diff --git a/dgl/NanoVG.hpp b/dgl/NanoVG.hpp
index 93e33a24..b7c1c4d6 100644
--- a/dgl/NanoVG.hpp
+++ b/dgl/NanoVG.hpp
@@ -180,9 +180,12 @@ public:
         ALIGN_BASELINE = 1 << 6  // Align vertically to baseline (default).
     };
 
-    enum Alpha {
-        STRAIGHT_ALPHA,
-        PREMULTIPLIED_ALPHA
+    enum ImageFlags {
+        IMAGE_GENERATE_MIPMAPS = 1 << 0, // Generate mipmaps during creation of the image.
+        IMAGE_REPEAT_X         = 1 << 1, // Repeat image in X direction.
+        IMAGE_REPEAT_Y         = 1 << 2, // Repeat image in Y direction.
+        IMAGE_FLIP_Y           = 1 << 3, // Flips (inverses) image in Y direction when rendered.
+        IMAGE_PREMULTIPLIED    = 1 << 4, // Image data has premultiplied alpha.
     };
 
     enum LineCap {
@@ -193,12 +196,6 @@ public:
         MITER
     };
 
-    enum PatternRepeat {
-        REPEAT_NONE = 0x0, // No repeat
-        REPEAT_X    = 0x1, // Repeat in X direction
-        REPEAT_Y    = 0x2  // Repeat in Y direction
-    };
-
     enum Solidity {
         SOLID = 1, // CCW
         HOLE  = 2  // CW
@@ -217,7 +214,6 @@ public:
         Color innerColor;
         Color outerColor;
         int   imageId;
-        PatternRepeat repeat;
 
         Paint() noexcept;
 
@@ -246,14 +242,8 @@ public:
 
    /**
       Constructor.
-      Uses 512x512 as default atlas size.
-    */
-    NanoVG();
-
-   /**
-      Constructor using custom text atlas size.
     */
-    NanoVG(const int textAtlasWidth, const int textAtlasHeight);
+    NanoVG(int textAtlasWidth = 512, int textAtlasHeight = 512);
 
    /**
       Destructor.
@@ -273,13 +263,18 @@ public:
       Begin drawing a new frame.
       @param withAlha Controls if drawing the shapes to the render target should be done using straight or pre-multiplied alpha.
     */
-    void beginFrame(const uint width, const uint height, const float scaleFactor = 1.0f, const Alpha alpha = PREMULTIPLIED_ALPHA);
+    void beginFrame(const uint width, const uint height, const float scaleFactor = 1.0f);
 
    /**
       Begin drawing a new frame inside a widget.
     */
     void beginFrame(Widget* const widget);
 
+   /**
+      Cancels drawing the current frame.
+    */
+    void cancelFrame();
+
    /**
       Ends drawing flushing remaining render state.
     */
@@ -374,6 +369,12 @@ public:
     */
     void lineJoin(LineCap join = MITER);
 
+   /**
+      Sets the transparency applied to all rendered shapes.
+      Already transparent paths will get proportionally more transparent as well.
+    */
+    void globalAlpha(float alpha);
+
    /* --------------------------------------------------------------------
     * Transforms */
 
@@ -539,7 +540,7 @@ public:
       and repeat tells if the image should be repeated across x or y.
       The gradient is transformed by the current transform when it is passed to fillPaint() or strokePaint().
     */
-    Paint imagePattern(float ox, float oy, float ex, float ey, float angle, const NanoImage* image, PatternRepeat repeat);
+    //Paint imagePattern(float ox, float oy, float ex, float ey, float angle, const NanoImage* image, PatternRepeat repeat);
 
    /* --------------------------------------------------------------------
     * Scissoring */
diff --git a/dgl/src/NanoVG.cpp b/dgl/src/NanoVG.cpp
index 9f8717cc..2c6e0af5 100644
--- a/dgl/src/NanoVG.cpp
+++ b/dgl/src/NanoVG.cpp
@@ -43,6 +43,7 @@
 // -----------------------------------------------------------------------
 // Include NanoVG OpenGL implementation
 
+#define STB_IMAGE_STATIC 1
 #define NANOVG_GL2_IMPLEMENTATION 1
 #include "nanovg/nanovg_gl.h"
 
@@ -158,15 +159,7 @@ void NanoImage::_updateSize()
 // -----------------------------------------------------------------------
 // NanoVG
 
-NanoVG::NanoVG()
-    : fContext(nvgCreateGL(512, 512, NVG_ANTIALIAS)),
-      fInFrame(false),
-      leakDetector_NanoVG()
-{
-    DISTRHO_SAFE_ASSERT_RETURN(fContext != nullptr,);
-}
-
-NanoVG::NanoVG(const int textAtlasWidth, const int textAtlasHeight)
+NanoVG::NanoVG(int textAtlasWidth, int textAtlasHeight)
     : fContext(nvgCreateGL(textAtlasWidth, textAtlasHeight, NVG_ANTIALIAS)),
       fInFrame(false),
       leakDetector_NanoVG()
diff --git a/dgl/src/nanovg/nanovg.c b/dgl/src/nanovg/nanovg.c
index 0697c9ba..44cf25f8 100644
--- a/dgl/src/nanovg/nanovg.c
+++ b/dgl/src/nanovg/nanovg.c
@@ -21,8 +21,19 @@
 #include "nanovg.h"
 #define FONTSTASH_IMPLEMENTATION
 #include "fontstash.h"
-#include "stb_image.c"
+#define STB_IMAGE_IMPLEMENTATION
+#include "stb_image.h"
 
+#ifdef _MSC_VER
+#pragma warning(disable: 4100)  // unreferenced formal parameter
+#pragma warning(disable: 4127)  // conditional expression is constant
+#pragma warning(disable: 4204)  // nonstandard extension used : non-constant aggregate initializer
+#pragma warning(disable: 4706)  // assignment within conditional expression
+#endif
+
+#define NVG_INIT_FONTIMAGE_SIZE  512
+#define NVG_MAX_FONTIMAGE_SIZE   2048
+#define NVG_MAX_FONTIMAGES       4
 
 #define NVG_INIT_COMMANDS_SIZE 256
 #define NVG_INIT_POINTS_SIZE 128
@@ -30,7 +41,7 @@
 #define NVG_INIT_VERTS_SIZE 256
 #define NVG_MAX_STATES 32
 
-#define NVG_KAPPA90 0.5522847493f	// Lenght proportional to radius of a cubic bezier handle for 90deg arcs.
+#define NVG_KAPPA90 0.5522847493f	// Length proportional to radius of a cubic bezier handle for 90deg arcs.
 
 #define NVG_COUNTOF(arr) (sizeof(arr) / sizeof(0[arr]))
 
@@ -51,21 +62,16 @@ enum NVGpointFlags
 	NVG_PR_INNERBEVEL = 0x08,
 };
 
-enum NVGexpandFeatures {
-	NVG_FILL = 0x01,
-	NVG_STROKE = 0x02,
-	NVG_CAPS = 0x04,
-};
-
 struct NVGstate {
-	struct NVGpaint fill;
-	struct NVGpaint stroke;
+	NVGpaint fill;
+	NVGpaint stroke;
 	float strokeWidth;
 	float miterLimit;
 	int lineJoin;
 	int lineCap;
+	float alpha;
 	float xform[6];
-	struct NVGscissor scissor;
+	NVGscissor scissor;
 	float fontSize;
 	float letterSpacing;
 	float lineHeight;
@@ -73,6 +79,7 @@ struct NVGstate {
 	int textAlign;
 	int fontId;
 };
+typedef struct NVGstate NVGstate;
 
 struct NVGpoint {
 	float x,y;
@@ -81,36 +88,38 @@ struct NVGpoint {
 	float dmx, dmy;
 	unsigned char flags;
 };
+typedef struct NVGpoint NVGpoint;
 
 struct NVGpathCache {
-	struct NVGpoint* points;
+	NVGpoint* points;
 	int npoints;
 	int cpoints;
-	struct NVGpath* paths;
+	NVGpath* paths;
 	int npaths;
 	int cpaths;
-	struct NVGvertex* verts;
+	NVGvertex* verts;
 	int nverts;
 	int cverts;
 	float bounds[4];
 };
+typedef struct NVGpathCache NVGpathCache;
 
 struct NVGcontext {
-	struct NVGparams params;
+	NVGparams params;
 	float* commands;
 	int ccommands;
 	int ncommands;
 	float commandx, commandy;
-	struct NVGstate states[NVG_MAX_STATES];
+	NVGstate states[NVG_MAX_STATES];
 	int nstates;
-	struct NVGpathCache* cache;
+	NVGpathCache* cache;
 	float tessTol;
 	float distTol;
 	float fringeWidth;
 	float devicePxRatio;
 	struct FONScontext* fs;
-	int fontImage;
-	int alphaBlend;
+	int fontImages[NVG_MAX_FONTIMAGES];
+	int fontImageIdx;
 	int drawCallCount;
 	int fillTriCount;
 	int strokeTriCount;
@@ -131,6 +140,7 @@ static int nvg__clampi(int a, int mn, int mx) { return a < mn ? mn : (a > mx ? m
 static float nvg__minf(float a, float b) { return a < b ? a : b; }
 static float nvg__maxf(float a, float b) { return a > b ? a : b; }
 static float nvg__absf(float a) { return a >= 0.0f ? a : -a; }
+static float nvg__signf(float a) { return a >= 0.0f ? 1.0f : -1.0f; }
 static float nvg__clampf(float a, float mn, float mx) { return a < mn ? mn : (a > mx ? mx : a); }
 static float nvg__cross(float dx0, float dy0, float dx1, float dy1) { return dx1*dy0 - dx0*dy1; }
 
@@ -146,7 +156,7 @@ static float nvg__normalize(float *x, float* y)
 }
 
 
-static void nvg__deletePathCache(struct NVGpathCache* c)
+static void nvg__deletePathCache(NVGpathCache* c)
 {
 	if (c == NULL) return;
 	if (c->points != NULL) free(c->points);
@@ -155,23 +165,23 @@ static void nvg__deletePathCache(struct NVGpathCache* c)
 	free(c);
 }
 
-static struct NVGpathCache* nvg__allocPathCache()
+static NVGpathCache* nvg__allocPathCache(void)
 {
-	struct NVGpathCache* c = (struct NVGpathCache*)malloc(sizeof(struct NVGpathCache));
+	NVGpathCache* c = (NVGpathCache*)malloc(sizeof(NVGpathCache));
 	if (c == NULL) goto error;
-	memset(c, 0, sizeof(struct NVGpathCache));
+	memset(c, 0, sizeof(NVGpathCache));
 
-	c->points = (struct NVGpoint*)malloc(sizeof(struct NVGpoint)*NVG_INIT_POINTS_SIZE);
+	c->points = (NVGpoint*)malloc(sizeof(NVGpoint)*NVG_INIT_POINTS_SIZE);
 	if (!c->points) goto error;
 	c->npoints = 0;
 	c->cpoints = NVG_INIT_POINTS_SIZE;
 
-	c->paths = (struct NVGpath*)malloc(sizeof(struct NVGpath)*NVG_INIT_PATHS_SIZE);
+	c->paths = (NVGpath*)malloc(sizeof(NVGpath)*NVG_INIT_PATHS_SIZE);
 	if (!c->paths) goto error;
 	c->npaths = 0;
 	c->cpaths = NVG_INIT_PATHS_SIZE;
 
-	c->verts = (struct NVGvertex*)malloc(sizeof(struct NVGvertex)*NVG_INIT_VERTS_SIZE);
+	c->verts = (NVGvertex*)malloc(sizeof(NVGvertex)*NVG_INIT_VERTS_SIZE);
 	if (!c->verts) goto error;
 	c->nverts = 0;
 	c->cverts = NVG_INIT_VERTS_SIZE;
@@ -182,30 +192,31 @@ error:
 	return NULL;
 }
 
-static void nvg__setDevicePixelRatio(struct NVGcontext* ctx, float ratio)
+static void nvg__setDevicePixelRatio(NVGcontext* ctx, float ratio)
 {
-	ctx->tessTol = 1.0f / ratio;
+	ctx->tessTol = 0.25f / ratio;
 	ctx->distTol = 0.01f / ratio;
 	ctx->fringeWidth = 1.0f / ratio;
 	ctx->devicePxRatio = ratio;
 }
 
-struct NVGcontext* nvgCreateInternal(struct NVGparams* params)
+NVGcontext* nvgCreateInternal(NVGparams* params)
 {
-	struct FONSparams fontParams;
-	struct NVGcontext* ctx = (struct NVGcontext*)malloc(sizeof(struct NVGcontext));
+	FONSparams fontParams;
+	NVGcontext* ctx = (NVGcontext*)malloc(sizeof(NVGcontext));
+	int i;
 	if (ctx == NULL) goto error;
-	memset(ctx, 0, sizeof(struct NVGcontext));
+	memset(ctx, 0, sizeof(NVGcontext));
 
 	ctx->params = *params;
+	for (i = 0; i < NVG_MAX_FONTIMAGES; i++)
+		ctx->fontImages[i] = 0;
 
 	ctx->commands = (float*)malloc(sizeof(float)*NVG_INIT_COMMANDS_SIZE);
 	if (!ctx->commands) goto error;
 	ctx->ncommands = 0;
 	ctx->ccommands = NVG_INIT_COMMANDS_SIZE;
 
-	ctx->alphaBlend = NVG_STRAIGHT_ALPHA;
-
 	ctx->cache = nvg__allocPathCache();
 	if (ctx->cache == NULL) goto error;
 
@@ -218,8 +229,8 @@ struct NVGcontext* nvgCreateInternal(struct NVGparams* params)
 
 	// Init font rendering
 	memset(&fontParams, 0, sizeof(fontParams));
-	fontParams.width = params->atlasWidth;
-	fontParams.height = params->atlasHeight;
+	fontParams.width = NVG_INIT_FONTIMAGE_SIZE;
+	fontParams.height = NVG_INIT_FONTIMAGE_SIZE;
 	fontParams.flags = FONS_ZERO_TOPLEFT;
 	fontParams.renderCreate = NULL;
 	fontParams.renderUpdate = NULL;
@@ -230,8 +241,9 @@ struct NVGcontext* nvgCreateInternal(struct NVGparams* params)
 	if (ctx->fs == NULL) goto error;
 
 	// Create font texture
-	ctx->fontImage = ctx->params.renderCreateTexture(ctx->params.userPtr, NVG_TEXTURE_ALPHA, fontParams.width, fontParams.height, NULL);
-	if (ctx->fontImage == 0) goto error;
+	ctx->fontImages[0] = ctx->params.renderCreateTexture(ctx->params.userPtr, NVG_TEXTURE_ALPHA, fontParams.width, fontParams.height, 0, NULL);
+	if (ctx->fontImages[0] == 0) goto error;
+	ctx->fontImageIdx = 0;
 
 	return ctx;
 
@@ -240,8 +252,14 @@ error:
 	return 0;
 }
 
-void nvgDeleteInternal(struct NVGcontext* ctx)
+NVGparams* nvgInternalParams(NVGcontext* ctx)
+{
+    return &ctx->params;
+}
+
+void nvgDeleteInternal(NVGcontext* ctx)
 {
+	int i;
 	if (ctx == NULL) return;
 	if (ctx->commands != NULL) free(ctx->commands);
 	if (ctx->cache != NULL) nvg__deletePathCache(ctx->cache);
@@ -249,13 +267,20 @@ void nvgDeleteInternal(struct NVGcontext* ctx)
 	if (ctx->fs)
 		fonsDeleteInternal(ctx->fs);
 
+	for (i = 0; i < NVG_MAX_FONTIMAGES; i++) {
+		if (ctx->fontImages[i] != 0) {
+			nvgDeleteImage(ctx, ctx->fontImages[i]);
+			ctx->fontImages[i] = 0;
+		}
+	}
+
 	if (ctx->params.renderDelete != NULL)
 		ctx->params.renderDelete(ctx->params.userPtr);
 
 	free(ctx);
 }
 
-void nvgBeginFrame(struct NVGcontext* ctx, int windowWidth, int windowHeight, float devicePixelRatio, int alphaBlend)
+void nvgBeginFrame(NVGcontext* ctx, int windowWidth, int windowHeight, float devicePixelRatio)
 {
 /*	printf("Tris: draws:%d  fill:%d  stroke:%d  text:%d  TOT:%d\n",
 		ctx->drawCallCount, ctx->fillTriCount, ctx->strokeTriCount, ctx->textTriCount,
@@ -266,9 +291,8 @@ void nvgBeginFrame(struct NVGcontext* ctx, int windowWidth, int windowHeight, fl
 	nvgReset(ctx);
 
 	nvg__setDevicePixelRatio(ctx, devicePixelRatio);
-	ctx->alphaBlend = alphaBlend;
-
-	ctx->params.renderViewport(ctx->params.userPtr, windowWidth, windowHeight, ctx->alphaBlend);
+	
+	ctx->params.renderViewport(ctx->params.userPtr, windowWidth, windowHeight);
 
 	ctx->drawCallCount = 0;
 	ctx->fillTriCount = 0;
@@ -276,24 +300,54 @@ void nvgBeginFrame(struct NVGcontext* ctx, int windowWidth, int windowHeight, fl
 	ctx->textTriCount = 0;
 }
 
-void nvgEndFrame(struct NVGcontext* ctx)
+void nvgCancelFrame(NVGcontext* ctx)
 {
-	ctx->params.renderFlush(ctx->params.userPtr, ctx->alphaBlend);
+	ctx->params.renderCancel(ctx->params.userPtr);
 }
 
-struct NVGcolor nvgRGB(unsigned char r, unsigned char g, unsigned char b)
+void nvgEndFrame(NVGcontext* ctx)
+{
+	ctx->params.renderFlush(ctx->params.userPtr);
+	if (ctx->fontImageIdx != 0) {
+		int fontImage = ctx->fontImages[ctx->fontImageIdx];
+		int i, j, iw, ih;
+		// delete images that smaller than current one
+		if (fontImage == 0)
+			return;
+		nvgImageSize(ctx, fontImage, &iw, &ih);
+		for (i = j = 0; i < ctx->fontImageIdx; i++) {
+			if (ctx->fontImages[i] != 0) {
+				int nw, nh;
+				nvgImageSize(ctx, ctx->fontImages[i], &nw, &nh);
+				if (nw < iw || nh < ih)
+					nvgDeleteImage(ctx, ctx->fontImages[i]);
+				else
+					ctx->fontImages[j++] = ctx->fontImages[i];
+			}
+		}
+		// make current font image to first
+		ctx->fontImages[j++] = ctx->fontImages[0];
+		ctx->fontImages[0] = fontImage;
+		ctx->fontImageIdx = 0;
+		// clear all images after j
+		for (i = j; i < NVG_MAX_FONTIMAGES; i++)
+			ctx->fontImages[i] = 0;
+	}
+}
+
+NVGcolor nvgRGB(unsigned char r, unsigned char g, unsigned char b)
 {
 	return nvgRGBA(r,g,b,255);
 }
 
-struct NVGcolor nvgRGBf(float r, float g, float b)
+NVGcolor nvgRGBf(float r, float g, float b)
 {
 	return nvgRGBAf(r,g,b,1.0f);
 }
 
-struct NVGcolor nvgRGBA(unsigned char r, unsigned char g, unsigned char b, unsigned char a)
+NVGcolor nvgRGBA(unsigned char r, unsigned char g, unsigned char b, unsigned char a)
 {
-	struct NVGcolor color;
+	NVGcolor color;
 	// Use longer initialization to suppress warning.
 	color.r = r / 255.0f;
 	color.g = g / 255.0f;
@@ -302,9 +356,9 @@ struct NVGcolor nvgRGBA(unsigned char r, unsigned char g, unsigned char b, unsig
 	return color;
 }
 
-struct NVGcolor nvgRGBAf(float r, float g, float b, float a)
+NVGcolor nvgRGBAf(float r, float g, float b, float a)
 {
-	struct NVGcolor color;
+	NVGcolor color;
 	// Use longer initialization to suppress warning.
 	color.r = r;
 	color.g = g;
@@ -313,35 +367,35 @@ struct NVGcolor nvgRGBAf(float r, float g, float b, float a)
 	return color;
 }
 
-struct NVGcolor nvgTransRGBA(struct NVGcolor c, unsigned char a)
+NVGcolor nvgTransRGBA(NVGcolor c, unsigned char a)
 {
 	c.a = a / 255.0f;
 	return c;
 }
 
-struct NVGcolor nvgTransRGBAf(struct NVGcolor c, float a)
+NVGcolor nvgTransRGBAf(NVGcolor c, float a)
 {
 	c.a = a;
 	return c;
 }
 
-struct NVGcolor nvgLerpRGBA(struct NVGcolor c0, struct NVGcolor c1, float u)
+NVGcolor nvgLerpRGBA(NVGcolor c0, NVGcolor c1, float u)
 {
 	int i;
 	float oneminu;
-	struct NVGcolor cint;
+	NVGcolor cint;
 
 	u = nvg__clampf(u, 0.0f, 1.0f);
 	oneminu = 1.0f - u;
-	for( i = 0; i <4; ++i )
+	for( i = 0; i <4; i++ )
 	{
 		cint.rgba[i] = c0.rgba[i] * oneminu + c1.rgba[i] * u;
 	}
-
+	 
 	return cint;
 }
 
-struct NVGcolor nvgHSL(float h, float s, float l)
+NVGcolor nvgHSL(float h, float s, float l)
 {
 	return nvgHSLA(h,s,l,255);
 }
@@ -359,10 +413,10 @@ static float nvg__hue(float h, float m1, float m2)
 	return m1;
 }
 
-struct NVGcolor nvgHSLA(float h, float s, float l, unsigned char a)
+NVGcolor nvgHSLA(float h, float s, float l, unsigned char a)
 {
 	float m1, m2;
-	struct NVGcolor col;
+	NVGcolor col;
 	h = nvg__modf(h, 1.0f);
 	if (h < 0.0f) h += 1.0f;
 	s = nvg__clampf(s, 0.0f, 1.0f);
@@ -377,7 +431,7 @@ struct NVGcolor nvgHSLA(float h, float s, float l, unsigned char a)
 }
 
 
-static struct NVGstate* nvg__getState(struct NVGcontext* ctx)
+static NVGstate* nvg__getState(NVGcontext* ctx)
 {
 	return &ctx->states[ctx->nstates-1];
 }
@@ -479,7 +533,7 @@ float nvgRadToDeg(float rad)
 	return rad / NVG_PI * 180.0f;
 }
 
-static void nvg__setPaintColor(struct NVGpaint* p, struct NVGcolor color)
+static void nvg__setPaintColor(NVGpaint* p, NVGcolor color)
 {
 	memset(p, 0, sizeof(*p));
 	nvgTransformIdentity(p->xform);
@@ -491,25 +545,25 @@ static void nvg__setPaintColor(struct NVGpaint* p, struct NVGcolor color)
 
 
 // State handling
-void nvgSave(struct NVGcontext* ctx)
+void nvgSave(NVGcontext* ctx)
 {
 	if (ctx->nstates >= NVG_MAX_STATES)
 		return;
 	if (ctx->nstates > 0)
-		memcpy(&ctx->states[ctx->nstates], &ctx->states[ctx->nstates-1], sizeof(struct NVGstate));
+		memcpy(&ctx->states[ctx->nstates], &ctx->states[ctx->nstates-1], sizeof(NVGstate));
 	ctx->nstates++;
 }
 
-void nvgRestore(struct NVGcontext* ctx)
+void nvgRestore(NVGcontext* ctx)
 {
 	if (ctx->nstates <= 1)
 		return;
 	ctx->nstates--;
 }
 
-void nvgReset(struct NVGcontext* ctx)
+void nvgReset(NVGcontext* ctx)
 {
-	struct NVGstate* state = nvg__getState(ctx);
+	NVGstate* state = nvg__getState(ctx);
 	memset(state, 0, sizeof(*state));
 
 	nvg__setPaintColor(&state->fill, nvgRGBA(255,255,255,255));
@@ -518,145 +572,154 @@ void nvgReset(struct NVGcontext* ctx)
 	state->miterLimit = 10.0f;
 	state->lineCap = NVG_BUTT;
 	state->lineJoin = NVG_MITER;
+	state->alpha = 1.0f;
 	nvgTransformIdentity(state->xform);
 
-	state->scissor.extent[0] = 0.0f;
-	state->scissor.extent[1] = 0.0f;
+	state->scissor.extent[0] = -1.0f;
+	state->scissor.extent[1] = -1.0f;
 
 	state->fontSize = 16.0f;
 	state->letterSpacing = 0.0f;
-	state->lineHeight = 0.0f;
+	state->lineHeight = 1.0f;
 	state->fontBlur = 0.0f;
 	state->textAlign = NVG_ALIGN_LEFT | NVG_ALIGN_BASELINE;
 	state->fontId = 0;
 }
 
 // State setting
-void nvgStrokeWidth(struct NVGcontext* ctx, float width)
+void nvgStrokeWidth(NVGcontext* ctx, float width)
 {
-	struct NVGstate* state = nvg__getState(ctx);
+	NVGstate* state = nvg__getState(ctx);
 	state->strokeWidth = width;
 }
 
-void nvgMiterLimit(struct NVGcontext* ctx, float limit)
+void nvgMiterLimit(NVGcontext* ctx, float limit)
 {
-	struct NVGstate* state = nvg__getState(ctx);
+	NVGstate* state = nvg__getState(ctx);
 	state->miterLimit = limit;
 }
 
-void nvgLineCap(struct NVGcontext* ctx, int cap)
+void nvgLineCap(NVGcontext* ctx, int cap)
 {
-	struct NVGstate* state = nvg__getState(ctx);
+	NVGstate* state = nvg__getState(ctx);
 	state->lineCap = cap;
 }
 
-void nvgLineJoin(struct NVGcontext* ctx, int join)
+void nvgLineJoin(NVGcontext* ctx, int join)
 {
-	struct NVGstate* state = nvg__getState(ctx);
+	NVGstate* state = nvg__getState(ctx);
 	state->lineJoin = join;
 }
 
+void nvgGlobalAlpha(NVGcontext* ctx, float alpha)
+{
+	NVGstate* state = nvg__getState(ctx);
+	state->alpha = alpha;
+}
 
-void nvgTransform(struct NVGcontext* ctx, float a, float b, float c, float d, float e, float f)
+void nvgTransform(NVGcontext* ctx, float a, float b, float c, float d, float e, float f)
 {
-	struct NVGstate* state = nvg__getState(ctx);
+	NVGstate* state = nvg__getState(ctx);
 	float t[6] = { a, b, c, d, e, f };
 	nvgTransformPremultiply(state->xform, t);
 }
 
-void nvgResetTransform(struct NVGcontext* ctx)
+void nvgResetTransform(NVGcontext* ctx)
 {
-	struct NVGstate* state = nvg__getState(ctx);
+	NVGstate* state = nvg__getState(ctx);
 	nvgTransformIdentity(state->xform);
 }
 
-void nvgTranslate(struct NVGcontext* ctx, float x, float y)
+void nvgTranslate(NVGcontext* ctx, float x, float y)
 {
-	struct NVGstate* state = nvg__getState(ctx);
+	NVGstate* state = nvg__getState(ctx);
 	float t[6];
 	nvgTransformTranslate(t, x,y);
 	nvgTransformPremultiply(state->xform, t);
 }
 
-void nvgRotate(struct NVGcontext* ctx, float angle)
+void nvgRotate(NVGcontext* ctx, float angle)
 {
-	struct NVGstate* state = nvg__getState(ctx);
+	NVGstate* state = nvg__getState(ctx);
 	float t[6];
 	nvgTransformRotate(t, angle);
 	nvgTransformPremultiply(state->xform, t);
 }
 
-void nvgSkewX(struct NVGcontext* ctx, float angle)
+void nvgSkewX(NVGcontext* ctx, float angle)
 {
-	struct NVGstate* state = nvg__getState(ctx);
+	NVGstate* state = nvg__getState(ctx);
 	float t[6];
 	nvgTransformSkewX(t, angle);
 	nvgTransformPremultiply(state->xform, t);
 }
 
-void nvgSkewY(struct NVGcontext* ctx, float angle)
+void nvgSkewY(NVGcontext* ctx, float angle)
 {
-	struct NVGstate* state = nvg__getState(ctx);
+	NVGstate* state = nvg__getState(ctx);
 	float t[6];
 	nvgTransformSkewY(t, angle);
 	nvgTransformPremultiply(state->xform, t);
 }
 
-void nvgScale(struct NVGcontext* ctx, float x, float y)
+void nvgScale(NVGcontext* ctx, float x, float y)
 {
-	struct NVGstate* state = nvg__getState(ctx);
+	NVGstate* state = nvg__getState(ctx);
 	float t[6];
 	nvgTransformScale(t, x,y);
 	nvgTransformPremultiply(state->xform, t);
 }
 
-void nvgCurrentTransform(struct NVGcontext* ctx, float* xform)
+void nvgCurrentTransform(NVGcontext* ctx, float* xform)
 {
-	struct NVGstate* state = nvg__getState(ctx);
+	NVGstate* state = nvg__getState(ctx);
 	if (xform == NULL) return;
 	memcpy(xform, state->xform, sizeof(float)*6);
 }
 
-void nvgStrokeColor(struct NVGcontext* ctx, struct NVGcolor color)
+void nvgStrokeColor(NVGcontext* ctx, NVGcolor color)
 {
-	struct NVGstate* state = nvg__getState(ctx);
+	NVGstate* state = nvg__getState(ctx);
 	nvg__setPaintColor(&state->stroke, color);
 }
 
-void nvgStrokePaint(struct NVGcontext* ctx, struct NVGpaint paint)
+void nvgStrokePaint(NVGcontext* ctx, NVGpaint paint)
 {
-	struct NVGstate* state = nvg__getState(ctx);
+	NVGstate* state = nvg__getState(ctx);
 	state->stroke = paint;
 	nvgTransformMultiply(state->stroke.xform, state->xform);
 }
 
-void nvgFillColor(struct NVGcontext* ctx, struct NVGcolor color)
+void nvgFillColor(NVGcontext* ctx, NVGcolor color)
 {
-	struct NVGstate* state = nvg__getState(ctx);
+	NVGstate* state = nvg__getState(ctx);
 	nvg__setPaintColor(&state->fill, color);
 }
 
-void nvgFillPaint(struct NVGcontext* ctx, struct NVGpaint paint)
+void nvgFillPaint(NVGcontext* ctx, NVGpaint paint)
 {
-	struct NVGstate* state = nvg__getState(ctx);
+	NVGstate* state = nvg__getState(ctx);
 	state->fill = paint;
 	nvgTransformMultiply(state->fill.xform, state->xform);
 }
 
-int nvgCreateImage(struct NVGcontext* ctx, const char* filename)
+int nvgCreateImage(NVGcontext* ctx, const char* filename, int imageFlags)
 {
 	int w, h, n, image;
-	unsigned char* img = stbi_load(filename, &w, &h, &n, 4);
+	unsigned char* img;
+	stbi_set_unpremultiply_on_load(1);
+	stbi_convert_iphone_png_to_rgb(1);
+	img = stbi_load(filename, &w, &h, &n, 4);
 	if (img == NULL) {
 //		printf("Failed to load %s - %s\n", filename, stbi_failure_reason());
 		return 0;
 	}
-	image = nvgCreateImageRGBA(ctx, w, h, img);
+	image = nvgCreateImageRGBA(ctx, w, h, imageFlags, img);
 	stbi_image_free(img);
 	return image;
 }
 
-int nvgCreateImageMem(struct NVGcontext* ctx, unsigned char* data, int ndata)
+int nvgCreateImageMem(NVGcontext* ctx, int imageFlags, unsigned char* data, int ndata)
 {
 	int w, h, n, image;
 	unsigned char* img = stbi_load_from_memory(data, ndata, &w, &h, &n, 4);
@@ -664,38 +727,38 @@ int nvgCreateImageMem(struct NVGcontext* ctx, unsigned char* data, int ndata)
 //		printf("Failed to load %s - %s\n", filename, stbi_failure_reason());
 		return 0;
 	}
-	image = nvgCreateImageRGBA(ctx, w, h, img);
+	image = nvgCreateImageRGBA(ctx, w, h, imageFlags, img);
 	stbi_image_free(img);
 	return image;
 }
 
-int nvgCreateImageRGBA(struct NVGcontext* ctx, int w, int h, const unsigned char* data)
+int nvgCreateImageRGBA(NVGcontext* ctx, int w, int h, int imageFlags, const unsigned char* data)
 {
-	return ctx->params.renderCreateTexture(ctx->params.userPtr, NVG_TEXTURE_RGBA, w, h, data);
+	return ctx->params.renderCreateTexture(ctx->params.userPtr, NVG_TEXTURE_RGBA, w, h, imageFlags, data);
 }
 
