e;
+    #if stbir__coder_min_num >= 2
+    e = encode[ stbir__encode_order1 ]; stbir_scalar_hi_clamp( e ); stbir_scalar_lo_clamp( e ); output[1] = e;
+    #endif
+    #if stbir__coder_min_num >= 3
+    e = encode[ stbir__encode_order2 ]; stbir_scalar_hi_clamp( e ); stbir_scalar_lo_clamp( e ); output[2] = e;
+    #endif
+    output += stbir__coder_min_num;
+    encode += stbir__coder_min_num;
+  }
+  #endif
+  
+  #endif
+}
+
+#undef stbir__decode_suffix 
+#undef stbir__decode_simdf8_flip
+#undef stbir__decode_simdf4_flip
+#undef stbir__decode_order0 
+#undef stbir__decode_order1
+#undef stbir__decode_order2
+#undef stbir__decode_order3
+#undef stbir__encode_order0 
+#undef stbir__encode_order1
+#undef stbir__encode_order2
+#undef stbir__encode_order3
+#undef stbir__encode_simdf8_unflip
+#undef stbir__encode_simdf4_unflip
+#undef stbir__encode_simdfX_unflip
+#undef STBIR__CODER_NAME
+#undef stbir__coder_min_num
+#undef stbir__decode_swizzle
+#undef stbir_scalar_hi_clamp
+#undef stbir_scalar_lo_clamp
+#undef STB_IMAGE_RESIZE_DO_CODERS
+
+#elif defined( STB_IMAGE_RESIZE_DO_VERTICALS)
+
+#ifdef STB_IMAGE_RESIZE_VERTICAL_CONTINUE
+#define STBIR_chans( start, end ) STBIR_strs_join14(start,STBIR__vertical_channels,end,_cont)
+#else
+#define STBIR_chans( start, end ) STBIR_strs_join1(start,STBIR__vertical_channels,end)
+#endif
+
+#if STBIR__vertical_channels >= 1
+#define stbIF0( code ) code
+#else
+#define stbIF0( code )
+#endif
+#if STBIR__vertical_channels >= 2
+#define stbIF1( code ) code
+#else
+#define stbIF1( code )
+#endif
+#if STBIR__vertical_channels >= 3
+#define stbIF2( code ) code
+#else
+#define stbIF2( code )
+#endif
+#if STBIR__vertical_channels >= 4
+#define stbIF3( code ) code
+#else
+#define stbIF3( code )
+#endif
+#if STBIR__vertical_channels >= 5
+#define stbIF4( code ) code
+#else
+#define stbIF4( code )
+#endif
+#if STBIR__vertical_channels >= 6
+#define stbIF5( code ) code
+#else
+#define stbIF5( code )
+#endif
+#if STBIR__vertical_channels >= 7
+#define stbIF6( code ) code
+#else
+#define stbIF6( code )
+#endif
+#if STBIR__vertical_channels >= 8
+#define stbIF7( code ) code
+#else
+#define stbIF7( code )
+#endif
+
+static void STBIR_chans( stbir__vertical_scatter_with_,_coeffs)( float ** outputs, float const * vertical_coefficients, float const * input, float const * input_end )
+{
+  stbIF0( float STBIR_SIMD_STREAMOUT_PTR( * ) output0 = outputs[0]; float c0s = vertical_coefficients[0]; )
+  stbIF1( float STBIR_SIMD_STREAMOUT_PTR( * ) output1 = outputs[1]; float c1s = vertical_coefficients[1]; )
+  stbIF2( float STBIR_SIMD_STREAMOUT_PTR( * ) output2 = outputs[2]; float c2s = vertical_coefficients[2]; )
+  stbIF3( float STBIR_SIMD_STREAMOUT_PTR( * ) output3 = outputs[3]; float c3s = vertical_coefficients[3]; )
+  stbIF4( float STBIR_SIMD_STREAMOUT_PTR( * ) output4 = outputs[4]; float c4s = vertical_coefficients[4]; )
+  stbIF5( float STBIR_SIMD_STREAMOUT_PTR( * ) output5 = outputs[5]; float c5s = vertical_coefficients[5]; )
+  stbIF6( float STBIR_SIMD_STREAMOUT_PTR( * ) output6 = outputs[6]; float c6s = vertical_coefficients[6]; )
+  stbIF7( float STBIR_SIMD_STREAMOUT_PTR( * ) output7 = outputs[7]; float c7s = vertical_coefficients[7]; )
+
+  #ifdef STBIR_SIMD
+  {
+    stbIF0(stbir__simdfX c0 = stbir__simdf_frepX( c0s ); )
+    stbIF1(stbir__simdfX c1 = stbir__simdf_frepX( c1s ); )
+    stbIF2(stbir__simdfX c2 = stbir__simdf_frepX( c2s ); )
+    stbIF3(stbir__simdfX c3 = stbir__simdf_frepX( c3s ); )
+    stbIF4(stbir__simdfX c4 = stbir__simdf_frepX( c4s ); )
+    stbIF5(stbir__simdfX c5 = stbir__simdf_frepX( c5s ); )
+    stbIF6(stbir__simdfX c6 = stbir__simdf_frepX( c6s ); )
+    stbIF7(stbir__simdfX c7 = stbir__simdf_frepX( c7s ); )
+    while ( ( (char*)input_end - (char*) input ) >= (16*stbir__simdfX_float_count) ) 
+    {
+      stbir__simdfX o0, o1, o2, o3, r0, r1, r2, r3;
+      STBIR_SIMD_NO_UNROLL(output0);
+
+      stbir__simdfX_load( r0, input );               stbir__simdfX_load( r1, input+stbir__simdfX_float_count );     stbir__simdfX_load( r2, input+(2*stbir__simdfX_float_count) );      stbir__simdfX_load( r3, input+(3*stbir__simdfX_float_count) );
+
+      #ifdef STB_IMAGE_RESIZE_VERTICAL_CONTINUE
+      stbIF0( stbir__simdfX_load( o0, output0 );     stbir__simdfX_load( o1, output0+stbir__simdfX_float_count );   stbir__simdfX_load( o2, output0+(2*stbir__simdfX_float_count) );    stbir__simdfX_load( o3, output0+(3*stbir__simdfX_float_count) );
+              stbir__simdfX_madd( o0, o0, r0, c0 );  stbir__simdfX_madd( o1, o1, r1, c0 );  stbir__simdfX_madd( o2, o2, r2, c0 );   stbir__simdfX_madd( o3, o3, r3, c0 );           
+              stbir__simdfX_store( output0, o0 );    stbir__simdfX_store( output0+stbir__simdfX_float_count, o1 );  stbir__simdfX_store( output0+(2*stbir__simdfX_float_count), o2 );   stbir__simdfX_store( output0+(3*stbir__simdfX_float_count), o3 ); )
+      stbIF1( stbir__simdfX_load( o0, output1 );     stbir__simdfX_load( o1, output1+stbir__simdfX_float_count );   stbir__simdfX_load( o2, output1+(2*stbir__simdfX_float_count) );    stbir__simdfX_load( o3, output1+(3*stbir__simdfX_float_count) );
+              stbir__simdfX_madd( o0, o0, r0, c1 );  stbir__simdfX_madd( o1, o1, r1, c1 );  stbir__simdfX_madd( o2, o2, r2, c1 );   stbir__simdfX_madd( o3, o3, r3, c1 );             
+              stbir__simdfX_store( output1, o0 );    stbir__simdfX_store( output1+stbir__simdfX_float_count, o1 );  stbir__simdfX_store( output1+(2*stbir__simdfX_float_count), o2 );   stbir__simdfX_store( output1+(3*stbir__simdfX_float_count), o3 ); )
+      stbIF2( stbir__simdfX_load( o0, output2 );     stbir__simdfX_load( o1, output2+stbir__simdfX_float_count );   stbir__simdfX_load( o2, output2+(2*stbir__simdfX_float_count) );    stbir__simdfX_load( o3, output2+(3*stbir__simdfX_float_count) );
+              stbir__simdfX_madd( o0, o0, r0, c2 );  stbir__simdfX_madd( o1, o1, r1, c2 );  stbir__simdfX_madd( o2, o2, r2, c2 );   stbir__simdfX_madd( o3, o3, r3, c2 );             
+              stbir__simdfX_store( output2, o0 );    stbir__simdfX_store( output2+stbir__simdfX_float_count, o1 );  stbir__simdfX_store( output2+(2*stbir__simdfX_float_count), o2 );   stbir__simdfX_store( output2+(3*stbir__simdfX_float_count), o3 ); )
+      stbIF3( stbir__simdfX_load( o0, output3 );     stbir__simdfX_load( o1, output3+stbir__simdfX_float_count );   stbir__simdfX_load( o2, output3+(2*stbir__simdfX_float_count) );    stbir__simdfX_load( o3, output3+(3*stbir__simdfX_float_count) );
+              stbir__simdfX_madd( o0, o0, r0, c3 );  stbir__simdfX_madd( o1, o1, r1, c3 );  stbir__simdfX_madd( o2, o2, r2, c3 );   stbir__simdfX_madd( o3, o3, r3, c3 );             
+              stbir__simdfX_store( output3, o0 );    stbir__simdfX_store( output3+stbir__simdfX_float_count, o1 );  stbir__simdfX_store( output3+(2*stbir__simdfX_float_count), o2 );   stbir__simdfX_store( output3+(3*stbir__simdfX_float_count), o3 ); )
+      stbIF4( stbir__simdfX_load( o0, output4 );     stbir__simdfX_load( o1, output4+stbir__simdfX_float_count );   stbir__simdfX_load( o2, output4+(2*stbir__simdfX_float_count) );    stbir__simdfX_load( o3, output4+(3*stbir__simdfX_float_count) );
+              stbir__simdfX_madd( o0, o0, r0, c4 );  stbir__simdfX_madd( o1, o1, r1, c4 );  stbir__simdfX_madd( o2, o2, r2, c4 );   stbir__simdfX_madd( o3, o3, r3, c4 );             
+              stbir__simdfX_store( output4, o0 );    stbir__simdfX_store( output4+stbir__simdfX_float_count, o1 );  stbir__simdfX_store( output4+(2*stbir__simdfX_float_count), o2 );   stbir__simdfX_store( output4+(3*stbir__simdfX_float_count), o3 ); )
+      stbIF5( stbir__simdfX_load( o0, output5 );     stbir__simdfX_load( o1, output5+stbir__simdfX_float_count );   stbir__simdfX_load( o2, output5+(2*stbir__simdfX_float_count));    stbir__simdfX_load( o3, output5+(3*stbir__simdfX_float_count) );
+              stbir__simdfX_madd( o0, o0, r0, c5 );  stbir__simdfX_madd( o1, o1, r1, c5 );  stbir__simdfX_madd( o2, o2, r2, c5 );   stbir__simdfX_madd( o3, o3, r3, c5 );             
+              stbir__simdfX_store( output5, o0 );    stbir__simdfX_store( output5+stbir__simdfX_float_count, o1 );  stbir__simdfX_store( output5+(2*stbir__simdfX_float_count), o2 );   stbir__simdfX_store( output5+(3*stbir__simdfX_float_count), o3 ); )
+      stbIF6( stbir__simdfX_load( o0, output6 );     stbir__simdfX_load( o1, output6+stbir__simdfX_float_count );   stbir__simdfX_load( o2, output6+(2*stbir__simdfX_float_count) );    stbir__simdfX_load( o3, output6+(3*stbir__simdfX_float_count) );
+              stbir__simdfX_madd( o0, o0, r0, c6 );  stbir__simdfX_madd( o1, o1, r1, c6 );  stbir__simdfX_madd( o2, o2, r2, c6 );   stbir__simdfX_madd( o3, o3, r3, c6 );             
+              stbir__simdfX_store( output6, o0 );    stbir__simdfX_store( output6+stbir__simdfX_float_count, o1 );  stbir__simdfX_store( output6+(2*stbir__simdfX_float_count), o2 );   stbir__simdfX_store( output6+(3*stbir__simdfX_float_count), o3 ); )
+      stbIF7( stbir__simdfX_load( o0, output7 );     stbir__simdfX_load( o1, output7+stbir__simdfX_float_count );   stbir__simdfX_load( o2, output7+(2*stbir__simdfX_float_count) );    stbir__simdfX_load( o3, output7+(3*stbir__simdfX_float_count) );
+              stbir__simdfX_madd( o0, o0, r0, c7 );  stbir__simdfX_madd( o1, o1, r1, c7 );  stbir__simdfX_madd( o2, o2, r2, c7 );   stbir__simdfX_madd( o3, o3, r3, c7 );             
+              stbir__simdfX_store( output7, o0 );    stbir__simdfX_store( output7+stbir__simdfX_float_count, o1 );  stbir__simdfX_store( output7+(2*stbir__simdfX_float_count), o2 );   stbir__simdfX_store( output7+(3*stbir__simdfX_float_count), o3 ); )
+      #else
+      stbIF0( stbir__simdfX_mult( o0, r0, c0 );      stbir__simdfX_mult( o1, r1, c0 );      stbir__simdfX_mult( o2, r2, c0 );       stbir__simdfX_mult( o3, r3, c0 );  
+              stbir__simdfX_store( output0, o0 );    stbir__simdfX_store( output0+stbir__simdfX_float_count, o1 );  stbir__simdfX_store( output0+(2*stbir__simdfX_float_count), o2 );   stbir__simdfX_store( output0+(3*stbir__simdfX_float_count), o3 ); )
+      stbIF1( stbir__simdfX_mult( o0, r0, c1 );      stbir__simdfX_mult( o1, r1, c1 );      stbir__simdfX_mult( o2, r2, c1 );       stbir__simdfX_mult( o3, r3, c1 );  
+              stbir__simdfX_store( output1, o0 );    stbir__simdfX_store( output1+stbir__simdfX_float_count, o1 );  stbir__simdfX_store( output1+(2*stbir__simdfX_float_count), o2 );   stbir__simdfX_store( output1+(3*stbir__simdfX_float_count), o3 ); )
+      stbIF2( stbir__simdfX_mult( o0, r0, c2 );      stbir__simdfX_mult( o1, r1, c2 );      stbir__simdfX_mult( o2, r2, c2 );       stbir__simdfX_mult( o3, r3, c2 );  
+              stbir__simdfX_store( output2, o0 );    stbir__simdfX_store( output2+stbir__simdfX_float_count, o1 );  stbir__simdfX_store( output2+(2*stbir__simdfX_float_count), o2 );   stbir__simdfX_store( output2+(3*stbir__simdfX_float_count), o3 ); )
+      stbIF3( stbir__simdfX_mult( o0, r0, c3 );      stbir__simdfX_mult( o1, r1, c3 );      stbir__simdfX_mult( o2, r2, c3 );       stbir__simdfX_mult( o3, r3, c3 );  
+              stbir__simdfX_store( output3, o0 );    stbir__simdfX_store( output3+stbir__simdfX_float_count, o1 );  stbir__simdfX_store( output3+(2*stbir__simdfX_float_count), o2 );   stbir__simdfX_store( output3+(3*stbir__simdfX_float_count), o3 ); )
+      stbIF4( stbir__simdfX_mult( o0, r0, c4 );      stbir__simdfX_mult( o1, r1, c4 );      stbir__simdfX_mult( o2, r2, c4 );       stbir__simdfX_mult( o3, r3, c4 );  
+              stbir__simdfX_store( output4, o0 );    stbir__simdfX_store( output4+stbir__simdfX_float_count, o1 );  stbir__simdfX_store( output4+(2*stbir__simdfX_float_count), o2 );   stbir__simdfX_store( output4+(3*stbir__simdfX_float_count), o3 ); )
+      stbIF5( stbir__simdfX_mult( o0, r0, c5 );      stbir__simdfX_mult( o1, r1, c5 );      stbir__simdfX_mult( o2, r2, c5 );       stbir__simdfX_mult( o3, r3, c5 );  
+              stbir__simdfX_store( output5, o0 );    stbir__simdfX_store( output5+stbir__simdfX_float_count, o1 );  stbir__simdfX_store( output5+(2*stbir__simdfX_float_count), o2 );   stbir__simdfX_store( output5+(3*stbir__simdfX_float_count), o3 ); )
+      stbIF6( stbir__simdfX_mult( o0, r0, c6 );      stbir__simdfX_mult( o1, r1, c6 );      stbir__simdfX_mult( o2, r2, c6 );       stbir__simdfX_mult( o3, r3, c6 );  
+              stbir__simdfX_store( output6, o0 );    stbir__simdfX_store( output6+stbir__simdfX_float_count, o1 );  stbir__simdfX_store( output6+(2*stbir__simdfX_float_count), o2 );   stbir__simdfX_store( output6+(3*stbir__simdfX_float_count), o3 ); )
+      stbIF7( stbir__simdfX_mult( o0, r0, c7 );      stbir__simdfX_mult( o1, r1, c7 );      stbir__simdfX_mult( o2, r2, c7 );       stbir__simdfX_mult( o3, r3, c7 );  
+              stbir__simdfX_store( output7, o0 );    stbir__simdfX_store( output7+stbir__simdfX_float_count, o1 );  stbir__simdfX_store( output7+(2*stbir__simdfX_float_count), o2 );   stbir__simdfX_store( output7+(3*stbir__simdfX_float_count), o3 ); )
+      #endif
+
+      input += (4*stbir__simdfX_float_count);
+      stbIF0( output0 += (4*stbir__simdfX_float_count); ) stbIF1( output1 += (4*stbir__simdfX_float_count); ) stbIF2( output2 += (4*stbir__simdfX_float_count); ) stbIF3( output3 += (4*stbir__simdfX_float_count); ) stbIF4( output4 += (4*stbir__simdfX_float_count); ) stbIF5( output5 += (4*stbir__simdfX_float_count); ) stbIF6( output6 += (4*stbir__simdfX_float_count); ) stbIF7( output7 += (4*stbir__simdfX_float_count); )
+    }
+    while ( ( (char*)input_end - (char*) input ) >= 16 ) 
+    {
+      stbir__simdf o0, r0;
+      STBIR_SIMD_NO_UNROLL(output0);
+
+      stbir__simdf_load( r0, input );
+
+      #ifdef STB_IMAGE_RESIZE_VERTICAL_CONTINUE
+      stbIF0( stbir__simdf_load( o0, output0 );  stbir__simdf_madd( o0, o0, r0, stbir__if_simdf8_cast_to_simdf4( c0 ) );  stbir__simdf_store( output0, o0 ); )
+      stbIF1( stbir__simdf_load( o0, output1 );  stbir__simdf_madd( o0, o0, r0, stbir__if_simdf8_cast_to_simdf4( c1 ) );  stbir__simdf_store( output1, o0 ); )
+      stbIF2( stbir__simdf_load( o0, output2 );  stbir__simdf_madd( o0, o0, r0, stbir__if_simdf8_cast_to_simdf4( c2 ) );  stbir__simdf_store( output2, o0 ); )
+      stbIF3( stbir__simdf_load( o0, output3 );  stbir__simdf_madd( o0, o0, r0, stbir__if_simdf8_cast_to_simdf4( c3 ) );  stbir__simdf_store( output3, o0 ); )
+      stbIF4( stbir__simdf_load( o0, output4 );  stbir__simdf_madd( o0, o0, r0, stbir__if_simdf8_cast_to_simdf4( c4 ) );  stbir__simdf_store( output4, o0 ); )
+      stbIF5( stbir__simdf_load( o0, output5 );  stbir__simdf_madd( o0, o0, r0, stbir__if_simdf8_cast_to_simdf4( c5 ) );  stbir__simdf_store( output5, o0 ); )
+      stbIF6( stbir__simdf_load( o0, output6 );  stbir__simdf_madd( o0, o0, r0, stbir__if_simdf8_cast_to_simdf4( c6 ) );  stbir__simdf_store( output6, o0 ); )
+      stbIF7( stbir__simdf_load( o0, output7 );  stbir__simdf_madd( o0, o0, r0, stbir__if_simdf8_cast_to_simdf4( c7 ) );  stbir__simdf_store( output7, o0 ); )
+      #else
+      stbIF0( stbir__simdf_mult( o0, r0, stbir__if_simdf8_cast_to_simdf4( c0 ) );   stbir__simdf_store( output0, o0 ); )
+      stbIF1( stbir__simdf_mult( o0, r0, stbir__if_simdf8_cast_to_simdf4( c1 ) );   stbir__simdf_store( output1, o0 ); )
+      stbIF2( stbir__simdf_mult( o0, r0, stbir__if_simdf8_cast_to_simdf4( c2 ) );   stbir__simdf_store( output2, o0 ); )
+      stbIF3( stbir__simdf_mult( o0, r0, stbir__if_simdf8_cast_to_simdf4( c3 ) );   stbir__simdf_store( output3, o0 ); )
+      stbIF4( stbir__simdf_mult( o0, r0, stbir__if_simdf8_cast_to_simdf4( c4 ) );   stbir__simdf_store( output4, o0 ); )
+      stbIF5( stbir__simdf_mult( o0, r0, stbir__if_simdf8_cast_to_simdf4( c5 ) );   stbir__simdf_store( output5, o0 ); )
+      stbIF6( stbir__simdf_mult( o0, r0, stbir__if_simdf8_cast_to_simdf4( c6 ) );   stbir__simdf_store( output6, o0 ); )
+      stbIF7( stbir__simdf_mult( o0, r0, stbir__if_simdf8_cast_to_simdf4( c7 ) );   stbir__simdf_store( output7, o0 ); )
+      #endif
+      
+      input += 4;