-void nvgUpdateImage(struct NVGcontext* ctx, int image, const unsigned char* data)
+void nvgUpdateImage(NVGcontext* ctx, int image, const unsigned char* data)
 {
 	int w, h;
 	ctx->params.renderGetTextureSize(ctx->params.userPtr, image, &w, &h);
 	ctx->params.renderUpdateTexture(ctx->params.userPtr, image, 0,0, w,h, data);
 }
 
-void nvgImageSize(struct NVGcontext* ctx, int image, int* w, int* h)
+void nvgImageSize(NVGcontext* ctx, int image, int* w, int* h)
 {
 	ctx->params.renderGetTextureSize(ctx->params.userPtr, image, w, h);
 }
 
-void nvgDeleteImage(struct NVGcontext* ctx, int image)
+void nvgDeleteImage(NVGcontext* ctx, int image)
 {
 	ctx->params.renderDeleteTexture(ctx->params.userPtr, image);
 }
 
-struct NVGpaint nvgLinearGradient(struct NVGcontext* ctx,
+NVGpaint nvgLinearGradient(NVGcontext* ctx,
 								  float sx, float sy, float ex, float ey,
-								  struct NVGcolor icol, struct NVGcolor ocol)
+								  NVGcolor icol, NVGcolor ocol)
 {
-	struct NVGpaint p;
+	NVGpaint p;
 	float dx, dy, d;
 	const float large = 1e5;
 	NVG_NOTUSED(ctx);
@@ -730,11 +793,11 @@ struct NVGpaint nvgLinearGradient(struct NVGcontext* ctx,
 	return p;
 }
 
-struct NVGpaint nvgRadialGradient(struct NVGcontext* ctx,
+NVGpaint nvgRadialGradient(NVGcontext* ctx,
 								  float cx, float cy, float inr, float outr,
-								  struct NVGcolor icol, struct NVGcolor ocol)
+								  NVGcolor icol, NVGcolor ocol)
 {
-	struct NVGpaint p;
+	NVGpaint p;
 	float r = (inr+outr)*0.5f;
 	float f = (outr-inr);
 	NVG_NOTUSED(ctx);
@@ -757,11 +820,11 @@ struct NVGpaint nvgRadialGradient(struct NVGcontext* ctx,
 	return p;
 }
 
-struct NVGpaint nvgBoxGradient(struct NVGcontext* ctx,
+NVGpaint nvgBoxGradient(NVGcontext* ctx,
 							   float x, float y, float w, float h, float r, float f,
-							   struct NVGcolor icol, struct NVGcolor ocol)
+							   NVGcolor icol, NVGcolor ocol)
 {
-	struct NVGpaint p;
+	NVGpaint p;
 	NVG_NOTUSED(ctx);
 	memset(&p, 0, sizeof(p));
 
@@ -783,11 +846,11 @@ struct NVGpaint nvgBoxGradient(struct NVGcontext* ctx,
 }
 
 
-struct NVGpaint nvgImagePattern(struct NVGcontext* ctx,
+NVGpaint nvgImagePattern(NVGcontext* ctx,
 								float cx, float cy, float w, float h, float angle,
-								int image, int repeat)
+								int image, float alpha)
 {
-	struct NVGpaint p;
+	NVGpaint p;
 	NVG_NOTUSED(ctx);
 	memset(&p, 0, sizeof(p));
 
@@ -799,15 +862,19 @@ struct NVGpaint nvgImagePattern(struct NVGcontext* ctx,
 	p.extent[1] = h;
 
 	p.image = image;
-	p.repeat = repeat;
+
+	p.innerColor = p.outerColor = nvgRGBAf(1,1,1,alpha);
 
 	return p;
 }
 
 // Scissoring
-void nvgScissor(struct NVGcontext* ctx, float x, float y, float w, float h)
+void nvgScissor(NVGcontext* ctx, float x, float y, float w, float h)
 {
-	struct NVGstate* state = nvg__getState(ctx);
+	NVGstate* state = nvg__getState(ctx);
+
+	w = nvg__maxf(0.0f, w);
+	h = nvg__maxf(0.0f, h);
 
 	nvgTransformIdentity(state->scissor.xform);
 	state->scissor.xform[4] = x+w*0.5f;
@@ -818,12 +885,55 @@ void nvgScissor(struct NVGcontext* ctx, float x, float y, float w, float h)
 	state->scissor.extent[1] = h*0.5f;
 }
 
-void nvgResetScissor(struct NVGcontext* ctx)
+static void nvg__isectRects(float* dst,
+							float ax, float ay, float aw, float ah,
+							float bx, float by, float bw, float bh)
+{
+	float minx = nvg__maxf(ax, bx);
+	float miny = nvg__maxf(ay, by);
+	float maxx = nvg__minf(ax+aw, bx+bw);
+	float maxy = nvg__minf(ay+ah, by+bh);
+	dst[0] = minx;
+	dst[1] = miny;
+	dst[2] = nvg__maxf(0.0f, maxx - minx);
+	dst[3] = nvg__maxf(0.0f, maxy - miny);
+}
+
+void nvgIntersectScissor(NVGcontext* ctx, float x, float y, float w, float h)
+{
+	NVGstate* state = nvg__getState(ctx);
+	float pxform[6], invxorm[6];
+	float rect[4];
+	float ex, ey, tex, tey;
+
+	// If no previous scissor has been set, set the scissor as current scissor.
+	if (state->scissor.extent[0] < 0) {
+		nvgScissor(ctx, x, y, w, h);
+		return;
+	}
+
+	// Transform the current scissor rect into current transform space.
+	// If there is difference in rotation, this will be approximation. 
+	memcpy(pxform, state->scissor.xform, sizeof(float)*6);
+	ex = state->scissor.extent[0];
+	ey = state->scissor.extent[1];
+	nvgTransformInverse(invxorm, state->xform);
+	nvgTransformMultiply(pxform, invxorm);
+	tex = ex*nvg__absf(pxform[0]) + ey*nvg__absf(pxform[2]);
+	tey = ex*nvg__absf(pxform[1]) + ey*nvg__absf(pxform[3]);
+
+	// Intersect rects.
+	nvg__isectRects(rect, pxform[4]-tex,pxform[5]-tey,tex*2,tey*2, x,y,w,h);
+
+	nvgScissor(ctx, rect[0], rect[1], rect[2], rect[3]);
+}
+
+void nvgResetScissor(NVGcontext* ctx)
 {
-	struct NVGstate* state = nvg__getState(ctx);
+	NVGstate* state = nvg__getState(ctx);
 	memset(state->scissor.xform, 0, sizeof(state->scissor.xform));
-	state->scissor.extent[0] = 0;
-	state->scissor.extent[1] = 0;
+	state->scissor.extent[0] = -1.0f;
+	state->scissor.extent[1] = -1.0f;
 }
 
 static int nvg__ptEquals(float x1, float y1, float x2, float y2, float tol)
@@ -850,9 +960,9 @@ static float nvg__distPtSeg(float x, float y, float px, float py, float qx, floa
 	return dx*dx + dy*dy;
 }
 
-static void nvg__appendCommands(struct NVGcontext* ctx, float* vals, int nvals)
+static void nvg__appendCommands(NVGcontext* ctx, float* vals, int nvals)
 {
-	struct NVGstate* state = nvg__getState(ctx);
+	NVGstate* state = nvg__getState(ctx);
 	int i;
 
 	if (ctx->ncommands+nvals > ctx->ccommands) {
@@ -864,6 +974,11 @@ static void nvg__appendCommands(struct NVGcontext* ctx, float* vals, int nvals)
 		ctx->ccommands = ccommands;
 	}
 
+	if ((int)vals[0] != NVG_CLOSE && (int)vals[0] != NVG_WINDING) {
+		ctx->commandx = vals[nvals-2];
+		ctx->commandy = vals[nvals-1];
+	}
+
 	// transform commands
 	i = 0;
 	while (i < nvals) {
@@ -897,34 +1012,29 @@ static void nvg__appendCommands(struct NVGcontext* ctx, float* vals, int nvals)
 	memcpy(&ctx->commands[ctx->ncommands], vals, nvals*sizeof(float));
 
 	ctx->ncommands += nvals;
-
-	if ((int)vals[0] != NVG_CLOSE && (int)vals[0] != NVG_WINDING) {
-		ctx->commandx = vals[nvals-2];
-		ctx->commandy = vals[nvals-1];
-	}
 }
 
 
-static void nvg__clearPathCache(struct NVGcontext* ctx)
+static void nvg__clearPathCache(NVGcontext* ctx)
 {
 	ctx->cache->npoints = 0;
 	ctx->cache->npaths = 0;
 }
 
-static struct NVGpath* nvg__lastPath(struct NVGcontext* ctx)
+static NVGpath* nvg__lastPath(NVGcontext* ctx)
 {
 	if (ctx->cache->npaths > 0)
 		return &ctx->cache->paths[ctx->cache->npaths-1];
 	return NULL;
 }
 
-static void nvg__addPath(struct NVGcontext* ctx)
+static void nvg__addPath(NVGcontext* ctx)
 {
-	struct NVGpath* path;
+	NVGpath* path;
 	if (ctx->cache->npaths+1 > ctx->cache->cpaths) {
-		struct NVGpath* paths;
+		NVGpath* paths;
 		int cpaths = ctx->cache->npaths+1 + ctx->cache->cpaths/2;
-		paths = (struct NVGpath*)realloc(ctx->cache->paths, sizeof(struct NVGpath)*cpaths);
+		paths = (NVGpath*)realloc(ctx->cache->paths, sizeof(NVGpath)*cpaths);
 		if (paths == NULL) return;
 		ctx->cache->paths = paths;
 		ctx->cache->cpaths = cpaths;
@@ -937,20 +1047,20 @@ static void nvg__addPath(struct NVGcontext* ctx)
 	ctx->cache->npaths++;
 }
 
-static struct NVGpoint* nvg__lastPoint(struct NVGcontext* ctx)
+static NVGpoint* nvg__lastPoint(NVGcontext* ctx)
 {
 	if (ctx->cache->npoints > 0)
 		return &ctx->cache->points[ctx->cache->npoints-1];
 	return NULL;
 }
 
-static void nvg__addPoint(struct NVGcontext* ctx, float x, float y, int flags)
+static void nvg__addPoint(NVGcontext* ctx, float x, float y, int flags)
 {
-	struct NVGpath* path = nvg__lastPath(ctx);
-	struct NVGpoint* pt;
+	NVGpath* path = nvg__lastPath(ctx);
+	NVGpoint* pt;
 	if (path == NULL) return;
 
-	if (ctx->cache->npoints > 0) {
+	if (path->count > 0 && ctx->cache->npoints > 0) {
 		pt = nvg__lastPoint(ctx);
 		if (nvg__ptEquals(pt->x,pt->y, x,y, ctx->distTol)) {
 			pt->flags |= flags;
@@ -959,9 +1069,9 @@ static void nvg__addPoint(struct NVGcontext* ctx, float x, float y, int flags)
 	}
 
 	if (ctx->cache->npoints+1 > ctx->cache->cpoints) {
-		struct NVGpoint* points;
+		NVGpoint* points;
 		int cpoints = ctx->cache->npoints+1 + ctx->cache->cpoints/2;
-		points = (struct NVGpoint*)realloc(ctx->cache->points, sizeof(struct NVGpoint)*cpoints);
+		points = (NVGpoint*)realloc(ctx->cache->points, sizeof(NVGpoint)*cpoints);
 		if (points == NULL) return;
 		ctx->cache->points = points;
 		ctx->cache->cpoints = cpoints;
@@ -971,22 +1081,22 @@ static void nvg__addPoint(struct NVGcontext* ctx, float x, float y, int flags)
 	memset(pt, 0, sizeof(*pt));
 	pt->x = x;
 	pt->y = y;
-	pt->flags = flags;
+	pt->flags = (unsigned char)flags;
 
 	ctx->cache->npoints++;
 	path->count++;
 }
 
-static void nvg__closePath(struct NVGcontext* ctx)
+static void nvg__closePath(NVGcontext* ctx)
 {
-	struct NVGpath* path = nvg__lastPath(ctx);
+	NVGpath* path = nvg__lastPath(ctx);
 	if (path == NULL) return;
 	path->closed = 1;
 }
 
-static void nvg__pathWinding(struct NVGcontext* ctx, int winding)
+static void nvg__pathWinding(NVGcontext* ctx, int winding)
 {
-	struct NVGpath* path = nvg__lastPath(ctx);
+	NVGpath* path = nvg__lastPath(ctx);
 	if (path == NULL) return;
 	path->winding = winding;
 }
@@ -998,12 +1108,12 @@ static float nvg__getAverageScale(float *t)
 	return (sx + sy) * 0.5f;
 }
 
-static struct NVGvertex* nvg__allocTempVerts(struct NVGcontext* ctx, int nverts)
+static NVGvertex* nvg__allocTempVerts(NVGcontext* ctx, int nverts)
 {
 	if (nverts > ctx->cache->cverts) {
-		struct NVGvertex* verts;
+		NVGvertex* verts;
 		int cverts = (nverts + 0xff) & ~0xff; // Round up to prevent allocations when things change just slightly.
-		verts = (struct NVGvertex*)realloc(ctx->cache->verts, sizeof(struct NVGvertex)*cverts);
+		verts = (NVGvertex*)realloc(ctx->cache->verts, sizeof(NVGvertex)*cverts);
 		if (verts == NULL) return NULL;
 		ctx->cache->verts = verts;
 		ctx->cache->cverts = cverts;
@@ -1021,22 +1131,22 @@ static float nvg__triarea2(float ax, float ay, float bx, float by, float cx, flo
 	return acx*aby - abx*acy;
 }
 
-static float nvg__polyArea(struct NVGpoint* pts, int npts)
+static float nvg__polyArea(NVGpoint* pts, int npts)
 {
 	int i;
 	float area = 0;
 	for (i = 2; i < npts; i++) {
-		struct NVGpoint* a = &pts[0];
-		struct NVGpoint* b = &pts[i-1];
-		struct NVGpoint* c = &pts[i];
+		NVGpoint* a = &pts[0];
+		NVGpoint* b = &pts[i-1];
+		NVGpoint* c = &pts[i];
 		area += nvg__triarea2(a->x,a->y, b->x,b->y, c->x,c->y);
 	}
 	return area * 0.5f;
 }
 
-static void nvg__polyReverse(struct NVGpoint* pts, int npts)
+static void nvg__polyReverse(NVGpoint* pts, int npts)
 {
-	struct NVGpoint tmp;
+	NVGpoint tmp;
 	int i = 0, j = npts-1;
 	while (i < j) {
 		tmp = pts[i];
@@ -1048,7 +1158,7 @@ static void nvg__polyReverse(struct NVGpoint* pts, int npts)
 }
 
 
-static void nvg__vset(struct NVGvertex* vtx, float x, float y, float u, float v)
+static void nvg__vset(NVGvertex* vtx, float x, float y, float u, float v)
 {
 	vtx->x = x;
 	vtx->y = y;
@@ -1056,14 +1166,14 @@ static void nvg__vset(struct NVGvertex* vtx, float x, float y, float u, float v)
 	vtx->v = v;
 }
 
-static void nvg__tesselateBezier(struct NVGcontext* ctx,
+static void nvg__tesselateBezier(NVGcontext* ctx,
 								 float x1, float y1, float x2, float y2,
 								 float x3, float y3, float x4, float y4,
 								 int level, int type)
 {
 	float x12,y12,x23,y23,x34,y34,x123,y123,x234,y234,x1234,y1234;
 	float dx,dy,d2,d3;
-
+	
 	if (level > 10) return;
 
 	x12 = (x1+x2)*0.5f;
@@ -1075,8 +1185,8 @@ static void nvg__tesselateBezier(struct NVGcontext* ctx,
 	x123 = (x12+x23)*0.5f;
 	y123 = (y12+y23)*0.5f;
 
-	dx = x3 - x1;
-	dy = y3 - y1;
+	dx = x4 - x1;
+	dy = y4 - y1;
 	d2 = nvg__absf(((x2 - x4) * dy - (y2 - y4) * dx));
 	d3 = nvg__absf(((x3 - x4) * dy - (y3 - y4) * dx));
 
@@ -1095,19 +1205,19 @@ static void nvg__tesselateBezier(struct NVGcontext* ctx,
 	x1234 = (x123+x234)*0.5f;
 	y1234 = (y123+y234)*0.5f;
 
-	nvg__tesselateBezier(ctx, x1,y1, x12,y12, x123,y123, x1234,y1234, level+1, 0);
-	nvg__tesselateBezier(ctx, x1234,y1234, x234,y234, x34,y34, x4,y4, level+1, type);
+	nvg__tesselateBezier(ctx, x1,y1, x12,y12, x123,y123, x1234,y1234, level+1, 0); 
+	nvg__tesselateBezier(ctx, x1234,y1234, x234,y234, x34,y34, x4,y4, level+1, type); 
 }
 
-static void nvg__flattenPaths(struct NVGcontext* ctx)
+static void nvg__flattenPaths(NVGcontext* ctx)
 {
-	struct NVGpathCache* cache = ctx->cache;
-//	struct NVGstate* state = nvg__getState(ctx);
-	struct NVGpoint* last;
-	struct NVGpoint* p0;
-	struct NVGpoint* p1;
-	struct NVGpoint* pts;
-	struct NVGpath* path;
+	NVGpathCache* cache = ctx->cache;
+//	NVGstate* state = nvg__getState(ctx);
+	NVGpoint* last;
+	NVGpoint* p0;
+	NVGpoint* p1;
+	NVGpoint* pts;
+	NVGpath* path;
 	int i, j;
 	float* cp1;
 	float* cp2;
@@ -1182,7 +1292,7 @@ static void nvg__flattenPaths(struct NVGcontext* ctx)
 				nvg__polyReverse(pts, path->count);
 		}
 
-		for(i = 0; i < path->count; ++i) {
+		for(i = 0; i < path->count; i++) {
 			// Calculate segment direction and length
 			p0->dx = p1->x - p0->x;
 			p0->dy = p1->y - p0->y;
@@ -1204,7 +1314,7 @@ static int nvg__curveDivs(float r, float arc, float tol)
 	return nvg__maxi(2, (int)ceilf(arc / da));
 }
 
-static void nvg__chooseBevel(int bevel, struct NVGpoint* p0, struct NVGpoint* p1, float w,
+static void nvg__chooseBevel(int bevel, NVGpoint* p0, NVGpoint* p1, float w,
 							float* x0, float* y0, float* x1, float* y1)
 {
 	if (bevel) {
@@ -1220,7 +1330,7 @@ static void nvg__chooseBevel(int bevel, struct NVGpoint* p0, struct NVGpoint* p1
 	}
 }
 
-static struct NVGvertex* nvg__roundJoin(struct NVGvertex* dst, struct NVGpoint* p0, struct NVGpoint* p1,
+static NVGvertex* nvg__roundJoin(NVGvertex* dst, NVGpoint* p0, NVGpoint* p1,
 										float lw, float rw, float lu, float ru, int ncap, float fringe)
 {
 	int i, n;
@@ -1280,12 +1390,11 @@ static struct NVGvertex* nvg__roundJoin(struct NVGvertex* dst, struct NVGpoint*
 	return dst;
 }
 
-static struct NVGvertex* nvg__bevelJoin(struct NVGvertex* dst, struct NVGpoint* p0, struct NVGpoint* p1,
+static NVGvertex* nvg__bevelJoin(NVGvertex* dst, NVGpoint* p0, NVGpoint* p1,
 										float lw, float rw, float lu, float ru, float fringe)
 {
 	float rx0,ry0,rx1,ry1;
 	float lx0,ly0,lx1,ly1;
-	float mx,my,len,mu;
 	float dlx0 = p0->dy;
 	float dly0 = -p0->dx;
 	float dlx1 = p1->dy;
@@ -1299,12 +1408,6 @@ static struct NVGvertex* nvg__bevelJoin(struct NVGvertex* dst, struct NVGpoint*
 		nvg__vset(dst, p1->x - dlx0*rw, p1->y - dly0*rw, ru,1); dst++;
 
 		if (p1->flags & NVG_PT_BEVEL) {
-			// TODO: this needs more work.
-			mx = (dlx0 + dlx1) * 0.5f;
-			my = (dly0 + dly1) * 0.5f;
-			len = sqrtf(mx*mx + my*my);
-			mu = ru + len*(lu-ru)*0.5f;
-
 			nvg__vset(dst, lx0, ly0, lu,1); dst++;
 			nvg__vset(dst, p1->x - dlx0*rw, p1->y - dly0*rw, ru,1); dst++;
 
@@ -1334,12 +1437,6 @@ static struct NVGvertex* nvg__bevelJoin(struct NVGvertex* dst, struct NVGpoint*
 		nvg__vset(dst, rx0, ry0, ru,1); dst++;
 
 		if (p1->flags & NVG_PT_BEVEL) {
-			// TODO: this needs more work.
-			mx = (dlx0 + dlx1) * 0.5f;
-			my = (dly0 + dly1) * 0.5f;
-			len = sqrtf(mx*mx + my*my);
-			mu = lu + len*(ru-lu)*0.5f;
-
 			nvg__vset(dst, p1->x + dlx0*lw, p1->y + dly0*lw, lu,1); dst++;
 			nvg__vset(dst, rx0, ry0, ru,1); dst++;
 
@@ -1364,35 +1461,96 @@ static struct NVGvertex* nvg__bevelJoin(struct NVGvertex* dst, struct NVGpoint*
 	}
 
 	return dst;
+}
 
-	NVG_NOTUSED(mu);
+static NVGvertex* nvg__buttCapStart(NVGvertex* dst, NVGpoint* p,
+										   float dx, float dy, float w, float d, float aa)
+{
+	float px = p->x - dx*d;
+	float py = p->y - dy*d;
+	float dlx = dy;
+	float dly = -dx;
+	nvg__vset(dst, px + dlx*w - dx*aa, py + dly*w - dy*aa, 0,0); dst++;
+	nvg__vset(dst, px - dlx*w - dx*aa, py - dly*w - dy*aa, 1,0); dst++;
+	nvg__vset(dst, px + dlx*w, py + dly*w, 0,1); dst++;
+	nvg__vset(dst, px - dlx*w, py - dly*w, 1,1); dst++;
+	return dst;
 }
 
-static int nvg__expandStrokeAndFill(struct NVGcontext* ctx, int feats, float w, int lineCap, int lineJoin, float miterLimit)
+static NVGvertex* nvg__buttCapEnd(NVGvertex* dst, NVGpoint* p,
+										   float dx, float dy, float w, float d, float aa)
 {
-	struct NVGpathCache* cache = ctx->cache;
-	struct NVGpath* path;
-	struct NVGpoint* pts;
-	struct NVGvertex* verts;
-	struct NVGvertex* dst;
-	struct NVGpoint* p0;
-	struct NVGpoint* p1;
-	int cverts, convex, i, j, s, e;
-	float wo = 0, iw = 0, aa = ctx->fringeWidth;
-	int ncap = nvg__curveDivs(w, NVG_PI, ctx->tessTol / 4.0f);
-	int nleft = 0;
+	float px = p->x + dx*d;
+	float py = p->y + dy*d;
+	float dlx = dy;
+	float dly = -dx;
+	nvg__vset(dst, px + dlx*w, py + dly*w, 0,1); dst++;
+	nvg__vset(dst, px - dlx*w, py - dly*w, 1,1); dst++;
+	nvg__vset(dst, px + dlx*w + dx*aa, py + dly*w + dy*aa, 0,0); dst++;
+	nvg__vset(dst, px - dlx*w + dx*aa, py - dly*w + dy*aa, 1,0); dst++;
+	return dst;
+}
+
+
+static NVGvertex* nvg__roundCapStart(NVGvertex* dst, NVGpoint* p,
+											float dx, float dy, float w, int ncap, float aa)
+{
+	int i;
+	float px = p->x;
+	float py = p->y;
+	float dlx = dy;
+	float dly = -dx;
+	NVG_NOTUSED(aa);
+	for (i = 0; i < ncap; i++) {
+		float a = i/(float)(ncap-1)*NVG_PI;
+		float ax = cosf(a) * w, ay = sinf(a) * w;
+		nvg__vset(dst, px - dlx*ax - dx*ay, py - dly*ax - dy*ay, 0,1); dst++;
+		nvg__vset(dst, px, py, 0.5f,1); dst++;
+	}
+	nvg__vset(dst, px + dlx*w, py + dly*w, 0,1); dst++;
+	nvg__vset(dst, px - dlx*w, py - dly*w, 1,1); dst++;
+	return dst;
+}
+
+static NVGvertex* nvg__roundCapEnd(NVGvertex* dst, NVGpoint* p,
+										  float dx, float dy, float w, int ncap, float aa)
+{
+	int i;
+	float px = p->x;
+	float py = p->y;
+	float dlx = dy;
+	float dly = -dx;
+	NVG_NOTUSED(aa);
+	nvg__vset(dst, px + dlx*w, py + dly*w, 0,1); dst++;
+	nvg__vset(dst, px - dlx*w, py - dly*w, 1,1); dst++;
+	for (i = 0; i < ncap; i++) {
+		float a = i/(float)(ncap-1)*NVG_PI;
+		float ax = cosf(a) * w, ay = sinf(a) * w;
+		nvg__vset(dst, px, py, 0.5f,1); dst++;
+		nvg__vset(dst, px - dlx*ax + dx*ay, py - dly*ax + dy*ay, 0,1); dst++;
+	}
+	return dst;
+}
+
+
+static void nvg__calculateJoins(NVGcontext* ctx, float w, int lineJoin, float miterLimit)
+{
+	NVGpathCache* cache = ctx->cache;
+	int i, j;
+	float iw = 0.0f;
 
 	if (w > 0.0f) iw = 1.0f / w;
 
 	// Calculate which joins needs extra vertices to append, and gather vertex count.
 	for (i = 0; i < cache->npaths; i++) {
-		path = &cache->paths[i];
-		pts = &cache->points[path->first];
+		NVGpath* path = &cache->paths[i];
+		NVGpoint* pts = &cache->points[path->first];
+		NVGpoint* p0 = &pts[path->count-1];
+		NVGpoint* p1 = &pts[0];
+		int nleft = 0;
+
 		path->nbevel = 0;
-		nleft = 0;
 
-		p0 = &pts[path->count-1];
-		p1 = &pts[0];
 		for (j = 0; j < path->count; j++) {
 			float dlx0, dly0, dlx1, dly1, dmr2, cross, limit;
 			dlx0 = p0->dy;
@@ -1442,160 +1600,222 @@ static int nvg__expandStrokeAndFill(struct NVGcontext* ctx, int feats, float w,
 
 		path->convex = (nleft == path->count) ? 1 : 0;
 	}
+}
+
+
+static int nvg__expandStroke(NVGcontext* ctx, float w, int lineCap, int lineJoin, float miterLimit)
+{	
+	NVGpathCache* cache = ctx->cache;
+	NVGvertex* verts;
+	NVGvertex* dst;
+	int cverts, i, j;
+	float aa = ctx->fringeWidth;
+	int ncap = nvg__curveDivs(w, NVG_PI, ctx->tessTol);	// Calculate divisions per half circle.
+
+	nvg__calculateJoins(ctx, w, lineJoin, miterLimit);
 
 	// Calculate max vertex usage.
 	cverts = 0;
 	for (i = 0; i < cache->npaths; i++) {
-		path = &cache->paths[i];
-		if (feats & NVG_FILL)
-			cverts += path->count + path->nbevel + 1;
-		if (feats & NVG_STROKE) {
-			int loop = ((feats & NVG_CAPS) && path->closed == 0) ? 0 : 1;
-			if (lineCap == NVG_ROUND)
-				cverts += (path->count + path->nbevel*(ncap+2) + 1) * 2; // plus one for loop
-			else
-				cverts += (path->count + path->nbevel*5 + 1) * 2; // plus one for loop
-			if (loop == 0) {
-				// space for caps
-				if (lineCap == NVG_ROUND) {
-					cverts += (ncap*2 + 2)*2;
+		NVGpath* path = &cache->paths[i];
+		int loop = (path->closed == 0) ? 0 : 1;
+		if (lineJoin == NVG_ROUND)
+			cverts += (path->count + path->nbevel*(ncap+2) + 1) * 2; // plus one for loop
+		else
+			cverts += (path->count + path->nbevel*5 + 1) * 2; // plus one for loop
+		if (loop == 0) {
+			// space for caps
+			if (lineCap == NVG_ROUND) {
+				cverts += (ncap*2 + 2)*2;
+			} else {
+				cverts += (3+3)*2;
+			}
+		}
+	}
+
+	verts = nvg__allocTempVerts(ctx, cverts);
+	if (verts == NULL) return 0;
+
+	for (i = 0; i < cache->npaths; i++) {
+		NVGpath* path = &cache->paths[i];
+		NVGpoint* pts = &cache->points[path->first];
+		NVGpoint* p0;
+		NVGpoint* p1;
+		int s, e, loop;
+		float dx, dy;
+
+		path->fill = 0;
+		path->nfill = 0;
+
+		// Calculate fringe or stroke
+		loop = (path->closed == 0) ? 0 : 1;
+		dst = verts;
+		path->stroke = dst;
+
+		if (loop) {
+			// Looping
+			p0 = &pts[path->count-1];
+			p1 = &pts[0];
+			s = 0;
+			e = path->count;
+		} else {
+			// Add cap
+			p0 = &pts[0];
+			p1 = &pts[1];
+			s = 1;
+			e = path->count-1;
+		}
+
+		if (loop == 0) {
+			// Add cap
+			dx = p1->x - p0->x;
+			dy = p1->y - p0->y;
+			nvg__normalize(&dx, &dy);
+			if (lineCap == NVG_BUTT)
+				dst = nvg__buttCapStart(dst, p0, dx, dy, w, -aa*0.5f, aa);
+			else if (lineCap == NVG_BUTT || lineCap == NVG_SQUARE)
+				dst = nvg__buttCapStart(dst, p0, dx, dy, w, w-aa, aa);
+			else if (lineCap == NVG_ROUND)
+				dst = nvg__roundCapStart(dst, p0, dx, dy, w, ncap, aa);
+		}
+
+		for (j = s; j < e; ++j) {
+			if ((p1->flags & (NVG_PT_BEVEL | NVG_PR_INNERBEVEL)) != 0) {
+				if (lineJoin == NVG_ROUND) {
+					dst = nvg__roundJoin(dst, p0, p1, w, w, 0, 1, ncap, aa);
 				} else {
-					cverts += (3+3)*2;
+					dst = nvg__bevelJoin(dst, p0, p1, w, w, 0, 1, aa);
 				}
+			} else {
+				nvg__vset(dst, p1->x + (p1->dmx * w), p1->y + (p1->dmy * w), 0,1); dst++;
+				nvg__vset(dst, p1->x - (p1->dmx * w), p1->y - (p1->dmy * w), 1,1); dst++;
 			}
+			p0 = p1++;
 		}
+
+		if (loop) {
+			// Loop it
+			nvg__vset(dst, verts[0].x, verts[0].y, 0,1); dst++;
+			nvg__vset(dst, verts[1].x, verts[1].y, 1,1); dst++;
+		} else {
+			// Add cap
+			dx = p1->x - p0->x;
+			dy = p1->y - p0->y;
+			nvg__normalize(&dx, &dy);
+			if (lineCap == NVG_BUTT)
+				dst = nvg__buttCapEnd(dst, p1, dx, dy, w, -aa*0.5f, aa);
+			else if (lineCap == NVG_BUTT || lineCap == NVG_SQUARE)
+				dst = nvg__buttCapEnd(dst, p1, dx, dy, w, w-aa, aa);
+			else if (lineCap == NVG_ROUND)
+				dst = nvg__roundCapEnd(dst, p1, dx, dy, w, ncap, aa);
+		}
+
+		path->nstroke = (int)(dst - verts);
+
+		verts = dst;
+	}
+
+	return 1;
+}
+
+static int nvg__expandFill(NVGcontext* ctx, float w, int lineJoin, float miterLimit)
+{
+	NVGpathCache* cache = ctx->cache;
+	NVGvertex* verts;
+	NVGvertex* dst;
+	int cverts, convex, i, j;
+	float aa = ctx->fringeWidth;
+	int fringe = w > 0.0f;
+
+	nvg__calculateJoins(ctx, w, lineJoin, miterLimit);
+
+	// Calculate max vertex usage.
+	cverts = 0;
+	for (i = 0; i < cache->npaths; i++) {
+		NVGpath* path = &cache->paths[i];
+		cverts += path->count + path->nbevel + 1;
+		if (fringe)
+			cverts += (path->count + path->nbevel*5 + 1) * 2; // plus one for loop
 	}
 
 	verts = nvg__allocTempVerts(ctx, cverts);
 	if (verts == NULL) return 0;
 
-	if ((feats & NVG_FILL) && cache->npaths == 1 && cache->paths[0].convex)
-		convex = 1;
-	else
-		convex = 0;
+	convex = cache->npaths == 1 && cache->paths[0].convex;
 
 	for (i = 0; i < cache->npaths; i++) {
-		path = &cache->paths[i];
-		pts = &cache->points[path->first];
+		NVGpath* path = &cache->paths[i];
+		NVGpoint* pts = &cache->points[path->first];
+		NVGpoint* p0;
+		NVGpoint* p1;
+		float rw, lw, woff;
+		float ru, lu;
 
 		// Calculate shape vertices.
-		if (feats & NVG_FILL) {
-			wo = 0.5f*aa;
-			dst = verts;
-			path->fill = dst;
+		woff = 0.5f*aa;
+		dst = verts;
+		path->fill = dst;
 
-			if (w == 0.0f) {
-				for (j = 0; j < path->count; ++j) {
-					nvg__vset(dst, pts[j].x, pts[j].y, 0.5f,1);
-					dst++;
-				}
-			} else {
-				// Looping
-				p0 = &pts[path->count-1];
-				p1 = &pts[0];
-				for (j = 0; j < path->count; ++j) {
-					if (p1->flags & NVG_PT_BEVEL) {
-						float dlx0 = p0->dy;
-						float dly0 = -p0->dx;
-						float dlx1 = p1->dy;
-						float dly1 = -p1->dx;
-						if (p1->flags & NVG_PT_LEFT) {
-							float lx = p1->x + p1->dmx * wo;
-							float ly = p1->y + p1->dmy * wo;
-							nvg__vset(dst, lx, ly, 0.5f,1); dst++;
-						} else {
-							float lx0 = p1->x + dlx0 * wo;
-							float ly0 = p1->y + dly0 * wo;
-							float lx1 = p1->x + dlx1 * wo;
-							float ly1 = p1->y + dly1 * wo;
-							nvg__vset(dst, lx0, ly0, 0.5f,1); dst++;
-							nvg__vset(dst, lx1, ly1, 0.5f,1); dst++;
-						}
+		if (fringe) {
+			// Looping
+			p0 = &pts[path->count-1];
+			p1 = &pts[0];
+			for (j = 0; j < path->count; ++j) {
+				if (p1->flags & NVG_PT_BEVEL) {
+					float dlx0 = p0->dy;
+					float dly0 = -p0->dx;
+					float dlx1 = p1->dy;
+					float dly1 = -p1->dx;
+					if (p1->flags & NVG_PT_LEFT) {
+						float lx = p1->x + p1->dmx * woff;
+						float ly = p1->y + p1->dmy * woff;
+						nvg__vset(dst, lx, ly, 0.5f,1); dst++;
 					} else {
-						nvg__vset(dst, p1->x + (p1->dmx * wo), p1->y + (p1->dmy * wo), 0.5f,1); dst++;
+						float lx0 = p1->x + dlx0 * woff;
+						float ly0 = p1->y + dly0 * woff;
+						float lx1 = p1->x + dlx1 * woff;
+						float ly1 = p1->y + dly1 * woff;
+						nvg__vset(dst, lx0, ly0, 0.5f,1); dst++;
+						nvg__vset(dst, lx1, ly1, 0.5f,1); dst++;
 					}
-					p0 = p1++;
+				} else {
+					nvg__vset(dst, p1->x + (p1->dmx * woff), p1->y + (p1->dmy * woff), 0.5f,1); dst++;
 				}
+				p0 = p1++;
 			}
-
-			path->nfill = (int)(dst - verts);
-			verts = dst;
 		} else {
-			wo = 0.0f;
-			path->fill = 0;
-			path->nfill = 0;
+			for (j = 0; j < path->count; ++j) {
+				nvg__vset(dst, pts[j].x, pts[j].y, 0.5f,1);
+				dst++;
+			}
 		}
 
-		// Calculate fringe or stroke
-		if (feats & NVG_STROKE) {
-			float lw = w + wo, rw = w - wo;
-			float lu = 0, ru = 1;
-			int loop = ((feats & NVG_CAPS) && path->closed == 0) ? 0 : 1;
+		path->nfill = (int)(dst - verts);
+		verts = dst;
+
+		// Calculate fringe
+		if (fringe) {
+			lw = w + woff;
+			rw = w - woff;
+			lu = 0;
+			ru = 1;
 			dst = verts;
 			path->stroke = dst;
 
 			// Create only half a fringe for convex shapes so that
 			// the shape can be rendered without stenciling.
 			if (convex) {
-				lw = wo;	// This should generate the same vertex as fill inset above.
+				lw = woff;	// This should generate the same vertex as fill inset above.
 				lu = 0.5f;	// Set outline fade at middle.
 			}
 
-			if (loop) {
-				// Looping
-				p0 = &pts[path->count-1];
-				p1 = &pts[0];
-				s = 0;
-				e = path->count;
-			} else {
-				// Add cap
-				p0 = &pts[0];
-				p1 = &pts[1];
-				s = 1;
-				e = path->count-1;
-			}
-
-			if (loop == 0) {
-				// Add cap
-				float dx, dy, dlx, dly, px, py;
-				dx = p1->x - p0->x;
-				dy = p1->y - p0->y;
-				nvg__normalize(&dx, &dy);
-				dlx = dy;
-				dly = -dx;
-				if (lineCap == NVG_BUTT || lineCap == NVG_SQUARE) {
-					if (lineCap == NVG_BUTT) {
-						px = p0->x + dx*ctx->fringeWidth*0.5f;
-						py = p0->y + dy*ctx->fringeWidth*0.5f;
-					} else /*if (lineCap == NVG_SQUARE)*/ {
-						px = p0->x - dx*(w - ctx->fringeWidth);
-						py = p0->y - dy*(w - ctx->fringeWidth);
-					}
-					nvg__vset(dst, px + dlx*lw - dx*aa, py + dly*lw - dy*aa, lu,0); dst++;
-					nvg__vset(dst, px - dlx*rw - dx*aa, py - dly*rw - dy*aa, ru,0); dst++;
-					nvg__vset(dst, px + dlx*lw, py + dly * lw, lu,1); dst++;
-					nvg__vset(dst, px - dlx*rw, py - dly * rw, ru,1); dst++;
-				} else if (lineCap == NVG_ROUND) {
-					px = p0->x;
-					py = p0->y;
-					for (j = 0; j < ncap; j++) {
-						float a = j/(float)(ncap-1)*NVG_PI;
-						float ax = cosf(a) * w, ay = sinf(a) * w;
-						nvg__vset(dst, px - dlx*ax - dx*ay, py - dly*ax - dy*ay, lu,1); dst++;
-						nvg__vset(dst, px, py, 0.5f,1); dst++;
-					}
-					nvg__vset(dst, px + dlx*lw, py + dly * lw, lu,1); dst++;
-					nvg__vset(dst, px - dlx*rw, py - dly * rw, ru,1); dst++;
-				}
-			}
+			// Looping
+			p0 = &pts[path->count-1];
+			p1 = &pts[0];
 
-			for (j = s; j < e; ++j) {
+			for (j = 0; j < path->count; ++j) {
 				if ((p1->flags & (NVG_PT_BEVEL | NVG_PR_INNERBEVEL)) != 0) {
-					if (lineJoin == NVG_ROUND) {
-						dst = nvg__roundJoin(dst, p0, p1, lw, rw, lu, ru, ncap, ctx->fringeWidth);
-					} else {
-						dst = nvg__bevelJoin(dst, p0, p1, lw, rw, lu, ru, ctx->fringeWidth);
-					}
+					dst = nvg__bevelJoin(dst, p0, p1, lw, rw, lu, ru, ctx->fringeWidth);
 				} else {
 					nvg__vset(dst, p1->x + (p1->dmx * lw), p1->y + (p1->dmy * lw), lu,1); dst++;
 					nvg__vset(dst, p1->x - (p1->dmx * rw), p1->y - (p1->dmy * rw), ru,1); dst++;
@@ -1603,49 +1823,14 @@ static int nvg__expandStrokeAndFill(struct NVGcontext* ctx, int feats, float w,
 				p0 = p1++;
 			}
 
-			if (loop) {
-				// Loop it
-				nvg__vset(dst, verts[0].x, verts[0].y, lu,1); dst++;
-				nvg__vset(dst, verts[1].x, verts[1].y, ru,1); dst++;
-			} else {
-				// Add cap
-				float dx, dy, dlx, dly, px, py;
-				dx = p1->x - p0->x;
-				dy = p1->y - p0->y;
-				nvg__normalize(&dx, &dy);
-				dlx = dy;
-				dly = -dx;
-				if (lineCap == NVG_BUTT || lineCap == NVG_SQUARE) {
-					if (lineCap == NVG_BUTT) {
-						px = p1->x - dx*ctx->fringeWidth*0.5f;
-						py = p1->y - dy*ctx->fringeWidth*0.5f;
-					} else /*if (lineCap == NVG_SQUARE)*/ {
-						px = p1->x + dx*(w - ctx->fringeWidth);
-						py = p1->y + dy*(w - ctx->fringeWidth);
-					}
-					nvg__vset(dst, px + dlx*lw, py + dly * lw, lu,1); dst++;
-					nvg__vset(dst, px - dlx*rw, py - dly * rw, ru,1); dst++;
-					nvg__vset(dst, px + dlx*lw + dx*aa, py + dly*lw + dy*aa, lu,0); dst++;
-					nvg__vset(dst, px - dlx*rw + dx*aa, py - dly*rw + dy*aa, ru,0); dst++;
-				} else if (lineCap == NVG_ROUND) {
-					px = p1->x;
-					py = p1->y;
-					nvg__vset(dst, px + dlx*lw, py + dly * lw, lu,1); dst++;
-					nvg__vset(dst, px - dlx*rw, py - dly * rw, ru,1); dst++;
-					for (j = 0; j < ncap; j++) {
-						float a = j/(float)(ncap-1)*NVG_PI;
-						float ax = cosf(a) * w, ay = sinf(a) * w;
-						nvg__vset(dst, px, py, 0.5f,1); dst++;
-						nvg__vset(dst, px - dlx*ax + dx*ay, py - dly*ax + dy*ay, lu,1); dst++;
-					}
-				}
-			}
+			// Loop it
+			nvg__vset(dst, verts[0].x, verts[0].y, lu,1); dst++;
+			nvg__vset(dst, verts[1].x, verts[1].y, ru,1); dst++;
 
 			path->nstroke = (int)(dst - verts);
-
 			verts = dst;
 		} else {
-			path->stroke = 0;
+			path->stroke = NULL;
 			path->nstroke = 0;
 		}
 	}
@@ -1655,31 +1840,42 @@ static int nvg__expandStrokeAndFill(struct NVGcontext* ctx, int feats, float w,
 
 
 // Draw
-void nvgBeginPath(struct NVGcontext* ctx)
+void nvgBeginPath(NVGcontext* ctx)
 {
 	ctx->ncommands = 0;
 	nvg__clearPathCache(ctx);
 }
 
-void nvgMoveTo(struct NVGcontext* ctx, float x, float y)
+void nvgMoveTo(NVGcontext* ctx, float x, float y)
 {
 	float vals[] = { NVG_MOVETO, x, y };
 	nvg__appendCommands(ctx, vals, NVG_COUNTOF(vals));
 }
 
-void nvgLineTo(struct NVGcontext* ctx, float x, float y)
+void nvgLineTo(NVGcontext* ctx, float x, float y)
 {
 	float vals[] = { NVG_LINETO, x, y };
 	nvg__appendCommands(ctx, vals, NVG_COUNTOF(vals));
 }
 
-void nvgBezierTo(struct NVGcontext* ctx, float c1x, float c1y, float c2x, float c2y, float x, float y)
+void nvgBezierTo(NVGcontext* ctx, float c1x, float c1y, float c2x, float c2y, float x, float y)
 {
 	float vals[] = { NVG_BEZIERTO, c1x, c1y, c2x, c2y, x, y };
 	nvg__appendCommands(ctx, vals, NVG_COUNTOF(vals));
 }
 
-void nvgArcTo(struct NVGcontext* ctx, float x1, float y1, float x2, float y2, float radius)
+void nvgQuadTo(NVGcontext* ctx, float cx, float cy, float x, float y)
+{
+    float x0 = ctx->commandx;
+    float y0 = ctx->commandy;
+    float vals[] = { NVG_BEZIERTO, 
+        x0 + 2.0f/3.0f*(cx - x0), y0 + 2.0f/3.0f*(cy - y0),
+        x + 2.0f/3.0f*(cx - x), y + 2.0f/3.0f*(cy - y),
+        x, y };
+    nvg__appendCommands(ctx, vals, NVG_COUNTOF(vals));
+}
+
+void nvgArcTo(NVGcontext* ctx, float x1, float y1, float x2, float y2, float radius)
 {
 	float x0 = ctx->commandx;
 	float y0 = ctx->commandy;
@@ -1735,26 +1931,26 @@ void nvgArcTo(struct NVGcontext* ctx, float x1, float y1, float x2, float y2, fl
 	nvgArc(ctx, cx, cy, radius, a0, a1, dir);
 }
 
-void nvgClosePath(struct NVGcontext* ctx)
+void nvgClosePath(NVGcontext* ctx)
 {
 	float vals[] = { NVG_CLOSE };
 	nvg__appendCommands(ctx, vals, NVG_COUNTOF(vals));
 }
 
-void nvgPathWinding(struct NVGcontext* ctx, int dir)
+void nvgPathWinding(NVGcontext* ctx, int dir)
 {
 	float vals[] = { NVG_WINDING, (float)dir };
 	nvg__appendCommands(ctx, vals, NVG_COUNTOF(vals));
 }
 
-void nvgArc(struct NVGcontext* ctx, float cx, float cy, float r, float a0, float a1, int dir)
+void nvgArc(NVGcontext* ctx, float cx, float cy, float r, float a0, float a1, int dir)
 {
 	float a = 0, da = 0, hda = 0, kappa = 0;
 	float dx = 0, dy = 0, x = 0, y = 0, tanx = 0, tany = 0;
 	float px = 0, py = 0, ptanx = 0, ptany = 0;
 	float vals[3 + 5*7 + 100];
 	int i, ndivs, nvals;
-	int move = ctx->ncommands > 0 ? NVG_LINETO : NVG_MOVETO;
+	int move = ctx->ncommands > 0 ? NVG_LINETO : NVG_MOVETO; 
 
 	// Clamp angles
 	da = a1 - a0;
@@ -1812,62 +2008,63 @@ void nvgArc(struct NVGcontext* ctx, float cx, float cy, float r, float a0, float
 	nvg__appendCommands(ctx, vals, nvals);
 }
 
-void nvgRect(struct NVGcontext* ctx, float x, float y, float w, float h)
+void nvgRect(NVGcontext* ctx, float x, float y, float w, float h)
 {
 	float vals[] = {
 		NVG_MOVETO, x,y,
-		NVG_LINETO, x+w,y,
-		NVG_LINETO, x+w,y+h,
 		NVG_LINETO, x,y+h,
+		NVG_LINETO, x+w,y+h,
+		NVG_LINETO, x+w,y,
 		NVG_CLOSE
 	};
 	nvg__appendCommands(ctx, vals, NVG_COUNTOF(vals));
 }
 
-void nvgRoundedRect(struct NVGcontext* ctx, float x, float y, float w, float h, float r)
+void nvgRoundedRect(NVGcontext* ctx, float x, float y, float w, float h, float r)
 {
 	if (r < 0.1f) {
 		nvgRect(ctx, x,y,w,h);
 		return;
 	}
 	else {
+		float rx = nvg__minf(r, nvg__absf(w)*0.5f) * nvg__signf(w), ry = nvg__minf(r, nvg__absf(h)*0.5f) * nvg__signf(h);
 		float vals[] = {
-			NVG_MOVETO, x+r, y,
-			NVG_LINETO, x+w-r, y,
-			NVG_BEZIERTO, x+w-r*(1-NVG_KAPPA90), y, x+w, y+r*(1-NVG_KAPPA90), x+w, y+r,
-			NVG_LINETO, x+w, y+h-r,
-			NVG_BEZIERTO, x+w, y+h-r*(1-NVG_KAPPA90), x+w-r*(1-NVG_KAPPA90), y+h, x+w-r, y+h,
-			NVG_LINETO, x+r, y+h,
-			NVG_BEZIERTO, x+r*(1-NVG_KAPPA90), y+h, x, y+h-r*(1-NVG_KAPPA90), x, y+h-r,
-			NVG_LINETO, x, y+r,
-			NVG_BEZIERTO, x, y+r*(1-NVG_KAPPA90), x+r*(1-NVG_KAPPA90), y, x+r, y,
+			NVG_MOVETO, x, y+ry,
+			NVG_LINETO, x, y+h-ry,
+			NVG_BEZIERTO, x, y+h-ry*(1-NVG_KAPPA90), x+rx*(1-NVG_KAPPA90), y+h, x+rx, y+h,
+			NVG_LINETO, x+w-rx, y+h,
+			NVG_BEZIERTO, x+w-rx*(1-NVG_KAPPA90), y+h, x+w, y+h-ry*(1-NVG_KAPPA90), x+w, y+h-ry,
+			NVG_LINETO, x+w, y+ry,
+			NVG_BEZIERTO, x+w, y+ry*(1-NVG_KAPPA90), x+w-rx*(1-NVG_KAPPA90), y, x+w-rx, y,
+			NVG_LINETO, x+rx, y,
+			NVG_BEZIERTO, x+rx*(1-NVG_KAPPA90), y, x, y+ry*(1-NVG_KAPPA90), x, y+ry,
 			NVG_CLOSE
 		};
 		nvg__appendCommands(ctx, vals, NVG_COUNTOF(vals));
 	}
 }
 
-void nvgEllipse(struct NVGcontext* ctx, float cx, float cy, float rx, float ry)
+void nvgEllipse(NVGcontext* ctx, float cx, float cy, float rx, float ry)
 {
 	float vals[] = {
-		NVG_MOVETO, cx+rx, cy,
-		NVG_BEZIERTO, cx+rx, cy+ry*NVG_KAPPA90, cx+rx*NVG_KAPPA90, cy+ry, cx, cy+ry,
-		NVG_BEZIERTO, cx-rx*NVG_KAPPA90, cy+ry, cx-rx, cy+ry*NVG_KAPPA90, cx-rx, cy,
-		NVG_BEZIERTO, cx-rx, cy-ry*NVG_KAPPA90, cx-rx*NVG_KAPPA90, cy-ry, cx, cy-ry,
-		NVG_BEZIERTO, cx+rx*NVG_KAPPA90, cy-ry, cx+rx, cy-ry*NVG_KAPPA90, cx+rx, cy,
+		NVG_MOVETO, cx-rx, cy,
+		NVG_BEZIERTO, cx-rx, cy+ry*NVG_KAPPA90, cx-rx*NVG_KAPPA90, cy+ry, cx, cy+ry,
+		NVG_BEZIERTO, cx+rx*NVG_KAPPA90, cy+ry, cx+rx, cy+ry*NVG_KAPPA90, cx+rx, cy,
+		NVG_BEZIERTO, cx+rx, cy-ry*NVG_KAPPA90, cx+rx*NVG_KAPPA90, cy-ry, cx, cy-ry,
+		NVG_BEZIERTO, cx-rx*NVG_KAPPA90, cy-ry, cx-rx, cy-ry*NVG_KAPPA90, cx-rx, cy,
 		NVG_CLOSE
 	};
 	nvg__appendCommands(ctx, vals, NVG_COUNTOF(vals));
 }
 
-void nvgCircle(struct NVGcontext* ctx, float cx, float cy, float r)
+void nvgCircle(NVGcontext* ctx, float cx, float cy, float r)
 {
 	nvgEllipse(ctx, cx,cy, r,r);
 }
 
-void nvgDebugDumpPathCache(struct NVGcontext* ctx)
+void nvgDebugDumpPathCache(NVGcontext* ctx)
 {
-	const struct NVGpath* path;
+	const NVGpath* path;
 	int i, j;
 
 	printf("Dumping %d cached paths\n", ctx->cache->npaths);
@@ -1887,19 +2084,24 @@ void nvgDebugDumpPathCache(struct NVGcontext* ctx)
 	}
 }
 
-void nvgFill(struct NVGcontext* ctx)
+void nvgFill(NVGcontext* ctx)
 {
-	struct NVGstate* state = nvg__getState(ctx);
-	const struct NVGpath* path;
+	NVGstate* state = nvg__getState(ctx);
+	const NVGpath* path;
+	NVGpaint fillPaint = state->fill;
 	int i;
 
 	nvg__flattenPaths(ctx);
 	if (ctx->params.edgeAntiAlias)
-		nvg__expandStrokeAndFill(ctx, NVG_FILL|NVG_STROKE, ctx->fringeWidth, NVG_BUTT, NVG_MITER, 3.6f);
+		nvg__expandFill(ctx, ctx->fringeWidth, NVG_MITER, 2.4f);
 	else
-		nvg__expandStrokeAndFill(ctx, NVG_FILL, 0.0f, NVG_BUTT, NVG_MITER, 1.2f);
+		nvg__expandFill(ctx, 0.0f, NVG_MITER, 2.4f);
+
+	// Apply global alpha
+	fillPaint.innerColor.a *= state->alpha;
+	fillPaint.outerColor.a *= state->alpha;
 
-	ctx->params.renderFill(ctx->params.userPtr, &state->fill, &state->scissor, ctx->fringeWidth,
+	ctx->params.renderFill(ctx->params.userPtr, &fillPaint, &state->scissor, ctx->fringeWidth,
 						   ctx->cache->bounds, ctx->cache->paths, ctx->cache->npaths);
 
 	// Count triangles
@@ -1911,17 +2113,17 @@ void nvgFill(struct NVGcontext* ctx)
 	}
 }
 
-void nvgStroke(struct NVGcontext* ctx)
+void nvgStroke(NVGcontext* ctx)
 {
-	struct NVGstate* state = nvg__getState(ctx);
+	NVGstate* state = nvg__getState(ctx);
 	float scale = nvg__getAverageScale(state->xform);
-	float strokeWidth = nvg__clampf(state->strokeWidth * scale, 0.0f, 20.0f);
-	struct NVGpaint strokePaint = state->stroke;
-	const struct NVGpath* path;
+	float strokeWidth = nvg__clampf(state->strokeWidth * scale, 0.0f, 200.0f);
+	NVGpaint strokePaint = state->stroke;
+	const NVGpath* path;
 	int i;
 
 	if (strokeWidth < ctx->fringeWidth) {
-		// If the stroke width is less than pixel size, use alpha to emulate coverate.
+		// If the stroke width is less than pixel size, use alpha to emulate coverage.
 		// Since coverage is area, scale by alpha*alpha.
 		float alpha = nvg__clampf(strokeWidth / ctx->fringeWidth, 0.0f, 1.0f);
 		strokePaint.innerColor.a *= alpha*alpha;
@@ -1929,11 +2131,16 @@ void nvgStroke(struct NVGcontext* ctx)
 		strokeWidth = ctx->fringeWidth;
 	}
 
+	// Apply global alpha
+	strokePaint.innerColor.a *= state->alpha;
+	strokePaint.outerColor.a *= state->alpha;
+
 	nvg__flattenPaths(ctx);
+
 	if (ctx->params.edgeAntiAlias)
-		nvg__expandStrokeAndFill(ctx, NVG_STROKE|NVG_CAPS, strokeWidth*0.5f + ctx->fringeWidth*0.5f, state->lineCap, state->lineJoin, state->miterLimit);
+		nvg__expandStroke(ctx, strokeWidth*0.5f + ctx->fringeWidth*0.5f, state->lineCap, state->lineJoin, state->miterLimit);
 	else
-		nvg__expandStrokeAndFill(ctx, NVG_STROKE|NVG_CAPS, strokeWidth*0.5f, state->lineCap, state->lineJoin, state->miterLimit);
+		nvg__expandStroke(ctx, strokeWidth*0.5f, state->lineCap, state->lineJoin, state->miterLimit);
 
 	ctx->params.renderStroke(ctx->params.userPtr, &strokePaint, &state->scissor, ctx->fringeWidth,
 							 strokeWidth, ctx->cache->paths, ctx->cache->npaths);
@@ -1947,62 +2154,62 @@ void nvgStroke(struct NVGcontext* ctx)
 }
 
 // Add fonts
-int nvgCreateFont(struct NVGcontext* ctx, const char* name, const char* path)
+int nvgCreateFont(NVGcontext* ctx, const char* name, const char* path)
 {
 	return fonsAddFont(ctx->fs, name, path);
 }
 
-int nvgCreateFontMem(struct NVGcontext* ctx, const char* name, unsigned char* data, int ndata, int freeData)
+int nvgCreateFontMem(NVGcontext* ctx, const char* name, unsigned char* data, int ndata, int freeData)
 {
 	return fonsAddFontMem(ctx->fs, name, data, ndata, freeData);
 }
 
-int nvgFindFont(struct NVGcontext* ctx, const char* name)
+int nvgFindFont(NVGcontext* ctx, const char* name)
 {
 	if (name == NULL) return -1;
 	return fonsGetFontByName(ctx->fs, name);
 }
 
 // State setting
-void nvgFontSize(struct NVGcontext* ctx, float size)
+void nvgFontSize(NVGcontext* ctx, float size)
 {
-	struct NVGstate* state = nvg__getState(ctx);
+	NVGstate* state = nvg__getState(ctx);
 	state->fontSize = size;
 }
 
-void nvgFontBlur(struct NVGcontext* ctx, float blur)
+void nvgFontBlur(NVGcontext* ctx, float blur)
 {
-	struct NVGstate* state = nvg__getState(ctx);
+	NVGstate* state = nvg__getState(ctx);
 	state->fontBlur = blur;
 }
 
-void nvgTextLetterSpacing(struct NVGcontext* ctx, float spacing)
+void nvgTextLetterSpacing(NVGcontext* ctx, float spacing)
 {
-	struct NVGstate* state = nvg__getState(ctx);
+	NVGstate* state = nvg__getState(ctx);
 	state->letterSpacing = spacing;
 }
 
-void nvgTextLineHeight(struct NVGcontext* ctx, float lineHeight)
+void nvgTextLineHeight(NVGcontext* ctx, float lineHeight)
 {
-	struct NVGstate* state = nvg__getState(ctx);
+	NVGstate* state = nvg__getState(ctx);
 	state->lineHeight = lineHeight;
 }
 
-void nvgTextAlign(struct NVGcontext* ctx, int align)
+void nvgTextAlign(NVGcontext* ctx, int align)
 {
-	struct NVGstate* state = nvg__getState(ctx);
+	NVGstate* state = nvg__getState(ctx);
 	state->textAlign = align;
 }
 
-void nvgFontFaceId(struct NVGcontext* ctx, int font)
+void nvgFontFaceId(NVGcontext* ctx, int font)
 {
-	struct NVGstate* state = nvg__getState(ctx);
+	NVGstate* state = nvg__getState(ctx);
 	state->fontId = font;
 }
 
-void nvgFontFace(struct NVGcontext* ctx, const char* font)
+void nvgFontFace(NVGcontext* ctx, const char* font)
 {
-	struct NVGstate* state = nvg__getState(ctx);
+	NVGstate* state = nvg__getState(ctx);
 	state->fontId = fonsGetFontByName(ctx->fs, font);
 }
 
@@ -2011,21 +2218,80 @@ static float nvg__quantize(float a, float d)
 	return ((int)(a / d + 0.5f)) * d;
 }
 
-static float nvg__getFontScale(struct NVGstate* state)
+static float nvg__getFontScale(NVGstate* state)
 {
 	return nvg__minf(nvg__quantize(nvg__getAverageScale(state->xform), 0.01f), 4.0f);
 }
 
-float nvgText(struct NVGcontext* ctx, float x, float y, const char* string, const char* end)
+static void nvg__flushTextTexture(NVGcontext* ctx)
+{
+	int dirty[4];
+
+	if (fonsValidateTexture(ctx->fs, dirty)) {
+		int fontImage = ctx->fontImages[ctx->fontImageIdx];
+		// Update texture
+		if (fontImage != 0) {
+			int iw, ih;
+			const unsigned char* data = fonsGetTextureData(ctx->fs, &iw, &ih);
+			int x = dirty[0];
+			int y = dirty[1];
+			int w = dirty[2] - dirty[0];
+			int h = dirty[3] - dirty[1];
+			ctx->params.renderUpdateTexture(ctx->params.userPtr, fontImage, x,y, w,h, data);
+		}
+	}
+}
+
+static int nvg__allocTextAtlas(NVGcontext* ctx)
+{
+	int iw, ih;
+	nvg__flushTextTexture(ctx);
+	if (ctx->fontImageIdx >= NVG_MAX_FONTIMAGES-1)
+		return 0;
+	// if next fontImage already have a texture
+	if (ctx->fontImages[ctx->fontImageIdx+1] != 0)
+		nvgImageSize(ctx, ctx->fontImages[ctx->fontImageIdx+1], &iw, &ih);
+	else { // calculate the new font image size and create it.
+		nvgImageSize(ctx, ctx->fontImages[ctx->fontImageIdx], &iw, &ih);
+		if (iw > ih)
+			ih *= 2;
+		else
+			iw *= 2;
+		if (iw > NVG_MAX_FONTIMAGE_SIZE || ih > NVG_MAX_FONTIMAGE_SIZE)
+			iw = ih = NVG_MAX_FONTIMAGE_SIZE;
+		ctx->fontImages[ctx->fontImageIdx+1] = ctx->params.renderCreateTexture(ctx->params.userPtr, NVG_TEXTURE_ALPHA, iw, ih, 0, NULL);
+	}
+	++ctx->fontImageIdx;
+	fonsResetAtlas(ctx->fs, iw, ih);
+	return 1;
+}
+
+static void nvg__renderText(NVGcontext* ctx, NVGvertex* verts, int nverts)
+{
+	NVGstate* state = nvg__getState(ctx);
+	NVGpaint paint = state->fill;
+
+	// Render triangles.
+	paint.image = ctx->fontImages[ctx->fontImageIdx];
+
+	// Apply global alpha
+	paint.innerColor.a *= state->alpha;
+	paint.outerColor.a *= state->alpha;
+
+	ctx->params.renderTriangles(ctx->params.userPtr, &paint, &state->scissor, verts, nverts);
+
+	ctx->drawCallCount++;
+	ctx->textTriCount += nverts/3;
+}
+
+float nvgText(NVGcontext* ctx, float x, float y, const char* string, const char* end)
 {
-	struct NVGstate* state = nvg__getState(ctx);
-	struct NVGpaint paint;
-	struct FONStextIter iter;
-	struct FONSquad q;
-	struct NVGvertex* verts;
+	NVGstate* state = nvg__getState(ctx);
+	FONStextIter iter, prevIter;
+	FONSquad q;
+	NVGvertex* verts;
 	float scale = nvg__getFontScale(state) * ctx->devicePxRatio;
 	float invscale = 1.0f / scale;
-	int dirty[4];
 	int cverts = 0;
 	int nverts = 0;
 
@@ -2045,9 +2311,23 @@ float nvgText(struct NVGcontext* ctx, float x, float y, const char* string, cons
 	if (verts == NULL) return x;
 
 	fonsTextIterInit(ctx->fs, &iter, x*scale, y*scale, string, end);
+	prevIter = iter;
 	while (fonsTextIterNext(ctx->fs, &iter, &q)) {
-		// Trasnform corners.
 		float c[4*2];
+		if (iter.prevGlyphIndex == -1) { // can not retrieve glyph?
+			if (!nvg__allocTextAtlas(ctx))
+				break; // no memory :(
+			if (nverts != 0) {
+				nvg__renderText(ctx, verts, nverts);
+				nverts = 0;
+			}
+			iter = prevIter;
+			fonsTextIterNext(ctx->fs, &iter, &q); // try again
+			if (iter.prevGlyphIndex == -1) // still can not find glyph?
+				break;
+		}
+		prevIter = iter;
+		// Transform corners.
 		nvgTransformPoint(&c[0],&c[1], state->xform, q.x0*invscale, q.y0*invscale);
 		nvgTransformPoint(&c[2],&c[3], state->xform, q.x1*invscale, q.y0*invscale);
 		nvgTransformPoint(&c[4],&c[5], state->xform, q.x1*invscale, q.y1*invscale);
@@ -2063,35 +2343,18 @@ float nvgText(struct NVGcontext* ctx, float x, float y, const char* string, cons
 		}
 	}
 