+      stbIF0( output0 += 4; ) stbIF1( output1 += 4; ) stbIF2( output2 += 4; ) stbIF3( output3 += 4; ) stbIF4( output4 += 4; ) stbIF5( output5 += 4; ) stbIF6( output6 += 4; ) stbIF7( output7 += 4; )
+    }
+  }
+  #else
+  while ( ( (char*)input_end - (char*) input ) >= 16 ) 
+  {
+    float r0, r1, r2, r3;
+    STBIR_NO_UNROLL(input);
+
+    r0 = input[0], r1 = input[1], r2 = input[2], r3 = input[3];
+
+    #ifdef STB_IMAGE_RESIZE_VERTICAL_CONTINUE
+    stbIF0( output0[0] += ( r0 * c0s ); output0[1] += ( r1 * c0s ); output0[2] += ( r2 * c0s ); output0[3] += ( r3 * c0s ); )
+    stbIF1( output1[0] += ( r0 * c1s ); output1[1] += ( r1 * c1s ); output1[2] += ( r2 * c1s ); output1[3] += ( r3 * c1s ); )
+    stbIF2( output2[0] += ( r0 * c2s ); output2[1] += ( r1 * c2s ); output2[2] += ( r2 * c2s ); output2[3] += ( r3 * c2s ); )
+    stbIF3( output3[0] += ( r0 * c3s ); output3[1] += ( r1 * c3s ); output3[2] += ( r2 * c3s ); output3[3] += ( r3 * c3s ); )
+    stbIF4( output4[0] += ( r0 * c4s ); output4[1] += ( r1 * c4s ); output4[2] += ( r2 * c4s ); output4[3] += ( r3 * c4s ); )
+    stbIF5( output5[0] += ( r0 * c5s ); output5[1] += ( r1 * c5s ); output5[2] += ( r2 * c5s ); output5[3] += ( r3 * c5s ); )
+    stbIF6( output6[0] += ( r0 * c6s ); output6[1] += ( r1 * c6s ); output6[2] += ( r2 * c6s ); output6[3] += ( r3 * c6s ); )
+    stbIF7( output7[0] += ( r0 * c7s ); output7[1] += ( r1 * c7s ); output7[2] += ( r2 * c7s ); output7[3] += ( r3 * c7s ); )
+    #else
+    stbIF0( output0[0]  = ( r0 * c0s ); output0[1]  = ( r1 * c0s ); output0[2]  = ( r2 * c0s ); output0[3]  = ( r3 * c0s ); )
+    stbIF1( output1[0]  = ( r0 * c1s ); output1[1]  = ( r1 * c1s ); output1[2]  = ( r2 * c1s ); output1[3]  = ( r3 * c1s ); )
+    stbIF2( output2[0]  = ( r0 * c2s ); output2[1]  = ( r1 * c2s ); output2[2]  = ( r2 * c2s ); output2[3]  = ( r3 * c2s ); )
+    stbIF3( output3[0]  = ( r0 * c3s ); output3[1]  = ( r1 * c3s ); output3[2]  = ( r2 * c3s ); output3[3]  = ( r3 * c3s ); )
+    stbIF4( output4[0]  = ( r0 * c4s ); output4[1]  = ( r1 * c4s ); output4[2]  = ( r2 * c4s ); output4[3]  = ( r3 * c4s ); )
+    stbIF5( output5[0]  = ( r0 * c5s ); output5[1]  = ( r1 * c5s ); output5[2]  = ( r2 * c5s ); output5[3]  = ( r3 * c5s ); )
+    stbIF6( output6[0]  = ( r0 * c6s ); output6[1]  = ( r1 * c6s ); output6[2]  = ( r2 * c6s ); output6[3]  = ( r3 * c6s ); )
+    stbIF7( output7[0]  = ( r0 * c7s ); output7[1]  = ( r1 * c7s ); output7[2]  = ( r2 * c7s ); output7[3]  = ( r3 * c7s ); )
+    #endif
+
+    input += 4;
+    stbIF0( output0 += 4; ) stbIF1( output1 += 4; ) stbIF2( output2 += 4; ) stbIF3( output3 += 4; ) stbIF4( output4 += 4; ) stbIF5( output5 += 4; ) stbIF6( output6 += 4; ) stbIF7( output7 += 4; )
+  }
+  #endif
+  while ( input < input_end ) 
+  {
+    float r = input[0];
+    STBIR_NO_UNROLL(output0);
+
+    #ifdef STB_IMAGE_RESIZE_VERTICAL_CONTINUE
+    stbIF0( output0[0] += ( r * c0s ); )
+    stbIF1( output1[0] += ( r * c1s ); )
+    stbIF2( output2[0] += ( r * c2s ); )
+    stbIF3( output3[0] += ( r * c3s ); )
+    stbIF4( output4[0] += ( r * c4s ); )
+    stbIF5( output5[0] += ( r * c5s ); )
+    stbIF6( output6[0] += ( r * c6s ); )
+    stbIF7( output7[0] += ( r * c7s ); )
+    #else
+    stbIF0( output0[0]  = ( r * c0s ); )
+    stbIF1( output1[0]  = ( r * c1s ); )
+    stbIF2( output2[0]  = ( r * c2s ); )
+    stbIF3( output3[0]  = ( r * c3s ); )
+    stbIF4( output4[0]  = ( r * c4s ); )
+    stbIF5( output5[0]  = ( r * c5s ); )
+    stbIF6( output6[0]  = ( r * c6s ); )
+    stbIF7( output7[0]  = ( r * c7s ); )
+    #endif
+
+    ++input;
+    stbIF0( ++output0; ) stbIF1( ++output1; ) stbIF2( ++output2; ) stbIF3( ++output3; ) stbIF4( ++output4; ) stbIF5( ++output5; ) stbIF6( ++output6; ) stbIF7( ++output7; )
+  }
+}
+
+static void STBIR_chans( stbir__vertical_gather_with_,_coeffs)( float * outputp, float const * vertical_coefficients, float const ** inputs, float const * input0_end )
+{
+  float STBIR_SIMD_STREAMOUT_PTR( * ) output = outputp;
+
+  stbIF0( float const * input0 = inputs[0]; float c0s = vertical_coefficients[0]; )
+  stbIF1( float const * input1 = inputs[1]; float c1s = vertical_coefficients[1]; )
+  stbIF2( float const * input2 = inputs[2]; float c2s = vertical_coefficients[2]; )
+  stbIF3( float const * input3 = inputs[3]; float c3s = vertical_coefficients[3]; )
+  stbIF4( float const * input4 = inputs[4]; float c4s = vertical_coefficients[4]; )
+  stbIF5( float const * input5 = inputs[5]; float c5s = vertical_coefficients[5]; )
+  stbIF6( float const * input6 = inputs[6]; float c6s = vertical_coefficients[6]; )
+  stbIF7( float const * input7 = inputs[7]; float c7s = vertical_coefficients[7]; )
+
+#if ( STBIR__vertical_channels == 1 ) && !defined(STB_IMAGE_RESIZE_VERTICAL_CONTINUE)
+  // check single channel one weight
+  if ( ( c0s >= (1.0f-0.000001f) ) && ( c0s <= (1.0f+0.000001f) ) )
+  {
+    STBIR_MEMCPY( output, input0, (char*)input0_end - (char*)input0 );
+    return;
+  }
+#endif  
+
+  #ifdef STBIR_SIMD
+  {
+    stbIF0(stbir__simdfX c0 = stbir__simdf_frepX( c0s ); )
+    stbIF1(stbir__simdfX c1 = stbir__simdf_frepX( c1s ); )
+    stbIF2(stbir__simdfX c2 = stbir__simdf_frepX( c2s ); )
+    stbIF3(stbir__simdfX c3 = stbir__simdf_frepX( c3s ); )
+    stbIF4(stbir__simdfX c4 = stbir__simdf_frepX( c4s ); )
+    stbIF5(stbir__simdfX c5 = stbir__simdf_frepX( c5s ); )
+    stbIF6(stbir__simdfX c6 = stbir__simdf_frepX( c6s ); )
+    stbIF7(stbir__simdfX c7 = stbir__simdf_frepX( c7s ); )
+    
+    while ( ( (char*)input0_end - (char*) input0 ) >= (16*stbir__simdfX_float_count) ) 
+    {
+      stbir__simdfX o0, o1, o2, o3, r0, r1, r2, r3;
+      STBIR_SIMD_NO_UNROLL(output);
+
+      // prefetch four loop iterations ahead (doesn't affect much for small resizes, but helps with big ones)
+      stbIF0( stbir__prefetch( input0 + (16*stbir__simdfX_float_count) ); ) 
+      stbIF1( stbir__prefetch( input1 + (16*stbir__simdfX_float_count) ); )
+      stbIF2( stbir__prefetch( input2 + (16*stbir__simdfX_float_count) ); )
+      stbIF3( stbir__prefetch( input3 + (16*stbir__simdfX_float_count) ); )
+      stbIF4( stbir__prefetch( input4 + (16*stbir__simdfX_float_count) ); )
+      stbIF5( stbir__prefetch( input5 + (16*stbir__simdfX_float_count) ); )
+      stbIF6( stbir__prefetch( input6 + (16*stbir__simdfX_float_count) ); )
+      stbIF7( stbir__prefetch( input7 + (16*stbir__simdfX_float_count) ); )
+
+      #ifdef STB_IMAGE_RESIZE_VERTICAL_CONTINUE
+      stbIF0( stbir__simdfX_load( o0, output );      stbir__simdfX_load( o1, output+stbir__simdfX_float_count );   stbir__simdfX_load( o2, output+(2*stbir__simdfX_float_count) );   stbir__simdfX_load( o3, output+(3*stbir__simdfX_float_count) );
+              stbir__simdfX_load( r0, input0 );      stbir__simdfX_load( r1, input0+stbir__simdfX_float_count );   stbir__simdfX_load( r2, input0+(2*stbir__simdfX_float_count) );   stbir__simdfX_load( r3, input0+(3*stbir__simdfX_float_count) );
+              stbir__simdfX_madd( o0, o0, r0, c0 );  stbir__simdfX_madd( o1, o1, r1, c0 );                         stbir__simdfX_madd( o2, o2, r2, c0 );                             stbir__simdfX_madd( o3, o3, r3, c0 ); )
+      #else
+      stbIF0( stbir__simdfX_load( r0, input0 );      stbir__simdfX_load( r1, input0+stbir__simdfX_float_count );   stbir__simdfX_load( r2, input0+(2*stbir__simdfX_float_count) );   stbir__simdfX_load( r3, input0+(3*stbir__simdfX_float_count) );
+              stbir__simdfX_mult( o0, r0, c0 );      stbir__simdfX_mult( o1, r1, c0 );                             stbir__simdfX_mult( o2, r2, c0 );                                 stbir__simdfX_mult( o3, r3, c0 );  )
+      #endif
+
+      stbIF1( stbir__simdfX_load( r0, input1 );      stbir__simdfX_load( r1, input1+stbir__simdfX_float_count );   stbir__simdfX_load( r2, input1+(2*stbir__simdfX_float_count) );   stbir__simdfX_load( r3, input1+(3*stbir__simdfX_float_count) );
+              stbir__simdfX_madd( o0, o0, r0, c1 );  stbir__simdfX_madd( o1, o1, r1, c1 );                         stbir__simdfX_madd( o2, o2, r2, c1 );                             stbir__simdfX_madd( o3, o3, r3, c1 ); )
+      stbIF2( stbir__simdfX_load( r0, input2 );      stbir__simdfX_load( r1, input2+stbir__simdfX_float_count );   stbir__simdfX_load( r2, input2+(2*stbir__simdfX_float_count) );   stbir__simdfX_load( r3, input2+(3*stbir__simdfX_float_count) );
+              stbir__simdfX_madd( o0, o0, r0, c2 );  stbir__simdfX_madd( o1, o1, r1, c2 );                         stbir__simdfX_madd( o2, o2, r2, c2 );                             stbir__simdfX_madd( o3, o3, r3, c2 ); )
+      stbIF3( stbir__simdfX_load( r0, input3 );      stbir__simdfX_load( r1, input3+stbir__simdfX_float_count );   stbir__simdfX_load( r2, input3+(2*stbir__simdfX_float_count) );   stbir__simdfX_load( r3, input3+(3*stbir__simdfX_float_count) );
+              stbir__simdfX_madd( o0, o0, r0, c3 );  stbir__simdfX_madd( o1, o1, r1, c3 );                         stbir__simdfX_madd( o2, o2, r2, c3 );                             stbir__simdfX_madd( o3, o3, r3, c3 ); )
+      stbIF4( stbir__simdfX_load( r0, input4 );      stbir__simdfX_load( r1, input4+stbir__simdfX_float_count );   stbir__simdfX_load( r2, input4+(2*stbir__simdfX_float_count) );   stbir__simdfX_load( r3, input4+(3*stbir__simdfX_float_count) );
+              stbir__simdfX_madd( o0, o0, r0, c4 );  stbir__simdfX_madd( o1, o1, r1, c4 );                         stbir__simdfX_madd( o2, o2, r2, c4 );                             stbir__simdfX_madd( o3, o3, r3, c4 ); )
+      stbIF5( stbir__simdfX_load( r0, input5 );      stbir__simdfX_load( r1, input5+stbir__simdfX_float_count );   stbir__simdfX_load( r2, input5+(2*stbir__simdfX_float_count) );   stbir__simdfX_load( r3, input5+(3*stbir__simdfX_float_count) );
+              stbir__simdfX_madd( o0, o0, r0, c5 );  stbir__simdfX_madd( o1, o1, r1, c5 );                         stbir__simdfX_madd( o2, o2, r2, c5 );                             stbir__simdfX_madd( o3, o3, r3, c5 ); )
+      stbIF6( stbir__simdfX_load( r0, input6 );      stbir__simdfX_load( r1, input6+stbir__simdfX_float_count );   stbir__simdfX_load( r2, input6+(2*stbir__simdfX_float_count) );   stbir__simdfX_load( r3, input6+(3*stbir__simdfX_float_count) );
+              stbir__simdfX_madd( o0, o0, r0, c6 );  stbir__simdfX_madd( o1, o1, r1, c6 );                         stbir__simdfX_madd( o2, o2, r2, c6 );                             stbir__simdfX_madd( o3, o3, r3, c6 ); )
+      stbIF7( stbir__simdfX_load( r0, input7 );      stbir__simdfX_load( r1, input7+stbir__simdfX_float_count );   stbir__simdfX_load( r2, input7+(2*stbir__simdfX_float_count) );   stbir__simdfX_load( r3, input7+(3*stbir__simdfX_float_count) );
+              stbir__simdfX_madd( o0, o0, r0, c7 );  stbir__simdfX_madd( o1, o1, r1, c7 );                         stbir__simdfX_madd( o2, o2, r2, c7 );                             stbir__simdfX_madd( o3, o3, r3, c7 ); )
+
+      stbir__simdfX_store( output, o0 );             stbir__simdfX_store( output+stbir__simdfX_float_count, o1 );  stbir__simdfX_store( output+(2*stbir__simdfX_float_count), o2 );  stbir__simdfX_store( output+(3*stbir__simdfX_float_count), o3 );
+      output += (4*stbir__simdfX_float_count);
+      stbIF0( input0 += (4*stbir__simdfX_float_count); ) stbIF1( input1 += (4*stbir__simdfX_float_count); ) stbIF2( input2 += (4*stbir__simdfX_float_count); ) stbIF3( input3 += (4*stbir__simdfX_float_count); ) stbIF4( input4 += (4*stbir__simdfX_float_count); ) stbIF5( input5 += (4*stbir__simdfX_float_count); ) stbIF6( input6 += (4*stbir__simdfX_float_count); ) stbIF7( input7 += (4*stbir__simdfX_float_count); )
+    }
+
+    while ( ( (char*)input0_end - (char*) input0 ) >= 16 ) 
+    {
+      stbir__simdf o0, r0;
+      STBIR_SIMD_NO_UNROLL(output);
+
+      #ifdef STB_IMAGE_RESIZE_VERTICAL_CONTINUE
+      stbIF0( stbir__simdf_load( o0, output );   stbir__simdf_load( r0, input0 ); stbir__simdf_madd( o0, o0, r0, stbir__if_simdf8_cast_to_simdf4( c0 ) ); )
+      #else
+      stbIF0( stbir__simdf_load( r0, input0 );  stbir__simdf_mult( o0, r0, stbir__if_simdf8_cast_to_simdf4( c0 ) ); )
+      #endif
+      stbIF1( stbir__simdf_load( r0, input1 );  stbir__simdf_madd( o0, o0, r0, stbir__if_simdf8_cast_to_simdf4( c1 ) ); )
+      stbIF2( stbir__simdf_load( r0, input2 );  stbir__simdf_madd( o0, o0, r0, stbir__if_simdf8_cast_to_simdf4( c2 ) ); )
+      stbIF3( stbir__simdf_load( r0, input3 );  stbir__simdf_madd( o0, o0, r0, stbir__if_simdf8_cast_to_simdf4( c3 ) ); )
+      stbIF4( stbir__simdf_load( r0, input4 );  stbir__simdf_madd( o0, o0, r0, stbir__if_simdf8_cast_to_simdf4( c4 ) ); )
+      stbIF5( stbir__simdf_load( r0, input5 );  stbir__simdf_madd( o0, o0, r0, stbir__if_simdf8_cast_to_simdf4( c5 ) ); )
+      stbIF6( stbir__simdf_load( r0, input6 );  stbir__simdf_madd( o0, o0, r0, stbir__if_simdf8_cast_to_simdf4( c6 ) ); )
+      stbIF7( stbir__simdf_load( r0, input7 );  stbir__simdf_madd( o0, o0, r0, stbir__if_simdf8_cast_to_simdf4( c7 ) ); )
+
+      stbir__simdf_store( output, o0 );
+      output += 4;
+      stbIF0( input0 += 4; ) stbIF1( input1 += 4; ) stbIF2( input2 += 4; ) stbIF3( input3 += 4; ) stbIF4( input4 += 4; ) stbIF5( input5 += 4; ) stbIF6( input6 += 4; ) stbIF7( input7 += 4; )
+    }
+  }
+  #else
+  while ( ( (char*)input0_end - (char*) input0 ) >= 16 ) 
+  {
+    float o0, o1, o2, o3;
+    STBIR_NO_UNROLL(output);
+    #ifdef STB_IMAGE_RESIZE_VERTICAL_CONTINUE
+    stbIF0( o0 = output[0] + input0[0] * c0s; o1 = output[1] + input0[1] * c0s; o2 = output[2] + input0[2] * c0s; o3 = output[3] + input0[3] * c0s; )
+    #else
+    stbIF0( o0  = input0[0] * c0s; o1  = input0[1] * c0s; o2  = input0[2] * c0s; o3  = input0[3] * c0s; )
+    #endif
+    stbIF1( o0 += input1[0] * c1s; o1 += input1[1] * c1s; o2 += input1[2] * c1s; o3 += input1[3] * c1s; )
+    stbIF2( o0 += input2[0] * c2s; o1 += input2[1] * c2s; o2 += input2[2] * c2s; o3 += input2[3] * c2s; )
+    stbIF3( o0 += input3[0] * c3s; o1 += input3[1] * c3s; o2 += input3[2] * c3s; o3 += input3[3] * c3s; )
+    stbIF4( o0 += input4[0] * c4s; o1 += input4[1] * c4s; o2 += input4[2] * c4s; o3 += input4[3] * c4s; )
+    stbIF5( o0 += input5[0] * c5s; o1 += input5[1] * c5s; o2 += input5[2] * c5s; o3 += input5[3] * c5s; )
+    stbIF6( o0 += input6[0] * c6s; o1 += input6[1] * c6s; o2 += input6[2] * c6s; o3 += input6[3] * c6s; )
+    stbIF7( o0 += input7[0] * c7s; o1 += input7[1] * c7s; o2 += input7[2] * c7s; o3 += input7[3] * c7s; )
+    output[0] = o0; output[1] = o1; output[2] = o2; output[3] = o3;
+    output += 4;
+    stbIF0( input0 += 4; ) stbIF1( input1 += 4; ) stbIF2( input2 += 4; ) stbIF3( input3 += 4; ) stbIF4( input4 += 4; ) stbIF5( input5 += 4; ) stbIF6( input6 += 4; ) stbIF7( input7 += 4; )
+  }
+  #endif
+  while ( input0 < input0_end ) 
+  {
+    float o0;
+    STBIR_NO_UNROLL(output);
+    #ifdef STB_IMAGE_RESIZE_VERTICAL_CONTINUE
+    stbIF0( o0 = output[0] + input0[0] * c0s; )
+    #else
+    stbIF0( o0  = input0[0] * c0s; )
+    #endif
+    stbIF1( o0 += input1[0] * c1s; )
+    stbIF2( o0 += input2[0] * c2s; )
+    stbIF3( o0 += input3[0] * c3s; )
+    stbIF4( o0 += input4[0] * c4s; )
+    stbIF5( o0 += input5[0] * c5s; )
+    stbIF6( o0 += input6[0] * c6s; )
+    stbIF7( o0 += input7[0] * c7s; )
+    output[0] = o0; 
+    ++output;
+    stbIF0( ++input0; ) stbIF1( ++input1; ) stbIF2( ++input2; ) stbIF3( ++input3; ) stbIF4( ++input4; ) stbIF5( ++input5; ) stbIF6( ++input6; ) stbIF7( ++input7; )
+  }
+}
+
+#undef stbIF0
+#undef stbIF1
+#undef stbIF2
+#undef stbIF3
+#undef stbIF4
+#undef stbIF5
+#undef stbIF6
+#undef stbIF7
+#undef STB_IMAGE_RESIZE_DO_VERTICALS
+#undef STBIR__vertical_channels
+#undef STB_IMAGE_RESIZE_DO_HORIZONTALS
+#undef STBIR_strs_join24
+#undef STBIR_strs_join14
+#undef STBIR_chans
+#ifdef STB_IMAGE_RESIZE_VERTICAL_CONTINUE
+#undef STB_IMAGE_RESIZE_VERTICAL_CONTINUE
+#endif
+
+#else // !STB_IMAGE_RESIZE_DO_VERTICALS
+
+#define STBIR_chans( start, end ) STBIR_strs_join1(start,STBIR__horizontal_channels,end)
+
+#ifndef stbir__2_coeff_only
+#define stbir__2_coeff_only()             \