-	// TODO: add back-end bit to do this just once per frame.
-	if (fonsValidateTexture(ctx->fs, dirty)) {
-		// Update texture
-		if (ctx->fontImage != 0) {
-			int iw, ih;
-			const unsigned char* data = fonsGetTextureData(ctx->fs, &iw, &ih);
-			int x = dirty[0];
-			int y = dirty[1];
-			int w = dirty[2] - dirty[0];
-			int h = dirty[3] - dirty[1];
-			ctx->params.renderUpdateTexture(ctx->params.userPtr, ctx->fontImage, x,y, w,h, data);
-		}
-	}
-
-	// Render triangles.
-	paint = state->fill;
-	paint.image = ctx->fontImage;
-	ctx->params.renderTriangles(ctx->params.userPtr, &paint, &state->scissor, verts, nverts);
+	// TODO: add back-end bit to do this just once per frame. 
+	nvg__flushTextTexture(ctx);
 
-	ctx->drawCallCount++;
-	ctx->textTriCount += nverts/3;
+	nvg__renderText(ctx, verts, nverts);
 
 	return iter.x;
 }
 
-void nvgTextBox(struct NVGcontext* ctx, float x, float y, float breakRowWidth, const char* string, const char* end)
+void nvgTextBox(NVGcontext* ctx, float x, float y, float breakRowWidth, const char* string, const char* end)
 {
-	struct NVGstate* state = nvg__getState(ctx);
-	struct NVGtextRow rows[2];
+	NVGstate* state = nvg__getState(ctx);
+	NVGtextRow rows[2];
 	int nrows = 0, i;
 	int oldAlign = state->textAlign;
 	int haling = state->textAlign & (NVG_ALIGN_LEFT | NVG_ALIGN_CENTER | NVG_ALIGN_RIGHT);
@@ -2106,7 +2369,7 @@ void nvgTextBox(struct NVGcontext* ctx, float x, float y, float breakRowWidth, c
 
 	while ((nrows = nvgTextBreakLines(ctx, string, end, breakRowWidth, rows, 2))) {
 		for (i = 0; i < nrows; i++) {
-			struct NVGtextRow* row = &rows[i];
+			NVGtextRow* row = &rows[i];
 			if (haling & NVG_ALIGN_LEFT)
 				nvgText(ctx, x, y, row->start, row->end);
 			else if (haling & NVG_ALIGN_CENTER)
@@ -2121,13 +2384,13 @@ void nvgTextBox(struct NVGcontext* ctx, float x, float y, float breakRowWidth, c
 	state->textAlign = oldAlign;
 }
 
-int nvgTextGlyphPositions(struct NVGcontext* ctx, float x, float y, const char* string, const char* end, struct NVGglyphPosition* positions, int maxPositions)
+int nvgTextGlyphPositions(NVGcontext* ctx, float x, float y, const char* string, const char* end, NVGglyphPosition* positions, int maxPositions)
 {
-	struct NVGstate* state = nvg__getState(ctx);
+	NVGstate* state = nvg__getState(ctx);
 	float scale = nvg__getFontScale(state) * ctx->devicePxRatio;
 	float invscale = 1.0f / scale;
-	struct FONStextIter iter;
-	struct FONSquad q;
+	FONStextIter iter, prevIter;
+	FONSquad q;
 	int npos = 0;
 
 	if (state->fontId == FONS_INVALID) return 0;
@@ -2145,11 +2408,17 @@ int nvgTextGlyphPositions(struct NVGcontext* ctx, float x, float y, const char*
 	fonsSetFont(ctx->fs, state->fontId);
 
 	fonsTextIterInit(ctx->fs, &iter, x*scale, y*scale, string, end);
+	prevIter = iter;
 	while (fonsTextIterNext(ctx->fs, &iter, &q)) {
+		if (iter.prevGlyphIndex < 0 && nvg__allocTextAtlas(ctx)) { // can not retrieve glyph?
+			iter = prevIter;
+			fonsTextIterNext(ctx->fs, &iter, &q); // try again
+		}
+		prevIter = iter;
 		positions[npos].str = iter.str;
 		positions[npos].x = iter.x * invscale;
-		positions[npos].minx = q.x0 * invscale;
-		positions[npos].maxx = q.x1 * invscale;
+		positions[npos].minx = nvg__minf(iter.x, q.x0) * invscale;
+		positions[npos].maxx = nvg__maxf(iter.nextx, q.x1) * invscale;
 		npos++;
 		if (npos >= maxPositions)
 			break;
@@ -2164,13 +2433,13 @@ enum NVGcodepointType {
 	NVG_CHAR,
 };
 
-int nvgTextBreakLines(struct NVGcontext* ctx, const char* string, const char* end, float breakRowWidth, struct NVGtextRow* rows, int maxRows)
+int nvgTextBreakLines(NVGcontext* ctx, const char* string, const char* end, float breakRowWidth, NVGtextRow* rows, int maxRows)
 {
-	struct NVGstate* state = nvg__getState(ctx);
+	NVGstate* state = nvg__getState(ctx);
 	float scale = nvg__getFontScale(state) * ctx->devicePxRatio;
 	float invscale = 1.0f / scale;
-	struct FONStextIter iter;
-	struct FONSquad q;
+	FONStextIter iter, prevIter;
+	FONSquad q;
 	int nrows = 0;
 	float rowStartX = 0;
 	float rowWidth = 0;
@@ -2204,7 +2473,13 @@ int nvgTextBreakLines(struct NVGcontext* ctx, const char* string, const char* en
 	breakRowWidth *= scale;
 
 	fonsTextIterInit(ctx->fs, &iter, 0, 0, string, end);
+	prevIter = iter;
 	while (fonsTextIterNext(ctx->fs, &iter, &q)) {
+		if (iter.prevGlyphIndex < 0 && nvg__allocTextAtlas(ctx)) { // can not retrieve glyph?
+			iter = prevIter;
+			fonsTextIterNext(ctx->fs, &iter, &q); // try again
+		}
+		prevIter = iter;
 		switch (iter.codepoint) {
 			case 9:			// \t
 			case 11:		// \v
@@ -2267,9 +2542,29 @@ int nvgTextBreakLines(struct NVGcontext* ctx, const char* string, const char* en
 					breakMaxX = 0.0;
 				}
 			} else {
-				float nextWidth = iter.nextx - rowStartX; //q.x1 - rowStartX;
+				float nextWidth = iter.nextx - rowStartX;
 
-				if (nextWidth > breakRowWidth) {
+				// track last non-white space character
+				if (type == NVG_CHAR) {
+					rowEnd = iter.next;
+					rowWidth = iter.nextx - rowStartX;
+					rowMaxX = q.x1 - rowStartX;
+				}
+				// track last end of a word
+				if (ptype == NVG_CHAR && type == NVG_SPACE) {
+					breakEnd = iter.str;
+					breakWidth = rowWidth;
+					breakMaxX = rowMaxX;
+				}
+				// track last beginning of a word
+				if (ptype == NVG_SPACE && type == NVG_CHAR) {
+					wordStart = iter.str;
+					wordStartX = iter.x;
+					wordMinX = q.x0 - rowStartX;
+				}
+
+				// Break to new line when a character is beyond break width.
+				if (type == NVG_CHAR && nextWidth > breakRowWidth) {
 					// The run length is too long, need to break to new line.
 					if (breakEnd == rowStart) {
 						// The current word is longer than the row length, just break it from here.
@@ -2292,7 +2587,7 @@ int nvgTextBreakLines(struct NVGcontext* ctx, const char* string, const char* en
 						wordStartX = iter.x;
 						wordMinX = q.x0 - rowStartX;
 					} else {
-						// Break the line from the end of the last word, and start new line from the begining of the new.
+						// Break the line from the end of the last word, and start new line from the beginning of the new.
 						rows[nrows].start = rowStart;
 						rows[nrows].end = breakEnd;
 						rows[nrows].width = breakWidth * invscale;
@@ -2305,7 +2600,7 @@ int nvgTextBreakLines(struct NVGcontext* ctx, const char* string, const char* en
 						rowStartX = wordStartX;
 						rowStart = wordStart;
 						rowEnd = iter.next;
-						rowWidth = iter.nextx - rowStartX; // q.x1 - rowStartX;
+						rowWidth = iter.nextx - rowStartX;
 						rowMinX = wordMinX;
 						rowMaxX = q.x1 - rowStartX;
 						// No change to the word start
@@ -2315,25 +2610,6 @@ int nvgTextBreakLines(struct NVGcontext* ctx, const char* string, const char* en
 					breakWidth = 0.0;
 					breakMaxX = 0.0;
 				}
-
-				// track last non-white space character
-				if (type == NVG_CHAR) {
-					rowEnd = iter.next;
-					rowWidth = iter.nextx - rowStartX; // q.x1 - rowStartX;
-					rowMaxX = q.x1 - rowStartX;
-				}
-				// track last end of a word
-				if (ptype == NVG_CHAR && (type == NVG_SPACE || type == NVG_SPACE)) {
-					breakEnd = iter.str;
-					breakWidth = rowWidth;
-					breakMaxX = rowMaxX;
-				}
-				// track last beginning of a word
-				if ((ptype == NVG_SPACE || ptype == NVG_SPACE) && type == NVG_CHAR) {
-					wordStart = iter.str;
-					wordStartX = iter.x;
-					wordMinX = q.x0 - rowStartX;
-				}
 			}
 		}
 
@@ -2341,7 +2617,7 @@ int nvgTextBreakLines(struct NVGcontext* ctx, const char* string, const char* en
 		ptype = type;
 	}
 
-	// Break the line from the end of the last word, and start new line from the begining of the new.
+	// Break the line from the end of the last word, and start new line from the beginning of the new.
 	if (rowStart != NULL) {
 		rows[nrows].start = rowStart;
 		rows[nrows].end = rowEnd;
@@ -2355,9 +2631,9 @@ int nvgTextBreakLines(struct NVGcontext* ctx, const char* string, const char* en
 	return nrows;
 }
 
-float nvgTextBounds(struct NVGcontext* ctx, float x, float y, const char* string, const char* end, float* bounds)
+float nvgTextBounds(NVGcontext* ctx, float x, float y, const char* string, const char* end, float* bounds)
 {
-	struct NVGstate* state = nvg__getState(ctx);
+	NVGstate* state = nvg__getState(ctx);
 	float scale = nvg__getFontScale(state) * ctx->devicePxRatio;
 	float invscale = 1.0f / scale;
 	float width;
@@ -2370,8 +2646,10 @@ float nvgTextBounds(struct NVGcontext* ctx, float x, float y, const char* string
 	fonsSetAlign(ctx->fs, state->textAlign);
 	fonsSetFont(ctx->fs, state->fontId);
 
-	width = fonsTextBounds(ctx->fs, x, y, string, end, bounds);
+	width = fonsTextBounds(ctx->fs, x*scale, y*scale, string, end, bounds);
 	if (bounds != NULL) {
+		// Use line bounds for height.
+		fonsLineBounds(ctx->fs, y*scale, &bounds[1], &bounds[3]);
 		bounds[0] *= invscale;
 		bounds[1] *= invscale;
 		bounds[2] *= invscale;
@@ -2380,10 +2658,10 @@ float nvgTextBounds(struct NVGcontext* ctx, float x, float y, const char* string
 	return width * invscale;
 }
 
-void nvgTextBoxBounds(struct NVGcontext* ctx, float x, float y, float breakRowWidth, const char* string, const char* end, float* bounds)
+void nvgTextBoxBounds(NVGcontext* ctx, float x, float y, float breakRowWidth, const char* string, const char* end, float* bounds)
 {
-	struct NVGstate* state = nvg__getState(ctx);
-	struct NVGtextRow rows[2];
+	NVGstate* state = nvg__getState(ctx);
+	NVGtextRow rows[2];
 	float scale = nvg__getFontScale(state) * ctx->devicePxRatio;
 	float invscale = 1.0f / scale;
 	int nrows = 0, i;
@@ -2401,8 +2679,6 @@ void nvgTextBoxBounds(struct NVGcontext* ctx, float x, float y, float breakRowWi
 
 	nvgTextMetrics(ctx, NULL, NULL, &lineh);
 
-	nvgTextMetrics(ctx, NULL, NULL, &lineh);
-
 	state->textAlign = NVG_ALIGN_LEFT | valign;
 
 	minx = maxx = x;
@@ -2419,7 +2695,7 @@ void nvgTextBoxBounds(struct NVGcontext* ctx, float x, float y, float breakRowWi
 
 	while ((nrows = nvgTextBreakLines(ctx, string, end, breakRowWidth, rows, 2))) {
 		for (i = 0; i < nrows; i++) {
-			struct NVGtextRow* row = &rows[i];
+			NVGtextRow* row = &rows[i];
 			float rminx, rmaxx, dx = 0;
 			// Horizontal bounds
 			if (haling & NVG_ALIGN_LEFT)
@@ -2451,9 +2727,9 @@ void nvgTextBoxBounds(struct NVGcontext* ctx, float x, float y, float breakRowWi
 	}
 }
 
-void nvgTextMetrics(struct NVGcontext* ctx, float* ascender, float* descender, float* lineh)
+void nvgTextMetrics(NVGcontext* ctx, float* ascender, float* descender, float* lineh)
 {
-	struct NVGstate* state = nvg__getState(ctx);
+	NVGstate* state = nvg__getState(ctx);
 	float scale = nvg__getFontScale(state) * ctx->devicePxRatio;
 	float invscale = 1.0f / scale;
 
@@ -2473,3 +2749,4 @@ void nvgTextMetrics(struct NVGcontext* ctx, float* ascender, float* descender, f
 	if (lineh != NULL)
 		*lineh *= invscale;
 }
+// vim: ft=c nu noet ts=4
diff --git a/dgl/src/nanovg/nanovg.h b/dgl/src/nanovg/nanovg.h
index 247a9eed..a54819c3 100644
--- a/dgl/src/nanovg/nanovg.h
+++ b/dgl/src/nanovg/nanovg.h
@@ -25,7 +25,12 @@ extern "C" {
 
 #define NVG_PI 3.14159265358979323846264338327f
 
-struct NVGcontext;
+#ifdef _MSC_VER
+#pragma warning(push)
+#pragma warning(disable: 4201)  // nonstandard extension used : nameless struct/union
+#endif
+
+typedef struct NVGcontext NVGcontext;
 
 struct NVGcolor {
 	union {
@@ -35,17 +40,18 @@ struct NVGcolor {
 		};
 	};
 };
+typedef struct NVGcolor NVGcolor;
 
 struct NVGpaint {
 	float xform[6];
 	float extent[2];
 	float radius;
 	float feather;
-	struct NVGcolor innerColor;
-	struct NVGcolor outerColor;
+	NVGcolor innerColor;
+	NVGcolor outerColor;
 	int image;
-	int repeat;
 };
+typedef struct NVGpaint NVGpaint;
 
 enum NVGwinding {
 	NVG_CCW = 1,			// Winding for solid shapes
@@ -65,11 +71,6 @@ enum NVGlineCap {
 	NVG_MITER,
 };
 
-enum NVGpatternRepeat {
-	NVG_REPEATX = 0x01,		// Repeat image pattern in X direction
-	NVG_REPEATY = 0x02,		// Repeat image pattern in Y direction
-};
-
 enum NVGalign {
 	// Horizontal align
 	NVG_ALIGN_LEFT 		= 1<<0,	// Default, align text horizontally to left.
@@ -82,16 +83,12 @@ enum NVGalign {
 	NVG_ALIGN_BASELINE	= 1<<6, // Default, align text vertically to baseline.
 };
 
-enum NVGalpha {
-	NVG_STRAIGHT_ALPHA,
-	NVG_PREMULTIPLIED_ALPHA,
-};
-
 struct NVGglyphPosition {
 	const char* str;	// Position of the glyph in the input string.
 	float x;			// The x-coordinate of the logical glyph position.
 	float minx, maxx;	// The bounds of the glyph shape.
 };
+typedef struct NVGglyphPosition NVGglyphPosition;
 
 struct NVGtextRow {
 	const char* start;	// Pointer to the input text where the row starts.
@@ -100,7 +97,15 @@ struct NVGtextRow {
 	float width;		// Logical width of the row.
 	float minx, maxx;	// Actual bounds of the row. Logical with and bounds can differ because of kerning and some parts over extending.
 };
-
+typedef struct NVGtextRow NVGtextRow;
+
+enum NVGimageFlags {
+	NVG_IMAGE_GENERATE_MIPMAPS		= 1<<0,		// Generate mipmaps during creation of the image.
+	NVG_IMAGE_REPEATX			= 1<<1,		// Repeat image in X direction.
+	NVG_IMAGE_REPEATY			= 1<<2,		// Repeat image in Y direction.
+	NVG_IMAGE_FLIPY				= 1<<3,		// Flips (inverses) image in Y direction when rendered.
+	NVG_IMAGE_PREMULTIPLIED			= 1<<4,		// Image data has premultiplied alpha.
+};
 
 // Begin drawing a new frame
 // Calls to nanovg drawing API should be wrapped in nvgBeginFrame() & nvgEndFrame()
@@ -110,14 +115,13 @@ struct NVGtextRow {
 // For example, GLFW returns two dimension for an opened window: window size and
 // frame buffer size. In that case you would set windowWidth/Height to the window size
 // devicePixelRatio to: frameBufferWidth / windowWidth.
-// AlphaBlend controls if drawing the shapes to the render target should be done using straight or
-// premultiplied alpha. If rendering directly to framebuffer you probably want to use NVG_STRAIGHT_ALPHA,
-// if rendering to texture which should contain transparent regions NVG_PREMULTIPLIED_ALPHA is the
-// right choice.
-void nvgBeginFrame(struct NVGcontext* ctx, int windowWidth, int windowHeight, float devicePixelRatio, int alphaBlend);
+void nvgBeginFrame(NVGcontext* ctx, int windowWidth, int windowHeight, float devicePixelRatio);
+
+// Cancels drawing the current frame.
+void nvgCancelFrame(NVGcontext* ctx);
 
 // Ends drawing flushing remaining render state.
-void nvgEndFrame(struct NVGcontext* ctx);
+void nvgEndFrame(NVGcontext* ctx);
 
 //
 // Color utils
@@ -125,35 +129,35 @@ void nvgEndFrame(struct NVGcontext* ctx);
 // Colors in NanoVG are stored as unsigned ints in ABGR format.
 
 // Returns a color value from red, green, blue values. Alpha will be set to 255 (1.0f).
-struct NVGcolor nvgRGB(unsigned char r, unsigned char g, unsigned char b);
+NVGcolor nvgRGB(unsigned char r, unsigned char g, unsigned char b);
 
 // Returns a color value from red, green, blue values. Alpha will be set to 1.0f.
-struct NVGcolor nvgRGBf(float r, float g, float b);
+NVGcolor nvgRGBf(float r, float g, float b);
 
 
 // Returns a color value from red, green, blue and alpha values.
-struct NVGcolor nvgRGBA(unsigned char r, unsigned char g, unsigned char b, unsigned char a);
+NVGcolor nvgRGBA(unsigned char r, unsigned char g, unsigned char b, unsigned char a);
 
 // Returns a color value from red, green, blue and alpha values.
-struct NVGcolor nvgRGBAf(float r, float g, float b, float a);
+NVGcolor nvgRGBAf(float r, float g, float b, float a);
 
 
-// Linearly interpoaltes from color c0 to c1, and returns resulting color value.
-struct NVGcolor nvgLerpRGBA(struct NVGcolor c0, struct NVGcolor c1, float u);
+// Linearly interpolates from color c0 to c1, and returns resulting color value.
+NVGcolor nvgLerpRGBA(NVGcolor c0, NVGcolor c1, float u);
 
 // Sets transparency of a color value.
-struct NVGcolor nvgTransRGBA(struct NVGcolor c0, unsigned char a);
+NVGcolor nvgTransRGBA(NVGcolor c0, unsigned char a);
 
 // Sets transparency of a color value.
-struct NVGcolor nvgTransRGBAf(struct NVGcolor c0, float a);
+NVGcolor nvgTransRGBAf(NVGcolor c0, float a);
 
 // Returns color value specified by hue, saturation and lightness.
 // HSL values are all in range [0..1], alpha will be set to 255.
-struct NVGcolor nvgHSL(float h, float s, float l);
+NVGcolor nvgHSL(float h, float s, float l);
 
 // Returns color value specified by hue, saturation and lightness and alpha.
 // HSL values are all in range [0..1], alpha in range [0..255]
-struct NVGcolor nvgHSLA(float h, float s, float l, unsigned char a);
+NVGcolor nvgHSLA(float h, float s, float l, unsigned char a);
 
 //
 // State Handling
@@ -164,13 +168,13 @@ struct NVGcolor nvgHSLA(float h, float s, float l, unsigned char a);
 
 // Pushes and saves the current render state into a state stack.
 // A matching nvgRestore() must be used to restore the state.
-void nvgSave(struct NVGcontext* ctx);
+void nvgSave(NVGcontext* ctx);
 
 // Pops and restores current render state.
-void nvgRestore(struct NVGcontext* ctx);
+void nvgRestore(NVGcontext* ctx);
 
 // Resets current render state to default values. Does not affect the render state stack.
-void nvgReset(struct NVGcontext* ctx);
+void nvgReset(NVGcontext* ctx);
 
 //
 // Render styles
@@ -182,31 +186,35 @@ void nvgReset(struct NVGcontext* ctx);
 // Current render style can be saved and restored using nvgSave() and nvgRestore().
 
 // Sets current stroke style to a solid color.
-void nvgStrokeColor(struct NVGcontext* ctx, struct NVGcolor color);
+void nvgStrokeColor(NVGcontext* ctx, NVGcolor color);
 
 // Sets current stroke style to a paint, which can be a one of the gradients or a pattern.
-void nvgStrokePaint(struct NVGcontext* ctx, struct NVGpaint paint);
+void nvgStrokePaint(NVGcontext* ctx, NVGpaint paint);
 
-// Sets current fill cstyle to a solid color.
-void nvgFillColor(struct NVGcontext* ctx, struct NVGcolor color);
+// Sets current fill style to a solid color.
+void nvgFillColor(NVGcontext* ctx, NVGcolor color);
 
 // Sets current fill style to a paint, which can be a one of the gradients or a pattern.
-void nvgFillPaint(struct NVGcontext* ctx, struct NVGpaint paint);
+void nvgFillPaint(NVGcontext* ctx, NVGpaint paint);
 
 // Sets the miter limit of the stroke style.
 // Miter limit controls when a sharp corner is beveled.
-void nvgMiterLimit(struct NVGcontext* ctx, float limit);
+void nvgMiterLimit(NVGcontext* ctx, float limit);
 
-// Sets the stroke witdth of the stroke style.
-void nvgStrokeWidth(struct NVGcontext* ctx, float size);
+// Sets the stroke width of the stroke style.
+void nvgStrokeWidth(NVGcontext* ctx, float size);
 
 // Sets how the end of the line (cap) is drawn,
 // Can be one of: NVG_BUTT (default), NVG_ROUND, NVG_SQUARE.
-void nvgLineCap(struct NVGcontext* ctx, int cap);
+void nvgLineCap(NVGcontext* ctx, int cap);
 
 // Sets how sharp path corners are drawn.
 // Can be one of NVG_MITER (default), NVG_ROUND, NVG_BEVEL.
-void nvgLineJoin(struct NVGcontext* ctx, int join);
+void nvgLineJoin(NVGcontext* ctx, int join);
+
+// Sets the transparency applied to all rendered shapes.
+// Already transparent paths will get proportionally more transparent as well.
+void nvgGlobalAlpha(NVGcontext* ctx, float alpha);
 
 //
 // Transforms
@@ -226,36 +234,36 @@ void nvgLineJoin(struct NVGcontext* ctx, int join);
 // Current coordinate system (transformation) can be saved and restored using nvgSave() and nvgRestore().
 
 // Resets current transform to a identity matrix.
-void nvgResetTransform(struct NVGcontext* ctx);
+void nvgResetTransform(NVGcontext* ctx);
 
 // Premultiplies current coordinate system by specified matrix.
 // The parameters are interpreted as matrix as follows:
 //   [a c e]
 //   [b d f]
 //   [0 0 1]
-void nvgTransform(struct NVGcontext* ctx, float a, float b, float c, float d, float e, float f);
+void nvgTransform(NVGcontext* ctx, float a, float b, float c, float d, float e, float f);
 
 // Translates current coordinate system.
-void nvgTranslate(struct NVGcontext* ctx, float x, float y);
+void nvgTranslate(NVGcontext* ctx, float x, float y);
 
 // Rotates current coordinate system. Angle is specifid in radians.
-void nvgRotate(struct NVGcontext* ctx, float angle);
+void nvgRotate(NVGcontext* ctx, float angle);
 
 // Skews the current coordinate system along X axis. Angle is specifid in radians.
-void nvgSkewX(struct NVGcontext* ctx, float angle);
+void nvgSkewX(NVGcontext* ctx, float angle);
 
 // Skews the current coordinate system along Y axis. Angle is specifid in radians.
-void nvgSkewY(struct NVGcontext* ctx, float angle);
+void nvgSkewY(NVGcontext* ctx, float angle);
 
 // Scales the current coordinat system.
-void nvgScale(struct NVGcontext* ctx, float x, float y);
+void nvgScale(NVGcontext* ctx, float x, float y);
 
 // Stores the top part (a-f) of the current transformation matrix in to the specified buffer.
 //   [a c e]
 //   [b d f]
 //   [0 0 1]
 // There should be space for 6 floats in the return buffer for the values a-f.
-void nvgCurrentTransform(struct NVGcontext* ctx, float* xform);
+void nvgCurrentTransform(NVGcontext* ctx, float* xform);
 
 
 // The following functions can be used to make calculations on 2x3 transformation matrices.
@@ -304,24 +312,24 @@ float nvgRadToDeg(float rad);
 
 // Creates image by loading it from the disk from specified file name.
 // Returns handle to the image.
-int nvgCreateImage(struct NVGcontext* ctx, const char* filename);
+int nvgCreateImage(NVGcontext* ctx, const char* filename);
 
 // Creates image by loading it from the specified chunk of memory.
 // Returns handle to the image.
-int nvgCreateImageMem(struct NVGcontext* ctx, unsigned char* data, int ndata);
+int nvgCreateImageMem(NVGcontext* ctx, unsigned char* data, int ndata);
 
 // Creates image from specified image data.
 // Returns handle to the image.
-int nvgCreateImageRGBA(struct NVGcontext* ctx, int w, int h, const unsigned char* data);
+int nvgCreateImageRGBA(NVGcontext* ctx, int w, int h, const unsigned char* data);
 
 // Updates image data specified by image handle.
-void nvgUpdateImage(struct NVGcontext* ctx, int image, const unsigned char* data);
+void nvgUpdateImage(NVGcontext* ctx, int image, const unsigned char* data);
 
 // Returns the domensions of a created image.
-void nvgImageSize(struct NVGcontext* ctx, int image, int* w, int* h);
+void nvgImageSize(NVGcontext* ctx, int image, int* w, int* h);
 
 // Deletes created image.
-void nvgDeleteImage(struct NVGcontext* ctx, int image);
+void nvgDeleteImage(NVGcontext* ctx, int image);
 
 //
 // Paints
@@ -332,28 +340,28 @@ void nvgDeleteImage(struct NVGcontext* ctx, int image);
 // Creates and returns a linear gradient. Parameters (sx,sy)-(ex,ey) specify the start and end coordinates
 // of the linear gradient, icol specifies the start color and ocol the end color.
 // The gradient is transformed by the current transform when it is passed to nvgFillPaint() or nvgStrokePaint().
-struct NVGpaint nvgLinearGradient(struct NVGcontext* ctx, float sx, float sy, float ex, float ey,
-								  struct NVGcolor icol, struct NVGcolor ocol);
+NVGpaint nvgLinearGradient(NVGcontext* ctx, float sx, float sy, float ex, float ey,
+								  NVGcolor icol, NVGcolor ocol);
 
 // Creates and returns a box gradient. Box gradient is a feathered rounded rectangle, it is useful for rendering
 // drop shadows or hilights for boxes. Parameters (x,y) define the top-left corner of the rectangle,
 // (w,h) define the size of the rectangle, r defines the corner radius, and f feather. Feather defines how blurry
 // the border of the rectangle is. Parameter icol specifies the inner color and ocol the outer color of the gradient.
 // The gradient is transformed by the current transform when it is passed to nvgFillPaint() or nvgStrokePaint().
-struct NVGpaint nvgBoxGradient(struct NVGcontext* ctx, float x, float y, float w, float h,
-							   float r, float f, struct NVGcolor icol, struct NVGcolor ocol);
+NVGpaint nvgBoxGradient(NVGcontext* ctx, float x, float y, float w, float h,
+							   float r, float f, NVGcolor icol, NVGcolor ocol);
 
 // Creates and returns a radial gradient. Parameters (cx,cy) specify the center, inr and outr specify
 // the inner and outer radius of the gradient, icol specifies the start color and ocol the end color.
 // The gradient is transformed by the current transform when it is passed to nvgFillPaint() or nvgStrokePaint().
-struct NVGpaint nvgRadialGradient(struct NVGcontext* ctx, float cx, float cy, float inr, float outr,
-								  struct NVGcolor icol, struct NVGcolor ocol);
+NVGpaint nvgRadialGradient(NVGcontext* ctx, float cx, float cy, float inr, float outr,
+								  NVGcolor icol, NVGcolor ocol);
 
 // Creates and returns an image patter. Parameters (ox,oy) specify the left-top location of the image pattern,
 // (ex,ey) the size of one image, angle rotation around the top-left corner, image is handle to the image to render,
 // and repeat is combination of NVG_REPEATX and NVG_REPEATY which tells if the image should be repeated across x or y.
 // The gradient is transformed by the current transform when it is passed to nvgFillPaint() or nvgStrokePaint().
-struct NVGpaint nvgImagePattern(struct NVGcontext* ctx, float ox, float oy, float ex, float ey,
+NVGpaint nvgImagePattern(NVGcontext* ctx, float ox, float oy, float ex, float ey,
 								float angle, int image, int repeat);
 
 //
@@ -364,10 +372,10 @@ struct NVGpaint nvgImagePattern(struct NVGcontext* ctx, float ox, float oy, floa
 
 // Sets the current
 // The scissor rectangle is transformed by the current transform.
-void nvgScissor(struct NVGcontext* ctx, float x, float y, float w, float h);
+void nvgScissor(NVGcontext* ctx, float x, float y, float w, float h);
 
 // Reset and disables scissoring.
-void nvgResetScissor(struct NVGcontext* ctx);
+void nvgResetScissor(NVGcontext* ctx);
 
 //
 // Paths
@@ -387,46 +395,46 @@ void nvgResetScissor(struct NVGcontext* ctx);
 // The curve segments and sub-paths are transformed by the current transform.
 