+    stbir__1_coeff_only();                \
+    stbir__1_coeff_remnant(1);            
+#endif
+
+#ifndef stbir__2_coeff_remnant
+#define stbir__2_coeff_remnant( ofs )     \
+    stbir__1_coeff_remnant(ofs);          \
+    stbir__1_coeff_remnant((ofs)+1);      
+#endif
+    
+#ifndef stbir__3_coeff_only
+#define stbir__3_coeff_only()             \
+    stbir__2_coeff_only();                \
+    stbir__1_coeff_remnant(2);            
+#endif
+    
+#ifndef stbir__3_coeff_remnant
+#define stbir__3_coeff_remnant( ofs )     \
+    stbir__2_coeff_remnant(ofs);          \
+    stbir__1_coeff_remnant((ofs)+2);      
+#endif
+
+#ifndef stbir__3_coeff_setup
+#define stbir__3_coeff_setup()
+#endif
+
+#ifndef stbir__4_coeff_start
+#define stbir__4_coeff_start()            \
+    stbir__2_coeff_only();                \
+    stbir__2_coeff_remnant(2);            
+#endif
+    
+#ifndef stbir__4_coeff_continue_from_4
+#define stbir__4_coeff_continue_from_4( ofs )     \
+    stbir__2_coeff_remnant(ofs);                  \
+    stbir__2_coeff_remnant((ofs)+2);      
+#endif
+
+#ifndef stbir__store_output_tiny
+#define stbir__store_output_tiny stbir__store_output
+#endif
+
+static void STBIR_chans( stbir__horizontal_gather_,_channels_with_1_coeff)( float * output_buffer, unsigned int output_sub_size, float const * decode_buffer, stbir__contributors const * horizontal_contributors, float const * horizontal_coefficients, int coefficient_width )
+{
+  float const * output_end = output_buffer + output_sub_size * STBIR__horizontal_channels;
+  float STBIR_SIMD_STREAMOUT_PTR( * ) output = output_buffer;
+  do {
+    float const * decode = decode_buffer + horizontal_contributors->n0 * STBIR__horizontal_channels; 
+    float const * hc = horizontal_coefficients;
+    stbir__1_coeff_only();
+    stbir__store_output_tiny();
+  } while ( output < output_end );
+}
+
+static void STBIR_chans( stbir__horizontal_gather_,_channels_with_2_coeffs)( float * output_buffer, unsigned int output_sub_size, float const * decode_buffer, stbir__contributors const * horizontal_contributors, float const * horizontal_coefficients, int coefficient_width )
+{
+  float const * output_end = output_buffer + output_sub_size * STBIR__horizontal_channels;
+  float STBIR_SIMD_STREAMOUT_PTR( * ) output = output_buffer;
+  do {
+    float const * decode = decode_buffer + horizontal_contributors->n0 * STBIR__horizontal_channels; 
+    float const * hc = horizontal_coefficients;
+    stbir__2_coeff_only();
+    stbir__store_output_tiny();
+  } while ( output < output_end );
+}
+
+static void STBIR_chans( stbir__horizontal_gather_,_channels_with_3_coeffs)( float * output_buffer, unsigned int output_sub_size, float const * decode_buffer, stbir__contributors const * horizontal_contributors, float const * horizontal_coefficients, int coefficient_width )
+{
+  float const * output_end = output_buffer + output_sub_size * STBIR__horizontal_channels;
+  float STBIR_SIMD_STREAMOUT_PTR( * ) output = output_buffer;
+  do {
+    float const * decode = decode_buffer + horizontal_contributors->n0 * STBIR__horizontal_channels; 
+    float const * hc = horizontal_coefficients;
+    stbir__3_coeff_only();
+    stbir__store_output_tiny();
+  } while ( output < output_end );
+}
+
+static void STBIR_chans( stbir__horizontal_gather_,_channels_with_4_coeffs)( float * output_buffer, unsigned int output_sub_size, float const * decode_buffer, stbir__contributors const * horizontal_contributors, float const * horizontal_coefficients, int coefficient_width )
+{
+  float const * output_end = output_buffer + output_sub_size * STBIR__horizontal_channels;
+  float STBIR_SIMD_STREAMOUT_PTR( * ) output = output_buffer;
+  do {
+    float const * decode = decode_buffer + horizontal_contributors->n0 * STBIR__horizontal_channels; 
+    float const * hc = horizontal_coefficients;
+    stbir__4_coeff_start();
+    stbir__store_output();
+  } while ( output < output_end );
+}
+
+static void STBIR_chans( stbir__horizontal_gather_,_channels_with_5_coeffs)( float * output_buffer, unsigned int output_sub_size, float const * decode_buffer, stbir__contributors const * horizontal_contributors, float const * horizontal_coefficients, int coefficient_width )
+{
+  float const * output_end = output_buffer + output_sub_size * STBIR__horizontal_channels;
+  float STBIR_SIMD_STREAMOUT_PTR( * ) output = output_buffer;
+  do {
+    float const * decode = decode_buffer + horizontal_contributors->n0 * STBIR__horizontal_channels; 
+    float const * hc = horizontal_coefficients;
+    stbir__4_coeff_start();
+    stbir__1_coeff_remnant(4);
+    stbir__store_output();
+  } while ( output < output_end );
+}
+
+static void STBIR_chans( stbir__horizontal_gather_,_channels_with_6_coeffs)( float * output_buffer, unsigned int output_sub_size, float const * decode_buffer, stbir__contributors const * horizontal_contributors, float const * horizontal_coefficients, int coefficient_width )
+{
+  float const * output_end = output_buffer + output_sub_size * STBIR__horizontal_channels;
+  float STBIR_SIMD_STREAMOUT_PTR( * ) output = output_buffer;
+  do {
+    float const * decode = decode_buffer + horizontal_contributors->n0 * STBIR__horizontal_channels; 
+    float const * hc = horizontal_coefficients;
+    stbir__4_coeff_start();
+    stbir__2_coeff_remnant(4);
+    stbir__store_output();
+  } while ( output < output_end );
+}
+
+static void STBIR_chans( stbir__horizontal_gather_,_channels_with_7_coeffs)( float * output_buffer, unsigned int output_sub_size, float const * decode_buffer, stbir__contributors const * horizontal_contributors, float const * horizontal_coefficients, int coefficient_width )
+{
+  float const * output_end = output_buffer + output_sub_size * STBIR__horizontal_channels;
+  float STBIR_SIMD_STREAMOUT_PTR( * ) output = output_buffer;
+  stbir__3_coeff_setup();
+  do {
+    float const * decode = decode_buffer + horizontal_contributors->n0 * STBIR__horizontal_channels; 
+    float const * hc = horizontal_coefficients;
+  
+    stbir__4_coeff_start();
+    stbir__3_coeff_remnant(4);
+    stbir__store_output();
+  } while ( output < output_end );
+}
+
+static void STBIR_chans( stbir__horizontal_gather_,_channels_with_8_coeffs)( float * output_buffer, unsigned int output_sub_size, float const * decode_buffer, stbir__contributors const * horizontal_contributors, float const * horizontal_coefficients, int coefficient_width )
+{
+  float const * output_end = output_buffer + output_sub_size * STBIR__horizontal_channels;
+  float STBIR_SIMD_STREAMOUT_PTR( * ) output = output_buffer;
+  do {
+    float const * decode = decode_buffer + horizontal_contributors->n0 * STBIR__horizontal_channels; 
+    float const * hc = horizontal_coefficients;
+    stbir__4_coeff_start();
+    stbir__4_coeff_continue_from_4(4);
+    stbir__store_output();
+  } while ( output < output_end );
+}
+
+static void STBIR_chans( stbir__horizontal_gather_,_channels_with_9_coeffs)( float * output_buffer, unsigned int output_sub_size, float const * decode_buffer, stbir__contributors const * horizontal_contributors, float const * horizontal_coefficients, int coefficient_width )
+{
+  float const * output_end = output_buffer + output_sub_size * STBIR__horizontal_channels;
+  float STBIR_SIMD_STREAMOUT_PTR( * ) output = output_buffer;
+  do {
+    float const * decode = decode_buffer + horizontal_contributors->n0 * STBIR__horizontal_channels; 
+    float const * hc = horizontal_coefficients;
+    stbir__4_coeff_start();
+    stbir__4_coeff_continue_from_4(4);
+    stbir__1_coeff_remnant(8);
+    stbir__store_output();
+  } while ( output < output_end );
+}
+
+static void STBIR_chans( stbir__horizontal_gather_,_channels_with_10_coeffs)( float * output_buffer, unsigned int output_sub_size, float const * decode_buffer, stbir__contributors const * horizontal_contributors, float const * horizontal_coefficients, int coefficient_width )
+{
+  float const * output_end = output_buffer + output_sub_size * STBIR__horizontal_channels;
+  float STBIR_SIMD_STREAMOUT_PTR( * ) output = output_buffer;
+  do {
+    float const * decode = decode_buffer + horizontal_contributors->n0 * STBIR__horizontal_channels; 
+    float const * hc = horizontal_coefficients;
+    stbir__4_coeff_start();
+    stbir__4_coeff_continue_from_4(4);
+    stbir__2_coeff_remnant(8);
+    stbir__store_output();
+  } while ( output < output_end );
+}
+
+static void STBIR_chans( stbir__horizontal_gather_,_channels_with_11_coeffs)( float * output_buffer, unsigned int output_sub_size, float const * decode_buffer, stbir__contributors const * horizontal_contributors, float const * horizontal_coefficients, int coefficient_width )
+{
+  float const * output_end = output_buffer + output_sub_size * STBIR__horizontal_channels;
+  float STBIR_SIMD_STREAMOUT_PTR( * ) output = output_buffer;
+  stbir__3_coeff_setup();
+  do {
+    float const * decode = decode_buffer + horizontal_contributors->n0 * STBIR__horizontal_channels; 
+    float const * hc = horizontal_coefficients;
+    stbir__4_coeff_start();
+    stbir__4_coeff_continue_from_4(4);
+    stbir__3_coeff_remnant(8);
+    stbir__store_output();
+  } while ( output < output_end );
+}
+
+static void STBIR_chans( stbir__horizontal_gather_,_channels_with_12_coeffs)( float * output_buffer, unsigned int output_sub_size, float const * decode_buffer, stbir__contributors const * horizontal_contributors, float const * horizontal_coefficients, int coefficient_width )
+{
+  float const * output_end = output_buffer + output_sub_size * STBIR__horizontal_channels;
+  float STBIR_SIMD_STREAMOUT_PTR( * ) output = output_buffer;
+  do {
+    float const * decode = decode_buffer + horizontal_contributors->n0 * STBIR__horizontal_channels; 
+    float const * hc = horizontal_coefficients;
+    stbir__4_coeff_start();
+    stbir__4_coeff_continue_from_4(4);
+    stbir__4_coeff_continue_from_4(8);
+    stbir__store_output();
+  } while ( output < output_end );
+}
+
+static void STBIR_chans( stbir__horizontal_gather_,_channels_with_n_coeffs_mod0 )( float * output_buffer, unsigned int output_sub_size, float const * decode_buffer, stbir__contributors const * horizontal_contributors, float const * horizontal_coefficients, int coefficient_width )
+{
+  float const * output_end = output_buffer + output_sub_size * STBIR__horizontal_channels;
+  float STBIR_SIMD_STREAMOUT_PTR( * ) output = output_buffer;
+  do {
+    float const * decode = decode_buffer + horizontal_contributors->n0 * STBIR__horizontal_channels; 
+    int n = ( ( horizontal_contributors->n1 - horizontal_contributors->n0 + 1 ) - 4 + 3 ) >> 2; 
+    float const * hc = horizontal_coefficients;
+
+    stbir__4_coeff_start();
+    do {
+      hc += 4;
+      decode += STBIR__horizontal_channels * 4;
+      stbir__4_coeff_continue_from_4( 0 );
+      --n;
+    } while ( n > 0 );
+    stbir__store_output();
+  } while ( output < output_end );
+}
+
+static void STBIR_chans( stbir__horizontal_gather_,_channels_with_n_coeffs_mod1 )( float * output_buffer, unsigned int output_sub_size, float const * decode_buffer, stbir__contributors const * horizontal_contributors, float const * horizontal_coefficients, int coefficient_width )
+{
+  float const * output_end = output_buffer + output_sub_size * STBIR__horizontal_channels;
+  float STBIR_SIMD_STREAMOUT_PTR( * ) output = output_buffer;
+  do {
+    float const * decode = decode_buffer + horizontal_contributors->n0 * STBIR__horizontal_channels; 
+    int n = ( ( horizontal_contributors->n1 - horizontal_contributors->n0 + 1 ) - 5 + 3 ) >> 2; 
+    float const * hc = horizontal_coefficients;
+
+    stbir__4_coeff_start();
+    do {
+      hc += 4;
+      decode += STBIR__horizontal_channels * 4;
+      stbir__4_coeff_continue_from_4( 0 );
+      --n;
+    } while ( n > 0 );
+    stbir__1_coeff_remnant( 4 ); 
+    stbir__store_output();
+  } while ( output < output_end );
+}
+
+static void STBIR_chans( stbir__horizontal_gather_,_channels_with_n_coeffs_mod2 )( float * output_buffer, unsigned int output_sub_size, float const * decode_buffer, stbir__contributors const * horizontal_contributors, float const * horizontal_coefficients, int coefficient_width )
+{
+  float const * output_end = output_buffer + output_sub_size * STBIR__horizontal_channels;
+  float STBIR_SIMD_STREAMOUT_PTR( * ) output = output_buffer;
+  do {
+    float const * decode = decode_buffer + horizontal_contributors->n0 * STBIR__horizontal_channels; 
+    int n = ( ( horizontal_contributors->n1 - horizontal_contributors->n0 + 1 ) - 6 + 3 ) >> 2; 
+    float const * hc = horizontal_coefficients;
+
+    stbir__4_coeff_start();
+    do {
+      hc += 4;
+      decode += STBIR__horizontal_channels * 4;
+      stbir__4_coeff_continue_from_4( 0 );
+      --n;
+    } while ( n > 0 );
+    stbir__2_coeff_remnant( 4 ); 
+
+    stbir__store_output();
+  } while ( output < output_end );
+}
+
+static void STBIR_chans( stbir__horizontal_gather_,_channels_with_n_coeffs_mod3 )( float * output_buffer, unsigned int output_sub_size, float const * decode_buffer, stbir__contributors const * horizontal_contributors, float const * horizontal_coefficients, int coefficient_width )
+{
+  float const * output_end = output_buffer + output_sub_size * STBIR__horizontal_channels;
+  float STBIR_SIMD_STREAMOUT_PTR( * ) output = output_buffer;
+  stbir__3_coeff_setup();
+  do {
+    float const * decode = decode_buffer + horizontal_contributors->n0 * STBIR__horizontal_channels; 
+    int n = ( ( horizontal_contributors->n1 - horizontal_contributors->n0 + 1 ) - 7 + 3 ) >> 2; 
+    float const * hc = horizontal_coefficients;
+
+    stbir__4_coeff_start();
+    do {
+      hc += 4;
+      decode += STBIR__horizontal_channels * 4;
+      stbir__4_coeff_continue_from_4( 0 );
+      --n;
+    } while ( n > 0 );
+    stbir__3_coeff_remnant( 4 ); 
+
+    stbir__store_output();
+  } while ( output < output_end );
+}
+
+static stbir__horizontal_gather_channels_func * STBIR_chans(stbir__horizontal_gather_,_channels_with_n_coeffs_funcs)[4]=
+{
+  STBIR_chans(stbir__horizontal_gather_,_channels_with_n_coeffs_mod0),  
+  STBIR_chans(stbir__horizontal_gather_,_channels_with_n_coeffs_mod1),  
+  STBIR_chans(stbir__horizontal_gather_,_channels_with_n_coeffs_mod2),  
+  STBIR_chans(stbir__horizontal_gather_,_channels_with_n_coeffs_mod3),  
+};
+
+static stbir__horizontal_gather_channels_func * STBIR_chans(stbir__horizontal_gather_,_channels_funcs)[12]=
+{
+  STBIR_chans(stbir__horizontal_gather_,_channels_with_1_coeff),  
+  STBIR_chans(stbir__horizontal_gather_,_channels_with_2_coeffs),  
+  STBIR_chans(stbir__horizontal_gather_,_channels_with_3_coeffs),
+  STBIR_chans(stbir__horizontal_gather_,_channels_with_4_coeffs),  
+  STBIR_chans(stbir__horizontal_gather_,_channels_with_5_coeffs),  
+  STBIR_chans(stbir__horizontal_gather_,_channels_with_6_coeffs),  
+  STBIR_chans(stbir__horizontal_gather_,_channels_with_7_coeffs),
+  STBIR_chans(stbir__horizontal_gather_,_channels_with_8_coeffs),  
+  STBIR_chans(stbir__horizontal_gather_,_channels_with_9_coeffs),  
+  STBIR_chans(stbir__horizontal_gather_,_channels_with_10_coeffs),  
+  STBIR_chans(stbir__horizontal_gather_,_channels_with_11_coeffs),  
+  STBIR_chans(stbir__horizontal_gather_,_channels_with_12_coeffs),  
+};
+
+#undef STBIR__horizontal_channels
+#undef STB_IMAGE_RESIZE_DO_HORIZONTALS
+#undef stbir__1_coeff_only
+#undef stbir__1_coeff_remnant
+#undef stbir__2_coeff_only
+#undef stbir__2_coeff_remnant
+#undef stbir__3_coeff_only
+#undef stbir__3_coeff_remnant
+#undef stbir__3_coeff_setup
+#undef stbir__4_coeff_start
+#undef stbir__4_coeff_continue_from_4
+#undef stbir__store_output
+#undef stbir__store_output_tiny
+#undef STBIR_chans
+
+#endif  // HORIZONALS
+
+#undef STBIR_strs_join2
+#undef STBIR_strs_join1
+
+#endif // STB_IMAGE_RESIZE_DO_HORIZONTALS/VERTICALS/CODERS
+
+/*
+------------------------------------------------------------------------------
+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.
+------------------------------------------------------------------------------
+*/

Engine/source/gfx/bitmap/loaders/stb/stb_image_write.h

@@ -0,0 +1,1724 @@
+/* stb_image_write - v1.16 - public domain - http://nothings.org/stb
+   writes out PNG/BMP/TGA/JPEG/HDR images to C stdio - Sean Barrett 2010-2015
+                                     no warranty implied; use at your own risk
+
+   Before #including,
+
+       #define STB_IMAGE_WRITE_IMPLEMENTATION
+
+   in the file that you want to have the implementation.
+
+   Will probably not work correctly with strict-aliasing optimizations.
+
+ABOUT:
+
+   This header file is a library for writing images to C stdio or a callback.
+
+   The PNG output is not optimal; it is 20-50% larger than the file
+   written by a decent optimizing implementation; though providing a custom
+   zlib compress function (see STBIW_ZLIB_COMPRESS) can mitigate that.
+   This library is designed for source code compactness and simplicity,
+   not optimal image file size or run-time performance.
+
+BUILDING:
+
+   You can #define STBIW_ASSERT(x) before the #include to avoid using assert.h.
+   You can #define STBIW_MALLOC(), STBIW_REALLOC(), and STBIW_FREE() to replace
+   malloc,realloc,free.
+   You can #define STBIW_MEMMOVE() to replace memmove()
+   You can #define STBIW_ZLIB_COMPRESS to use a custom zlib-style compress function
+   for PNG compression (instead of the builtin one), it must have the following signature:
+   unsigned char * my_compress(unsigned char *data, int data_len, int *out_len, int quality);
+   The returned data will be freed with STBIW_FREE() (free() by default),
+   so it must be heap allocated with STBIW_MALLOC() (malloc() by default),
+
+UNICODE:
+
+   If compiling for Windows and you wish to use Unicode filenames, compile
+   with
+       #define STBIW_WINDOWS_UTF8
+   and pass utf8-encoded filenames. Call stbiw_convert_wchar_to_utf8 to convert
+   Windows wchar_t filenames to utf8.
+
+USAGE:
+
+   There are five functions, one for each image file format:
+
+     int stbi_write_png(char const *filename, int w, int h, int comp, const void *data, int stride_in_bytes);
+     int stbi_write_bmp(char const *filename, int w, int h, int comp, const void *data);
+     int stbi_write_tga(char const *filename, int w, int h, int comp, const void *data);
+     int stbi_write_jpg(char const *filename, int w, int h, int comp, const void *data, int quality);
+     int stbi_write_hdr(char const *filename, int w, int h, int comp, const float *data);
+
+     void stbi_flip_vertically_on_write(int flag); // flag is non-zero to flip data vertically
+
+   There are also five equivalent functions that use an arbitrary write function. You are
+   expected to open/close your file-equivalent before and after calling these:
+
+     int stbi_write_png_to_func(stbi_write_func *func, void *context, int w, int h, int comp, const void  *data, int stride_in_bytes);
+     int stbi_write_bmp_to_func(stbi_write_func *func, void *context, int w, int h, int comp, const void  *data);
+     int stbi_write_tga_to_func(stbi_write_func *func, void *context, int w, int h, int comp, const void  *data);
+     int stbi_write_hdr_to_func(stbi_write_func *func, void *context, int w, int h, int comp, const float *data);
+     int stbi_write_jpg_to_func(stbi_write_func *func, void *context, int x, int y, int comp, const void *data, int quality);
+
+   where the callback is:
+      void stbi_write_func(void *context, void *data, int size);
+
+   You can configure it with these global variables:
+      int stbi_write_tga_with_rle;             // defaults to true; set to 0 to disable RLE
+      int stbi_write_png_compression_level;    // defaults to 8; set to higher for more compression
+      int stbi_write_force_png_filter;         // defaults to -1; set to 0..5 to force a filter mode
+
+
+   You can define STBI_WRITE_NO_STDIO to disable the file variant of these
+   functions, so the library will not use stdio.h at all. However, this will
+   also disable HDR writing, because it requires stdio for formatted output.
+
+   Each function returns 0 on failure and non-0 on success.
+
+   The functions create an image file defined by the parameters. The image
+   is a rectangle of pixels stored from left-to-right, top-to-bottom.
+   Each pixel contains 'comp' channels of data stored interleaved with 8-bits
+   per channel, in the following order: 1=Y, 2=YA, 3=RGB, 4=RGBA. (Y is
+   monochrome color.) The rectangle is 'w' pixels wide and 'h' pixels tall.
+   The *data pointer points to the first byte of the top-left-most pixel.
+   For PNG, "stride_in_bytes" is the distance in bytes from the first byte of
+   a row of pixels to the first byte of the next row of pixels.
+
+   PNG creates output files with the same number of components as the input.
+   The BMP format expands Y to RGB in the file format and does not
+   output alpha.
+
+   PNG supports writing rectangles of data even when the bytes storing rows of
+   data are not consecutive in memory (e.g. sub-rectangles of a larger image),
+   by supplying the stride between the beginning of adjacent rows. The other
+   formats do not. (Thus you cannot write a native-format BMP through the BMP
+   writer, both because it is in BGR order and because it may have padding
+   at the end of the line.)
+
+   PNG allows you to set the deflate compression level by setting the global
+   variable 'stbi_write_png_compression_level' (it defaults to 8).
+
+   HDR expects linear float data. Since the format is always 32-bit rgb(e)
+   data, alpha (if provided) is discarded, and for monochrome data it is
+   replicated across all three channels.
+
+   TGA supports RLE or non-RLE compressed data. To use non-RLE-compressed
+   data, set the global variable 'stbi_write_tga_with_rle' to 0.
+
+   JPEG does ignore alpha channels in input data; quality is between 1 and 100.
+   Higher quality looks better but results in a bigger image.
+   JPEG baseline (no JPEG progressive).
+
+CREDITS:
+
+
+   Sean Barrett           -    PNG/BMP/TGA
+   Baldur Karlsson        -    HDR
+   Jean-Sebastien Guay    -    TGA monochrome
+   Tim Kelsey             -    misc enhancements
+   Alan Hickman           -    TGA RLE
+   Emmanuel Julien        -    initial file IO callback implementation
+   Jon Olick              -    original jo_jpeg.cpp code
+   Daniel Gibson          -    integrate JPEG, allow external zlib
+   Aarni Koskela          -    allow choosing PNG filter
+
+   bugfixes:
+      github:Chribba
+      Guillaume Chereau
+      github:jry2
+      github:romigrou
+      Sergio Gonzalez
+      Jonas Karlsson
+      Filip Wasil
+      Thatcher Ulrich
+      github:poppolopoppo
+      Patrick Boettcher
+      github:xeekworx
+      Cap Petschulat
+      Simon Rodriguez
+      Ivan Tikhonov
+      github:ignotion
+      Adam Schackart
+      Andrew Kensler
+
+LICENSE
+
+  See end of file for license information.
+
+*/
+
+#ifndef INCLUDE_STB_IMAGE_WRITE_H
+#define INCLUDE_STB_IMAGE_WRITE_H
+
+#include <stdlib.h>
+
+// if STB_IMAGE_WRITE_STATIC causes problems, try defining STBIWDEF to 'inline' or 'static inline'
+#ifndef STBIWDEF
+#ifdef STB_IMAGE_WRITE_STATIC
+#define STBIWDEF  static
+#else
+#ifdef __cplusplus
+#define STBIWDEF  extern "C"
+#else
+#define STBIWDEF  extern
+#endif
+#endif
+#endif
+
+#ifndef STB_IMAGE_WRITE_STATIC  // C++ forbids static forward declarations
+STBIWDEF int stbi_write_tga_with_rle;
+STBIWDEF int stbi_write_png_compression_level;
+STBIWDEF int stbi_write_force_png_filter;
+#endif
+
+#ifndef STBI_WRITE_NO_STDIO
+STBIWDEF int stbi_write_png(char const *filename, int w, int h, int comp, const void  *data, int stride_in_bytes);
+STBIWDEF int stbi_write_bmp(char const *filename, int w, int h, int comp, const void  *data);
+STBIWDEF int stbi_write_tga(char const *filename, int w, int h, int comp, const void  *data);
+STBIWDEF int stbi_write_hdr(char const *filename, int w, int h, int comp, const float *data);
+STBIWDEF int stbi_write_jpg(char const *filename, int x, int y, int comp, const void  *data, int quality);
+
+#ifdef STBIW_WINDOWS_UTF8
+STBIWDEF int stbiw_convert_wchar_to_utf8(char *buffer, size_t bufferlen, const wchar_t* input);
+#endif
+#endif
+
+typedef void stbi_write_func(void *context, void *data, int size);
+
+STBIWDEF int stbi_write_png_to_func(stbi_write_func *func, void *context, int w, int h, int comp, const void  *data, int stride_in_bytes);
+STBIWDEF int stbi_write_bmp_to_func(stbi_write_func *func, void *context, int w, int h, int comp, const void  *data);
+STBIWDEF int stbi_write_tga_to_func(stbi_write_func *func, void *context, int w, int h, int comp, const void  *data);
+STBIWDEF int stbi_write_hdr_to_func(stbi_write_func *func, void *context, int w, int h, int comp, const float *data);
+STBIWDEF int stbi_write_jpg_to_func(stbi_write_func *func, void *context, int x, int y, int comp, const void  *data, int quality);
+
+STBIWDEF void stbi_flip_vertically_on_write(int flip_boolean);
+
+#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)
+
+#ifdef STB_IMAGE_WRITE_STATIC
+static int stbi_write_png_compression_level = 8;
+static int stbi_write_tga_with_rle = 1;
+static int stbi_write_force_png_filter = -1;
+#else
+int stbi_write_png_compression_level = 8;
+int stbi_write_tga_with_rle = 1;
+int stbi_write_force_png_filter = -1;
+#endif
+
+static int stbi__flip_vertically_on_write = 0;
+
+STBIWDEF void stbi_flip_vertically_on_write(int flag)
+{
+   stbi__flip_vertically_on_write = flag;
+}
+
+typedef struct
+{
+   stbi_write_func *func;
+   void *context;
+   unsigned char buffer[64];
+   int buf_used;
+} 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);
+}
+
+#if defined(_WIN32) && defined(STBIW_WINDOWS_UTF8)
+#ifdef __cplusplus
+#define STBIW_EXTERN extern "C"
+#else
+#define STBIW_EXTERN extern
+#endif
+STBIW_EXTERN __declspec(dllimport) int __stdcall MultiByteToWideChar(unsigned int cp, unsigned long flags, const char *str, int cbmb, wchar_t *widestr, int cchwide);
+STBIW_EXTERN __declspec(dllimport) int __stdcall WideCharToMultiByte(unsigned int cp, unsigned long flags, const wchar_t *widestr, int cchwide, char *str, int cbmb, const char *defchar, int *used_default);
+
+STBIWDEF int stbiw_convert_wchar_to_utf8(char *buffer, size_t bufferlen, const wchar_t* input)
+{
+   return WideCharToMultiByte(65001 /* UTF8 */, 0, input, -1, buffer, (int) bufferlen, NULL, NULL);
+}
+#endif
+
+static FILE *stbiw__fopen(char const *filename, char const *mode)
+{
+   FILE *f;
+#if defined(_WIN32) && defined(STBIW_WINDOWS_UTF8)
+   wchar_t wMode[64];
+   wchar_t wFilename[1024];
+   if (0 == MultiByteToWideChar(65001 /* UTF8 */, 0, filename, -1, wFilename, sizeof(wFilename)/sizeof(*wFilename)))
+      return 0;
+
+   if (0 == MultiByteToWideChar(65001 /* UTF8 */, 0, mode, -1, wMode, sizeof(wMode)/sizeof(*wMode)))
+      return 0;
+
+#if defined(_MSC_VER) && _MSC_VER >= 1400
+   if (0 != _wfopen_s(&f, wFilename, wMode))
+      f = 0;
+#else
+   f = _wfopen(wFilename, wMode);
+#endif
+
+#elif defined(_MSC_VER) && _MSC_VER >= 1400
+   if (0 != fopen_s(&f, filename, mode))
+      f=0;
+#else
+   f = fopen(filename, mode);
+#endif
+   return f;
+}
+
+static int stbi__start_write_file(stbi__write_context *s, const char *filename)
+{
+   FILE *f = stbiw__fopen(filename, "wb");
+   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];
+
+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__write_flush(stbi__write_context *s)
+{
+   if (s->buf_used) {
+      s->func(s->context, &s->buffer, s->buf_used);
+      s->buf_used = 0;
+   }
+}
+
+static void stbiw__putc(stbi__write_context *s, unsigned char c)
+{
+   s->func(s->context, &c, 1);
+}
+
+static void stbiw__write1(stbi__write_context *s, unsigned char a)
+{
+   if ((size_t)s->buf_used + 1 > sizeof(s->buffer))
+      stbiw__write_flush(s);
+   s->buffer[s->buf_used++] = a;
+}
+
+static void stbiw__write3(stbi__write_context *s, unsigned char a, unsigned char b, unsigned char c)
+{
+   int n;
+   if ((size_t)s->buf_used + 3 > sizeof(s->buffer))
+      stbiw__write_flush(s);
+   n = s->buf_used;
+   s->buf_used = n+3;
+   s->buffer[n+0] = a;
+   s->buffer[n+1] = b;
+   s->buffer[n+2] = c;
+}
+
+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)
+      stbiw__write1(s, d[comp - 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
+            stbiw__write1(s, d[0]);  // 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)
+      stbiw__write1(s, d[comp - 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 (stbi__flip_vertically_on_write)
+      vdir *= -1;
+
+   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);
+      }
+      stbiw__write_flush(s);
+      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)
+{
+   if (comp != 4) {
+      // write RGB bitmap
+      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
+   } else {
+      // RGBA bitmaps need a v4 header
+      // use BI_BITFIELDS mode with 32bpp and alpha mask
+      // (straight BI_RGB with alpha mask doesn't work in most readers)
+      return stbiw__outfile(s,-1,-1,x,y,comp,1,(void *)data,1,0,
+         "11 4 22 4" "4 44 22 444444 4444 4 444 444 444 444",
+         'B', 'M', 14+108+x*y*4, 0, 0, 14+108, // file header
+         108, x,y, 1,32, 3,0,0,0,0,0, 0xff0000,0xff00,0xff,0xff000000u, 0, 0,0,0, 0,0,0, 0,0,0, 0,0,0); // bitmap V4 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 = { 0 };
+   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 = { 0 };
+   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;
+      int jend, jdir;
+
+      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);
+
+      if (stbi__flip_vertically_on_write) {
+         j = 0;
+         jend = y;
+         jdir = 1;
+      } else {
+         j = y-1;
+         jend = -1;
+         jdir = -1;
+      }
+      for (; j != jend; j += jdir) {
+         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);
+               stbiw__write1(s, header);
+               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);
+               stbiw__write1(s, header);
+               stbiw__write_pixel(s, -1, comp, has_alpha, 0, begin);
+            }
+         }
+      }
+      stbiw__write_flush(s);
+   }
+   return 1;
+}
+
+STBIWDEF 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 = { 0 };
+   stbi__start_write_callbacks(&s, func, context);
+   return stbi_write_tga_core(&s, x, y, comp, (void *) data);
+}
+
+#ifndef STBI_WRITE_NO_STDIO
+STBIWDEF int stbi_write_tga(char const *filename, int x, int y, int comp, const void *data)
+{
+   stbi__write_context s = { 0 };
+   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))
+
+#ifndef STBI_WRITE_NO_STDIO
+
+static 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);
+   }
+}
+
+static 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);
+}
+
+static 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);
+}
+
+static 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);
+
+#ifdef __STDC_LIB_EXT1__
+      len = sprintf_s(buffer, sizeof(buffer), "EXPOSURE=          1.0000000000000\n\n-Y %d +X %d\n", y, x);
+#else
+      len = sprintf(buffer, "EXPOSURE=          1.0000000000000\n\n-Y %d +X %d\n", y, x);
+#endif
+      s->func(s->context, buffer, len);
+
+      for(i=0; i < y; i++)
+         stbiw__write_hdr_scanline(s, x, comp, scratch, data + comp*x*(stbi__flip_vertically_on_write ? y-1-i : i));
+      STBIW_FREE(scratch);
+      return 1;
+   }
+}
+
+STBIWDEF 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 = { 0 };
+   stbi__start_write_callbacks(&s, func, context);
+   return stbi_write_hdr_core(&s, x, y, comp, (float *) data);
+}
+
+STBIWDEF int stbi_write_hdr(char const *filename, int x, int y, int comp, const float *data)
+{
+   stbi__write_context s = { 0 };
+   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
+//
+
+#ifndef STBIW_ZLIB_COMPRESS
+// stretchy buffer; stbiw__sbpush() == vector<>::push_back() -- stbiw__sbcount() == vector<>::size()
+#define stbiw__sbraw(a) ((int *) (void *) (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
+
+#endif // STBIW_ZLIB_COMPRESS
+
+STBIWDEF unsigned char * stbi_zlib_compress(unsigned char *data, int data_len, int *out_len, int quality)
+{
+#ifdef STBIW_ZLIB_COMPRESS
+   // user provided a zlib compress implementation, use that
+   return STBIW_ZLIB_COMPRESS(data, data_len, out_len, quality);
+#else // use builtin
+   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(unsigned char**));
+   if (hash_table == NULL)
+      return NULL;
+   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);
+
+   // store uncompressed instead if compression was worse
+   if (stbiw__sbn(out) > data_len + 2 + ((data_len+32766)/32767)*5) {
+      stbiw__sbn(out) = 2;  // truncate to DEFLATE 32K window and FLEVEL = 1
+      for (j = 0; j < data_len;) {
+         int blocklen = data_len - j;
+         if (blocklen > 32767) blocklen = 32767;
+         stbiw__sbpush(out, data_len - j == blocklen); // BFINAL = ?, BTYPE = 0 -- no compression
+         stbiw__sbpush(out, STBIW_UCHAR(blocklen)); // LEN
+         stbiw__sbpush(out, STBIW_UCHAR(blocklen >> 8));
+         stbiw__sbpush(out, STBIW_UCHAR(~blocklen)); // NLEN
+         stbiw__sbpush(out, STBIW_UCHAR(~blocklen >> 8));
+         memcpy(out+stbiw__sbn(out), data+j, blocklen);
+         stbiw__sbn(out) += blocklen;
+         j += blocklen;
+      }
+   }
+
+   {
+      // 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);
+#endif // STBIW_ZLIB_COMPRESS
+}
+
+static unsigned int stbiw__crc32(unsigned char *buffer, int len)
+{
+#ifdef STBIW_CRC32
+    return STBIW_CRC32(buffer, len);
+#else
+   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;
+#endif
+}
+
+#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
+static void stbiw__encode_png_line(unsigned char *pixels, int stride_bytes, int width, int height, int y, int n, int filter_type, signed char *line_buffer)
+{
+   static int mapping[] = { 0,1,2,3,4 };
+   static int firstmap[] = { 0,1,0,5,6 };
+   int *mymap = (y != 0) ? mapping : firstmap;
+   int i;
+   int type = mymap[filter_type];
+   unsigned char *z = pixels + stride_bytes * (stbi__flip_vertically_on_write ? height-1-y : y);
+   int signed_stride = stbi__flip_vertically_on_write ? -stride_bytes : stride_bytes;
+
+   if (type==0) {
+      memcpy(line_buffer, z, width*n);
+      return;
+   }
+
+   // first loop isn't optimized since it's just one pixel
+   for (i = 0; i < n; ++i) {
+      switch (type) {
+         case 1: line_buffer[i] = z[i]; break;
+         case 2: line_buffer[i] = z[i] - z[i-signed_stride]; break;
+         case 3: line_buffer[i] = z[i] - (z[i-signed_stride]>>1); break;
+         case 4: line_buffer[i] = (signed char) (z[i] - stbiw__paeth(0,z[i-signed_stride],0)); break;
+         case 5: line_buffer[i] = z[i]; break;
+         case 6: line_buffer[i] = z[i]; break;
+      }
+   }
+   switch (type) {
+      case 1: for (i=n; i < width*n; ++i) line_buffer[i] = z[i] - z[i-n]; break;
+      case 2: for (i=n; i < width*n; ++i) line_buffer[i] = z[i] - z[i-signed_stride]; break;
+      case 3: for (i=n; i < width*n; ++i) line_buffer[i] = z[i] - ((z[i-n] + z[i-signed_stride])>>1); break;
+      case 4: for (i=n; i < width*n; ++i) line_buffer[i] = z[i] - stbiw__paeth(z[i-n], z[i-signed_stride], z[i-signed_stride-n]); break;
+      case 5: for (i=n; i < width*n; ++i) line_buffer[i] = z[i] - (z[i-n]>>1); break;
+      case 6: for (i=n; i < width*n; ++i) line_buffer[i] = z[i] - stbiw__paeth(z[i-n], 0,0); break;
+   }
+}
+
+STBIWDEF unsigned char *stbi_write_png_to_mem(const unsigned char *pixels, int stride_bytes, int x, int y, int n, int *out_len)
+{
+   int force_filter = stbi_write_force_png_filter;
+   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 j,zlen;
+
+   if (stride_bytes == 0)
+      stride_bytes = x * n;
+
+   if (force_filter >= 5) {
+      force_filter = -1;
+   }
+
+   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) {
+      int filter_type;
+      if (force_filter > -1) {
+         filter_type = force_filter;
+         stbiw__encode_png_line((unsigned char*)(pixels), stride_bytes, x, y, j, n, force_filter, line_buffer);
+      } else { // Estimate the best filter by running through all of them:
+         int best_filter = 0, best_filter_val = 0x7fffffff, est, i;
+         for (filter_type = 0; filter_type < 5; filter_type++) {
+            stbiw__encode_png_line((unsigned char*)(pixels), stride_bytes, x, y, j, n, filter_type, line_buffer);
+
+            // Estimate the entropy of the line using this filter; the less, the better.
+            est = 0;
+            for (i = 0; i < x*n; ++i) {
+               est += abs((signed char) line_buffer[i]);
+            }
+            if (est < best_filter_val) {
+               best_filter_val = est;
+               best_filter = filter_type;
+            }
+         }
+         if (filter_type != best_filter) {  // If the last iteration already got us the best filter, don't redo it
+            stbiw__encode_png_line((unsigned char*)(pixels), stride_bytes, x, y, j, n, best_filter, line_buffer);
+            filter_type = best_filter;
+         }
+      }
+      // when we get here, filter_type contains the filter type, and line_buffer contains the data
+      filt[j*(x*n+1)] = (unsigned char) filter_type;
+      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, stbi_write_png_compression_level);
+   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((const unsigned char *) data, stride_bytes, x, y, comp, &len);
+   if (png == NULL) return 0;
+
+   f = stbiw__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((const unsigned char *) data, stride_bytes, x, y, comp, &len);
+   if (png == NULL) return 0;
+   func(context, png, len);
+   STBIW_FREE(png);
+   return 1;
+}
+
+
+/* ***************************************************************************
+ *
+ * JPEG writer
+ *
+ * This is based on Jon Olick's jo_jpeg.cpp:
+ * public domain Simple, Minimalistic JPEG writer - http://www.jonolick.com/code.html
+ */
+
+static const unsigned char stbiw__jpg_ZigZag[] = { 0,1,5,6,14,15,27,28,2,4,7,13,16,26,29,42,3,8,12,17,25,30,41,43,9,11,18,
+      24,31,40,44,53,10,19,23,32,39,45,52,54,20,22,33,38,46,51,55,60,21,34,37,47,50,56,59,61,35,36,48,49,57,58,62,63 };
+
+static void stbiw__jpg_writeBits(stbi__write_context *s, int *bitBufP, int *bitCntP, const unsigned short *bs) {
+   int bitBuf = *bitBufP, bitCnt = *bitCntP;
+   bitCnt += bs[1];
+   bitBuf |= bs[0] << (24 - bitCnt);
+   while(bitCnt >= 8) {
+      unsigned char c = (bitBuf >> 16) & 255;
+      stbiw__putc(s, c);
+      if(c == 255) {
+         stbiw__putc(s, 0);
+      }
+      bitBuf <<= 8;
+      bitCnt -= 8;
+   }
+   *bitBufP = bitBuf;
+   *bitCntP = bitCnt;
+}
+
+static void stbiw__jpg_DCT(float *d0p, float *d1p, float *d2p, float *d3p, float *d4p, float *d5p, float *d6p, float *d7p) {