 // Clears the current path and sub-paths.
-void nvgBeginPath(struct NVGcontext* ctx);
+void nvgBeginPath(NVGcontext* ctx);
 
 // Starts new sub-path with specified point as first point.
-void nvgMoveTo(struct NVGcontext* ctx, float x, float y);
+void nvgMoveTo(NVGcontext* ctx, float x, float y);
 
 // Adds line segment from the last point in the path to the specified point.
-void nvgLineTo(struct NVGcontext* ctx, float x, float y);
+void nvgLineTo(NVGcontext* ctx, float x, float y);
 
 // Adds bezier segment from last point in the path via two control points to the specified point.
-void nvgBezierTo(struct NVGcontext* ctx, float c1x, float c1y, float c2x, float c2y, float x, float y);
+void nvgBezierTo(NVGcontext* ctx, float c1x, float c1y, float c2x, float c2y, float x, float y);
 
 // Adds an arc segment at the corner defined by the last path point, and two specified points.
-void nvgArcTo(struct NVGcontext* ctx, float x1, float y1, float x2, float y2, float radius);
+void nvgArcTo(NVGcontext* ctx, float x1, float y1, float x2, float y2, float radius);
 
 // Closes current sub-path with a line segment.
-void nvgClosePath(struct NVGcontext* ctx);
+void nvgClosePath(NVGcontext* ctx);
 
 // Sets the current sub-path winding, see NVGwinding and NVGsolidity.
-void nvgPathWinding(struct NVGcontext* ctx, int dir);
+void nvgPathWinding(NVGcontext* ctx, int dir);
 
 // Creates new arc shaped sub-path.
-void nvgArc(struct NVGcontext* ctx, float cx, float cy, float r, float a0, float a1, int dir);
+void nvgArc(NVGcontext* ctx, float cx, float cy, float r, float a0, float a1, int dir);
 
 // Creates new rectangle shaped sub-path.
-void nvgRect(struct NVGcontext* ctx, float x, float y, float w, float h);
+void nvgRect(NVGcontext* ctx, float x, float y, float w, float h);
 
 // Creates new rounded rectangle shaped sub-path.
-void nvgRoundedRect(struct NVGcontext* ctx, float x, float y, float w, float h, float r);
+void nvgRoundedRect(NVGcontext* ctx, float x, float y, float w, float h, float r);
 
 // Creates new ellipse shaped sub-path.
-void nvgEllipse(struct NVGcontext* ctx, float cx, float cy, float rx, float ry);
+void nvgEllipse(NVGcontext* ctx, float cx, float cy, float rx, float ry);
 
 // Creates new circle shaped sub-path.
-void nvgCircle(struct NVGcontext* ctx, float cx, float cy, float r);
+void nvgCircle(NVGcontext* ctx, float cx, float cy, float r);
 
 // Fills the current path with current fill style.
-void nvgFill(struct NVGcontext* ctx);
+void nvgFill(NVGcontext* ctx);
 
 // Fills the current path with current stroke style.
-void nvgStroke(struct NVGcontext* ctx);
+void nvgStroke(NVGcontext* ctx);
 
 
 //
@@ -464,67 +472,67 @@ void nvgStroke(struct NVGcontext* ctx);
 
 // Creates font by loading it from the disk from specified file name.
 // Returns handle to the font.
-int nvgCreateFont(struct NVGcontext* ctx, const char* name, const char* filename);
+int nvgCreateFont(NVGcontext* ctx, const char* name, const char* filename);
 
 // Creates image by loading it from the specified memory chunk.
 // Returns handle to the font.
-int nvgCreateFontMem(struct NVGcontext* ctx, const char* name, unsigned char* data, int ndata, int freeData);
+int nvgCreateFontMem(NVGcontext* ctx, const char* name, unsigned char* data, int ndata, int freeData);
 
 // Finds a loaded font of specified name, and returns handle to it, or -1 if the font is not found.
-int nvgFindFont(struct NVGcontext* ctx, const char* name);
+int nvgFindFont(NVGcontext* ctx, const char* name);
 
 // Sets the font size of current text style.
-void nvgFontSize(struct NVGcontext* ctx, float size);
+void nvgFontSize(NVGcontext* ctx, float size);
 
 // Sets the blur of current text style.
-void nvgFontBlur(struct NVGcontext* ctx, float blur);
+void nvgFontBlur(NVGcontext* ctx, float blur);
 
 // Sets the letter spacing of current text style.
-void nvgTextLetterSpacing(struct NVGcontext* ctx, float spacing);
+void nvgTextLetterSpacing(NVGcontext* ctx, float spacing);
 
 // Sets the proportional line height of current text style. The line height is specified as multiple of font size.
-void nvgTextLineHeight(struct NVGcontext* ctx, float lineHeight);
+void nvgTextLineHeight(NVGcontext* ctx, float lineHeight);
 
 // Sets the text align of current text style, see NVGaling for options.
-void nvgTextAlign(struct NVGcontext* ctx, int align);
+void nvgTextAlign(NVGcontext* ctx, int align);
 
 // Sets the font face based on specified id of current text style.
-void nvgFontFaceId(struct NVGcontext* ctx, int font);
+void nvgFontFaceId(NVGcontext* ctx, int font);
 
 // Sets the font face based on specified name of current text style.
-void nvgFontFace(struct NVGcontext* ctx, const char* font);
+void nvgFontFace(NVGcontext* ctx, const char* font);
 
 // Draws text string at specified location. If end is specified only the sub-string up to the end is drawn.
-float nvgText(struct NVGcontext* ctx, float x, float y, const char* string, const char* end);
+float nvgText(NVGcontext* ctx, float x, float y, const char* string, const char* end);
 
 // Draws multi-line text string at specified location wrapped at the specified width. If end is specified only the sub-string up to the end is drawn.
 // White space is stripped at the beginning of the rows, the text is split at word boundaries or when new-line characters are encountered.
 // Words longer than the max width are slit at nearest character (i.e. no hyphenation).
-void nvgTextBox(struct NVGcontext* ctx, float x, float y, float breakRowWidth, const char* string, const char* end);
+void nvgTextBox(NVGcontext* ctx, float x, float y, float breakRowWidth, const char* string, const char* end);
 
 // Measures the specified text string. Parameter bounds should be a pointer to float[4],
 // if the bounding box of the text should be returned. The bounds value are [xmin,ymin, xmax,ymax]
 // Returns the horizontal advance of the measured text (i.e. where the next character should drawn).
 // Measured values are returned in local coordinate space.
-float nvgTextBounds(struct NVGcontext* ctx, float x, float y, const char* string, const char* end, float* bounds);
+float nvgTextBounds(NVGcontext* ctx, float x, float y, const char* string, const char* end, float* bounds);
 
 // Measures the specified multi-text string. Parameter bounds should be a pointer to float[4],
 // if the bounding box of the text should be returned. The bounds value are [xmin,ymin, xmax,ymax]
 // Measured values are returned in local coordinate space.
-void nvgTextBoxBounds(struct NVGcontext* ctx, float x, float y, float breakRowWidth, const char* string, const char* end, float* bounds);
+void nvgTextBoxBounds(NVGcontext* ctx, float x, float y, float breakRowWidth, const char* string, const char* end, float* bounds);
 
 // Calculates the glyph x positions of the specified text. If end is specified only the sub-string will be used.
 // Measured values are returned in local coordinate space.
-int nvgTextGlyphPositions(struct NVGcontext* ctx, float x, float y, const char* string, const char* end, struct NVGglyphPosition* positions, int maxPositions);
+int nvgTextGlyphPositions(NVGcontext* ctx, float x, float y, const char* string, const char* end, NVGglyphPosition* positions, int maxPositions);
 
 // Returns the vertical metrics based on the current text style.
 // Measured values are returned in local coordinate space.
-void nvgTextMetrics(struct NVGcontext* ctx, float* ascender, float* descender, float* lineh);
+void nvgTextMetrics(NVGcontext* ctx, float* ascender, float* descender, float* lineh);
 
 // Breaks the specified text into lines. If end is specified only the sub-string will be used.
 // White space is stripped at the beginning of the rows, the text is split at word boundaries or when new-line characters are encountered.
 // Words longer than the max width are slit at nearest character (i.e. no hyphenation).
-int nvgTextBreakLines(struct NVGcontext* ctx, const char* string, const char* end, float breakRowWidth, struct NVGtextRow* rows, int maxRows);
+int nvgTextBreakLines(NVGcontext* ctx, const char* string, const char* end, float breakRowWidth, NVGtextRow* rows, int maxRows);
 
 //
 // Internal Render API
@@ -534,54 +542,63 @@ enum NVGtexture {
 	NVG_TEXTURE_RGBA = 0x02,
 };
 
-struct NVGscissor
-{
+struct NVGscissor {
 	float xform[6];
 	float extent[2];
 };
+typedef struct NVGscissor NVGscissor;
 
 struct NVGvertex {
 	float x,y,u,v;
 };
+typedef struct NVGvertex NVGvertex;
 
 struct NVGpath {
 	int first;
 	int count;
 	unsigned char closed;
 	int nbevel;
-	struct NVGvertex* fill;
+	NVGvertex* fill;
 	int nfill;
-	struct NVGvertex* stroke;
+	NVGvertex* stroke;
 	int nstroke;
 	int winding;
 	int convex;
 };
+typedef struct NVGpath NVGpath;
 
 struct NVGparams {
 	void* userPtr;
-	int atlasWidth, atlasHeight;
 	int edgeAntiAlias;
 	int (*renderCreate)(void* uptr);
-	int (*renderCreateTexture)(void* uptr, int type, int w, int h, const unsigned char* data);
+	int (*renderCreateTexture)(void* uptr, int type, int w, int h, int imageFlags, const unsigned char* data);
 	int (*renderDeleteTexture)(void* uptr, int image);
 	int (*renderUpdateTexture)(void* uptr, int image, int x, int y, int w, int h, const unsigned char* data);
 	int (*renderGetTextureSize)(void* uptr, int image, int* w, int* h);
-	void (*renderViewport)(void* uptr, int width, int height, int alphaBlend);
-	void (*renderFlush)(void* uptr, int alphaBlend);
-	void (*renderFill)(void* uptr, struct NVGpaint* paint, struct NVGscissor* scissor, float fringe, const float* bounds, const struct NVGpath* paths, int npaths);
-	void (*renderStroke)(void* uptr, struct NVGpaint* paint, struct NVGscissor* scissor, float fringe, float strokeWidth, const struct NVGpath* paths, int npaths);
-	void (*renderTriangles)(void* uptr, struct NVGpaint* paint, struct NVGscissor* scissor, const struct NVGvertex* verts, int nverts);
+	void (*renderViewport)(void* uptr, int width, int height);
+	void (*renderCancel)(void* uptr);
+	void (*renderFlush)(void* uptr);
+	void (*renderFill)(void* uptr, NVGpaint* paint, NVGscissor* scissor, float fringe, const float* bounds, const NVGpath* paths, int npaths);
+	void (*renderStroke)(void* uptr, NVGpaint* paint, NVGscissor* scissor, float fringe, float strokeWidth, const NVGpath* paths, int npaths);
+	void (*renderTriangles)(void* uptr, NVGpaint* paint, NVGscissor* scissor, const NVGvertex* verts, int nverts);
 	void (*renderDelete)(void* uptr);
 };
+typedef struct NVGparams NVGparams;
 
-// Contructor and destructor, called by the render back-end.
-struct NVGcontext* nvgCreateInternal(struct NVGparams* params);
-void nvgDeleteInternal(struct NVGcontext* ctx);
+// Constructor and destructor, called by the render back-end.
+NVGcontext* nvgCreateInternal(NVGparams* params);
+void nvgDeleteInternal(NVGcontext* ctx);
+
+NVGparams* nvgInternalParams(NVGcontext* ctx);
 
 // Debug function to dump cached path data.
-void nvgDebugDumpPathCache(struct NVGcontext* ctx);
+void nvgDebugDumpPathCache(NVGcontext* ctx);
+
+#ifdef _MSC_VER
+#pragma warning(pop)
+#endif
 
-#define NVG_NOTUSED(v) do { (void)(1 ? (void)0 : ( (void)(v) ) ); } while(0)
+#define NVG_NOTUSED(v) for (;;) { (void)(1 ? (void)0 : ( (void)(v) ) ); break; }
 
 #ifdef __cplusplus
 }
diff --git a/dgl/src/nanovg/nanovg_gl.h b/dgl/src/nanovg/nanovg_gl.h
index c3c80cfe..52357951 100644
--- a/dgl/src/nanovg/nanovg_gl.h
+++ b/dgl/src/nanovg/nanovg_gl.h
@@ -15,14 +15,24 @@
 //    misrepresented as being the original software.
 // 3. This notice may not be removed or altered from any source distribution.
 //
-#ifndef NANOVG_GL3_H
-#define NANOVG_GL3_H
+#ifndef NANOVG_GL_H
+#define NANOVG_GL_H
 
 #ifdef __cplusplus
 extern "C" {
 #endif
 
-#define NVG_ANTIALIAS 1
+// Create flags
+
+enum NVGcreateFlags {
+	// Flag indicating if geometry based anti-aliasing is used (may not be needed when using MSAA).
+	NVG_ANTIALIAS 		= 1<<0,
+	// Flag indicating if strokes should be drawn using stencil buffer. The rendering will be a little
+	// slower, but path overlaps (i.e. self-intersecting or sharp turns) will be drawn just once.
+	NVG_STENCIL_STROKES	= 1<<1,
+	// Flag indicating that additional debug checks are done.
+	NVG_DEBUG 			= 1<<2,
+};
 
 #if defined NANOVG_GL2_IMPLEMENTATION
 #  define NANOVG_GL2 1
@@ -38,35 +48,54 @@ extern "C" {
 #  define NANOVG_GLES3 1
 #  define NANOVG_GL_IMPLEMENTATION 1
 #endif
-	
+
+#define NANOVG_GL_USE_STATE_FILTER (1)
+
+// Creates NanoVG contexts for different OpenGL (ES) versions.
+// Flags should be combination of the create flags above.
 
 #if defined NANOVG_GL2
 
-struct NVGcontext* nvgCreateGL2(int atlasw, int atlash, int edgeaa);
-void nvgDeleteGL2(struct NVGcontext* ctx);
+NVGcontext* nvgCreateGL2(int flags);
+void nvgDeleteGL2(NVGcontext* ctx);
 
-#elif defined NANOVG_GL3
+#endif
 
-struct NVGcontext* nvgCreateGL3(int atlasw, int atlash, int edgeaa);
-void nvgDeleteGL3(struct NVGcontext* ctx);
+#if defined NANOVG_GL3
 
-#elif defined NANOVG_GLES2
+NVGcontext* nvgCreateGL3(int flags);
+void nvgDeleteGL3(NVGcontext* ctx);
 
-struct NVGcontext* nvgCreateGLES2(int atlasw, int atlash, int edgeaa);
-void nvgDeleteGLES2(struct NVGcontext* ctx);
+#endif
 
-#elif defined NANOVG_GLES3
+#if defined NANOVG_GLES2
 
-struct NVGcontext* nvgCreateGLES3(int atlasw, int atlash, int edgeaa);
-void nvgDeleteGLES3(struct NVGcontext* ctx);
+NVGcontext* nvgCreateGLES2(int flags);
+void nvgDeleteGLES2(NVGcontext* ctx);
 
 #endif
 
+#if defined NANOVG_GLES3
+
+NVGcontext* nvgCreateGLES3(int flags);
+void nvgDeleteGLES3(NVGcontext* ctx);
+
+#endif
+
+// These are additional flags on top of NVGimageFlags.
+enum NVGimageFlagsGL {
+	NVG_IMAGE_NODELETE			= 1<<16,	// Do not delete GL texture handle.
+};
+
+int nvglCreateImageFromHandle(NVGcontext* ctx, GLuint textureId, int w, int h, int flags);
+GLuint nvglImageHandle(NVGcontext* ctx, int image);
+
+
 #ifdef __cplusplus
 }
 #endif
 
-#endif
+#endif /* NANOVG_GL_H */
 
 #ifdef NANOVG_GL_IMPLEMENTATION
 
@@ -79,22 +108,7 @@ void nvgDeleteGLES3(struct NVGcontext* ctx);
 enum GLNVGuniformLoc {
 	GLNVG_LOC_VIEWSIZE,
 	GLNVG_LOC_TEX,
-#if NANOVG_GL_USE_UNIFORMBUFFER
 	GLNVG_LOC_FRAG,
-#else
-	GLNVG_LOC_SCISSORMAT,
-	GLNVG_LOC_SCISSOREXT,
-	GLNVG_LOC_SCISSORSCALE,
-	GLNVG_LOC_PAINTMAT,
-	GLNVG_LOC_EXTENT,
-	GLNVG_LOC_RADIUS,
-	GLNVG_LOC_FEATHER,
-	GLNVG_LOC_INNERCOL,
-	GLNVG_LOC_OUTERCOL,
-	GLNVG_LOC_STROKEMULT,
-	GLNVG_LOC_TEXTYPE,
-	GLNVG_LOC_TYPE,
-#endif
 	GLNVG_MAX_LOCS
 };
 
@@ -117,13 +131,16 @@ struct GLNVGshader {
 	GLuint vert;
 	GLint loc[GLNVG_MAX_LOCS];
 };
+typedef struct GLNVGshader GLNVGshader;
 
 struct GLNVGtexture {
 	int id;
 	GLuint tex;
 	int width, height;
 	int type;
+	int flags;
 };
+typedef struct GLNVGtexture GLNVGtexture;
 
 enum GLNVGcallType {
 	GLNVG_NONE = 0,
@@ -142,6 +159,7 @@ struct GLNVGcall {
 	int triangleCount;
 	int uniformOffset;
 };
+typedef struct GLNVGcall GLNVGcall;
 
 struct GLNVGpath {
 	int fillOffset;
@@ -149,25 +167,52 @@ struct GLNVGpath {
 	int strokeOffset;
 	int strokeCount;
 };
+typedef struct GLNVGpath GLNVGpath;
 
 struct GLNVGfragUniforms {
-   float scissorMat[12]; // matrices are actually 3 vec4s
-   float paintMat[12];
-   struct NVGcolor innerCol;
-   struct NVGcolor outerCol;
-   float scissorExt[2];
-   float scissorScale[2];
-   float extent[2];
-   float radius;
-   float feather;
-   float strokeMult;
-   int texType;
-   int type;
+	#if NANOVG_GL_USE_UNIFORMBUFFER
+		float scissorMat[12]; // matrices are actually 3 vec4s
+		float paintMat[12];
+		struct NVGcolor innerCol;
+		struct NVGcolor outerCol;
+		float scissorExt[2];
+		float scissorScale[2];
+		float extent[2];
+		float radius;
+		float feather;
+		float strokeMult;
+		float strokeThr;
+		int texType;
+		int type;
+	#else
+		// note: after modifying layout or size of uniform array,
+		// don't forget to also update the fragment shader source!
+		#define NANOVG_GL_UNIFORMARRAY_SIZE 11
+		union {
+			struct {
+				float scissorMat[12]; // matrices are actually 3 vec4s
+				float paintMat[12];
+				struct NVGcolor innerCol;
+				struct NVGcolor outerCol;
+				float scissorExt[2];
+				float scissorScale[2];
+				float extent[2];
+				float radius;
+				float feather;
+				float strokeMult;
+				float strokeThr;
+				float texType;
+				float type;
+			};
+			float uniformArray[NANOVG_GL_UNIFORMARRAY_SIZE][4];
+		};
+	#endif
 };
+typedef struct GLNVGfragUniforms GLNVGfragUniforms;
 
 struct GLNVGcontext {
-	struct GLNVGshader shader;
-	struct GLNVGtexture* textures;
+	GLNVGshader shader;
+	GLNVGtexture* textures;
 	float view[2];
 	int ntextures;
 	int ctextures;
@@ -180,13 +225,13 @@ struct GLNVGcontext {
 	GLuint fragBuf;
 #endif
 	int fragSize;
-	int edgeAntiAlias;
+	int flags;
 
 	// Per frame buffers
-	struct GLNVGcall* calls;
+	GLNVGcall* calls;
 	int ccalls;
 	int ncalls;
-	struct GLNVGpath* paths;
+	GLNVGpath* paths;
 	int cpaths;
 	int npaths;
 	struct NVGvertex* verts;
@@ -195,13 +240,78 @@ struct GLNVGcontext {
 	unsigned char* uniforms;
 	int cuniforms;
 	int nuniforms;
+
+	// cached state
+	#if NANOVG_GL_USE_STATE_FILTER
+	GLuint boundTexture;
+	GLuint stencilMask;
+	GLenum stencilFunc;
+	GLint stencilFuncRef;
+	GLuint stencilFuncMask;
+	#endif
 };
+typedef struct GLNVGcontext GLNVGcontext;
 
 static int glnvg__maxi(int a, int b) { return a > b ? a : b; }
 
-static struct GLNVGtexture* glnvg__allocTexture(struct GLNVGcontext* gl)
+#ifdef NANOVG_GLES2
+static unsigned int glnvg__nearestPow2(unsigned int num)
+{
+	unsigned n = num > 0 ? num - 1 : 0;
+	n |= n >> 1;
+	n |= n >> 2;
+	n |= n >> 4;
+	n |= n >> 8;
+	n |= n >> 16;
+	n++;
+	return n;
+}
+#endif
+
+static void glnvg__bindTexture(GLNVGcontext* gl, GLuint tex)
+{
+#if NANOVG_GL_USE_STATE_FILTER
+	if (gl->boundTexture != tex) {
+		gl->boundTexture = tex;
+		glBindTexture(GL_TEXTURE_2D, tex);
+	}
+#else
+	glBindTexture(GL_TEXTURE_2D, tex);
+#endif
+}
+
+static void glnvg__stencilMask(GLNVGcontext* gl, GLuint mask)
+{
+#if NANOVG_GL_USE_STATE_FILTER
+	if (gl->stencilMask != mask) {
+		gl->stencilMask = mask;
+		glStencilMask(mask);
+	}
+#else
+	glStencilMask(mask);
+#endif
+}
+
+static void glnvg__stencilFunc(GLNVGcontext* gl, GLenum func, GLint ref, GLuint mask)
 {
-	struct GLNVGtexture* tex = NULL;
+#if NANOVG_GL_USE_STATE_FILTER
+	if ((gl->stencilFunc != func) ||
+		(gl->stencilFuncRef != ref) ||
+		(gl->stencilFuncMask != mask)) {
+		
+		gl->stencilFunc = func;
+		gl->stencilFuncRef = ref;
+		gl->stencilFuncMask = mask;
+		glStencilFunc(func, ref, mask);
+	}
+#else
+	glStencilFunc(func, ref, mask);
+#endif
+}
+
+static GLNVGtexture* glnvg__allocTexture(GLNVGcontext* gl)
+{
+	GLNVGtexture* tex = NULL;
 	int i;
 
 	for (i = 0; i < gl->ntextures; i++) {
@@ -212,9 +322,9 @@ static struct GLNVGtexture* glnvg__allocTexture(struct GLNVGcontext* gl)
 	}
 	if (tex == NULL) {
 		if (gl->ntextures+1 > gl->ctextures) {
-			struct GLNVGtexture* textures;
+			GLNVGtexture* textures;
 			int ctextures = glnvg__maxi(gl->ntextures+1, 4) +  gl->ctextures/2; // 1.5x Overallocate
-			textures = (struct GLNVGtexture*)realloc(gl->textures, sizeof(struct GLNVGtexture)*ctextures);
+			textures = (GLNVGtexture*)realloc(gl->textures, sizeof(GLNVGtexture)*ctextures);
 			if (textures == NULL) return NULL;
 			gl->textures = textures;
 			gl->ctextures = ctextures;
@@ -228,7 +338,7 @@ static struct GLNVGtexture* glnvg__allocTexture(struct GLNVGcontext* gl)
 	return tex;
 }
 
-static struct GLNVGtexture* glnvg__findTexture(struct GLNVGcontext* gl, int id)
+static GLNVGtexture* glnvg__findTexture(GLNVGcontext* gl, int id)
 {
 	int i;
 	for (i = 0; i < gl->ntextures; i++)
@@ -237,12 +347,12 @@ static struct GLNVGtexture* glnvg__findTexture(struct GLNVGcontext* gl, int id)
 	return NULL;
 }
 
-static int glnvg__deleteTexture(struct GLNVGcontext* gl, int id)
+static int glnvg__deleteTexture(GLNVGcontext* gl, int id)
 {
 	int i;
 	for (i = 0; i < gl->ntextures; i++) {
 		if (gl->textures[i].id == id) {
-			if (gl->textures[i].tex != 0)
+			if (gl->textures[i].tex != 0 && (gl->textures[i].flags & NVG_IMAGE_NODELETE) == 0)
 				glDeleteTextures(1, &gl->textures[i].tex);
 			memset(&gl->textures[i], 0, sizeof(gl->textures[i]));
 			return 1;
@@ -271,17 +381,18 @@ static void glnvg__dumpProgramError(GLuint prog, const char* name)
 	printf("Program %s error:\n%s\n", name, str);
 }
 
-static int glnvg__checkError(const char* str)
+static void glnvg__checkError(GLNVGcontext* gl, const char* str)
 {
-	GLenum err = glGetError();
+	GLenum err;
+	if ((gl->flags & NVG_DEBUG) == 0) return;
+	err = glGetError();
 	if (err != GL_NO_ERROR) {
 		printf("Error %08x after %s\n", err, str);
-		return 1;
+		return;
 	}
-	return 0;
 }
 
-static int glnvg__createShader(struct GLNVGshader* shader, const char* name, const char* header, const char* opts, const char* vshader, const char* fshader)
+static int glnvg__createShader(GLNVGshader* shader, const char* name, const char* header, const char* opts, const char* vshader, const char* fshader)
 {
 	GLint status;
 	GLuint prog, vert, frag;
@@ -333,7 +444,7 @@ static int glnvg__createShader(struct GLNVGshader* shader, const char* name, con
 	return 1;
 }
 
-static void glnvg__deleteShader(struct GLNVGshader* shader)
+static void glnvg__deleteShader(GLNVGshader* shader)
 {
 	if (shader->prog != 0)
 		glDeleteProgram(shader->prog);
@@ -343,7 +454,7 @@ static void glnvg__deleteShader(struct GLNVGshader* shader)
 		glDeleteShader(shader->frag);
 }
 
-static void glnvg__getUniforms(struct GLNVGshader* shader)
+static void glnvg__getUniforms(GLNVGshader* shader)
 {
 	shader->loc[GLNVG_LOC_VIEWSIZE] = glGetUniformLocation(shader->prog, "viewSize");
 	shader->loc[GLNVG_LOC_TEX] = glGetUniformLocation(shader->prog, "tex");
@@ -351,46 +462,37 @@ static void glnvg__getUniforms(struct GLNVGshader* shader)
 #if NANOVG_GL_USE_UNIFORMBUFFER
 	shader->loc[GLNVG_LOC_FRAG] = glGetUniformBlockIndex(shader->prog, "frag");
 #else
-	shader->loc[GLNVG_LOC_SCISSORMAT] = glGetUniformLocation(shader->prog, "scissorMat");
-	shader->loc[GLNVG_LOC_SCISSOREXT] = glGetUniformLocation(shader->prog, "scissorExt");
-	shader->loc[GLNVG_LOC_SCISSORSCALE] = glGetUniformLocation(shader->prog, "scissorScale");
-	shader->loc[GLNVG_LOC_PAINTMAT] = glGetUniformLocation(shader->prog, "paintMat");
-	shader->loc[GLNVG_LOC_EXTENT] = glGetUniformLocation(shader->prog, "extent");
-	shader->loc[GLNVG_LOC_RADIUS] = glGetUniformLocation(shader->prog, "radius");
-	shader->loc[GLNVG_LOC_FEATHER] = glGetUniformLocation(shader->prog, "feather");
-	shader->loc[GLNVG_LOC_INNERCOL] = glGetUniformLocation(shader->prog, "innerCol");
-	shader->loc[GLNVG_LOC_OUTERCOL] = glGetUniformLocation(shader->prog, "outerCol");
-	shader->loc[GLNVG_LOC_STROKEMULT] = glGetUniformLocation(shader->prog, "strokeMult");
-	shader->loc[GLNVG_LOC_TEXTYPE] = glGetUniformLocation(shader->prog, "texType");
-	shader->loc[GLNVG_LOC_TYPE] = glGetUniformLocation(shader->prog, "type");
+	shader->loc[GLNVG_LOC_FRAG] = glGetUniformLocation(shader->prog, "frag");
 #endif
 }
 
 static int glnvg__renderCreate(void* uptr)
 {
-	struct GLNVGcontext* gl = (struct GLNVGcontext*)uptr;
+	GLNVGcontext* gl = (GLNVGcontext*)uptr;
 	int align = 4;
 
 	// TODO: mediump float may not be enough for GLES2 in iOS.
 	// see the following discussion: https://github.com/memononen/nanovg/issues/46
 	static const char* shaderHeader =
 #if defined NANOVG_GL2
-		"#define NANOVG_GL2 1\n";
+		"#define NANOVG_GL2 1\n"
 #elif defined NANOVG_GL3
 		"#version 150 core\n"
-#if NANOVG_GL_USE_UNIFORMBUFFER
-		"#define USE_UNIFORMBUFFER 1\n"
-#endif
-		"#define NANOVG_GL3 1\n";
+		"#define NANOVG_GL3 1\n"
 #elif defined NANOVG_GLES2
 		"#version 100\n"
-		"precision mediump float;\n"
-		"#define NANOVG_GL2 1\n";
+		"#define NANOVG_GL2 1\n"
 #elif defined NANOVG_GLES3
 		"#version 300 es\n"
-		"precision mediump float;\n"
-		"#define NANOVG_GL3 1\n";
+		"#define NANOVG_GL3 1\n"
+#endif
+
+#if NANOVG_GL_USE_UNIFORMBUFFER
+	"#define USE_UNIFORMBUFFER 1\n"
+#else
+	"#define UNIFORMARRAY_SIZE 11\n"
 #endif
+	"\n";
 
 	static const char* fillVertShader =
 		"#ifdef NANOVG_GL3\n"
@@ -413,6 +515,13 @@ static int glnvg__renderCreate(void* uptr)
 		"}\n";
 
 	static const char* fillFragShader = 
+		"#ifdef GL_ES\n"
+		"#if defined(GL_FRAGMENT_PRECISION_HIGH) || defined(NANOVG_GL3)\n"
+		" precision highp float;\n"
+		"#else\n"
+		" precision mediump float;\n"
+		"#endif\n"
+		"#endif\n"
 		"#ifdef NANOVG_GL3\n"
 		"#ifdef USE_UNIFORMBUFFER\n"
 		"	layout(std140) uniform frag {\n"
@@ -426,44 +535,38 @@ static int glnvg__renderCreate(void* uptr)
 		"		float radius;\n"
 		"		float feather;\n"
 		"		float strokeMult;\n"
+		"		float strokeThr;\n"
 		"		int texType;\n"
 		"		int type;\n"
 		"	};\n"
-		"#else\n"
-		"	uniform mat3 scissorMat;\n"
-		"	uniform mat3 paintMat;\n"
-		"	uniform vec4 innerCol;\n"
-		"	uniform vec4 outerCol;\n"
-		"	uniform vec2 scissorExt;\n"
-		"	uniform vec2 scissorScale;\n"
-		"	uniform vec2 extent;\n"
-		"	uniform float radius;\n"
-		"	uniform float feather;\n"
-		"	uniform float strokeMult;\n"
-		"	uniform int texType;\n"
-		"	uniform int type;\n"
+		"#else\n" // NANOVG_GL3 && !USE_UNIFORMBUFFER
+		"	uniform vec4 frag[UNIFORMARRAY_SIZE];\n"
 		"#endif\n"
 		"	uniform sampler2D tex;\n"
 		"	in vec2 ftcoord;\n"
 		"	in vec2 fpos;\n"
 		"	out vec4 outColor;\n"
-		"#else\n"
-		"	uniform mat3 scissorMat;\n"
-		"	uniform mat3 paintMat;\n"
-		"	uniform vec4 innerCol;\n"
-		"	uniform vec4 outerCol;\n"
-		"	uniform vec2 scissorExt;\n"
-		"	uniform vec2 scissorScale;\n"
-		"	uniform vec2 extent;\n"
-		"	uniform float radius;\n"
-		"	uniform float feather;\n"
-		"	uniform float strokeMult;\n"
-		"	uniform int texType;\n"
-		"	uniform int type;\n"
+		"#else\n" // !NANOVG_GL3
+		"	uniform vec4 frag[UNIFORMARRAY_SIZE];\n"
 		"	uniform sampler2D tex;\n"
 		"	varying vec2 ftcoord;\n"
 		"	varying vec2 fpos;\n"
 		"#endif\n"
+		"#ifndef USE_UNIFORMBUFFER\n"
+		"	#define scissorMat mat3(frag[0].xyz, frag[1].xyz, frag[2].xyz)\n"
+		"	#define paintMat mat3(frag[3].xyz, frag[4].xyz, frag[5].xyz)\n"
+		"	#define innerCol frag[6]\n"
+		"	#define outerCol frag[7]\n"
+		"	#define scissorExt frag[8].xy\n"
+		"	#define scissorScale frag[8].zw\n"
+		"	#define extent frag[9].xy\n"
+		"	#define radius frag[9].z\n"
+		"	#define feather frag[9].w\n"
+		"	#define strokeMult frag[10].x\n"
+		"	#define strokeThr frag[10].y\n"
+		"	#define texType int(frag[10].z)\n"
+		"	#define type int(frag[10].w)\n"
+		"#endif\n"
 		"\n"
 		"float sdroundrect(vec2 pt, vec2 ext, float rad) {\n"
 		"	vec2 ext2 = ext - vec2(rad,rad);\n"
@@ -498,7 +601,7 @@ static int glnvg__renderCreate(void* uptr)
 		"		float d = clamp((sdroundrect(pt, extent, radius) + feather*0.5) / feather, 0.0, 1.0);\n"
 		"		vec4 color = mix(innerCol,outerCol,d);\n"
 		"		// Combine alpha\n"
-		"		color.w *= strokeAlpha * scissor;\n"
+		"		color *= strokeAlpha * scissor;\n"
 		"		result = color;\n"
 		"	} else if (type == 1) {		// Image\n"
 		"		// Calculate color fron texture\n"
@@ -508,9 +611,12 @@ static int glnvg__renderCreate(void* uptr)
 		"#else\n"
 		"		vec4 color = texture2D(tex, pt);\n"
 		"#endif\n"
-		"   	color = texType == 0 ? color : vec4(1,1,1,color.x);\n"
+		"		if (texType == 1) color = vec4(color.xyz*color.w,color.w);"
+		"		if (texType == 2) color = vec4(color.x);"
+		"		// Apply color tint and alpha.\n"
+		"		color *= innerCol;\n"
 		"		// Combine alpha\n"
-		"		color.w *= strokeAlpha * scissor;\n"
+		"		color *= strokeAlpha * scissor;\n"
 		"		result = color;\n"
 		"	} else if (type == 2) {		// Stencil fill\n"
 		"		result = vec4(1,1,1,1);\n"
@@ -520,10 +626,14 @@ static int glnvg__renderCreate(void* uptr)
 		"#else\n"
 		"		vec4 color = texture2D(tex, ftcoord);\n"
 		"#endif\n"
-		"   	color = texType == 0 ? color : vec4(1,1,1,color.x);\n"
-		"		color.w *= scissor;\n"
+		"		if (texType == 1) color = vec4(color.xyz*color.w,color.w);"
+		"		if (texType == 2) color = vec4(color.x);"
+		"		color *= scissor;\n"
 		"		result = color * innerCol;\n"
 		"	}\n"
+		"#ifdef EDGE_AA\n"
+		"	if (strokeAlpha < strokeThr) discard;\n"
+		"#endif\n"
 		"#ifdef NANOVG_GL3\n"
 		"	outColor = result;\n"
 		"#else\n"
@@ -531,9 +641,9 @@ static int glnvg__renderCreate(void* uptr)
 		"#endif\n"
 		"}\n";
 