+   float d0 = *d0p, d1 = *d1p, d2 = *d2p, d3 = *d3p, d4 = *d4p, d5 = *d5p, d6 = *d6p, d7 = *d7p;
+   float z1, z2, z3, z4, z5, z11, z13;
+
+   float tmp0 = d0 + d7;
+   float tmp7 = d0 - d7;
+   float tmp1 = d1 + d6;
+   float tmp6 = d1 - d6;
+   float tmp2 = d2 + d5;
+   float tmp5 = d2 - d5;
+   float tmp3 = d3 + d4;
+   float tmp4 = d3 - d4;
+
+   // Even part
+   float tmp10 = tmp0 + tmp3;   // phase 2
+   float tmp13 = tmp0 - tmp3;
+   float tmp11 = tmp1 + tmp2;
+   float tmp12 = tmp1 - tmp2;
+
+   d0 = tmp10 + tmp11;       // phase 3
+   d4 = tmp10 - tmp11;
+
+   z1 = (tmp12 + tmp13) * 0.707106781f; // c4
+   d2 = tmp13 + z1;       // phase 5
+   d6 = tmp13 - z1;
+
+   // Odd part
+   tmp10 = tmp4 + tmp5;       // phase 2
+   tmp11 = tmp5 + tmp6;
+   tmp12 = tmp6 + tmp7;
+
+   // The rotator is modified from fig 4-8 to avoid extra negations.
+   z5 = (tmp10 - tmp12) * 0.382683433f; // c6
+   z2 = tmp10 * 0.541196100f + z5; // c2-c6
+   z4 = tmp12 * 1.306562965f + z5; // c2+c6
+   z3 = tmp11 * 0.707106781f; // c4
+
+   z11 = tmp7 + z3;      // phase 5
+   z13 = tmp7 - z3;
+
+   *d5p = z13 + z2;         // phase 6
+   *d3p = z13 - z2;
+   *d1p = z11 + z4;
+   *d7p = z11 - z4;
+
+   *d0p = d0;  *d2p = d2;  *d4p = d4;  *d6p = d6;
+}
+
+static void stbiw__jpg_calcBits(int val, unsigned short bits[2]) {
+   int tmp1 = val < 0 ? -val : val;
+   val = val < 0 ? val-1 : val;
+   bits[1] = 1;
+   while(tmp1 >>= 1) {
+      ++bits[1];
+   }
+   bits[0] = val & ((1<<bits[1])-1);
+}
+
+static int stbiw__jpg_processDU(stbi__write_context *s, int *bitBuf, int *bitCnt, float *CDU, int du_stride, float *fdtbl, int DC, const unsigned short HTDC[256][2], const unsigned short HTAC[256][2]) {
+   const unsigned short EOB[2] = { HTAC[0x00][0], HTAC[0x00][1] };
+   const unsigned short M16zeroes[2] = { HTAC[0xF0][0], HTAC[0xF0][1] };
+   int dataOff, i, j, n, diff, end0pos, x, y;
+   int DU[64];
+
+   // DCT rows
+   for(dataOff=0, n=du_stride*8; dataOff<n; dataOff+=du_stride) {
+      stbiw__jpg_DCT(&CDU[dataOff], &CDU[dataOff+1], &CDU[dataOff+2], &CDU[dataOff+3], &CDU[dataOff+4], &CDU[dataOff+5], &CDU[dataOff+6], &CDU[dataOff+7]);
+   }
+   // DCT columns
+   for(dataOff=0; dataOff<8; ++dataOff) {
+      stbiw__jpg_DCT(&CDU[dataOff], &CDU[dataOff+du_stride], &CDU[dataOff+du_stride*2], &CDU[dataOff+du_stride*3], &CDU[dataOff+du_stride*4],
+                     &CDU[dataOff+du_stride*5], &CDU[dataOff+du_stride*6], &CDU[dataOff+du_stride*7]);
+   }
+   // Quantize/descale/zigzag the coefficients
+   for(y = 0, j=0; y < 8; ++y) {
+      for(x = 0; x < 8; ++x,++j) {
+         float v;
+         i = y*du_stride+x;
+         v = CDU[i]*fdtbl[j];
+         // DU[stbiw__jpg_ZigZag[j]] = (int)(v < 0 ? ceilf(v - 0.5f) : floorf(v + 0.5f));
+         // ceilf() and floorf() are C99, not C89, but I /think/ they're not needed here anyway?
+         DU[stbiw__jpg_ZigZag[j]] = (int)(v < 0 ? v - 0.5f : v + 0.5f);
+      }
+   }
+
+   // Encode DC
+   diff = DU[0] - DC;
+   if (diff == 0) {
+      stbiw__jpg_writeBits(s, bitBuf, bitCnt, HTDC[0]);
+   } else {
+      unsigned short bits[2];
+      stbiw__jpg_calcBits(diff, bits);
+      stbiw__jpg_writeBits(s, bitBuf, bitCnt, HTDC[bits[1]]);
+      stbiw__jpg_writeBits(s, bitBuf, bitCnt, bits);
+   }
+   // Encode ACs
+   end0pos = 63;
+   for(; (end0pos>0)&&(DU[end0pos]==0); --end0pos) {
+   }
+   // end0pos = first element in reverse order !=0
+   if(end0pos == 0) {
+      stbiw__jpg_writeBits(s, bitBuf, bitCnt, EOB);
+      return DU[0];
+   }
+   for(i = 1; i <= end0pos; ++i) {
+      int startpos = i;
+      int nrzeroes;
+      unsigned short bits[2];
+      for (; DU[i]==0 && i<=end0pos; ++i) {
+      }
+      nrzeroes = i-startpos;
+      if ( nrzeroes >= 16 ) {
+         int lng = nrzeroes>>4;
+         int nrmarker;
+         for (nrmarker=1; nrmarker <= lng; ++nrmarker)
+            stbiw__jpg_writeBits(s, bitBuf, bitCnt, M16zeroes);
+         nrzeroes &= 15;
+      }
+      stbiw__jpg_calcBits(DU[i], bits);
+      stbiw__jpg_writeBits(s, bitBuf, bitCnt, HTAC[(nrzeroes<<4)+bits[1]]);
+      stbiw__jpg_writeBits(s, bitBuf, bitCnt, bits);
+   }
+   if(end0pos != 63) {
+      stbiw__jpg_writeBits(s, bitBuf, bitCnt, EOB);
+   }
+   return DU[0];
+}
+
+static int stbi_write_jpg_core(stbi__write_context *s, int width, int height, int comp, const void* data, int quality) {
+   // Constants that don't pollute global namespace
+   static const unsigned char std_dc_luminance_nrcodes[] = {0,0,1,5,1,1,1,1,1,1,0,0,0,0,0,0,0};
+   static const unsigned char std_dc_luminance_values[] = {0,1,2,3,4,5,6,7,8,9,10,11};
+   static const unsigned char std_ac_luminance_nrcodes[] = {0,0,2,1,3,3,2,4,3,5,5,4,4,0,0,1,0x7d};
+   static const unsigned char std_ac_luminance_values[] = {
+      0x01,0x02,0x03,0x00,0x04,0x11,0x05,0x12,0x21,0x31,0x41,0x06,0x13,0x51,0x61,0x07,0x22,0x71,0x14,0x32,0x81,0x91,0xa1,0x08,
+      0x23,0x42,0xb1,0xc1,0x15,0x52,0xd1,0xf0,0x24,0x33,0x62,0x72,0x82,0x09,0x0a,0x16,0x17,0x18,0x19,0x1a,0x25,0x26,0x27,0x28,
+      0x29,0x2a,0x34,0x35,0x36,0x37,0x38,0x39,0x3a,0x43,0x44,0x45,0x46,0x47,0x48,0x49,0x4a,0x53,0x54,0x55,0x56,0x57,0x58,0x59,
+      0x5a,0x63,0x64,0x65,0x66,0x67,0x68,0x69,0x6a,0x73,0x74,0x75,0x76,0x77,0x78,0x79,0x7a,0x83,0x84,0x85,0x86,0x87,0x88,0x89,
+      0x8a,0x92,0x93,0x94,0x95,0x96,0x97,0x98,0x99,0x9a,0xa2,0xa3,0xa4,0xa5,0xa6,0xa7,0xa8,0xa9,0xaa,0xb2,0xb3,0xb4,0xb5,0xb6,
+      0xb7,0xb8,0xb9,0xba,0xc2,0xc3,0xc4,0xc5,0xc6,0xc7,0xc8,0xc9,0xca,0xd2,0xd3,0xd4,0xd5,0xd6,0xd7,0xd8,0xd9,0xda,0xe1,0xe2,
+      0xe3,0xe4,0xe5,0xe6,0xe7,0xe8,0xe9,0xea,0xf1,0xf2,0xf3,0xf4,0xf5,0xf6,0xf7,0xf8,0xf9,0xfa
+   };
+   static const unsigned char std_dc_chrominance_nrcodes[] = {0,0,3,1,1,1,1,1,1,1,1,1,0,0,0,0,0};
+   static const unsigned char std_dc_chrominance_values[] = {0,1,2,3,4,5,6,7,8,9,10,11};
+   static const unsigned char std_ac_chrominance_nrcodes[] = {0,0,2,1,2,4,4,3,4,7,5,4,4,0,1,2,0x77};
+   static const unsigned char std_ac_chrominance_values[] = {
+      0x00,0x01,0x02,0x03,0x11,0x04,0x05,0x21,0x31,0x06,0x12,0x41,0x51,0x07,0x61,0x71,0x13,0x22,0x32,0x81,0x08,0x14,0x42,0x91,
+      0xa1,0xb1,0xc1,0x09,0x23,0x33,0x52,0xf0,0x15,0x62,0x72,0xd1,0x0a,0x16,0x24,0x34,0xe1,0x25,0xf1,0x17,0x18,0x19,0x1a,0x26,
+      0x27,0x28,0x29,0x2a,0x35,0x36,0x37,0x38,0x39,0x3a,0x43,0x44,0x45,0x46,0x47,0x48,0x49,0x4a,0x53,0x54,0x55,0x56,0x57,0x58,
+      0x59,0x5a,0x63,0x64,0x65,0x66,0x67,0x68,0x69,0x6a,0x73,0x74,0x75,0x76,0x77,0x78,0x79,0x7a,0x82,0x83,0x84,0x85,0x86,0x87,
+      0x88,0x89,0x8a,0x92,0x93,0x94,0x95,0x96,0x97,0x98,0x99,0x9a,0xa2,0xa3,0xa4,0xa5,0xa6,0xa7,0xa8,0xa9,0xaa,0xb2,0xb3,0xb4,
+      0xb5,0xb6,0xb7,0xb8,0xb9,0xba,0xc2,0xc3,0xc4,0xc5,0xc6,0xc7,0xc8,0xc9,0xca,0xd2,0xd3,0xd4,0xd5,0xd6,0xd7,0xd8,0xd9,0xda,
+      0xe2,0xe3,0xe4,0xe5,0xe6,0xe7,0xe8,0xe9,0xea,0xf2,0xf3,0xf4,0xf5,0xf6,0xf7,0xf8,0xf9,0xfa
+   };
+   // Huffman tables
+   static const unsigned short YDC_HT[256][2] = { {0,2},{2,3},{3,3},{4,3},{5,3},{6,3},{14,4},{30,5},{62,6},{126,7},{254,8},{510,9}};
+   static const unsigned short UVDC_HT[256][2] = { {0,2},{1,2},{2,2},{6,3},{14,4},{30,5},{62,6},{126,7},{254,8},{510,9},{1022,10},{2046,11}};
+   static const unsigned short YAC_HT[256][2] = {
+      {10,4},{0,2},{1,2},{4,3},{11,4},{26,5},{120,7},{248,8},{1014,10},{65410,16},{65411,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
+      {12,4},{27,5},{121,7},{502,9},{2038,11},{65412,16},{65413,16},{65414,16},{65415,16},{65416,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
+      {28,5},{249,8},{1015,10},{4084,12},{65417,16},{65418,16},{65419,16},{65420,16},{65421,16},{65422,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
+      {58,6},{503,9},{4085,12},{65423,16},{65424,16},{65425,16},{65426,16},{65427,16},{65428,16},{65429,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
+      {59,6},{1016,10},{65430,16},{65431,16},{65432,16},{65433,16},{65434,16},{65435,16},{65436,16},{65437,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
+      {122,7},{2039,11},{65438,16},{65439,16},{65440,16},{65441,16},{65442,16},{65443,16},{65444,16},{65445,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
+      {123,7},{4086,12},{65446,16},{65447,16},{65448,16},{65449,16},{65450,16},{65451,16},{65452,16},{65453,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
+      {250,8},{4087,12},{65454,16},{65455,16},{65456,16},{65457,16},{65458,16},{65459,16},{65460,16},{65461,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
+      {504,9},{32704,15},{65462,16},{65463,16},{65464,16},{65465,16},{65466,16},{65467,16},{65468,16},{65469,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
+      {505,9},{65470,16},{65471,16},{65472,16},{65473,16},{65474,16},{65475,16},{65476,16},{65477,16},{65478,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
+      {506,9},{65479,16},{65480,16},{65481,16},{65482,16},{65483,16},{65484,16},{65485,16},{65486,16},{65487,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
+      {1017,10},{65488,16},{65489,16},{65490,16},{65491,16},{65492,16},{65493,16},{65494,16},{65495,16},{65496,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
+      {1018,10},{65497,16},{65498,16},{65499,16},{65500,16},{65501,16},{65502,16},{65503,16},{65504,16},{65505,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
+      {2040,11},{65506,16},{65507,16},{65508,16},{65509,16},{65510,16},{65511,16},{65512,16},{65513,16},{65514,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
+      {65515,16},{65516,16},{65517,16},{65518,16},{65519,16},{65520,16},{65521,16},{65522,16},{65523,16},{65524,16},{0,0},{0,0},{0,0},{0,0},{0,0},
+      {2041,11},{65525,16},{65526,16},{65527,16},{65528,16},{65529,16},{65530,16},{65531,16},{65532,16},{65533,16},{65534,16},{0,0},{0,0},{0,0},{0,0},{0,0}
+   };
+   static const unsigned short UVAC_HT[256][2] = {
+      {0,2},{1,2},{4,3},{10,4},{24,5},{25,5},{56,6},{120,7},{500,9},{1014,10},{4084,12},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
+      {11,4},{57,6},{246,8},{501,9},{2038,11},{4085,12},{65416,16},{65417,16},{65418,16},{65419,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
+      {26,5},{247,8},{1015,10},{4086,12},{32706,15},{65420,16},{65421,16},{65422,16},{65423,16},{65424,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
+      {27,5},{248,8},{1016,10},{4087,12},{65425,16},{65426,16},{65427,16},{65428,16},{65429,16},{65430,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
+      {58,6},{502,9},{65431,16},{65432,16},{65433,16},{65434,16},{65435,16},{65436,16},{65437,16},{65438,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
+      {59,6},{1017,10},{65439,16},{65440,16},{65441,16},{65442,16},{65443,16},{65444,16},{65445,16},{65446,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
+      {121,7},{2039,11},{65447,16},{65448,16},{65449,16},{65450,16},{65451,16},{65452,16},{65453,16},{65454,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
+      {122,7},{2040,11},{65455,16},{65456,16},{65457,16},{65458,16},{65459,16},{65460,16},{65461,16},{65462,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
+      {249,8},{65463,16},{65464,16},{65465,16},{65466,16},{65467,16},{65468,16},{65469,16},{65470,16},{65471,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
+      {503,9},{65472,16},{65473,16},{65474,16},{65475,16},{65476,16},{65477,16},{65478,16},{65479,16},{65480,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
+      {504,9},{65481,16},{65482,16},{65483,16},{65484,16},{65485,16},{65486,16},{65487,16},{65488,16},{65489,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
+      {505,9},{65490,16},{65491,16},{65492,16},{65493,16},{65494,16},{65495,16},{65496,16},{65497,16},{65498,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
+      {506,9},{65499,16},{65500,16},{65501,16},{65502,16},{65503,16},{65504,16},{65505,16},{65506,16},{65507,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
+      {2041,11},{65508,16},{65509,16},{65510,16},{65511,16},{65512,16},{65513,16},{65514,16},{65515,16},{65516,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0},
+      {16352,14},{65517,16},{65518,16},{65519,16},{65520,16},{65521,16},{65522,16},{65523,16},{65524,16},{65525,16},{0,0},{0,0},{0,0},{0,0},{0,0},
+      {1018,10},{32707,15},{65526,16},{65527,16},{65528,16},{65529,16},{65530,16},{65531,16},{65532,16},{65533,16},{65534,16},{0,0},{0,0},{0,0},{0,0},{0,0}
+   };
+   static const int YQT[] = {16,11,10,16,24,40,51,61,12,12,14,19,26,58,60,55,14,13,16,24,40,57,69,56,14,17,22,29,51,87,80,62,18,22,
+                             37,56,68,109,103,77,24,35,55,64,81,104,113,92,49,64,78,87,103,121,120,101,72,92,95,98,112,100,103,99};
+   static const int UVQT[] = {17,18,24,47,99,99,99,99,18,21,26,66,99,99,99,99,24,26,56,99,99,99,99,99,47,66,99,99,99,99,99,99,
+                              99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99};
+   static const float aasf[] = { 1.0f * 2.828427125f, 1.387039845f * 2.828427125f, 1.306562965f * 2.828427125f, 1.175875602f * 2.828427125f,
+                                 1.0f * 2.828427125f, 0.785694958f * 2.828427125f, 0.541196100f * 2.828427125f, 0.275899379f * 2.828427125f };
+
+   int row, col, i, k, subsample;
+   float fdtbl_Y[64], fdtbl_UV[64];
+   unsigned char YTable[64], UVTable[64];
+
+   if(!data || !width || !height || comp > 4 || comp < 1) {
+      return 0;
+   }
+
+   quality = quality ? quality : 90;
+   subsample = quality <= 90 ? 1 : 0;
+   quality = quality < 1 ? 1 : quality > 100 ? 100 : quality;
+   quality = quality < 50 ? 5000 / quality : 200 - quality * 2;
+
+   for(i = 0; i < 64; ++i) {
+      int uvti, yti = (YQT[i]*quality+50)/100;
+      YTable[stbiw__jpg_ZigZag[i]] = (unsigned char) (yti < 1 ? 1 : yti > 255 ? 255 : yti);
+      uvti = (UVQT[i]*quality+50)/100;
+      UVTable[stbiw__jpg_ZigZag[i]] = (unsigned char) (uvti < 1 ? 1 : uvti > 255 ? 255 : uvti);
+   }
+
+   for(row = 0, k = 0; row < 8; ++row) {
+      for(col = 0; col < 8; ++col, ++k) {
+         fdtbl_Y[k]  = 1 / (YTable [stbiw__jpg_ZigZag[k]] * aasf[row] * aasf[col]);
+         fdtbl_UV[k] = 1 / (UVTable[stbiw__jpg_ZigZag[k]] * aasf[row] * aasf[col]);
+      }
+   }
+
+   // Write Headers
+   {
+      static const unsigned char head0[] = { 0xFF,0xD8,0xFF,0xE0,0,0x10,'J','F','I','F',0,1,1,0,0,1,0,1,0,0,0xFF,0xDB,0,0x84,0 };
+      static const unsigned char head2[] = { 0xFF,0xDA,0,0xC,3,1,0,2,0x11,3,0x11,0,0x3F,0 };
+      const unsigned char head1[] = { 0xFF,0xC0,0,0x11,8,(unsigned char)(height>>8),STBIW_UCHAR(height),(unsigned char)(width>>8),STBIW_UCHAR(width),
+                                      3,1,(unsigned char)(subsample?0x22:0x11),0,2,0x11,1,3,0x11,1,0xFF,0xC4,0x01,0xA2,0 };
+      s->func(s->context, (void*)head0, sizeof(head0));
+      s->func(s->context, (void*)YTable, sizeof(YTable));
+      stbiw__putc(s, 1);
+      s->func(s->context, UVTable, sizeof(UVTable));
+      s->func(s->context, (void*)head1, sizeof(head1));
+      s->func(s->context, (void*)(std_dc_luminance_nrcodes+1), sizeof(std_dc_luminance_nrcodes)-1);
+      s->func(s->context, (void*)std_dc_luminance_values, sizeof(std_dc_luminance_values));
+      stbiw__putc(s, 0x10); // HTYACinfo
+      s->func(s->context, (void*)(std_ac_luminance_nrcodes+1), sizeof(std_ac_luminance_nrcodes)-1);
+      s->func(s->context, (void*)std_ac_luminance_values, sizeof(std_ac_luminance_values));
+      stbiw__putc(s, 1); // HTUDCinfo
+      s->func(s->context, (void*)(std_dc_chrominance_nrcodes+1), sizeof(std_dc_chrominance_nrcodes)-1);
+      s->func(s->context, (void*)std_dc_chrominance_values, sizeof(std_dc_chrominance_values));
+      stbiw__putc(s, 0x11); // HTUACinfo
+      s->func(s->context, (void*)(std_ac_chrominance_nrcodes+1), sizeof(std_ac_chrominance_nrcodes)-1);
+      s->func(s->context, (void*)std_ac_chrominance_values, sizeof(std_ac_chrominance_values));
+      s->func(s->context, (void*)head2, sizeof(head2));
+   }
+
+   // Encode 8x8 macroblocks
+   {
+      static const unsigned short fillBits[] = {0x7F, 7};
+      int DCY=0, DCU=0, DCV=0;
+      int bitBuf=0, bitCnt=0;
+      // comp == 2 is grey+alpha (alpha is ignored)
+      int ofsG = comp > 2 ? 1 : 0, ofsB = comp > 2 ? 2 : 0;
+      const unsigned char *dataR = (const unsigned char *)data;
+      const unsigned char *dataG = dataR + ofsG;
+      const unsigned char *dataB = dataR + ofsB;
+      int x, y, pos;
+      if(subsample) {
+         for(y = 0; y < height; y += 16) {
+            for(x = 0; x < width; x += 16) {
+               float Y[256], U[256], V[256];
+               for(row = y, pos = 0; row < y+16; ++row) {
+                  // row >= height => use last input row
+                  int clamped_row = (row < height) ? row : height - 1;
+                  int base_p = (stbi__flip_vertically_on_write ? (height-1-clamped_row) : clamped_row)*width*comp;
+                  for(col = x; col < x+16; ++col, ++pos) {
+                     // if col >= width => use pixel from last input column
+                     int p = base_p + ((col < width) ? col : (width-1))*comp;
+                     float r = dataR[p], g = dataG[p], b = dataB[p];
+                     Y[pos]= +0.29900f*r + 0.58700f*g + 0.11400f*b - 128;
+                     U[pos]= -0.16874f*r - 0.33126f*g + 0.50000f*b;
+                     V[pos]= +0.50000f*r - 0.41869f*g - 0.08131f*b;
+                  }
+               }
+               DCY = stbiw__jpg_processDU(s, &bitBuf, &bitCnt, Y+0,   16, fdtbl_Y, DCY, YDC_HT, YAC_HT);
+               DCY = stbiw__jpg_processDU(s, &bitBuf, &bitCnt, Y+8,   16, fdtbl_Y, DCY, YDC_HT, YAC_HT);
+               DCY = stbiw__jpg_processDU(s, &bitBuf, &bitCnt, Y+128, 16, fdtbl_Y, DCY, YDC_HT, YAC_HT);
+               DCY = stbiw__jpg_processDU(s, &bitBuf, &bitCnt, Y+136, 16, fdtbl_Y, DCY, YDC_HT, YAC_HT);
+
+               // subsample U,V
+               {
+                  float subU[64], subV[64];
+                  int yy, xx;
+                  for(yy = 0, pos = 0; yy < 8; ++yy) {
+                     for(xx = 0; xx < 8; ++xx, ++pos) {
+                        int j = yy*32+xx*2;
+                        subU[pos] = (U[j+0] + U[j+1] + U[j+16] + U[j+17]) * 0.25f;
+                        subV[pos] = (V[j+0] + V[j+1] + V[j+16] + V[j+17]) * 0.25f;
+                     }
+                  }
+                  DCU = stbiw__jpg_processDU(s, &bitBuf, &bitCnt, subU, 8, fdtbl_UV, DCU, UVDC_HT, UVAC_HT);
+                  DCV = stbiw__jpg_processDU(s, &bitBuf, &bitCnt, subV, 8, fdtbl_UV, DCV, UVDC_HT, UVAC_HT);
+               }
+            }
+         }
+      } else {
+         for(y = 0; y < height; y += 8) {
+            for(x = 0; x < width; x += 8) {
+               float Y[64], U[64], V[64];
+               for(row = y, pos = 0; row < y+8; ++row) {
+                  // row >= height => use last input row
+                  int clamped_row = (row < height) ? row : height - 1;
+                  int base_p = (stbi__flip_vertically_on_write ? (height-1-clamped_row) : clamped_row)*width*comp;
+                  for(col = x; col < x+8; ++col, ++pos) {
+                     // if col >= width => use pixel from last input column
+                     int p = base_p + ((col < width) ? col : (width-1))*comp;
+                     float r = dataR[p], g = dataG[p], b = dataB[p];
+                     Y[pos]= +0.29900f*r + 0.58700f*g + 0.11400f*b - 128;
+                     U[pos]= -0.16874f*r - 0.33126f*g + 0.50000f*b;
+                     V[pos]= +0.50000f*r - 0.41869f*g - 0.08131f*b;
+                  }
+               }
+
+               DCY = stbiw__jpg_processDU(s, &bitBuf, &bitCnt, Y, 8, fdtbl_Y,  DCY, YDC_HT, YAC_HT);
+               DCU = stbiw__jpg_processDU(s, &bitBuf, &bitCnt, U, 8, fdtbl_UV, DCU, UVDC_HT, UVAC_HT);
+               DCV = stbiw__jpg_processDU(s, &bitBuf, &bitCnt, V, 8, fdtbl_UV, DCV, UVDC_HT, UVAC_HT);
+            }
+         }
+      }
+
+      // Do the bit alignment of the EOI marker
+      stbiw__jpg_writeBits(s, &bitBuf, &bitCnt, fillBits);
+   }
+
+   // EOI
+   stbiw__putc(s, 0xFF);
+   stbiw__putc(s, 0xD9);
+
+   return 1;
+}
+
+STBIWDEF int stbi_write_jpg_to_func(stbi_write_func *func, void *context, int x, int y, int comp, const void *data, int quality)
+{
+   stbi__write_context s = { 0 };
+   stbi__start_write_callbacks(&s, func, context);
+   return stbi_write_jpg_core(&s, x, y, comp, (void *) data, quality);
+}
+
+
+#ifndef STBI_WRITE_NO_STDIO
+STBIWDEF int stbi_write_jpg(char const *filename, int x, int y, int comp, const void *data, int quality)
+{
+   stbi__write_context s = { 0 };
+   if (stbi__start_write_file(&s,filename)) {
+      int r = stbi_write_jpg_core(&s, x, y, comp, data, quality);
+      stbi__end_write_file(&s);
+      return r;
+   } else
+      return 0;
+}
+#endif
+
+#endif // STB_IMAGE_WRITE_IMPLEMENTATION
+
+/* Revision history
+      1.16  (2021-07-11)
+             make Deflate code emit uncompressed blocks when it would otherwise expand
+             support writing BMPs with alpha channel
+      1.15  (2020-07-13) unknown
+      1.14  (2020-02-02) updated JPEG writer to downsample chroma channels
+      1.13
+      1.12
+      1.11  (2019-08-11)
+
+      1.10  (2019-02-07)
+             support utf8 filenames in Windows; fix warnings and platform ifdefs
+      1.09  (2018-02-11)
+             fix typo in zlib quality API, improve STB_I_W_STATIC in C++
+      1.08  (2018-01-29)
+             add stbi__flip_vertically_on_write, external zlib, zlib quality, choose PNG filter
+      1.07  (2017-07-24)
+             doc fix
+      1.06 (2017-07-23)
+             writing JPEG (using Jon Olick's code)
+      1.05   ???
+      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)
+             STBIW_REALLOC_SIZED: support allocators with no realloc support
+             avoid race-condition in crc initialization
+             minor compile issues
+      1.00 (2015-09-14)
+             installable file IO function
+      0.99 (2015-09-13)
+             warning fixes; TGA rle support
+      0.98 (2015-04-08)
+             added STBIW_MALLOC, STBIW_ASSERT etc
+      0.97 (2015-01-18)
+             fixed HDR asserts, rewrote HDR rle logic
+      0.96 (2015-01-17)
+             add HDR output
+             fix monochrome BMP
+      0.95 (2014-08-17)
+             add monochrome TGA output
+      0.94 (2014-05-31)
+             rename private functions to avoid conflicts with stb_image.h
+      0.93 (2014-05-27)
+             warning fixes
+      0.92 (2010-08-01)
+             casts to unsigned char to fix warnings
+      0.91 (2010-07-17)
+             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.
+------------------------------------------------------------------------------
+*/