-	glnvg__checkError("init");
+	glnvg__checkError(gl, "init");
 
-	if (gl->edgeAntiAlias) {
+	if (gl->flags & NVG_ANTIALIAS) {
 		if (glnvg__createShader(&gl->shader, "shader", shaderHeader, "#define EDGE_AA 1\n", fillVertShader, fillFragShader) == 0)
 			return 0;
 	} else {
@@ -541,7 +651,7 @@ static int glnvg__renderCreate(void* uptr)
 			return 0;
 	}
 
-	glnvg__checkError("uniform locations");
+	glnvg__checkError(gl, "uniform locations");
 	glnvg__getUniforms(&gl->shader);
 
 	// Create dynamic vertex array
@@ -556,27 +666,44 @@ static int glnvg__renderCreate(void* uptr)
 	glGenBuffers(1, &gl->fragBuf); 
 	glGetIntegerv(GL_UNIFORM_BUFFER_OFFSET_ALIGNMENT, &align);
 #endif
-	gl->fragSize = sizeof(struct GLNVGfragUniforms) + align - sizeof(struct GLNVGfragUniforms) % align;
+	gl->fragSize = sizeof(GLNVGfragUniforms) + align - sizeof(GLNVGfragUniforms) % align;
 
-	glnvg__checkError("create done");
+	glnvg__checkError(gl, "create done");
 
 	glFinish();
 
 	return 1;
 }
 
-static int glnvg__renderCreateTexture(void* uptr, int type, int w, int h, const unsigned char* data)
+static int glnvg__renderCreateTexture(void* uptr, int type, int w, int h, int imageFlags, const unsigned char* data)
 {
-	struct GLNVGcontext* gl = (struct GLNVGcontext*)uptr;
-	struct GLNVGtexture* tex = glnvg__allocTexture(gl);
+	GLNVGcontext* gl = (GLNVGcontext*)uptr;
+	GLNVGtexture* tex = glnvg__allocTexture(gl);
 
 	if (tex == NULL) return 0;
 
+#ifdef NANOVG_GLES2
+	// Check for non-power of 2.
+	if (glnvg__nearestPow2(w) != (unsigned int)w || glnvg__nearestPow2(h) != (unsigned int)h) {
+		// No repeat
+		if ((imageFlags & NVG_IMAGE_REPEATX) != 0 || (imageFlags & NVG_IMAGE_REPEATY) != 0) {
+			printf("Repeat X/Y is not supported for non power-of-two textures (%d x %d)\n", w, h);
+			imageFlags &= ~(NVG_IMAGE_REPEATX | NVG_IMAGE_REPEATY);
+		}
+		// No mips. 
+		if (imageFlags & NVG_IMAGE_GENERATE_MIPMAPS) {
+			printf("Mip-maps is not support for non power-of-two textures (%d x %d)\n", w, h);
+			imageFlags &= ~NVG_IMAGE_GENERATE_MIPMAPS;
+		}
+	}
+#endif
+
 	glGenTextures(1, &tex->tex);
 	tex->width = w;
 	tex->height = h;
 	tex->type = type;
-	glBindTexture(GL_TEXTURE_2D, tex->tex);
+	tex->flags = imageFlags;
+	glnvg__bindTexture(gl, tex->tex);
 
 	glPixelStorei(GL_UNPACK_ALIGNMENT,1);
 #ifndef NANOVG_GLES2
@@ -585,6 +712,13 @@ static int glnvg__renderCreateTexture(void* uptr, int type, int w, int h, const
 	glPixelStorei(GL_UNPACK_SKIP_ROWS, 0);
 #endif
 
+#if defined (NANOVG_GL2)
+	// GL 1.4 and later has support for generating mipmaps using a tex parameter.
+	if (imageFlags & NVG_IMAGE_GENERATE_MIPMAPS) {
+		glTexParameteri(GL_TEXTURE_2D, GL_GENERATE_MIPMAP, GL_TRUE);
+	}
+#endif
+
 	if (type == NVG_TEXTURE_RGBA)
 		glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA, w, h, 0, GL_RGBA, GL_UNSIGNED_BYTE, data);
 	else
@@ -596,9 +730,23 @@ static int glnvg__renderCreateTexture(void* uptr, int type, int w, int h, const
 		glTexImage2D(GL_TEXTURE_2D, 0, GL_RED, w, h, 0, GL_RED, GL_UNSIGNED_BYTE, data);
 #endif
 
-	glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
+	if (imageFlags & NVG_IMAGE_GENERATE_MIPMAPS) {
+		glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR_MIPMAP_LINEAR);
+	} else {
+		glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
+	}
 	glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
 
+	if (imageFlags & NVG_IMAGE_REPEATX)
+		glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_REPEAT);
+	else
+		glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
+
+	if (imageFlags & NVG_IMAGE_REPEATY)
+		glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_REPEAT);
+	else
+		glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
+
 	glPixelStorei(GL_UNPACK_ALIGNMENT, 4);
 #ifndef NANOVG_GLES2
 	glPixelStorei(GL_UNPACK_ROW_LENGTH, 0);
@@ -606,10 +754,15 @@ static int glnvg__renderCreateTexture(void* uptr, int type, int w, int h, const
 	glPixelStorei(GL_UNPACK_SKIP_ROWS, 0);
 #endif
 
-	if (glnvg__checkError("create tex"))
-		return 0;
+	// The new way to build mipmaps on GLES and GL3
+#if !defined(NANOVG_GL2)
+	if (imageFlags & NVG_IMAGE_GENERATE_MIPMAPS) {
+		glGenerateMipmap(GL_TEXTURE_2D);
+	}
+#endif
 
-	glBindTexture(GL_TEXTURE_2D, 0);
+	glnvg__checkError(gl, "create tex");
+	glnvg__bindTexture(gl, 0);
 
 	return tex->id;
 }
@@ -617,17 +770,17 @@ static int glnvg__renderCreateTexture(void* uptr, int type, int w, int h, const
 
 static int glnvg__renderDeleteTexture(void* uptr, int image)
 {
-	struct GLNVGcontext* gl = (struct GLNVGcontext*)uptr;
+	GLNVGcontext* gl = (GLNVGcontext*)uptr;
 	return glnvg__deleteTexture(gl, image);
 }
 
 static int glnvg__renderUpdateTexture(void* uptr, int image, int x, int y, int w, int h, const unsigned char* data)
 {
-	struct GLNVGcontext* gl = (struct GLNVGcontext*)uptr;
-	struct GLNVGtexture* tex = glnvg__findTexture(gl, image);
+	GLNVGcontext* gl = (GLNVGcontext*)uptr;
+	GLNVGtexture* tex = glnvg__findTexture(gl, image);
 
 	if (tex == NULL) return 0;
-	glBindTexture(GL_TEXTURE_2D, tex->tex);
+	glnvg__bindTexture(gl, tex->tex);
 
 	glPixelStorei(GL_UNPACK_ALIGNMENT,1);
 
@@ -661,15 +814,15 @@ static int glnvg__renderUpdateTexture(void* uptr, int image, int x, int y, int w
 	glPixelStorei(GL_UNPACK_SKIP_ROWS, 0);
 #endif
 
-	glBindTexture(GL_TEXTURE_2D, 0);
+	glnvg__bindTexture(gl, 0);
 
 	return 1;
 }
 
 static int glnvg__renderGetTextureSize(void* uptr, int image, int* w, int* h)
 {
-	struct GLNVGcontext* gl = (struct GLNVGcontext*)uptr;
-	struct GLNVGtexture* tex = glnvg__findTexture(gl, image);
+	GLNVGcontext* gl = (GLNVGcontext*)uptr;
+	GLNVGtexture* tex = glnvg__findTexture(gl, image);
 	if (tex == NULL) return 0;
 	*w = tex->width;
 	*h = tex->height;
@@ -692,21 +845,26 @@ static void glnvg__xformToMat3x4(float* m3, float* t)
 	m3[11] = 0.0f;
 }
 
-static int glnvg__convertPaint(struct GLNVGcontext* gl, struct GLNVGfragUniforms* frag, struct NVGpaint* paint,
-							   struct NVGscissor* scissor, float width, float fringe)
+static NVGcolor glnvg__premulColor(NVGcolor c)
 {
-	struct GLNVGtexture* tex = NULL;
+	c.r *= c.a;
+	c.g *= c.a;
+	c.b *= c.a;
+	return c;
+}
+
+static int glnvg__convertPaint(GLNVGcontext* gl, GLNVGfragUniforms* frag, NVGpaint* paint,
+							   NVGscissor* scissor, float width, float fringe, float strokeThr)
+{
+	GLNVGtexture* tex = NULL;
 	float invxform[6];
 
 	memset(frag, 0, sizeof(*frag));
 
-	frag->innerCol = paint->innerColor;
-	frag->outerCol = paint->outerColor;
+	frag->innerCol = glnvg__premulColor(paint->innerColor);
+	frag->outerCol = glnvg__premulColor(paint->outerColor);
 
-	nvgTransformInverse(invxform, paint->xform);
-	glnvg__xformToMat3x4(frag->paintMat, invxform);
-
-	if (scissor->extent[0] < 0.5f || scissor->extent[1] < 0.5f) {
+	if (scissor->extent[0] < -0.5f || scissor->extent[1] < -0.5f) {
 		memset(frag->scissorMat, 0, sizeof(frag->scissorMat));
 		frag->scissorExt[0] = 1.0f;
 		frag->scissorExt[1] = 1.0f;
@@ -720,95 +878,82 @@ static int glnvg__convertPaint(struct GLNVGcontext* gl, struct GLNVGfragUniforms
 		frag->scissorScale[0] = sqrtf(scissor->xform[0]*scissor->xform[0] + scissor->xform[2]*scissor->xform[2]) / fringe;
 		frag->scissorScale[1] = sqrtf(scissor->xform[1]*scissor->xform[1] + scissor->xform[3]*scissor->xform[3]) / fringe;
 	}
+
 	memcpy(frag->extent, paint->extent, sizeof(frag->extent));
 	frag->strokeMult = (width*0.5f + fringe*0.5f) / fringe;
+	frag->strokeThr = strokeThr;
 
 	if (paint->image != 0) {
 		tex = glnvg__findTexture(gl, paint->image);
 		if (tex == NULL) return 0;
+		if ((tex->flags & NVG_IMAGE_FLIPY) != 0) {
+			float flipped[6];
+			nvgTransformScale(flipped, 1.0f, -1.0f);
+			nvgTransformMultiply(flipped, paint->xform);
+			nvgTransformInverse(invxform, flipped);
+		} else {
+			nvgTransformInverse(invxform, paint->xform);
+		}
 		frag->type = NSVG_SHADER_FILLIMG;
-		frag->texType = tex->type == NVG_TEXTURE_RGBA ? 0 : 1;
+
+		if (tex->type == NVG_TEXTURE_RGBA)
+			frag->texType = (tex->flags & NVG_IMAGE_PREMULTIPLIED) ? 0 : 1;
+		else
+			frag->texType = 2;
+//		printf("frag->texType = %d\n", frag->texType);
 	} else {
 		frag->type = NSVG_SHADER_FILLGRAD;
 		frag->radius = paint->radius;
 		frag->feather = paint->feather;
+		nvgTransformInverse(invxform, paint->xform);
 	}
-	return 1;
-}
 
-static struct GLNVGfragUniforms* nvg__fragUniformPtr(struct GLNVGcontext* gl, int i);
-
-#if !NANOVG_GL_USE_UNIFORMBUFFER
-static void glnvg__mat3(float* dst, float* src)
-{
-	dst[0] = src[0];
-	dst[1] = src[1];
-	dst[2] = src[2];
-
-	dst[3] = src[4];
-	dst[4] = src[5];
-	dst[5] = src[6];
+	glnvg__xformToMat3x4(frag->paintMat, invxform);
 
-	dst[6] = src[8];
-	dst[7] = src[9];
-	dst[8] = src[10];
+	return 1;
 }
-#endif
 
-static void glnvg__setUniforms(struct GLNVGcontext* gl, int uniformOffset, int image)
+static GLNVGfragUniforms* nvg__fragUniformPtr(GLNVGcontext* gl, int i);
+
+static void glnvg__setUniforms(GLNVGcontext* gl, int uniformOffset, int image)
 {
 #if NANOVG_GL_USE_UNIFORMBUFFER
-	glBindBufferRange(GL_UNIFORM_BUFFER, GLNVG_FRAG_BINDING, gl->fragBuf, uniformOffset, sizeof(struct GLNVGfragUniforms));
+	glBindBufferRange(GL_UNIFORM_BUFFER, GLNVG_FRAG_BINDING, gl->fragBuf, uniformOffset, sizeof(GLNVGfragUniforms));
 #else
-	struct GLNVGfragUniforms* frag = nvg__fragUniformPtr(gl, uniformOffset);
-	float tmp[9]; // Maybe there's a way to get rid of this...
-	glnvg__mat3(tmp, frag->scissorMat);
-	glUniformMatrix3fv(gl->shader.loc[GLNVG_LOC_SCISSORMAT], 1, GL_FALSE, tmp);
-	glnvg__mat3(tmp, frag->paintMat);
-	glUniformMatrix3fv(gl->shader.loc[GLNVG_LOC_PAINTMAT], 1, GL_FALSE, tmp);
-	glUniform4fv(gl->shader.loc[GLNVG_LOC_INNERCOL], 1, frag->innerCol.rgba);
-	glUniform4fv(gl->shader.loc[GLNVG_LOC_OUTERCOL], 1, frag->outerCol.rgba);
-	glUniform2fv(gl->shader.loc[GLNVG_LOC_SCISSOREXT], 1, frag->scissorExt);
-	glUniform2fv(gl->shader.loc[GLNVG_LOC_SCISSORSCALE], 1, frag->scissorScale);
-	glUniform2fv(gl->shader.loc[GLNVG_LOC_EXTENT], 1, frag->extent);
-	glUniform1f(gl->shader.loc[GLNVG_LOC_RADIUS], frag->radius);
-	glUniform1f(gl->shader.loc[GLNVG_LOC_FEATHER], frag->feather);
-	glUniform1f(gl->shader.loc[GLNVG_LOC_STROKEMULT], frag->strokeMult);
-	glUniform1i(gl->shader.loc[GLNVG_LOC_TEXTYPE], frag->texType);
-	glUniform1i(gl->shader.loc[GLNVG_LOC_TYPE], frag->type);
+	GLNVGfragUniforms* frag = nvg__fragUniformPtr(gl, uniformOffset);
+	glUniform4fv(gl->shader.loc[GLNVG_LOC_FRAG], NANOVG_GL_UNIFORMARRAY_SIZE, &(frag->uniformArray[0][0]));
 #endif
 
 	if (image != 0) {
-		struct GLNVGtexture* tex = glnvg__findTexture(gl, image);
-		glBindTexture(GL_TEXTURE_2D, tex != NULL ? tex->tex : 0);
-		glnvg__checkError("tex paint tex");
+		GLNVGtexture* tex = glnvg__findTexture(gl, image);
+		glnvg__bindTexture(gl, tex != NULL ? tex->tex : 0);
+		glnvg__checkError(gl, "tex paint tex");
 	} else {
-		glBindTexture(GL_TEXTURE_2D, 0);
+		glnvg__bindTexture(gl, 0);
 	}
 }
 
-static void glnvg__renderViewport(void* uptr, int width, int height, int alphaBlend)
+static void glnvg__renderViewport(void* uptr, int width, int height)
 {
-	struct GLNVGcontext* gl = (struct GLNVGcontext*)uptr;
-	NVG_NOTUSED(alphaBlend);
+	GLNVGcontext* gl = (GLNVGcontext*)uptr;
 	gl->view[0] = (float)width;
 	gl->view[1] = (float)height;
 }
 
-static void glnvg__fill(struct GLNVGcontext* gl, struct GLNVGcall* call)
+static void glnvg__fill(GLNVGcontext* gl, GLNVGcall* call)
 {
-	struct GLNVGpath* paths = &gl->paths[call->pathOffset];
+	GLNVGpath* paths = &gl->paths[call->pathOffset];
 	int i, npaths = call->pathCount;
 
 	// Draw shapes
 	glEnable(GL_STENCIL_TEST);
-	glStencilMask(0xff);
-	glStencilFunc(GL_ALWAYS, 0, ~0L);
+	glnvg__stencilMask(gl, 0xff);
+	glnvg__stencilFunc(gl, GL_ALWAYS, 0, 0xff);
 	glColorMask(GL_FALSE, GL_FALSE, GL_FALSE, GL_FALSE);
 
 	// set bindpoint for solid loc
 	glnvg__setUniforms(gl, call->uniformOffset, 0);
-	glnvg__checkError("fill simple");
+	glnvg__checkError(gl, "fill simple");
 
 	glStencilOpSeparate(GL_FRONT, GL_KEEP, GL_KEEP, GL_INCR_WRAP);
 	glStencilOpSeparate(GL_BACK, GL_KEEP, GL_KEEP, GL_DECR_WRAP);
@@ -817,14 +962,14 @@ static void glnvg__fill(struct GLNVGcontext* gl, struct GLNVGcall* call)
 		glDrawArrays(GL_TRIANGLE_FAN, paths[i].fillOffset, paths[i].fillCount);
 	glEnable(GL_CULL_FACE);
 
-	// Draw aliased off-pixels
+	// Draw anti-aliased pixels
 	glColorMask(GL_TRUE, GL_TRUE, GL_TRUE, GL_TRUE);
 
 	glnvg__setUniforms(gl, call->uniformOffset + gl->fragSize, call->image);
-	glnvg__checkError("fill fill");
+	glnvg__checkError(gl, "fill fill");
 
-	if (gl->edgeAntiAlias) {
-		glStencilFunc(GL_EQUAL, 0x00, 0xff);
+	if (gl->flags & NVG_ANTIALIAS) {
+		glnvg__stencilFunc(gl, GL_EQUAL, 0x00, 0xff);
 		glStencilOp(GL_KEEP, GL_KEEP, GL_KEEP);
 		// Draw fringes
 		for (i = 0; i < npaths; i++)
@@ -832,54 +977,96 @@ static void glnvg__fill(struct GLNVGcontext* gl, struct GLNVGcall* call)
 	}
 
 	// Draw fill
-	glStencilFunc(GL_NOTEQUAL, 0x0, 0xff);
+	glnvg__stencilFunc(gl, GL_NOTEQUAL, 0x0, 0xff);
 	glStencilOp(GL_ZERO, GL_ZERO, GL_ZERO);
 	glDrawArrays(GL_TRIANGLES, call->triangleOffset, call->triangleCount);
 
 	glDisable(GL_STENCIL_TEST);
 }
 
-static void glnvg__convexFill(struct GLNVGcontext* gl, struct GLNVGcall* call)
+static void glnvg__convexFill(GLNVGcontext* gl, GLNVGcall* call)
 {
-	struct GLNVGpath* paths = &gl->paths[call->pathOffset];
+	GLNVGpath* paths = &gl->paths[call->pathOffset];
 	int i, npaths = call->pathCount;
 
 	glnvg__setUniforms(gl, call->uniformOffset, call->image);
-	glnvg__checkError("convex fill");
+	glnvg__checkError(gl, "convex fill");
 
 	for (i = 0; i < npaths; i++)
 		glDrawArrays(GL_TRIANGLE_FAN, paths[i].fillOffset, paths[i].fillCount);
-	if (gl->edgeAntiAlias) {
+	if (gl->flags & NVG_ANTIALIAS) {
 		// Draw fringes
 		for (i = 0; i < npaths; i++)
 			glDrawArrays(GL_TRIANGLE_STRIP, paths[i].strokeOffset, paths[i].strokeCount);
 	}
 }
 
-static void glnvg__stroke(struct GLNVGcontext* gl, struct GLNVGcall* call)
+static void glnvg__stroke(GLNVGcontext* gl, GLNVGcall* call)
 {
-	struct GLNVGpath* paths = &gl->paths[call->pathOffset];
+	GLNVGpath* paths = &gl->paths[call->pathOffset];
 	int npaths = call->pathCount, i;
 
-	glnvg__setUniforms(gl, call->uniformOffset, call->image);
-	glnvg__checkError("stroke fill");
+	if (gl->flags & NVG_STENCIL_STROKES) {
 
-	// Draw Strokes
-	for (i = 0; i < npaths; i++)
-		glDrawArrays(GL_TRIANGLE_STRIP, paths[i].strokeOffset, paths[i].strokeCount);
+		glEnable(GL_STENCIL_TEST);
+		glnvg__stencilMask(gl, 0xff);
+
+		// Fill the stroke base without overlap
+		glnvg__stencilFunc(gl, GL_EQUAL, 0x0, 0xff);
+		glStencilOp(GL_KEEP, GL_KEEP, GL_INCR);
+		glnvg__setUniforms(gl, call->uniformOffset + gl->fragSize, call->image);
+		glnvg__checkError(gl, "stroke fill 0");
+		for (i = 0; i < npaths; i++)
+			glDrawArrays(GL_TRIANGLE_STRIP, paths[i].strokeOffset, paths[i].strokeCount);
+
+		// Draw anti-aliased pixels.
+		glnvg__setUniforms(gl, call->uniformOffset, call->image);
+		glnvg__stencilFunc(gl, GL_EQUAL, 0x00, 0xff);
+		glStencilOp(GL_KEEP, GL_KEEP, GL_KEEP);
+		for (i = 0; i < npaths; i++)
+			glDrawArrays(GL_TRIANGLE_STRIP, paths[i].strokeOffset, paths[i].strokeCount);
+
+		// Clear stencil buffer.		
+		glColorMask(GL_FALSE, GL_FALSE, GL_FALSE, GL_FALSE);
+		glnvg__stencilFunc(gl, GL_ALWAYS, 0x0, 0xff);
+		glStencilOp(GL_ZERO, GL_ZERO, GL_ZERO);
+		glnvg__checkError(gl, "stroke fill 1");
+		for (i = 0; i < npaths; i++)
+			glDrawArrays(GL_TRIANGLE_STRIP, paths[i].strokeOffset, paths[i].strokeCount);
+		glColorMask(GL_TRUE, GL_TRUE, GL_TRUE, GL_TRUE);
+
+		glDisable(GL_STENCIL_TEST);
+
+//		glnvg__convertPaint(gl, nvg__fragUniformPtr(gl, call->uniformOffset + gl->fragSize), paint, scissor, strokeWidth, fringe, 1.0f - 0.5f/255.0f);
+
+	} else {
+		glnvg__setUniforms(gl, call->uniformOffset, call->image);
+		glnvg__checkError(gl, "stroke fill");
+		// Draw Strokes
+		for (i = 0; i < npaths; i++)
+			glDrawArrays(GL_TRIANGLE_STRIP, paths[i].strokeOffset, paths[i].strokeCount);
+	}
 }
 
-static void glnvg__triangles(struct GLNVGcontext* gl, struct GLNVGcall* call)
+static void glnvg__triangles(GLNVGcontext* gl, GLNVGcall* call)
 {
 	glnvg__setUniforms(gl, call->uniformOffset, call->image);
-	glnvg__checkError("triangles fill");
+	glnvg__checkError(gl, "triangles fill");
 
 	glDrawArrays(GL_TRIANGLES, call->triangleOffset, call->triangleCount);
 }
 
-static void glnvg__renderFlush(void* uptr, int alphaBlend)
+static void glnvg__renderCancel(void* uptr) {
+	GLNVGcontext* gl = (GLNVGcontext*)uptr;
+	gl->nverts = 0;
+	gl->npaths = 0;
+	gl->ncalls = 0;
+	gl->nuniforms = 0;
+}
+
+static void glnvg__renderFlush(void* uptr)
 {
-	struct GLNVGcontext* gl = (struct GLNVGcontext*)uptr;
+	GLNVGcontext* gl = (GLNVGcontext*)uptr;
 	int i;
 
 	if (gl->ncalls > 0) {
@@ -887,10 +1074,7 @@ static void glnvg__renderFlush(void* uptr, int alphaBlend)
 		// Setup require GL state.
 		glUseProgram(gl->shader.prog);
 
-		if (alphaBlend == NVG_PREMULTIPLIED_ALPHA)
-			glBlendFuncSeparate(GL_SRC_ALPHA, GL_ONE_MINUS_SRC_ALPHA, GL_ONE, GL_ONE_MINUS_SRC_ALPHA);
-		else
-			glBlendFunc(GL_SRC_ALPHA, GL_ONE_MINUS_SRC_ALPHA);
+		glBlendFunc(GL_ONE, GL_ONE_MINUS_SRC_ALPHA);
 		glEnable(GL_CULL_FACE);
 		glCullFace(GL_BACK);
 		glFrontFace(GL_CCW);
@@ -902,6 +1086,14 @@ static void glnvg__renderFlush(void* uptr, int alphaBlend)
 		glStencilOp(GL_KEEP, GL_KEEP, GL_KEEP);
 		glStencilFunc(GL_ALWAYS, 0, 0xffffffff);
 		glActiveTexture(GL_TEXTURE0);
+		glBindTexture(GL_TEXTURE_2D, 0);
+		#if NANOVG_GL_USE_STATE_FILTER
+		gl->boundTexture = 0;
+		gl->stencilMask = 0xffffffff;
+		gl->stencilFunc = GL_ALWAYS;
+		gl->stencilFuncRef = 0;
+		gl->stencilFuncMask = 0xffffffff;
+		#endif
 
 #if NANOVG_GL_USE_UNIFORMBUFFER
 		// Upload ubo for frag shaders
@@ -914,13 +1106,13 @@ static void glnvg__renderFlush(void* uptr, int alphaBlend)
 		glBindVertexArray(gl->vertArr);
 #endif
 		glBindBuffer(GL_ARRAY_BUFFER, gl->vertBuf);
-		glBufferData(GL_ARRAY_BUFFER, gl->nverts * sizeof(struct NVGvertex), gl->verts, GL_STREAM_DRAW);
+		glBufferData(GL_ARRAY_BUFFER, gl->nverts * sizeof(NVGvertex), gl->verts, GL_STREAM_DRAW);
 		glEnableVertexAttribArray(0);
 		glEnableVertexAttribArray(1);
-		glVertexAttribPointer(0, 2, GL_FLOAT, GL_FALSE, sizeof(struct NVGvertex), (const GLvoid*)(size_t)0);
-		glVertexAttribPointer(1, 2, GL_FLOAT, GL_FALSE, sizeof(struct NVGvertex), (const GLvoid*)(0 + 2*sizeof(float)));
+		glVertexAttribPointer(0, 2, GL_FLOAT, GL_FALSE, sizeof(NVGvertex), (const GLvoid*)(size_t)0);
+		glVertexAttribPointer(1, 2, GL_FLOAT, GL_FALSE, sizeof(NVGvertex), (const GLvoid*)(0 + 2*sizeof(float)));
 
-		// Set view and texture just once per frame.		
+		// Set view and texture just once per frame.
 		glUniform1i(gl->shader.loc[GLNVG_LOC_TEX], 0);
 		glUniform2fv(gl->shader.loc[GLNVG_LOC_VIEWSIZE], 1, gl->view);
 
@@ -929,7 +1121,7 @@ static void glnvg__renderFlush(void* uptr, int alphaBlend)
 #endif
 
 		for (i = 0; i < gl->ncalls; i++) {
-			struct GLNVGcall* call = &gl->calls[i];
+			GLNVGcall* call = &gl->calls[i];
 			if (call->type == GLNVG_FILL)
 				glnvg__fill(gl, call);
 			else if (call->type == GLNVG_CONVEXFILL)
@@ -944,9 +1136,11 @@ static void glnvg__renderFlush(void* uptr, int alphaBlend)
 		glDisableVertexAttribArray(1);
 #if defined NANOVG_GL3
 		glBindVertexArray(0);
-#endif
+#endif	
+		glDisable(GL_CULL_FACE);
+			glBindBuffer(GL_ARRAY_BUFFER, 0);
 		glUseProgram(0);
-		glBindTexture(GL_TEXTURE_2D, 0);
+		glnvg__bindTexture(gl, 0);
 	}
 
 	// Reset calls
@@ -956,7 +1150,7 @@ static void glnvg__renderFlush(void* uptr, int alphaBlend)
 	gl->nuniforms = 0;
 }
 
-static int glnvg__maxVertCount(const struct NVGpath* paths, int npaths)
+static int glnvg__maxVertCount(const NVGpath* paths, int npaths)
 {
 	int i, count = 0;
 	for (i = 0; i < npaths; i++) {
@@ -966,29 +1160,29 @@ static int glnvg__maxVertCount(const struct NVGpath* paths, int npaths)
 	return count;
 }
 
-static struct GLNVGcall* glnvg__allocCall(struct GLNVGcontext* gl)
+static GLNVGcall* glnvg__allocCall(GLNVGcontext* gl)
 {
-	struct GLNVGcall* ret = NULL;
+	GLNVGcall* ret = NULL;
 	if (gl->ncalls+1 > gl->ccalls) {
-		struct GLNVGcall* calls;
-		int ccalls = glnvg__maxi(gl->ncalls+1, 128) + gl->ccalls; // 1.5x Overallocate
-		calls = (struct GLNVGcall*)realloc(gl->calls, sizeof(struct GLNVGcall) * ccalls);
+		GLNVGcall* calls;
+		int ccalls = glnvg__maxi(gl->ncalls+1, 128) + gl->ccalls/2; // 1.5x Overallocate
+		calls = (GLNVGcall*)realloc(gl->calls, sizeof(GLNVGcall) * ccalls);
 		if (calls == NULL) return NULL;
 		gl->calls = calls;
 		gl->ccalls = ccalls;
 	}
 	ret = &gl->calls[gl->ncalls++];
-	memset(ret, 0, sizeof(struct GLNVGcall));
+	memset(ret, 0, sizeof(GLNVGcall));
 	return ret;
 }
 
-static int glnvg__allocPaths(struct GLNVGcontext* gl, int n)
+static int glnvg__allocPaths(GLNVGcontext* gl, int n)
 {
 	int ret = 0;
 	if (gl->npaths+n > gl->cpaths) {
-		struct GLNVGpath* paths;
-		int cpaths = glnvg__maxi(gl->npaths + n, 128) + gl->cpaths; // 1.5x Overallocate
-		paths = (struct GLNVGpath*)realloc(gl->paths, sizeof(struct GLNVGpath) * cpaths);
+		GLNVGpath* paths;
+		int cpaths = glnvg__maxi(gl->npaths + n, 128) + gl->cpaths/2; // 1.5x Overallocate
+		paths = (GLNVGpath*)realloc(gl->paths, sizeof(GLNVGpath) * cpaths);
 		if (paths == NULL) return -1;
 		gl->paths = paths;
 		gl->cpaths = cpaths;
@@ -998,13 +1192,13 @@ static int glnvg__allocPaths(struct GLNVGcontext* gl, int n)
 	return ret;
 }
 