Engine/source/gfx/gFont.cpp

@@ -699,7 +699,8 @@ bool GFont::read(Stream& io_rStream)
    for(i = 0; i < numSheets; i++)
    {
        GBitmap *bmp = new GBitmap;
-       if(!bmp->readBitmap("png", io_rStream))
+       String path = String::ToString("%s/%s %d %d (%s).png", Con::getVariable("$GUI::fontCacheDirectory"), mFaceName.c_str(), mSize, i, getCharSetName(mCharSet));
+       if(!bmp->readBitmap("png", path))
        {
            delete bmp;
            return false;
@@ -775,8 +776,11 @@ bool GFont::write(Stream& stream)
    }
 
    stream.write(mTextureSheets.size());
-   for(i = 0; i < mTextureSheets.size(); i++)
-       mTextureSheets[i].getBitmap()->writeBitmap("png", stream);
+   for (i = 0; i < mTextureSheets.size(); i++)
+   {
+      String path = String::ToString("%s/%s %d %d (%s).png", Con::getVariable("$GUI::fontCacheDirectory"), mFaceName.c_str(), mSize, i, getCharSetName(mCharSet));
+      mTextureSheets[i].getBitmap()->writeBitmap("png", path);
+   }
 
    stream.write(mCurX);
    stream.write(mCurY);
@@ -865,20 +869,9 @@ void GFont::exportStrip(const char *fileName, U32 padding, U32 kerning)
       // Advance.
       curWidth +=  mCharInfoList[i].width + kerning + 2*padding;
    }
-
-   // Write the image!
-   FileStream fs;
-   
-   fs.open( fileName, Torque::FS::File::Write );
-
-   if(fs.getStatus() != Stream::Ok)
-   {
-      Con::errorf("GFont::exportStrip - failed to open '%s' for writing.", fileName);
-      return;
-   }
  
    // Done!
-   gb.writeBitmap("png", fs);
+   gb.writeBitmap("png", fileName);
 }
 
 void  GFont::setPlatformFont(PlatformFont *inPlatformFont)

Engine/source/gfx/gfxTextureObject.cpp

@@ -246,14 +246,10 @@ U32 GFXTextureObject::getEstimatedSizeInBytes() const
 
 bool GFXTextureObject::dumpToDisk( const String &bmType, const String &path )
 {   
-   FileStream stream;
-   if ( !stream.open( path, Torque::FS::File::Write ) )
-      return false;
-
    if ( mBitmap )
-      return mBitmap->writeBitmap( bmType, stream );
+      return mBitmap->writeBitmap( bmType, path);
 
    GBitmap bitmap( getWidth(), getHeight(), false, getFormat() );
    copyToBmp( &bitmap );
-   return bitmap.writeBitmap( bmType, stream );
+   return bitmap.writeBitmap( bmType, path);
 }

Engine/source/gfx/screenshot.cpp

@@ -126,89 +126,75 @@ void ScreenShot::capture( GuiCanvas *canvas )
 
    // Open up the file on disk.
    dSprintf( filename, 256, "%s.%s", mFilename, "png" );
-   FileStream fs;
-   if ( !fs.open( filename, Torque::FS::File::Write ) )
-      Con::errorf( "ScreenShot::capture() - Failed to open output file '%s'!", filename );
 
-   // Open a PNG stream for the final image
-   DeferredPNGWriter pngWriter;
-   pngWriter.begin(outBuffer->getFormat(), outBuffer->getWidth(), canvasSize.y * mTiles - overlapPixels.y * mTiles * 2, fs, 0);
-   
-   // Render each tile to generate a huge screenshot.
-   for( U32 ty=0; ty < mTiles; ty++ )
+   //// Render each tile to generate a huge screenshot.
+   for (U32 ty = 0; ty < mTiles; ty++)
    {
-      for( S32 tx=0; tx < mTiles; tx++ )
+      for (S32 tx = 0; tx < mTiles; tx++)
       {
          // Set the current tile offset for tileFrustum().
-         mCurrTile.set( tx, mTiles - ty - 1 );
+         mCurrTile.set(tx, mTiles - ty - 1);
 
          // Let the canvas render the scene.
-         canvas->renderFrame( false );
-
+         canvas->renderFrame(false);
          // Now grab the current back buffer.
-         GBitmap *gb = _captureBackBuffer();
+         GBitmap* gb = _captureBackBuffer();
 
-			// The current GFX device couldn't capture the backbuffer or it's unable of doing so.
-			if (gb == NULL)
-				return;
+         // The current GFX device couldn't capture the backbuffer or it's unable of doing so.
+         if (gb == NULL)
+            return;
 
-                  
          // Copy the captured bitmap into its tile
          // within the output bitmap.         
-         const U32 inStride = gb->getWidth() * gb->getBytesPerPixel();         
-         const U8 *inColor = gb->getBits() + inStride * overlapPixels.y;
+         const U32 inStride = gb->getWidth() * gb->getBytesPerPixel();
+         const U8* inColor = gb->getBits() + inStride * overlapPixels.y;
          const U32 outStride = outBuffer->getWidth() * outBuffer->getBytesPerPixel();
-         const U32 inOverlapOffset = overlapPixels.x * gb->getBytesPerPixel();         
-         const U32 inOverlapStride = overlapPixels.x * gb->getBytesPerPixel()*2;
-         const U32 outOffset = (tx * (gb->getWidth() - overlapPixels.x*2 )) * gb->getBytesPerPixel();
-         U8 *outColor = outBuffer->getWritableBits() + outOffset;
-         for( U32 row=0; row < gb->getHeight() - overlapPixels.y; row++ )
+         const U32 inOverlapOffset = overlapPixels.x * gb->getBytesPerPixel();
+         const U32 inOverlapStride = overlapPixels.x * gb->getBytesPerPixel() * 2;
+         const U32 outOffset = (tx * (gb->getWidth() - overlapPixels.x * 2)) * gb->getBytesPerPixel();
+         U8 * outColor = outBuffer->getWritableBits() + outOffset;
+         for (U32 row = 0; row < gb->getHeight() - overlapPixels.y; row++)
          {
-            dMemcpy( outColor, inColor + inOverlapOffset, inStride - inOverlapStride );
-            
-            //Grandient blend the left overlap area of this tile over the previous tile left border
+            dMemcpy(outColor, inColor + inOverlapOffset, inStride - inOverlapStride);
+
+            //Grandient blend the left overlap area of this tile over the previous tile left borde
             if (tx && !(ty && row < overlapPixels.y))
             {
-               U8 *blendOverlapSrc = (U8*)inColor;
-               U8 *blendOverlapDst = outColor - inOverlapOffset;
-               for ( U32 px=0; px < overlapPixels.x; px++)
+               U8* blendOverlapSrc = (U8*)inColor;
+               U8* blendOverlapDst = outColor - inOverlapOffset;
+               for (U32 px = 0; px < overlapPixels.x; px++)
                {
                   F32 blendFactor = (F32)px / (F32)overlapPixels.x;
-                  sBlendPixelRGB888(blendOverlapSrc, blendOverlapDst, blendFactor);                 
+                  sBlendPixelRGB888(blendOverlapSrc, blendOverlapDst, blendFactor);
 
                   blendOverlapSrc += gb->getBytesPerPixel();
-                  blendOverlapDst += outBuffer->getBytesPerPixel();                   
-               }               
+                  blendOverlapDst += outBuffer->getBytesPerPixel();
+               }
             }
 
-            //Gradient blend against the rows the excess overlap rows already in the buffer            
+            //Gradient blend against the rows the excess overlap rows already in the buffer       
             if (ty && row < overlapPixels.y)
             {
                F32 rowBlendFactor = (F32)row / (F32)overlapPixels.y;
-               U8 *blendSrc = outColor + outStride * (outBuffer->getHeight() - overlapPixels.y);
-               U8 *blendDst = outColor;               
-               for ( U32 px=0; px < gb->getWidth() - overlapPixels.x*2; px++)
-               {                  
-                  sBlendPixelRGB888(blendSrc, blendDst, 1.0-rowBlendFactor); 
+               U8* blendSrc = outColor + outStride * (outBuffer->getHeight() - overlapPixels.y);
+               U8* blendDst = outColor;
+               for (U32 px = 0; px < gb->getWidth() - overlapPixels.x * 2; px++)
+               {
+                  sBlendPixelRGB888(blendSrc, blendDst, 1.0 - rowBlendFactor);
                   blendSrc += gb->getBytesPerPixel();
-                  blendDst += outBuffer->getBytesPerPixel();                   
-               }                              
+                  blendDst += outBuffer->getBytesPerPixel();
+               }
             }
 
-            
             inColor += inStride;
             outColor += outStride;
          }
 
          delete gb;
       }
-
-      // Write the captured tile row into the PNG stream
-      pngWriter.append(outBuffer, outBuffer->getHeight()-overlapPixels.y);
    }
 
-   //Close the PNG stream
-   pngWriter.end();
+   outBuffer->writeBitmap("png", filename);
    
    // We captured... clear the flag.
    mPending = false;
@@ -234,20 +220,11 @@ void ScreenShot::_singleCapture( GuiCanvas *canvas )
    char filename[256];
    dSprintf( filename, 256, "%s.%s", mFilename, mWriteJPG ? "jpg" : "png" );
 
-   // Open up the file on disk.
-   FileStream fs;
-   if ( !fs.open( filename, Torque::FS::File::Write ) )
-      Con::errorf( "ScreenShot::_singleCapture() - Failed to open output file '%s'!", filename );
+   // Write it and close.
+   if ( mWriteJPG )
+      bitmap->writeBitmap( "jpg", filename);
    else
-   {
-      // Write it and close.
-      if ( mWriteJPG )
-         bitmap->writeBitmap( "jpg", fs );
-      else
-         bitmap->writeBitmap( "png", fs );
-
-      fs.close();
-   }
+      bitmap->writeBitmap( "png", filename);
 
    // Cleanup.
    delete bitmap;

Engine/source/gfx/video/videoEncoderPNG.cpp

@@ -44,18 +44,12 @@ public:
    /// Pushes a new frame into the video stream
    bool pushFrame( GBitmap * bitmap )
    {
-      FileStream fs;
       String framePath = mPath + String::ToString("%.6u.png", mCurrentFrame);
-      if ( !fs.open( framePath, Torque::FS::File::Write ) )
-      {
-         Con::errorf( "VideoEncoderPNG::pushFrame() - Failed to open output file '%s'!", framePath.c_str() );
-         return false;
-      }
 
       //Increment
       mCurrentFrame++;
 
-      bool result = bitmap->writeBitmap("png", fs, 0);
+      bool result = bitmap->writeBitmap("png", framePath, 0);
       pushProcessedBitmap(bitmap);
  
       return result;
@@ -73,4 +67,4 @@ public:
    }
 };
 
-REGISTER_VIDEO_ENCODER(VideoEncoderPNG, PNG)
+REGISTER_VIDEO_ENCODER(VideoEncoderPNG, PNG)

Engine/source/gui/containers/guiWindowCtrl.cpp

@@ -79,7 +79,7 @@ GuiWindowCtrl::GuiWindowCtrl()
       mResizeEdge(edgeNone),
       mResizeHeight(true),
       mCanMove(true),
-      mResizeMargin(2.f),
+      mResizeMargin(5.f),
       mCanClose(true),
       mCanMinimize(true),
       mCanMaximize(true),
@@ -143,6 +143,8 @@ void GuiWindowCtrl::initPersistFields()
          "Whether the window can be resized horizontally." );
       addField( "resizeHeight",      TypeBool,         Offset( mResizeHeight, GuiWindowCtrl ),
          "Whether the window can be resized vertically." );
+      addField("resizeMargin", TypeF32, Offset(mResizeMargin, GuiWindowCtrl),
+         "Margin along the window edge to allow grabbing.");
       addField( "canMove",           TypeBool,         Offset( mCanMove, GuiWindowCtrl ),
          "Whether the window can be moved by dragging its titlebar." );
       addField( "canClose",          TypeBool,         Offset( mCanClose, GuiWindowCtrl ),
@@ -1490,7 +1492,7 @@ const RectI GuiWindowCtrl::getClientRect()
    // Finally, inset it by padding
    // Inset by padding.  margin is specified for all t/b/l/r but 
    // uses only pointx pointy uniformly on both ends. This should be fixed. - JDD
-   // winRect.inset( mSizingOptions.mPadding.point.x, mSizingOptions.mPadding.point.y );
+   winRect.inset(mResizeMargin, mResizeMargin);
 
    return winRect;
 }

Engine/source/lighting/basic/blTerrainSystem.cpp

@@ -71,7 +71,7 @@ bool blTerrainChunk::read(Stream & stream)
       return(false);
 
    mLightmap = new GBitmap();
-   return mLightmap->readBitmap("png",stream);
+   return false;//mLightmap->readBitmap("png",stream);
 }
 
 bool blTerrainChunk::write(Stream & stream)
@@ -82,8 +82,8 @@ bool blTerrainChunk::write(Stream & stream)
    if(!mLightmap)
       return(false);
 
-   if(!mLightmap->writeBitmap("png",stream))
-      return(false);
+   //if(!mLightmap->writeBitmap("png",stream))
+      //return(false);
 
    return(true);
 }

Engine/source/sfx/sfxMemoryStream.cpp

@@ -39,9 +39,10 @@ SFXMemoryStream::SFXMemoryStream( const SFXFormat& format,
 
 void SFXMemoryStream::reset()
 {
-   if( dynamic_cast< IResettable* >( getSourceStream() ) )
+   IResettable* rStream = dynamic_cast<IResettable*>(getSourceStream());
+   if(rStream )
    {
-      reinterpret_cast< IResettable* >( getSourceStream() )->reset();
+      rStream->reset();
       
       if( mCurrentPacket )
          destructSingle( mCurrentPacket );

Engine/source/terrain/terrExport.cpp

@@ -67,14 +67,7 @@ bool TerrainBlock::exportHeightMap( const UTF8 *filePath, const String &format )
       }
    }
 
-   FileStream stream;
-   if ( !stream.open( filePath, Torque::FS::File::Write ) )
-   {
-      Con::errorf( "TerrainBlock::exportHeightMap() - Error opening file for writing: %s !", filePath );
-      return false;
-   }
-
-   if ( !output.writeBitmap( format, stream ) )
+   if ( !output.writeBitmap( format, filePath) )
    {
       Con::errorf( "TerrainBlock::exportHeightMap() - Error writing %s: %s !", format.c_str(), filePath );
       return false;
@@ -120,14 +113,7 @@ bool TerrainBlock::exportLayerMaps( const UTF8 *filePrefix, const String &format
       UTF8 filePath[1024];
       dSprintf( filePath, 1024, "%s_%d_%s.%s", filePrefix, i, mFile->mMaterials[i]->getInternalName(), format.c_str() );
 
-      FileStream stream;
-      if ( !stream.open( filePath, Torque::FS::File::Write ) )
-      {
-         Con::errorf( "TerrainBlock::exportLayerMaps() - Error opening file for writing: %s !", filePath );
-         return false;
-      }
-
-      if ( !output.writeBitmap( format, stream ) )
+      if ( !output.writeBitmap( format, filePath) )
       {
          Con::errorf( "TerrainBlock::exportLayerMaps() - Error writing %s: %s !", format.c_str(), filePath );
          return false;

Engine/source/terrain/terrRender.cpp

@@ -477,16 +477,9 @@ void TerrainBlock::_updateBaseTexture(bool writeToCache)
    }
    else
    {
-      FileStream stream;
-      if (!stream.open(_getBaseTexCacheFileName(), Torque::FS::File::Write))
-      {
-         mBaseTex = blendTex;
-         return;
-      }
-
       GBitmap bitmap(blendTex->getWidth(), blendTex->getHeight(), false, GFXFormatR8G8B8A8);
       blendTex->copyToBmp(&bitmap);
-      bitmap.writeBitmap(formatToExtension(mBaseTexFormat), stream);
+      bitmap.writeBitmap(formatToExtension(mBaseTexFormat), _getBaseTexCacheFileName());
    }
 }
 

Engine/source/ts/tsLastDetail.cpp

@@ -504,14 +504,10 @@ void TSLastDetail::_update()
       String imposterPath = _getDiffuseMapPath();
       String normalsPath = _getNormalMapPath();
 
-      FileStream stream;
-      if ( stream.open( imposterPath, Torque::FS::File::Write  ) )
-         destBmp.writeBitmap( "png", stream );
-      stream.close();
-
-      if ( stream.open( normalsPath, Torque::FS::File::Write ) )
-         destNormal.writeBitmap( "png", stream );
-      stream.close();
+      if (!destBmp.writeBitmap("png", imposterPath))
+         Con::errorf("TSLastDetail::_update() - failed to write imposter %s", imposterPath.c_str());
+      if (!destNormal.writeBitmap("png", normalsPath))
+         Con::errorf("TSLastDetail::_update() - failed to write normal %s", normalsPath.c_str());
    }
 
    // DEBUG: Some code to force usage of a test image.