-static int glnvg__allocVerts(struct GLNVGcontext* gl, int n)
+static int glnvg__allocVerts(GLNVGcontext* gl, int n)
 {
 	int ret = 0;
 	if (gl->nverts+n > gl->cverts) {
-		struct NVGvertex* verts;
+		NVGvertex* verts;
 		int cverts = glnvg__maxi(gl->nverts + n, 4096) + gl->cverts/2; // 1.5x Overallocate
-		verts = (struct NVGvertex*)realloc(gl->verts, sizeof(struct NVGvertex) * cverts);
+		verts = (NVGvertex*)realloc(gl->verts, sizeof(NVGvertex) * cverts);
 		if (verts == NULL) return -1;
 		gl->verts = verts;
 		gl->cverts = cverts;
@@ -1014,7 +1208,7 @@ static int glnvg__allocVerts(struct GLNVGcontext* gl, int n)
 	return ret;
 }
 
-static int glnvg__allocFragUniforms(struct GLNVGcontext* gl, int n)
+static int glnvg__allocFragUniforms(GLNVGcontext* gl, int n)
 {
 	int ret = 0, structSize = gl->fragSize;
 	if (gl->nuniforms+n > gl->cuniforms) {
@@ -1030,12 +1224,12 @@ static int glnvg__allocFragUniforms(struct GLNVGcontext* gl, int n)
 	return ret;
 }
 
-static struct GLNVGfragUniforms* nvg__fragUniformPtr(struct GLNVGcontext* gl, int i)
+static GLNVGfragUniforms* nvg__fragUniformPtr(GLNVGcontext* gl, int i)
 {
-	return (struct GLNVGfragUniforms*)&gl->uniforms[i];
+	return (GLNVGfragUniforms*)&gl->uniforms[i];
 }
 
-static void glnvg__vset(struct NVGvertex* vtx, float x, float y, float u, float v)
+static void glnvg__vset(NVGvertex* vtx, float x, float y, float u, float v)
 {
 	vtx->x = x;
 	vtx->y = y;
@@ -1043,13 +1237,13 @@ static void glnvg__vset(struct NVGvertex* vtx, float x, float y, float u, float
 	vtx->v = v;
 }
 
-static void glnvg__renderFill(void* uptr, struct NVGpaint* paint, struct NVGscissor* scissor, float fringe,
-							  const float* bounds, const struct NVGpath* paths, int npaths)
+static void glnvg__renderFill(void* uptr, NVGpaint* paint, NVGscissor* scissor, float fringe,
+							  const float* bounds, const NVGpath* paths, int npaths)
 {
-	struct GLNVGcontext* gl = (struct GLNVGcontext*)uptr;
-	struct GLNVGcall* call = glnvg__allocCall(gl);
-	struct NVGvertex* quad;
-	struct GLNVGfragUniforms* frag;
+	GLNVGcontext* gl = (GLNVGcontext*)uptr;
+	GLNVGcall* call = glnvg__allocCall(gl);
+	NVGvertex* quad;
+	GLNVGfragUniforms* frag;
 	int i, maxverts, offset;
 
 	if (call == NULL) return;
@@ -1069,19 +1263,19 @@ static void glnvg__renderFill(void* uptr, struct NVGpaint* paint, struct NVGscis
 	if (offset == -1) goto error;
 
 	for (i = 0; i < npaths; i++) {
-		struct GLNVGpath* copy = &gl->paths[call->pathOffset + i];
-		const struct NVGpath* path = &paths[i];
-		memset(copy, 0, sizeof(struct GLNVGpath));
+		GLNVGpath* copy = &gl->paths[call->pathOffset + i];
+		const NVGpath* path = &paths[i];
+		memset(copy, 0, sizeof(GLNVGpath));
 		if (path->nfill > 0) {
 			copy->fillOffset = offset;
 			copy->fillCount = path->nfill;
-			memcpy(&gl->verts[offset], path->fill, sizeof(struct NVGvertex) * path->nfill);
+			memcpy(&gl->verts[offset], path->fill, sizeof(NVGvertex) * path->nfill);
 			offset += path->nfill;
 		}
 		if (path->nstroke > 0) {
 			copy->strokeOffset = offset;
 			copy->strokeCount = path->nstroke;
-			memcpy(&gl->verts[offset], path->stroke, sizeof(struct NVGvertex) * path->nstroke);
+			memcpy(&gl->verts[offset], path->stroke, sizeof(NVGvertex) * path->nstroke);
 			offset += path->nstroke;
 		}
 	}
@@ -1105,14 +1299,15 @@ static void glnvg__renderFill(void* uptr, struct NVGpaint* paint, struct NVGscis
 		// Simple shader for stencil
 		frag = nvg__fragUniformPtr(gl, call->uniformOffset);
 		memset(frag, 0, sizeof(*frag));
+		frag->strokeThr = -1.0f;
 		frag->type = NSVG_SHADER_SIMPLE;
 		// Fill shader
-		glnvg__convertPaint(gl, nvg__fragUniformPtr(gl, call->uniformOffset + gl->fragSize), paint, scissor, fringe, fringe);
+		glnvg__convertPaint(gl, nvg__fragUniformPtr(gl, call->uniformOffset + gl->fragSize), paint, scissor, fringe, fringe, -1.0f);
 	} else {
 		call->uniformOffset = glnvg__allocFragUniforms(gl, 1);
 		if (call->uniformOffset == -1) goto error;
 		// Fill shader
-		glnvg__convertPaint(gl, nvg__fragUniformPtr(gl, call->uniformOffset), paint, scissor, fringe, fringe);
+		glnvg__convertPaint(gl, nvg__fragUniformPtr(gl, call->uniformOffset), paint, scissor, fringe, fringe, -1.0f);
 	}
 
 	return;
@@ -1123,11 +1318,11 @@ error:
 	if (gl->ncalls > 0) gl->ncalls--;
 }
 
-static void glnvg__renderStroke(void* uptr, struct NVGpaint* paint, struct NVGscissor* scissor, float fringe,
-								float strokeWidth, const struct NVGpath* paths, int npaths)
+static void glnvg__renderStroke(void* uptr, NVGpaint* paint, NVGscissor* scissor, float fringe,
+								float strokeWidth, const NVGpath* paths, int npaths)
 {
-	struct GLNVGcontext* gl = (struct GLNVGcontext*)uptr;
-	struct GLNVGcall* call = glnvg__allocCall(gl);
+	GLNVGcontext* gl = (GLNVGcontext*)uptr;
+	GLNVGcall* call = glnvg__allocCall(gl);
 	int i, maxverts, offset;
 
 	if (call == NULL) return;
@@ -1144,21 +1339,31 @@ static void glnvg__renderStroke(void* uptr, struct NVGpaint* paint, struct NVGsc
 	if (offset == -1) goto error;
 
 	for (i = 0; i < npaths; i++) {
-		struct GLNVGpath* copy = &gl->paths[call->pathOffset + i];
-		const struct NVGpath* path = &paths[i];
-		memset(copy, 0, sizeof(struct GLNVGpath));
+		GLNVGpath* copy = &gl->paths[call->pathOffset + i];
+		const NVGpath* path = &paths[i];
+		memset(copy, 0, sizeof(GLNVGpath));
 		if (path->nstroke) {
 			copy->strokeOffset = offset;
 			copy->strokeCount = path->nstroke;
-			memcpy(&gl->verts[offset], path->stroke, sizeof(struct NVGvertex) * path->nstroke);
+			memcpy(&gl->verts[offset], path->stroke, sizeof(NVGvertex) * path->nstroke);
 			offset += path->nstroke;
 		}
 	}
 
-	// Fill shader
-	call->uniformOffset = glnvg__allocFragUniforms(gl, 1);
-	if (call->uniformOffset == -1) goto error;
-	glnvg__convertPaint(gl, nvg__fragUniformPtr(gl, call->uniformOffset), paint, scissor, strokeWidth, fringe);
+	if (gl->flags & NVG_STENCIL_STROKES) {
+		// Fill shader
+		call->uniformOffset = glnvg__allocFragUniforms(gl, 2);
+		if (call->uniformOffset == -1) goto error;
+
+		glnvg__convertPaint(gl, nvg__fragUniformPtr(gl, call->uniformOffset), paint, scissor, strokeWidth, fringe, -1.0f);
+		glnvg__convertPaint(gl, nvg__fragUniformPtr(gl, call->uniformOffset + gl->fragSize), paint, scissor, strokeWidth, fringe, 1.0f - 0.5f/255.0f);
+
+	} else {
+		// Fill shader
+		call->uniformOffset = glnvg__allocFragUniforms(gl, 1);
+		if (call->uniformOffset == -1) goto error;
+		glnvg__convertPaint(gl, nvg__fragUniformPtr(gl, call->uniformOffset), paint, scissor, strokeWidth, fringe, -1.0f);
+	}
 
 	return;
 
@@ -1168,12 +1373,12 @@ error:
 	if (gl->ncalls > 0) gl->ncalls--;
 }
 
-static void glnvg__renderTriangles(void* uptr, struct NVGpaint* paint, struct NVGscissor* scissor,
-								   const struct NVGvertex* verts, int nverts)
+static void glnvg__renderTriangles(void* uptr, NVGpaint* paint, NVGscissor* scissor,
+								   const NVGvertex* verts, int nverts)
 {
-	struct GLNVGcontext* gl = (struct GLNVGcontext*)uptr;
-	struct GLNVGcall* call = glnvg__allocCall(gl);
-	struct GLNVGfragUniforms* frag;
+	GLNVGcontext* gl = (GLNVGcontext*)uptr;
+	GLNVGcall* call = glnvg__allocCall(gl);
+	GLNVGfragUniforms* frag;
 
 	if (call == NULL) return;
 
@@ -1185,13 +1390,13 @@ static void glnvg__renderTriangles(void* uptr, struct NVGpaint* paint, struct NV
 	if (call->triangleOffset == -1) goto error;
 	call->triangleCount = nverts;
 
-	memcpy(&gl->verts[call->triangleOffset], verts, sizeof(struct NVGvertex) * nverts);
+	memcpy(&gl->verts[call->triangleOffset], verts, sizeof(NVGvertex) * nverts);
 
 	// Fill shader
 	call->uniformOffset = glnvg__allocFragUniforms(gl, 1);
 	if (call->uniformOffset == -1) goto error;
 	frag = nvg__fragUniformPtr(gl, call->uniformOffset);
-	glnvg__convertPaint(gl, frag, paint, scissor, 1.0f, 1.0f);
+	glnvg__convertPaint(gl, frag, paint, scissor, 1.0f, 1.0f, -1.0f);
 	frag->type = NSVG_SHADER_IMG;
 
 	return;
@@ -1204,7 +1409,7 @@ error:
 
 static void glnvg__renderDelete(void* uptr)
 {
-	struct GLNVGcontext* gl = (struct GLNVGcontext*)uptr;
+	GLNVGcontext* gl = (GLNVGcontext*)uptr;
 	int i;
 	if (gl == NULL) return;
 
@@ -1222,30 +1427,35 @@ static void glnvg__renderDelete(void* uptr)
 		glDeleteBuffers(1, &gl->vertBuf);
 
 	for (i = 0; i < gl->ntextures; i++) {
-		if (gl->textures[i].tex != 0)
+		if (gl->textures[i].tex != 0 && (gl->textures[i].flags & NVG_IMAGE_NODELETE) == 0)
 			glDeleteTextures(1, &gl->textures[i].tex);
 	}
 	free(gl->textures);
 
+	free(gl->paths);
+	free(gl->verts);
+	free(gl->uniforms);
+	free(gl->calls);
+
 	free(gl);
 }
 
 
 #if defined NANOVG_GL2
-struct NVGcontext* nvgCreateGL2(int atlasw, int atlash, int edgeaa)
+NVGcontext* nvgCreateGL2(int flags)
 #elif defined NANOVG_GL3
-struct NVGcontext* nvgCreateGL3(int atlasw, int atlash, int edgeaa)
+NVGcontext* nvgCreateGL3(int flags)
 #elif defined NANOVG_GLES2
-struct NVGcontext* nvgCreateGLES2(int atlasw, int atlash, int edgeaa)
+NVGcontext* nvgCreateGLES2(int flags)
 #elif defined NANOVG_GLES3
-struct NVGcontext* nvgCreateGLES3(int atlasw, int atlash, int edgeaa)
+NVGcontext* nvgCreateGLES3(int flags)
 #endif
 {
-	struct NVGparams params;
-	struct NVGcontext* ctx = NULL;
-	struct GLNVGcontext* gl = (struct GLNVGcontext*)malloc(sizeof(struct GLNVGcontext));
+	NVGparams params;
+	NVGcontext* ctx = NULL;
+	GLNVGcontext* gl = (GLNVGcontext*)malloc(sizeof(GLNVGcontext));
 	if (gl == NULL) goto error;
-	memset(gl, 0, sizeof(struct GLNVGcontext));
+	memset(gl, 0, sizeof(GLNVGcontext));
 
 	memset(&params, 0, sizeof(params));
 	params.renderCreate = glnvg__renderCreate;
@@ -1254,17 +1464,16 @@ struct NVGcontext* nvgCreateGLES3(int atlasw, int atlash, int edgeaa)
 	params.renderUpdateTexture = glnvg__renderUpdateTexture;
 	params.renderGetTextureSize = glnvg__renderGetTextureSize;
 	params.renderViewport = glnvg__renderViewport;
+	params.renderCancel = glnvg__renderCancel;
 	params.renderFlush = glnvg__renderFlush;
 	params.renderFill = glnvg__renderFill;
 	params.renderStroke = glnvg__renderStroke;
 	params.renderTriangles = glnvg__renderTriangles;
 	params.renderDelete = glnvg__renderDelete;
 	params.userPtr = gl;
-	params.atlasWidth = atlasw;
-	params.atlasHeight = atlash;
-	params.edgeAntiAlias = edgeaa;
+	params.edgeAntiAlias = flags & NVG_ANTIALIAS ? 1 : 0;
 
-	gl->edgeAntiAlias = edgeaa;
+	gl->flags = flags;
 
 	ctx = nvgCreateInternal(&params);
 	if (ctx == NULL) goto error;
@@ -1277,17 +1486,40 @@ error:
 	return NULL;
 }
 
-#if NANOVG_GL2
-void nvgDeleteGL2(struct NVGcontext* ctx)
-#elif NANOVG_GL3
-void nvgDeleteGL3(struct NVGcontext* ctx)
-#elif NANOVG_GLES2
-void nvgDeleteGLES2(struct NVGcontext* ctx)
-#elif NANOVG_GLES3
-void nvgDeleteGLES3(struct NVGcontext* ctx)
+#if defined NANOVG_GL2
+void nvgDeleteGL2(NVGcontext* ctx)
+#elif defined NANOVG_GL3
+void nvgDeleteGL3(NVGcontext* ctx)
+#elif defined NANOVG_GLES2
+void nvgDeleteGLES2(NVGcontext* ctx)
+#elif defined NANOVG_GLES3
+void nvgDeleteGLES3(NVGcontext* ctx)
 #endif
 {
 	nvgDeleteInternal(ctx);
 }
 