Engine/source/util/imposterCapture.cpp

@@ -316,16 +316,9 @@ void ImposterCapture::_separateAlpha( GBitmap *imposterOut )
       
       if ( 0 )
       {
-         FileStream fs;
-         if ( fs.open( "./imposterout.png", Torque::FS::File::Write ) )
-            imposterOut->writeBitmap( "png", fs );
+         imposterOut->writeBitmap("png", "./imposterout.png");
 
-         fs.close();
-
-         if ( fs.open( "./temp.png", Torque::FS::File::Write ) )
-            bmp->writeBitmap( "png", fs );
-
-         fs.close();
+         bmp->writeBitmap("png", "./temp.png");
       }
    
 
@@ -482,26 +475,13 @@ void ImposterCapture::capture(   const MatrixF &rotMatrix,
    if ( 0 )
    {
       // Render out the bitmaps for debug purposes.
-      FileStream fs;
-      if ( fs.open( "./blackbmp.png", Torque::FS::File::Write ) )
-         mBlackBmp->writeBitmap( "png", fs );
-
-      fs.close();
-
-      if ( fs.open( "./whitebmp.png", Torque::FS::File::Write ) )
-         mWhiteBmp->writeBitmap( "png", fs );
-
-      fs.close();
-
-      if ( fs.open( "./normalbmp.png", Torque::FS::File::Write ) )
-         (*normalMapOut)->writeBitmap( "png", fs );
+      mBlackBmp->writeBitmap( "png", "./blackbmp.png" );
 
-      fs.close();
+      mWhiteBmp->writeBitmap( "png", "./whitebmp.png" );
 
-      if ( fs.open( "./finalimposter.png", Torque::FS::File::Write ) )
-         (*imposterOut)->writeBitmap( "png", fs );
+      (*normalMapOut)->writeBitmap( "png", "./normalbmp.png" );
 
-      fs.close();
+      (*imposterOut)->writeBitmap( "png", "./finalimposter.png" );
    }
 }
 

Templates/BaseGame/game/data/Prototyping/Prototyping.tscript

@@ -37,6 +37,9 @@ function Prototyping::onDestroyGameServer(%this)
 //This is called when the client is initially set up by the game application
 function Prototyping::initClient(%this)
 {
+   //class method prototyping
+   %this.queueExec("./UI/classPrototyping");
+   %this.queueExec("./UI/classPrototyping.gui");
 }
 
 //This is called when a client connects to a server

Templates/BaseGame/game/data/Prototyping/UI/classPrototyping.asset.taml

@@ -0,0 +1,5 @@
+<GUIAsset
+    AssetName="classPrototyping"
+    ScriptFile="@assetFile=classPrototyping.tscript"
+    GUIFile="@assetFile=classPrototyping.gui"
+    VersionId="1"/>

Templates/BaseGame/game/data/Prototyping/UI/classPrototyping.gui

@@ -0,0 +1,115 @@
+//--- OBJECT WRITE BEGIN ---
+$guiContent = new GuiControl(classPrototyping) {
+   extent = "1024 768";
+   profile = "GuiDefaultProfile";
+   tooltipProfile = "GuiToolTipProfile";
+   isContainer = "1";
+   canSaveDynamicFields = "1";
+      originalAssetName = "classPrototyping";
+
+   new GuiWindowCtrl() {
+      text = "Class Prototyping";
+      position = "216 124";
+      extent = "592 519";
+      horizSizing = "center";
+      vertSizing = "center";
+      profile = "ToolsGuiWindowProfile";
+      tooltipProfile = "GuiToolTipProfile";
+      closeCommand = "Canvas.popDialog(classPrototyping);";
+      new GuiScrollCtrl() {
+         lockVertScroll = "1";
+         position = "14 30";
+         extent = "564 33";
+         profile = "ToolsGuiScrollProfile";
+         tooltipProfile = "GuiToolTipProfile";
+
+         new GuiDynamicCtrlArrayControl(ClassInheritanceListCtrl) {
+            colCount = "1";
+            colSize = "80";
+            rowCount = "1";
+            rowSize = "18";
+            autoCellSize = "1";
+            fillRowFirst = "0";
+            dynamicSize = "1";
+            position = "1 1";
+            extent = "400 18";
+            profile = "GuiDefaultProfile";
+            tooltipProfile = "GuiToolTipProfile";
+         };
+      };
+      new GuiTextCtrl() {
+         text = "Callbacks";
+         position = "24 66";
+         extent = "54 14";
+         profile = "ToolsGuiTextProfile";
+         tooltipProfile = "GuiToolTipProfile";
+      };
+      new GuiScrollCtrl() {
+         position = "19 80";
+         extent = "552 326";
+         profile = "ToolsGuiScrollProfile";
+         tooltipProfile = "GuiToolTipProfile";
+
+         new GuiDynamicCtrlArrayControl(ClassMethodListCtrl) {
+            colCount = "1";
+            colSize = "8000";
+            rowCount = "1";
+            rowSize = "18";
+            dynamicSize = "1";
+            extent = "552 326";
+            profile = "GuiDefaultProfile";
+            tooltipProfile = "GuiToolTipProfile";
+         };
+      };
+      new GuiBitmapButtonCtrl() {
+         BitmapAsset = "ToolsModule:iconOpen_image";
+         bitmapMode = "Centered";
+         position = "348 467";
+         extent = "22 22";
+         horizSizing = "left";
+         profile = "ToolsGuiButtonProfile";
+         command = "SelectAssetPath.showDialog(AssetBrowser.dirHandler.currentAddress, \"setProtoTypeFilePath\");\nSelectAssetPathWindow.selectWindow();";
+         tooltipProfile = "GuiToolTipProfile";
+         tooltip = "New Module";
+      };
+      new GuiTextEditCtrl() {
+         text = "data/ExampleModule";
+         position = "143 470";
+         extent = "201 20";
+         horizSizing = "width";
+         profile = "ToolsGuiTextEditProfile";
+         active = "0";
+         tooltipProfile = "GuiToolTipProfile";
+         internalName = "targetPath";
+      };
+      new GuiTextCtrl() {
+         text = "Target Path:";
+         position = "20 470";
+         extent = "116 17";
+         profile = "ToolsGuiTextProfile";
+         tooltipProfile = "GuiToolTipProfile";
+      };
+      new GuiButtonCtrl() {
+         text = "Save";
+         position = "431 465";
+         profile = "ToolsGuiButtonProfile";
+         command = "classPrototyping.writeResults();";
+         tooltipProfile = "GuiToolTipProfile";
+      };
+      new GuiCheckBoxCtrl(ReportCommands) {
+         text = "Report Commands";
+         position = "16 420";
+         extent = "125 30";
+         profile = "ToolsGuiCheckBoxProfile";
+         tooltipProfile = "GuiToolTipProfile";
+      };
+      new GuiCheckBoxCtrl(ReportVariables) {
+         text = "Report Stock Variables";
+         position = "152 420";
+         extent = "125 30";
+         profile = "ToolsGuiCheckBoxProfile";
+         tooltipProfile = "GuiToolTipProfile";
+      };
+   };
+};
+//--- OBJECT WRITE END ---

Templates/BaseGame/game/data/Prototyping/UI/classPrototyping.tscript

@@ -0,0 +1,204 @@
+function classPrototyping::onWake(%this)
+{
+    
+}
+
+function classPrototyping::onSleep(%this)
+{
+    
+}
+//PrototypeClass(MainMenuGui)
+//PrototypeClass(GuiChunkedBitmapCtrl)
+function PrototypeClass(%classInstance)
+{
+    if (!isObject(%classInstance)) return;
+    Canvas.pushDialog(classPrototyping);
+    classPrototyping.fillClasslist(%classInstance);
+    classPrototyping.SetNamespaceUsed(%classInstance);
+}
+
+function classPrototyping::fillClasslist(%this, %classInstance)
+{
+    ClassInheritanceListCtrl.deleteAllObjects();
+    %this.instanceName = %classInstance.getName();
+    
+    //get potentially scripted namespaces
+    %class = %classInstance.getClassName();    
+    %prepend = "";
+    if (%classInstance.getName() !$= "")
+        %prepend = %classInstance.getName();
+    if (%classInstance.class !$= "")
+        %prepend = %prepend SPC %classInstance.class;        
+    if (%classInstance.superclass !$= "")
+        %prepend = %prepend SPC %classInstance.superclass;
+        
+    //append to hardcoded potential namespaces
+    %this.classlist = %prepend SPC getClassHierarchy(%class);
+    %this.classCount = getWordCount(%this.classlist);
+    for (%i=0; %i<%this.classCount; %i++)
+    {
+        %inheritanceOrder = %this.classCount-(%i+1);
+        %className = getWord(%this.classlist,%inheritanceOrder);
+        if (%i<%this.classCount-1)
+            %className = %className @"->";
+        %elemClass = new GuiRadioCtrl("ProtoClassSelect"@ %i) {
+               text = %className;
+               entry = strreplace(%className,"->","");
+               groupNum = "1";
+               extent = "80 18";
+               profile = "ToolsGuiRadioProfile";
+               tooltipProfile = "GuiToolTipProfile";
+               };
+        eval("function ProtoClassSelect"@ %i @"::onClick(%this){classPrototyping.SetNamespaceUsed(%this.entry);}");
+        ClassInheritanceListCtrl.addGuiControl(%elemClass);
+    }
+    %lastElem = "ProtoClassSelect"@ %this.classCount-1;
+    %lastElem.setStateOn(true);
+}
+
+function classPrototyping::SetNamespaceUsed(%this, %nameSpaceUsed)
+{
+    ClassMethodListCtrl.deleteAllObjects();
+    %this.fillMethodlist(%nameSpaceUsed);
+}
+
+function classPrototyping::fillMethodlist(%this, %nameSpaceUsed)
+{
+    ClassMethodListCtrl.deleteAllObjects();
+    %this.nameSpaceUsed = %nameSpaceUsed;
+    %this.methodArray = getMethodSigsNS(%nameSpaceUsed);
+    %this.methodCount = %this.methodArray.count();
+    
+    for (%i=0; %i<%this.methodCount; %i++)
+    {
+         %methodDef = getRecord(%this.methodArray.getValue(%i),0);
+         %methodName = strreplace(%methodDef,"::"," ");
+         %methodName = getWord(strreplace(%methodName,"("," "),2);
+         
+        %elemMethod = new GuiCheckBoxCtrl("ProtoMethodSelect"@ %i) {
+            text = %methodName;
+            position = "1 1";
+            profile = "ToolsGuiCheckBoxProfile";
+            tooltipProfile = "GuiToolTipProfile";
+         };
+        ClassMethodListCtrl.addGuiControl(%elemMethod);
+    }
+}
+
+function setProtoTypeFilePath(%targetPath)
+{
+   classPrototyping-->targetPath.text = %targetPath;
+}
+
+function classPrototyping::readExistingLayout(%this)
+{
+    for (%i=0; %i<%this.classCount; %i++)
+    {
+        %inheritanceOrder = %this.classCount-(%i+1);
+        %obj = "ProtoClassSelect"@ %i;
+        if (%obj.isStateOn())
+            %namespaceUsed = getWord(%this.classlist,%inheritanceOrder);
+    }
+    
+	%file = new FileObject();
+    %filename = classPrototyping-->targetPath.text @"/"@ %namespaceUsed @"."@ $TorqueScriptFileExtension;
+    
+    if (!isObject(%this.callbacksDefined))
+        %this.callbacksDefined = new arrayobject();
+    %this.callbacksDefined.empty();
+    %this.reportedCommands = false;
+    %this.reportedVariables = false;
+    %this.callbackBlockDefined = false;
+    
+    %key=0;
+	if(%file.openForRead(%filename))
+    {
+        while (!%file.isEof())
+        {
+            %line = %file.readLine();
+            
+            //have we already reported commands?
+            if (startsWith(%line,"/* Available Commands:") )
+                %this.reportedCommands = true;
+                
+            //have we already reported variables?
+            if (startsWith(%line,"/* HardCoded Variables") )
+                %this.reportedVariables = true;
+                
+            if (startsWith(%line,"/*--- Callbacks ---*/") )
+                %this.callbackBlockDefined = true;
+            
+            //get list of methods already existing
+            if (startswith(%line,"function "@ %namespaceUsed) )
+            {
+                %methodName = strreplace(%line,"::"," ");
+                %methodName = getWord(strreplace(%methodName,"("," "),2);
+                %this.callbacksDefined.add(%key++,%methodName);
+            }
+        }
+    }
+	%file.delete();
+}
+
+function classPrototyping::writeResults(%this)
+{
+    %namespaceUsed = "";
+    for (%i=0; %i<%this.classCount; %i++)
+    {
+        %inheritanceOrder = %this.classCount-(%i+1);
+        %obj = "ProtoClassSelect"@ %i;
+        if (%obj.isStateOn())
+            %namespaceUsed = getWord(%this.classlist,%inheritanceOrder);
+    }
+    %this.readExistingLayout();
+	%file = new FileObject();
+    %filename = classPrototyping-->targetPath.text @"/"@ %namespaceUsed @"."@ $TorqueScriptFileExtension;
+	if(%file.openForAppend(%filename))
+    {
+        if (ReportCommands.isStateOn() && %this.reportedCommands == false)
+        {
+            %this.commandArray = getMethodSigsNS(%this.nameSpaceUsed,true);
+            %this.commandCount = %this.commandArray.count();
+            %file.writeLine("/* Available Commands:");
+            for (%i=0; %i< %this.commandCount; %i++)
+            {
+                %file.writeLine(getRecord(%this.commandArray.getValue(%i),0));
+            }
+            %file.writeLine("*/");
+        }
+    
+        if (ReportVariables.isStateOn() && %this.reportedVariables == false)
+        {
+            %file.writeLine("/* HardCoded Variables");
+            for (%i=0; %i< getFieldCountNS(%this.nameSpaceUsed); %i++)
+            {
+                %file.writeLine(getFieldNS(%this.nameSpaceUsed,%i));
+            }
+            %file.writeLine("*/");
+        }
+        
+        if (%this.callbackBlockDefined == false)
+            %file.writeLine("\n/*--- Callbacks ---*/\n");
+            
+        for (%i=0; %i<%this.methodCount; %i++)
+        {
+            %obj = "ProtoMethodSelect"@ %i;
+            if (%obj.isStateOn())
+            {
+                %methodDef = getRecord(%this.methodArray.getValue(%i),0);
+                %methodName = strreplace(%methodDef,"::"," ");
+                %methodName = getWord(strreplace(%methodName,"("," "),2);
+                if (%this.callbacksDefined.countValue(%methodName)==0)
+                {
+                    echo(%methodName @ "not found. defining...");
+                    %file.writeLine("\n" @ strreplace(%this.methodArray.getValue(%i),%this.instanceName,%namespaceUsed));
+                }
+            }
+        }
+    }
+	else
+	{
+		error( "Failed to open " @ %filename @ " for writing" );
+	}
+	%file.delete();
+}

Templates/BaseGame/game/tools/assetBrowser/scripts/assetBrowser.tscript

@@ -2582,12 +2582,24 @@ function GuiEditor::onControlDropped(%this, %payload, %position)
       //dealing with an actual asset, so build the assetName
       %assetId = %payload.moduleName @ ":" @ %payload.assetName;
       %assetType = AssetDatabase.getAssetType(%assetId);
-      
-      if(%assetType $= "ImageAsset")
+ 
+      if(AssetBrowser.isMethod("on" @ %assetType @ "GUIEditorDropped"))
+      {
+        %module = %payload.moduleName;
+        %asset = %payload.assetName;
+        %assetDef = AssetDatabase.acquireAsset(%module @ ":" @ %asset);
+        %buildCommand = AssetBrowser @ ".on" @ %assetType @ "GUIEditorDropped(" @ %assetDef @ ",\"" @ %pos @ "\");";  
+        eval(%buildCommand);    
+      }
+      else
       {
-         %cmd = "return new guiBitmapCtrl();";
-         %ctrl = eval( %cmd ); 
-         %ctrl.bitmap = %assetId;
+        //fallback example
+        if(%assetType $= "ImageAsset")
+        {
+            %cmd = "return new guiBitmapCtrl();";
+            %ctrl = eval( %cmd ); 
+            %ctrl.bitmap = %assetId;
+        }
       }
    }
    else

Templates/BaseGame/game/tools/assetBrowser/scripts/assetImport.tscript

@@ -94,7 +94,7 @@ function ImportAssetWindow::onWake(%this)
 //
 function isImageFormat(%fileExt)
 {
-   if( (%fileExt $= ".png") || (%fileExt $= ".jpg") || (%fileExt $= ".bmp") || (%fileExt $= ".dds") || (%fileExt $= ".tif"))
+   if( (%fileExt $= ".png") || (%fileExt $= ".jpg") || (%fileExt $= ".bmp") || (%fileExt $= ".dds") || (%fileExt $= ".tif") || (%fileExt $= ".psd") || (%fileExt $= ".gif") || (%fileExt $= ".hdr"))
       return true;
       
    return false;

Templates/BaseGame/game/tools/assetBrowser/scripts/assetTypes/sound.tscript

@@ -91,6 +91,17 @@ function AssetBrowser::onSoundAssetEditorDropped(%this, %assetDef, %position)
    
 }
 
+function AssetBrowser::onSoundAssetGUIEditorDropped(%this, %assetDef, %position)
+{    
+   %assetId = %assetDef.getAssetId();
+   %cmd = "new GuiAudioCtrl(){";
+   %cmd = %cmd @ "SoundAsset =\""@ %assetId @"\";";
+   %cmd = %cmd @ "position =\""@ %position @"\";";
+   %cmd = %cmd @ "};";
+   %ctrl = GuiEditCanvas.createObject(%cmd);
+   echo(%ctrl SPC "created");
+}
+
 function GuiInspectorTypeSoundAssetPtr::onControlDropped( %this, %payload, %position )
 {
    Canvas.popDialog(EditorDragAndDropLayer);

Templates/BaseGame/game/tools/guiEditor/scripts/guiEditorCanvas.ed.tscript

@@ -211,6 +211,8 @@ function GuiEditCanvas::onCreateMenu(%this)
       };
    };
    
+   %this.buildAddMenu();
+   
    // Workaround (for some reason it doesn't size to the width of the canvas)
    // TODO: After a canvas resize it still messes up the width
    %position = %this.menubar.position.x SPC %this.menubar.position.y;
@@ -220,6 +222,88 @@ function GuiEditCanvas::onCreateMenu(%this)
    %this.menuBar.attachToCanvas( Canvas, 0 );
 }
 
+function GuiEditCanvas::buildAddMenu(%this)
+{
+   %addMenu = MenuBuilder::newMenu("Add");
+
+   %enumeratedClasses = enumerateConsoleClasses("GuiControl");
+   for(%c=0; %c < getFieldCount(%enumeratedClasses); %c++)
+   {
+      %class = getField(%enumeratedClasses, %c);  
+
+      if( GuiEditor.isFilteredClass( %class )
+          || !isMemberOfClass( %class, "GuiControl" ) )
+         continue;
+         
+      %category = getCategoryOfClass(%class);
+
+      if(%category $= "")
+      {
+         error("Attempted to fetch category of class " @ %class @ " but none were found.");
+         continue;  
+      }
+
+      %parentMenu = %addMenu; //start at the top
+      for(%cat=0; %cat < getFieldCount(%category); %cat++)
+      {
+         %subCat = getField(%category, %cat);
+         %targetSubmenu = %parentMenu.findMenu(%subCat);
+         if(!isObject(%targetSubmenu))
+         {
+            %targetSubmenu = %parentMenu.newSubmenu(%subCat);
+         }
+
+         %parentMenu = %targetSubmenu;
+      }
+
+      %buildfunc = "";
+      %class = %class;
+      %method = "build" @ %buildfunc;
+      if( !GuiEditCanvas.isMethod( %method ) )
+         %method = "build" @ %class;
+
+      if( !GuiEditCanvas.isMethod( %method ) )
+         %cmd = "return new " @ %class @ "();";
+      else
+         %cmd = "GuiEditCanvas." @ %method @ "();";
+
+      %createCmd = "GuiEditCanvas.newObjectCallback = \"GuiEditCanvas.onFinishCreateObject\"; GuiEditCanvas.createObject( \"" @ %cmd @ "\" );";
+
+      %targetSubmenu.newItem(%class, %createCmd, "");
+   }
+
+   MenuBuilder::addMenuToMenubar(%this.menubar, %addMenu, 4);
+}
+
+function GuiEditCanvas::createObject( %this, %cmd )
+{   
+   if(startsWith(%cmd, "return "))
+      %objId = eval(%cmd);
+   else
+      %objId = eval("return " @ %cmd);
+   
+   if( isObject( %objId ) )
+      %this.onFinishCreateObject( %objId );
+      
+   return %objId;
+}
+
+function GuiEditCanvas::onFinishCreateObject( %this, %objId )
+{
+   GuiEditor.getCurrentAddSet().add( %objId );   
+   %this.onObjectCreated( %objId );
+}
+
+function GuiEditCanvas::onObjectCreated( %this, %objId )
+{
+   // Can we submit an undo action?
+   if ( isObject( %objId ) )
+      GuiEditor.onAddNewCtrl( %objId );
+      
+   GuiEditorTreeView.clearSelection();
+   GuiEditor.select(%objId);
+}
+
 $GUI_EDITOR_MENU_EDGESNAP_INDEX = 0;
 $GUI_EDITOR_MENU_CENTERSNAP_INDEX = 1;
 $GUI_EDITOR_MENU_GUIDESNAP_INDEX = 3;

Templates/BaseGame/game/tools/guiEditor/scripts/guiEditorUndo.ed.tscript

@@ -128,6 +128,8 @@ function UndoActionAddDelete::trashObjects(%this)
    
    if( isObject( %this.tree ) )
       %this.tree.update();
+
+   GuiEditor.clearSelection();
 }
 
 function UndoActionAddDelete::restoreObjects(%this)