-#endif
+int nvglCreateImageFromHandle(NVGcontext* ctx, GLuint textureId, int w, int h, int imageFlags)
+{
+	GLNVGcontext* gl = (GLNVGcontext*)nvgInternalParams(ctx)->userPtr;
+	GLNVGtexture* tex = glnvg__allocTexture(gl);
+
+	if (tex == NULL) return 0;
+
+	tex->type = NVG_TEXTURE_RGBA;
+	tex->tex = textureId;
+	tex->flags = imageFlags;
+	tex->width = w;
+	tex->height = h;
+
+	return tex->id;
+}
+
+GLuint nvglImageHandle(NVGcontext* ctx, int image)
+{
+	GLNVGcontext* gl = (GLNVGcontext*)nvgInternalParams(ctx)->userPtr;
+	GLNVGtexture* tex = glnvg__findTexture(gl, image);
+	return tex->tex;
+}
+
+#endif /* NANOVG_GL_IMPLEMENTATION */
diff --git a/dgl/src/nanovg/stb_image.h b/dgl/src/nanovg/stb_image.h
index d72ee6a4..4da7ec65 100644
--- a/dgl/src/nanovg/stb_image.h
+++ b/dgl/src/nanovg/stb_image.h
@@ -1,4676 +1,4647 @@
-/* stbi-1.33 - public domain JPEG/PNG reader - http://nothings.org/stb_image.c
-   when you control the images you're loading
-                                     no warranty implied; use at your own risk
-
-   QUICK NOTES:
-      Primarily of interest to game developers and other people who can
-          avoid problematic images and only need the trivial interface
-
-      JPEG baseline (no JPEG progressive)
-      PNG 8-bit-per-channel only
-
-      TGA (not sure what subset, if a subset)
-      BMP non-1bpp, non-RLE
-      PSD (composited view only, no extra channels)
-
-      GIF (*comp always reports as 4-channel)
-      HDR (radiance rgbE format)
-      PIC (Softimage PIC)
-
-      - decode from memory or through FILE (define STBI_NO_STDIO to remove code)
-      - decode from arbitrary I/O callbacks
-      - overridable dequantizing-IDCT, YCbCr-to-RGB conversion (define STBI_SIMD)
-
-   Latest revisions:
-      1.33 (2011-07-14) minor fixes suggested by Dave Moore
-      1.32 (2011-07-13) info support for all filetypes (SpartanJ)
-      1.31 (2011-06-19) a few more leak fixes, bug in PNG handling (SpartanJ)
-      1.30 (2011-06-11) added ability to load files via io callbacks (Ben Wenger)
-      1.29 (2010-08-16) various warning fixes from Aurelien Pocheville
-      1.28 (2010-08-01) fix bug in GIF palette transparency (SpartanJ)
-      1.27 (2010-08-01) cast-to-uint8 to fix warnings (Laurent Gomila)
-                        allow trailing 0s at end of image data (Laurent Gomila)
-      1.26 (2010-07-24) fix bug in file buffering for PNG reported by SpartanJ
-
-   See end of file for full revision history.
-
-   TODO:
-      stbi_info support for BMP,PSD,HDR,PIC
-
-
- ============================    Contributors    =========================
-
- Image formats                                Optimizations & bugfixes
-    Sean Barrett (jpeg, png, bmp)                Fabian "ryg" Giesen
-    Nicolas Schulz (hdr, psd)
-    Jonathan Dummer (tga)                     Bug fixes & warning fixes
-    Jean-Marc Lienher (gif)                      Marc LeBlanc
-    Tom Seddon (pic)                             Christpher Lloyd
-    Thatcher Ulrich (psd)                        Dave Moore
-                                                 Won Chun
-                                                 the Horde3D community
- Extensions, features                            Janez Zemva
-    Jetro Lauha (stbi_info)                      Jonathan Blow
-    James "moose2000" Brown (iPhone PNG)         Laurent Gomila
-    Ben "Disch" Wenger (io callbacks)            Aruelien Pocheville
-    Martin "SpartanJ" Golini                     Ryamond Barbiero
-                                                 David Woo
-
-
- If your name should be here but isn't, let Sean know.
-
-*/
-
-#ifndef STBI_INCLUDE_STB_IMAGE_H
-#define STBI_INCLUDE_STB_IMAGE_H
-
-// To get a header file for this, either cut and paste the header,
-// or create stb_image.h, #define STBI_HEADER_FILE_ONLY, and
-// then include stb_image.c from it.
-
-////   begin header file  ////////////////////////////////////////////////////
-//
-// Limitations:
-//    - no jpeg progressive support
-//    - non-HDR formats support 8-bit samples only (jpeg, png)
-//    - no delayed line count (jpeg) -- IJG doesn't support either
-//    - no 1-bit BMP
-//    - GIF always returns *comp=4
-//
-// Basic usage (see HDR discussion below):
-//    int x,y,n;
-//    unsigned char *data = stbi_load(filename, &x, &y, &n, 0);
-//    // ... process data if not NULL ...
-//    // ... x = width, y = height, n = # 8-bit components per pixel ...
-//    // ... replace '0' with '1'..'4' to force that many components per pixel
-//    // ... but 'n' will always be the number that it would have been if you said 0
-//    stbi_image_free(data)
-//
-// Standard parameters:
-//    int *x       -- outputs image width in pixels
-//    int *y       -- outputs image height in pixels
-//    int *comp    -- outputs # of image components in image file
-//    int req_comp -- if non-zero, # of image components requested in result
-//
-// The return value from an image loader is an 'unsigned char *' which points
-// to the pixel data. The pixel data consists of *y scanlines of *x pixels,
-// with each pixel consisting of N interleaved 8-bit components; the first
-// pixel pointed to is top-left-most in the image. There is no padding between
-// image scanlines or between pixels, regardless of format. The number of
-// components N is 'req_comp' if req_comp is non-zero, or *comp otherwise.
-// If req_comp is non-zero, *comp has the number of components that _would_
-// have been output otherwise. E.g. if you set req_comp to 4, you will always
-// get RGBA output, but you can check *comp to easily see if it's opaque.
-//
-// An output image with N components has the following components interleaved
-// in this order in each pixel:
-//
-//     N=#comp     components
-//       1           grey
-//       2           grey, alpha
-//       3           red, green, blue
-//       4           red, green, blue, alpha
-//
-// If image loading fails for any reason, the return value will be NULL,
-// and *x, *y, *comp will be unchanged. The function stbi_failure_reason()
-// can be queried for an extremely brief, end-user unfriendly explanation
-// of why the load failed. Define STBI_NO_FAILURE_STRINGS to avoid
-// compiling these strings at all, and STBI_FAILURE_USERMSG to get slightly
-// more user-friendly ones.
-//
-// Paletted PNG, BMP, GIF, and PIC images are automatically depalettized.
-//
-// ===========================================================================
-//
-// iPhone PNG support:
-//
-// By default we convert iphone-formatted PNGs back to RGB; nominally they
-// would silently load as BGR, except the existing code should have just
-// failed on such iPhone PNGs. But you can disable this conversion by
-// by calling stbi_convert_iphone_png_to_rgb(0), in which case
-// you will always just get the native iphone "format" through.
-//
-// Call stbi_set_unpremultiply_on_load(1) as well to force a divide per
-// pixel to remove any premultiplied alpha *only* if the image file explicitly
-// says there's premultiplied data (currently only happens in iPhone images,
-// and only if iPhone convert-to-rgb processing is on).
-//
-// ===========================================================================
-//
-// HDR image support   (disable by defining STBI_NO_HDR)
-//
-// stb_image now supports loading HDR images in general, and currently
-// the Radiance .HDR file format, although the support is provided
-// generically. You can still load any file through the existing interface;
-// if you attempt to load an HDR file, it will be automatically remapped to
-// LDR, assuming gamma 2.2 and an arbitrary scale factor defaulting to 1;
-// both of these constants can be reconfigured through this interface:
-//
-//     stbi_hdr_to_ldr_gamma(2.2f);
-//     stbi_hdr_to_ldr_scale(1.0f);
-//
-// (note, do not use _inverse_ constants; stbi_image will invert them
-// appropriately).
-//
-// Additionally, there is a new, parallel interface for loading files as
-// (linear) floats to preserve the full dynamic range:
-//
-//    float *data = stbi_loadf(filename, &x, &y, &n, 0);
-//
-// If you load LDR images through this interface, those images will
-// be promoted to floating point values, run through the inverse of
-// constants corresponding to the above:
-//
-//     stbi_ldr_to_hdr_scale(1.0f);
-//     stbi_ldr_to_hdr_gamma(2.2f);
-//
-// Finally, given a filename (or an open file or memory block--see header
-// file for details) containing image data, you can query for the "most
-// appropriate" interface to use (that is, whether the image is HDR or
-// not), using:
-//
-//     stbi_is_hdr(char *filename);
-//
-// ===========================================================================
-//
-// 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).
-
-
-#ifndef STBI_NO_STDIO
-
-#if defined(_MSC_VER) && _MSC_VER >= 0x1400
-#define _CRT_SECURE_NO_WARNINGS // suppress bogus warnings about fopen()
-#endif
-
-#include <stdio.h>
-#endif
-
-#define STBI_VERSION 1
-
-enum
-{
-   STBI_default = 0, // only used for req_comp
-
-   STBI_grey       = 1,
-   STBI_grey_alpha = 2,
-   STBI_rgb        = 3,
-   STBI_rgb_alpha  = 4
-};
-
-typedef unsigned char stbi_uc;
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-//////////////////////////////////////////////////////////////////////////////
-//
-// PRIMARY API - works on images of any type
-//
-
-//
-// load image by filename, open file, or memory buffer
-//
-
-extern stbi_uc *stbi_load_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp);
-
-#ifndef STBI_NO_STDIO
-extern stbi_uc *stbi_load            (char const *filename,     int *x, int *y, int *comp, int req_comp);
-extern stbi_uc *stbi_load_from_file  (FILE *f,                  int *x, int *y, int *comp, int req_comp);
-// for stbi_load_from_file, file pointer is left pointing immediately after image
-#endif
-
-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,unsigned n);            // skip the next 'n' bytes
-   int      (*eof)   (void *user);                       // returns nonzero if we are at end of file/data
-} stbi_io_callbacks;
-
-extern stbi_uc *stbi_load_from_callbacks  (stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp, int req_comp);
-
-#ifndef STBI_NO_HDR
-   extern float *stbi_loadf_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp);
-
-   #ifndef STBI_NO_STDIO
-   extern float *stbi_loadf            (char const *filename,   int *x, int *y, int *comp, int req_comp);
-   extern float *stbi_loadf_from_file  (FILE *f,                int *x, int *y, int *comp, int req_comp);
-   #endif
-
-   extern float *stbi_loadf_from_callbacks  (stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp, int req_comp);
-
-   extern void   stbi_hdr_to_ldr_gamma(float gamma);
-   extern void   stbi_hdr_to_ldr_scale(float scale);
-
-   extern void   stbi_ldr_to_hdr_gamma(float gamma);
-   extern void   stbi_ldr_to_hdr_scale(float scale);
-#endif // STBI_NO_HDR
-
-// stbi_is_hdr is always defined
-extern int    stbi_is_hdr_from_callbacks(stbi_io_callbacks const *clbk, void *user);
-extern int    stbi_is_hdr_from_memory(stbi_uc const *buffer, int len);
-#ifndef STBI_NO_STDIO
-extern int      stbi_is_hdr          (char const *filename);
-extern int      stbi_is_hdr_from_file(FILE *f);
-#endif // STBI_NO_STDIO
-
-
-// get a VERY brief reason for failure
-// NOT THREADSAFE
-extern const char *stbi_failure_reason  (void);
-
-// free the loaded image -- this is just free()
-extern void     stbi_image_free      (void *retval_from_stbi_load);
-
-// get image dimensions & components without fully decoding
-extern int      stbi_info_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp);
-extern int      stbi_info_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp);
-
-#ifndef STBI_NO_STDIO
-extern int      stbi_info            (char const *filename,     int *x, int *y, int *comp);
-extern int      stbi_info_from_file  (FILE *f,                  int *x, int *y, int *comp);
-
-#endif
-
-
-
-// for image formats that explicitly notate that they have premultiplied alpha,
-// we just return the colors as stored in the file. set this flag to force
-// unpremultiplication. results are undefined if the unpremultiply overflow.
-extern void stbi_set_unpremultiply_on_load(int flag_true_if_should_unpremultiply);
-
-// indicate whether we should process iphone images back to canonical format,
-// or just pass them through "as-is"
-extern void stbi_convert_iphone_png_to_rgb(int flag_true_if_should_convert);
-
-
-// ZLIB client - used by PNG, available for other purposes
-
-extern char *stbi_zlib_decode_malloc_guesssize(const char *buffer, int len, int initial_size, int *outlen);
-extern char *stbi_zlib_decode_malloc(const char *buffer, int len, int *outlen);
-extern int   stbi_zlib_decode_buffer(char *obuffer, int olen, const char *ibuffer, int ilen);
-
-extern char *stbi_zlib_decode_noheader_malloc(const char *buffer, int len, int *outlen);
-extern int   stbi_zlib_decode_noheader_buffer(char *obuffer, int olen, const char *ibuffer, int ilen);
-
-
-// define faster low-level operations (typically SIMD support)
-#ifdef STBI_SIMD
-typedef void (*stbi_idct_8x8)(stbi_uc *out, int out_stride, short data[64], unsigned short *dequantize);
-// compute an integer IDCT on "input"
-//     input[x] = data[x] * dequantize[x]
-//     write results to 'out': 64 samples, each run of 8 spaced by 'out_stride'
-//                             CLAMP results to 0..255
-typedef void (*stbi_YCbCr_to_RGB_run)(stbi_uc *output, stbi_uc const  *y, stbi_uc const *cb, stbi_uc const *cr, int count, int step);
-// compute a conversion from YCbCr to RGB
-//     'count' pixels
-//     write pixels to 'output'; each pixel is 'step' bytes (either 3 or 4; if 4, write '255' as 4th), order R,G,B
-//     y: Y input channel
-//     cb: Cb input channel; scale/biased to be 0..255
-//     cr: Cr input channel; scale/biased to be 0..255
-
-extern void stbi_install_idct(stbi_idct_8x8 func);
-extern void stbi_install_YCbCr_to_RGB(stbi_YCbCr_to_RGB_run func);
-#endif // STBI_SIMD
-
-
-#ifdef __cplusplus
-}
-#endif
-
-//
-//
-////   end header file   /////////////////////////////////////////////////////
-#endif // STBI_INCLUDE_STB_IMAGE_H
-
-#ifndef STBI_HEADER_FILE_ONLY
-
-#ifndef STBI_NO_HDR
-#include <math.h>  // ldexp
-#include <string.h> // strcmp, strtok
-#endif
-
-#ifndef STBI_NO_STDIO
-#include <stdio.h>
-#endif
-#include <stdlib.h>
-#include <memory.h>
-#include <assert.h>
-#include <stdarg.h>
-
-#ifndef _MSC_VER
-   #ifdef __cplusplus
-   #define stbi_inline inline
-   #else
-   #define stbi_inline
-   #endif
-#else
-   #define stbi_inline __forceinline
-#endif
-
-
-// implementation:
-typedef unsigned char  uint8;
-typedef unsigned short uint16;
-typedef   signed short  int16;
-typedef unsigned int   uint32;
-typedef   signed int    int32;
-typedef unsigned int   uint;
-
-// should produce compiler error if size is wrong
-typedef unsigned char validate_uint32[sizeof(uint32)==4 ? 1 : -1];
-
-#if defined(STBI_NO_STDIO) && !defined(STBI_NO_WRITE)
-#define STBI_NO_WRITE
-#endif
-
-#define STBI_NOTUSED(v)  (void)sizeof(v)
-
-#ifdef _MSC_VER
-#define STBI_HAS_LROTL
-#endif
-
-#ifdef STBI_HAS_LROTL
-   #define stbi_lrot(x,y)  _lrotl(x,y)
-#else
-   #define stbi_lrot(x,y)  (((x) << (y)) | ((x) >> (32 - (y))))
-#endif
-
-///////////////////////////////////////////////
-//
-//  stbi struct and start_xxx functions
-
-// stbi structure is our basic context used by all images, so it
-// contains all the IO context, plus some basic image information
-typedef struct
-{
-   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;
-   uint8 buffer_start[128];
-
-   uint8 *img_buffer, *img_buffer_end;
-   uint8 *img_buffer_original;
-} stbi;
-
-
-static void refill_buffer(stbi *s);
-
-// initialize a memory-decode context
-static void start_mem(stbi *s, uint8 const *buffer, int len)
-{
-   s->io.read = NULL;
-   s->read_from_callbacks = 0;
-   s->img_buffer = s->img_buffer_original = (uint8 *) buffer;
-   s->img_buffer_end = (uint8 *) buffer+len;
-}
-
-// initialize a callback-based context
-static void start_callbacks(stbi *s, stbi_io_callbacks *c, void *user)
-{
-   s->io = *c;
-   s->io_user_data = user;
-   s->buflen = sizeof(s->buffer_start);
-   s->read_from_callbacks = 1;
-   s->img_buffer_original = s->buffer_start;
-   refill_buffer(s);
-}
-
-#ifndef STBI_NO_STDIO
-
-static int stdio_read(void *user, char *data, int size)
-{
-   return (int) fread(data,1,size,(FILE*) user);
-}
-
-static void stdio_skip(void *user, unsigned n)
-{
-   fseek((FILE*) user, n, SEEK_CUR);
-}
-
-static int stdio_eof(void *user)
-{
-   return feof((FILE*) user);
-}
-
-static stbi_io_callbacks stbi_stdio_callbacks =
-{
-   stdio_read,
-   stdio_skip,
-   stdio_eof,
-};
-
-static void start_file(stbi *s, FILE *f)
-{
-   start_callbacks(s, &stbi_stdio_callbacks, (void *) f);
-}
-
-//static void stop_file(stbi *s) { }
-
-#endif // !STBI_NO_STDIO
-
-static void stbi_rewind(stbi *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;
-}
-
-static int      stbi_jpeg_test(stbi *s);
-static stbi_uc *stbi_jpeg_load(stbi *s, int *x, int *y, int *comp, int req_comp);
-static int      stbi_jpeg_info(stbi *s, int *x, int *y, int *comp);
-static int      stbi_png_test(stbi *s);
-static stbi_uc *stbi_png_load(stbi *s, int *x, int *y, int *comp, int req_comp);
-static int      stbi_png_info(stbi *s, int *x, int *y, int *comp);
-static int      stbi_bmp_test(stbi *s);
-static stbi_uc *stbi_bmp_load(stbi *s, int *x, int *y, int *comp, int req_comp);
-static int      stbi_tga_test(stbi *s);
-static stbi_uc *stbi_tga_load(stbi *s, int *x, int *y, int *comp, int req_comp);
-static int      stbi_tga_info(stbi *s, int *x, int *y, int *comp);
-static int      stbi_psd_test(stbi *s);
-static stbi_uc *stbi_psd_load(stbi *s, int *x, int *y, int *comp, int req_comp);
-static int      stbi_hdr_test(stbi *s);
-static float   *stbi_hdr_load(stbi *s, int *x, int *y, int *comp, int req_comp);
-static int      stbi_pic_test(stbi *s);
-static stbi_uc *stbi_pic_load(stbi *s, int *x, int *y, int *comp, int req_comp);
-static int      stbi_gif_test(stbi *s);
-static stbi_uc *stbi_gif_load(stbi *s, int *x, int *y, int *comp, int req_comp);
-static int      stbi_gif_info(stbi *s, int *x, int *y, int *comp);
-
-
-// this is not threadsafe
-static const char *failure_reason;
-
-const char *stbi_failure_reason(void)
-{
-   return failure_reason;
-}
-
-static int e(const char *str)
-{
-   failure_reason = str;
-   return 0;
-}
-
-// e - error
-// epf - error returning pointer to float
-// epuc - error returning pointer to unsigned char
-
-#ifdef STBI_NO_FAILURE_STRINGS
-   #define e(x,y)  0
-#elif defined(STBI_FAILURE_USERMSG)
-   #define e(x,y)  e(y)
-#else
-   #define e(x,y)  e(x)
-#endif
-
-#define epf(x,y)   ((float *) (e(x,y)?NULL:NULL))
-#define epuc(x,y)  ((unsigned char *) (e(x,y)?NULL:NULL))
-
-void stbi_image_free(void *retval_from_stbi_load)
-{
-   free(retval_from_stbi_load);
-}
-
-#ifndef STBI_NO_HDR
-static float   *ldr_to_hdr(stbi_uc *data, int x, int y, int comp);
-static stbi_uc *hdr_to_ldr(float   *data, int x, int y, int comp);
-#endif
-
-static unsigned char *stbi_load_main(stbi *s, int *x, int *y, int *comp, int req_comp)
-{
-   if (stbi_jpeg_test(s)) return stbi_jpeg_load(s,x,y,comp,req_comp);
-   if (stbi_png_test(s))  return stbi_png_load(s,x,y,comp,req_comp);
-   if (stbi_bmp_test(s))  return stbi_bmp_load(s,x,y,comp,req_comp);
-   if (stbi_gif_test(s))  return stbi_gif_load(s,x,y,comp,req_comp);
-   if (stbi_psd_test(s))  return stbi_psd_load(s,x,y,comp,req_comp);
-   if (stbi_pic_test(s))  return stbi_pic_load(s,x,y,comp,req_comp);
-
-   #ifndef STBI_NO_HDR
-   if (stbi_hdr_test(s)) {
-      float *hdr = stbi_hdr_load(s, x,y,comp,req_comp);
-      return hdr_to_ldr(hdr, *x, *y, req_comp ? req_comp : *comp);
-   }
-   #endif
-
-   // test tga last because it's a crappy test!
-   if (stbi_tga_test(s))
-      return stbi_tga_load(s,x,y,comp,req_comp);
-   return epuc("unknown image type", "Image not of any known type, or corrupt");
-}
-
-#ifndef STBI_NO_STDIO
-unsigned char *stbi_load(char const *filename, int *x, int *y, int *comp, int req_comp)
-{
-   FILE *f = fopen(filename, "rb");
-   unsigned char *result;
-   if (!f) return epuc("can't fopen", "Unable to open file");
-   result = stbi_load_from_file(f,x,y,comp,req_comp);
-   fclose(f);
-   return result;
-}
-
-unsigned char *stbi_load_from_file(FILE *f, int *x, int *y, int *comp, int req_comp)
-{
-   stbi s;
-   start_file(&s,f);
-   return stbi_load_main(&s,x,y,comp,req_comp);
-}
-#endif //!STBI_NO_STDIO
-
-unsigned char *stbi_load_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp)
-{
-   stbi s;
-   start_mem(&s,buffer,len);
-   return stbi_load_main(&s,x,y,comp,req_comp);
-}
-
-unsigned char *stbi_load_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp, int req_comp)
-{
-   stbi s;
-   start_callbacks(&s, (stbi_io_callbacks *) clbk, user);
-   return stbi_load_main(&s,x,y,comp,req_comp);
-}
-
-#ifndef STBI_NO_HDR
-
-float *stbi_loadf_main(stbi *s, int *x, int *y, int *comp, int req_comp)
-{
-   unsigned char *data;
-   #ifndef STBI_NO_HDR
-   if (stbi_hdr_test(s))
-      return stbi_hdr_load(s,x,y,comp,req_comp);
-   #endif
-   data = stbi_load_main(s, x, y, comp, req_comp);
-   if (data)
-      return ldr_to_hdr(data, *x, *y, req_comp ? req_comp : *comp);
-   return epf("unknown image type", "Image not of any known type, or corrupt");
-}
-
-float *stbi_loadf_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp)
-{
-   stbi s;
-   start_mem(&s,buffer,len);
-   return stbi_loadf_main(&s,x,y,comp,req_comp);
-}
-
-float *stbi_loadf_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp, int req_comp)
-{
-   stbi s;
-   start_callbacks(&s, (stbi_io_callbacks *) clbk, user);
-   return stbi_loadf_main(&s,x,y,comp,req_comp);
-}
-
-#ifndef STBI_NO_STDIO
-float *stbi_loadf(char const *filename, int *x, int *y, int *comp, int req_comp)
-{
-   FILE *f = fopen(filename, "rb");
-   float *result;
-   if (!f) return epf("can't fopen", "Unable to open file");
-   result = stbi_loadf_from_file(f,x,y,comp,req_comp);
-   fclose(f);
-   return result;
-}
-
-float *stbi_loadf_from_file(FILE *f, int *x, int *y, int *comp, int req_comp)
-{
-   stbi s;
-   start_file(&s,f);
-   return stbi_loadf_main(&s,x,y,comp,req_comp);
-}
-#endif // !STBI_NO_STDIO
-
-#endif // !STBI_NO_HDR
-
-// these is-hdr-or-not is defined independent of whether STBI_NO_HDR is
-// defined, for API simplicity; if STBI_NO_HDR is defined, it always
-// reports false!
-
-int stbi_is_hdr_from_memory(stbi_uc const *buffer, int len)
-{
-   #ifndef STBI_NO_HDR
-   stbi s;
-   start_mem(&s,buffer,len);
-   return stbi_hdr_test(&s);
-   #else
-   STBI_NOTUSED(buffer);
-   STBI_NOTUSED(len);
-   return 0;
-   #endif
-}
-
-#ifndef STBI_NO_STDIO
-extern int      stbi_is_hdr          (char const *filename)
-{
-   FILE *f = fopen(filename, "rb");
-   int result=0;
-   if (f) {
-      result = stbi_is_hdr_from_file(f);
-      fclose(f);
-   }
-   return result;
-}
-
-extern int      stbi_is_hdr_from_file(FILE *f)
-{
-   #ifndef STBI_NO_HDR
-   stbi s;
-   start_file(&s,f);
-   return stbi_hdr_test(&s);
-   #else
-   return 0;
-   #endif
-}
-#endif // !STBI_NO_STDIO
-
-extern int      stbi_is_hdr_from_callbacks(stbi_io_callbacks const *clbk, void *user)
-{
-   #ifndef STBI_NO_HDR
-   stbi s;
-   start_callbacks(&s, (stbi_io_callbacks *) clbk, user);
-   return stbi_hdr_test(&s);
-   #else
-   return 0;
-   #endif
-}
-
-#ifndef STBI_NO_HDR
-static float h2l_gamma_i=1.0f/2.2f, h2l_scale_i=1.0f;
-static float l2h_gamma=2.2f, l2h_scale=1.0f;
-
-void   stbi_hdr_to_ldr_gamma(float gamma) { h2l_gamma_i = 1/gamma; }
-void   stbi_hdr_to_ldr_scale(float scale) { h2l_scale_i = 1/scale; }
-
-void   stbi_ldr_to_hdr_gamma(float gamma) { l2h_gamma = gamma; }
-void   stbi_ldr_to_hdr_scale(float scale) { l2h_scale = scale; }
-#endif
-
-
-//////////////////////////////////////////////////////////////////////////////
-//
-// Common code used by all image loaders
-//
-
-enum
-{
-   SCAN_load=0,
-   SCAN_type,
-   SCAN_header
-};
-
-static void refill_buffer(stbi *s)
-{
-   int n = (s->io.read)(s->io_user_data,(char*)s->buffer_start,s->buflen);
-   if (n == 0) {
-      // at end of file, treat same as if from memory
-      s->read_from_callbacks = 0;
-      s->img_buffer = s->img_buffer_end-1;
-      *s->img_buffer = 0;
-   } else {
-      s->img_buffer = s->buffer_start;
-      s->img_buffer_end = s->buffer_start + n;
-   }
-}
-
-stbi_inline static int get8(stbi *s)
-{
-   if (s->img_buffer < s->img_buffer_end)
-      return *s->img_buffer++;
-   if (s->read_from_callbacks) {
-      refill_buffer(s);
-      return *s->img_buffer++;
-   }
-   return 0;
-}
-
-stbi_inline static int at_eof(stbi *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;
-}
-
-stbi_inline static uint8 get8u(stbi *s)
-{
-   return (uint8) get8(s);
-}
-
-static void skip(stbi *s, int n)
-{
-   if (s->io.read) {
-      int blen = s->img_buffer_end - s->img_buffer;
-      if (blen < n) {
-         s->img_buffer = s->img_buffer_end;
-         (s->io.skip)(s->io_user_data, n - blen);
-         return;
-      }
-   }
-   s->img_buffer += n;
-}
-
-static int getn(stbi *s, stbi_uc *buffer, int n)
-{
-   if (s->io.read) {
-      int blen = s->img_buffer_end - s->img_buffer;
-      if (blen < n) {
-         int res, count;
-
-         memcpy(buffer, s->img_buffer, blen);
-
-         count = (s->io.read)(s->io_user_data, (char*) buffer + blen, n - blen);
-         res = (count == (n-blen));
-         s->img_buffer = s->img_buffer_end;
-         return res;
-      }
-   }
-
-   if (s->img_buffer+n <= s->img_buffer_end) {
-      memcpy(buffer, s->img_buffer, n);
-      s->img_buffer += n;
-      return 1;
-   } else
-      return 0;
-}
-
-static int get16(stbi *s)
-{
-   int z = get8(s);
-   return (z << 8) + get8(s);
-}
-
-static uint32 get32(stbi *s)
-{
-   uint32 z = get16(s);
-   return (z << 16) + get16(s);
-}
-
-static int get16le(stbi *s)
-{
-   int z = get8(s);
-   return z + (get8(s) << 8);
-}
-
-static uint32 get32le(stbi *s)
-{
-   uint32 z = get16le(s);
-   return z + (get16le(s) << 16);
-}
-
-//////////////////////////////////////////////////////////////////////////////
-//
-//  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 uint8 compute_y(int r, int g, int b)
-{
-   return (uint8) (((r*77) + (g*150) +  (29*b)) >> 8);
-}
-
-static unsigned char *convert_format(unsigned char *data, int img_n, int req_comp, uint x, uint y)
-{
-   int i,j;
-   unsigned char *good;
-
-   if (req_comp == img_n) return data;
-   assert(req_comp >= 1 && req_comp <= 4);
-
-   good = (unsigned char *) malloc(req_comp * x * y);
-   if (good == NULL) {
-      free(data);
-      return epuc("outofmem", "Out of memory");
-   }
-
-   for (j=0; j < (int) y; ++j) {
-      unsigned char *src  = data + j * x * img_n   ;
-      unsigned char *dest = good + j * x * req_comp;
-
-      #define COMBO(a,b)  ((a)*8+(b))
-      #define CASE(a,b)   case COMBO(a,b): for(i=x-1; i >= 0; --i, src += a, dest += b)
-      // convert source image with img_n components to one with req_comp components;
-      // avoid switch per pixel, so use switch per scanline and massive macros
-      switch (COMBO(img_n, req_comp)) {
-         CASE(1,2) dest[0]=src[0], dest[1]=255; break;
-         CASE(1,3) dest[0]=dest[1]=dest[2]=src[0]; break;
-         CASE(1,4) dest[0]=dest[1]=dest[2]=src[0], dest[3]=255; break;
-         CASE(2,1) dest[0]=src[0]; break;
-         CASE(2,3) dest[0]=dest[1]=dest[2]=src[0]; break;
-         CASE(2,4) dest[0]=dest[1]=dest[2]=src[0], dest[3]=src[1]; break;
-         CASE(3,4) dest[0]=src[0],dest[1]=src[1],dest[2]=src[2],dest[3]=255; break;
-         CASE(3,1) dest[0]=compute_y(src[0],src[1],src[2]); break;
-         CASE(3,2) dest[0]=compute_y(src[0],src[1],src[2]), dest[1] = 255; break;
-         CASE(4,1) dest[0]=compute_y(src[0],src[1],src[2]); break;
-         CASE(4,2) dest[0]=compute_y(src[0],src[1],src[2]), dest[1] = src[3]; break;
-         CASE(4,3) dest[0]=src[0],dest[1]=src[1],dest[2]=src[2]; break;
-         default: assert(0);
-      }
-      #undef CASE
-   }
-
-   free(data);
-   return good;
-}
-
-#ifndef STBI_NO_HDR
-static float   *ldr_to_hdr(stbi_uc *data, int x, int y, int comp)
-{
-   int i,k,n;
-   float *output = (float *) malloc(x * y * comp * sizeof(float));
-   if (output == NULL) { free(data); return epf("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, l2h_gamma) * l2h_scale;
-      }
-      if (k < comp) output[i*comp + k] = data[i*comp+k]/255.0f;
-   }
-   free(data);
-   return output;
-}
-
-#define float2int(x)   ((int) (x))
-static stbi_uc *hdr_to_ldr(float   *data, int x, int y, int comp)
-{
-   int i,k,n;
-   stbi_uc *output = (stbi_uc *) malloc(x * y * comp);
-   if (output == NULL) { free(data); return epuc("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]*h2l_scale_i, h2l_gamma_i) * 255 + 0.5f;
-         if (z < 0) z = 0;
-         if (z > 255) z = 255;
-         output[i*comp + k] = (uint8) 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] = (uint8) float2int(z);
-      }
-   }
-   free(data);
-   return output;
-}
-#endif
-
-//////////////////////////////////////////////////////////////////////////////
-//
-//  "baseline" JPEG/JFIF decoder (not actually fully baseline implementation)
-//
-//    simple implementation
-//      - channel subsampling of at most 2 in each dimension
-//      - 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
-//      - uses a lot of intermediate memory, could cache poorly
-//      - load http://nothings.org/remote/anemones.jpg 3 times on 2.8Ghz P4
-//          stb_jpeg:   1.34 seconds (MSVC6, default release build)
-//          stb_jpeg:   1.06 seconds (MSVC6, processor = Pentium Pro)
-//          IJL11.dll:  1.08 seconds (compiled by intel)
-//          IJG 1998:   0.98 seconds (MSVC6, makefile provided by IJG)
-//          IJG 1998:   0.95 seconds (MSVC6, makefile + proc=PPro)
-
-// huffman decoding acceleration
-#define FAST_BITS   9  // larger handles more cases; smaller stomps less cache
-
-typedef struct
-{
-   uint8  fast[1 << FAST_BITS];
-   // weirdly, repacking this into AoS is a 10% speed loss, instead of a win
-   uint16 code[256];
-   uint8  values[256];
-   uint8  size[257];
-   unsigned int maxcode[18];
-   int    delta[17];   // old 'firstsymbol' - old 'firstcode'
-} huffman;
-
-typedef struct
-{
-   #ifdef STBI_SIMD
-   unsigned short dequant2[4][64];
-   #endif
-   stbi *s;
-   huffman huff_dc[4];
-   huffman huff_ac[4];
-   uint8 dequant[4][64];
-
-// 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;
-      uint8 *data;
-      void *raw_data;
-      uint8 *linebuf;
-   } img_comp[4];
-
-   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 scan_n, order[4];
-   int restart_interval, todo;
-} jpeg;
-
-static int build_huffman(huffman *h, int *count)
-{
-   int i,j,k=0,code;
-   // build size list for each symbol (from JPEG spec)
-   for (i=0; i < 16; ++i)
-      for (j=0; j < count[i]; ++j)
-         h->size[k++] = (uint8) (i+1);
-   h->size[k] = 0;
-
-   // compute actual symbols (from jpeg spec)
-   code = 0;
-   k = 0;
-   for(j=1; j <= 16; ++j) {
-      // compute delta to add to code to compute symbol id
-      h->delta[j] = k - code;
-      if (h->size[k] == j) {
-         while (h->size[k] == j)
-            h->code[k++] = (uint16) (code++);
-         if (code-1 >= (1 << j)) return e("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] = (uint8) i;
-         }
-      }
-   }
-   return 1;
-}
-
-static void grow_buffer_unsafe(jpeg *j)
-{
-   do {
-      int b = j->nomore ? 0 : get8(j->s);
-      if (b == 0xff) {
-         int c = get8(j->s);
-         if (c != 0) {
-            j->marker = (unsigned char) c;
-            j->nomore = 1;
-            return;
-         }
-      }
-      j->code_buffer |= b << (24 - j->code_bits);
-      j->code_bits += 8;
-   } while (j->code_bits <= 24);
-}
-
-// (1 << n) - 1
-static uint32 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 decode(jpeg *j, huffman *h)
-{
-   unsigned int temp;
-   int c,k;
-
-   if (j->code_bits < 16) 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)) & bmask[k]) + h->delta[k];
-   assert((((j->code_buffer) >> (32 - h->size[c])) & 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];
-}
-
-// combined JPEG 'receive' and JPEG 'extend', since baseline
-// always extends everything it receives.
-stbi_inline static int extend_receive(jpeg *j, int n)
-{
-   unsigned int m = 1 << (n-1);
-   unsigned int k;
-   if (j->code_bits < n) grow_buffer_unsafe(j);
-
-   #if 1
-   k = stbi_lrot(j->code_buffer, n);
-   j->code_buffer = k & ~bmask[n];
-   k &= bmask[n];
-   j->code_bits -= n;
-   #else
-   k = (j->code_buffer >> (32 - n)) & bmask[n];
-   j->code_bits -= n;
-   j->code_buffer <<= n;
-   #endif
-   // the following test is probably a random branch that won't
-   // predict well. I tried to table accelerate it but failed.
-   // maybe it's compiling as a conditional move?
-   if (k < m)
-      return (-1 << n) + k + 1;
-   else
-      return k;
-}
-
-// 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 uint8 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 decode_block(jpeg *j, short data[64], huffman *hdc, huffman *hac, int b)
-{
-   int diff,dc,k;
-   int t = decode(j, hdc);
-   if (t < 0) return e("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 ? extend_receive(j, t) : 0;
-   dc = j->img_comp[b].dc_pred + diff;
-   j->img_comp[b].dc_pred = dc;
-   data[0] = (short) dc;
-
-   // decode AC components, see JPEG spec
-   k = 1;
-   do {
-      int r,s;
-      int rs = decode(j, hac);
-      if (rs < 0) return e("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
-         data[dezigzag[k++]] = (short) extend_receive(j,s);
-      }
-   } while (k < 64);
-   return 1;
-}
-
-// take a -128..127 value and clamp it and convert to 0..255
-stbi_inline static uint8 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 (uint8) x;
-}
-
-#define f2f(x)  (int) (((x) * 4096 + 0.5))
-#define fsh(x)  ((x) << 12)
-
-// derived from jidctint -- DCT_ISLOW
-#define 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) * f2f(0.5411961f);             \
-   t2 = p1 + p3*f2f(-1.847759065f);            \
-   t3 = p1 + p2*f2f( 0.765366865f);            \
-   p2 = s0;                                    \
-   p3 = s4;                                    \
-   t0 = fsh(p2+p3);                            \
-   t1 = 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)*f2f( 1.175875602f);            \
-   t0 = t0*f2f( 0.298631336f);                 \
-   t1 = t1*f2f( 2.053119869f);                 \
-   t2 = t2*f2f( 3.072711026f);                 \
-   t3 = t3*f2f( 1.501321110f);                 \
-   p1 = p5 + p1*f2f(-0.899976223f);            \
-   p2 = p5 + p2*f2f(-2.562915447f);            \
-   p3 = p3*f2f(-1.961570560f);                 \
-   p4 = p4*f2f(-0.390180644f);                 \
-   t3 += p1+p4;                                \
-   t2 += p2+p3;                                \
-   t1 += p2+p4;                                \
-   t0 += p1+p3;
-
-#ifdef STBI_SIMD
-typedef unsigned short stbi_dequantize_t;
-#else
-typedef uint8 stbi_dequantize_t;
-#endif
-
-// .344 seconds on 3*anemones.jpg
-static void idct_block(uint8 *out, int out_stride, short data[64], stbi_dequantize_t *dequantize)
-{
-   int i,val[64],*v=val;
-   stbi_dequantize_t *dq = dequantize;
-   uint8 *o;
-   short *d = data;
-
-   // columns
-   for (i=0; i < 8; ++i,++d,++dq, ++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] * dq[0] << 2;
-         v[0] = v[8] = v[16] = v[24] = v[32] = v[40] = v[48] = v[56] = dcterm;
-      } else {
-         IDCT_1D(d[ 0]*dq[ 0],d[ 8]*dq[ 8],d[16]*dq[16],d[24]*dq[24],
-                 d[32]*dq[32],d[40]*dq[40],d[48]*dq[48],d[56]*dq[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
-      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] = clamp((x0+t3) >> 17);
-      o[7] = clamp((x0-t3) >> 17);
-      o[1] = clamp((x1+t2) >> 17);
-      o[6] = clamp((x1-t2) >> 17);
-      o[2] = clamp((x2+t1) >> 17);
-      o[5] = clamp((x2-t1) >> 17);
-      o[3] = clamp((x3+t0) >> 17);
-      o[4] = clamp((x3-t0) >> 17);
-   }
-}
-
-#ifdef STBI_SIMD
-static stbi_idct_8x8 stbi_idct_installed = idct_block;
-
-void stbi_install_idct(stbi_idct_8x8 func)
-{
-   stbi_idct_installed = func;
-}
-#endif
-
-#define 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 uint8 get_marker(jpeg *j)
-{
-   uint8 x;
-   if (j->marker != MARKER_none) { x = j->marker; j->marker = MARKER_none; return x; }
-   x = get8u(j->s);
-   if (x != 0xff) return MARKER_none;
-   while (x == 0xff)
-      x = get8u(j->s);
-   return x;
-}
-
-// in each scan, we'll have scan_n components, and the order
-// of the components is specified by order[]
-#define RESTART(x)     ((x) >= 0xd0 && (x) <= 0xd7)
-
-// after a restart interval, reset the entropy decoder and
-// the dc prediction
-static void reset(jpeg *j)
-{
-   j->code_bits = 0;
-   j->code_buffer = 0;
-   j->nomore = 0;
-   j->img_comp[0].dc_pred = j->img_comp[1].dc_pred = j->img_comp[2].dc_pred = 0;
-   j->marker = MARKER_none;
-   j->todo = j->restart_interval ? j->restart_interval : 0x7fffffff;
-   // 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 parse_entropy_coded_data(jpeg *z)
-{
-   reset(z);
-   if (z->scan_n == 1) {
-      int i,j;
-      #ifdef STBI_SIMD
-      __declspec(align(16))
-      #endif
-      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) {
-            if (!decode_block(z, data, z->huff_dc+z->img_comp[n].hd, z->huff_ac+z->img_comp[n].ha, n)) return 0;
-            #ifdef STBI_SIMD
-            stbi_idct_installed(z->img_comp[n].data+z->img_comp[n].w2*j*8+i*8, z->img_comp[n].w2, data, z->dequant2[z->img_comp[n].tq]);
-            #else
-            idct_block(z->img_comp[n].data+z->img_comp[n].w2*j*8+i*8, z->img_comp[n].w2, data, z->dequant[z->img_comp[n].tq]);
-            #endif
-            // every data block is an MCU, so countdown the restart interval
-            if (--z->todo <= 0) {
-               if (z->code_bits < 24) grow_buffer_unsafe(z);
-               // if it's NOT a restart, then just bail, so we get corrupt data
-               // rather than no data
-               if (!RESTART(z->marker)) return 1;
-               reset(z);
-            }
-         }
-      }
-   } else { // interleaved!
-      int i,j,k,x,y;
-      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;
-                     if (!decode_block(z, data, z->huff_dc+z->img_comp[n].hd, z->huff_ac+z->img_comp[n].ha, n)) return 0;
-                     #ifdef STBI_SIMD
-                     stbi_idct_installed(z->img_comp[n].data+z->img_comp[n].w2*y2+x2, z->img_comp[n].w2, data, z->dequant2[z->img_comp[n].tq]);
-                     #else
-                     idct_block(z->img_comp[n].data+z->img_comp[n].w2*y2+x2, z->img_comp[n].w2, data, z->dequant[z->img_comp[n].tq]);
-                     #endif
-                  }
-               }
-            }
-            // 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) grow_buffer_unsafe(z);
-               // if it's NOT a restart, then just bail, so we get corrupt data
-               // rather than no data
-               if (!RESTART(z->marker)) return 1;
-               reset(z);
-            }
-         }
-      }
-   }
-   return 1;
-}
-
-static int process_marker(jpeg *z, int m)
-{
-   int L;
-   switch (m) {
-      case MARKER_none: // no marker found
-         return e("expected marker","Corrupt JPEG");
-
-      case 0xC2: // SOF - progressive
-         return e("progressive jpeg","JPEG format not supported (progressive)");
-
-      case 0xDD: // DRI - specify restart interval
-         if (get16(z->s) != 4) return e("bad DRI len","Corrupt JPEG");
-         z->restart_interval = get16(z->s);
-         return 1;
-
-      case 0xDB: // DQT - define quantization table
-         L = get16(z->s)-2;
-         while (L > 0) {
-            int q = get8(z->s);
-            int p = q >> 4;
-            int t = q & 15,i;
-            if (p != 0) return e("bad DQT type","Corrupt JPEG");
-            if (t > 3) return e("bad DQT table","Corrupt JPEG");
-            for (i=0; i < 64; ++i)
-               z->dequant[t][dezigzag[i]] = get8u(z->s);
-            #ifdef STBI_SIMD
-            for (i=0; i < 64; ++i)
-               z->dequant2[t][i] = z->dequant[t][i];
-            #endif
-            L -= 65;
-         }
-         return L==0;
-
-      case 0xC4: // DHT - define huffman table
-         L = get16(z->s)-2;
-         while (L > 0) {
-            uint8 *v;
-            int sizes[16],i,m=0;
-            int q = get8(z->s);
-            int tc = q >> 4;
-            int th = q & 15;
-            if (tc > 1 || th > 3) return e("bad DHT header","Corrupt JPEG");
-            for (i=0; i < 16; ++i) {
-               sizes[i] = get8(z->s);
-               m += sizes[i];
-            }
-            L -= 17;
-            if (tc == 0) {
-               if (!build_huffman(z->huff_dc+th, sizes)) return 0;
-               v = z->huff_dc[th].values;
-            } else {
-               if (!build_huffman(z->huff_ac+th, sizes)) return 0;
-               v = z->huff_ac[th].values;
-            }
-            for (i=0; i < m; ++i)
-               v[i] = get8u(z->s);
-            L -= m;
-         }
-         return L==0;
-   }
-   // check for comment block or APP blocks
-   if ((m >= 0xE0 && m <= 0xEF) || m == 0xFE) {
-      skip(z->s, get16(z->s)-2);
-      return 1;
-   }
-   return 0;
-}
-
-// after we see SOS
-static int process_scan_header(jpeg *z)
-{
-   int i;
-   int Ls = get16(z->s);
-   z->scan_n = get8(z->s);
-   if (z->scan_n < 1 || z->scan_n > 4 || z->scan_n > (int) z->s->img_n) return e("bad SOS component count","Corrupt JPEG");
-   if (Ls != 6+2*z->scan_n) return e("bad SOS len","Corrupt JPEG");
-   for (i=0; i < z->scan_n; ++i) {
-      int id = get8(z->s), which;
-      int q = 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;
-      z->img_comp[which].hd = q >> 4;   if (z->img_comp[which].hd > 3) return e("bad DC huff","Corrupt JPEG");
-      z->img_comp[which].ha = q & 15;   if (z->img_comp[which].ha > 3) return e("bad AC huff","Corrupt JPEG");
-      z->order[i] = which;
-   }
-   if (get8(z->s) != 0) return e("bad SOS","Corrupt JPEG");
-   get8(z->s); // should be 63, but might be 0
-   if (get8(z->s) != 0) return e("bad SOS","Corrupt JPEG");
-
-   return 1;
-}
-
-static int process_frame_header(jpeg *z, int scan)
-{
-   stbi *s = z->s;
-   int Lf,p,i,q, h_max=1,v_max=1,c;
-   Lf = get16(s);         if (Lf < 11) return e("bad SOF len","Corrupt JPEG"); // JPEG
-   p  = get8(s);          if (p != 8) return e("only 8-bit","JPEG format not supported: 8-bit only"); // JPEG baseline
-   s->img_y = get16(s);   if (s->img_y == 0) return e("no header height", "JPEG format not supported: delayed height"); // Legal, but we don't handle it--but neither does IJG
-   s->img_x = get16(s);   if (s->img_x == 0) return e("0 width","Corrupt JPEG"); // JPEG requires
-   c = get8(s);
-   if (c != 3 && c != 1) return e("bad component count","Corrupt JPEG");    // JFIF requires
-   s->img_n = c;
-   for (i=0; i < c; ++i) {
-      z->img_comp[i].data = NULL;
-      z->img_comp[i].linebuf = NULL;
-   }
-
-   if (Lf != 8+3*s->img_n) return e("bad SOF len","Corrupt JPEG");
-
-   for (i=0; i < s->img_n; ++i) {
-      z->img_comp[i].id = get8(s);
-      if (z->img_comp[i].id != i+1)   // JFIF requires
-         if (z->img_comp[i].id != i)  // some version of jpegtran outputs non-JFIF-compliant files!
-            return e("bad component ID","Corrupt JPEG");
-      q = get8(s);
-      z->img_comp[i].h = (q >> 4);  if (!z->img_comp[i].h || z->img_comp[i].h > 4) return e("bad H","Corrupt JPEG");
-      z->img_comp[i].v = q & 15;    if (!z->img_comp[i].v || z->img_comp[i].v > 4) return e("bad V","Corrupt JPEG");
-      z->img_comp[i].tq = get8(s);  if (z->img_comp[i].tq > 3) return e("bad TQ","Corrupt JPEG");
-   }
-
-   if (scan != SCAN_load) return 1;
-
-   if ((1 << 30) / s->img_x / s->img_n < s->img_y) return e("too large", "Image too large to decode");
-
-   for (i=0; i < s->img_n; ++i) {
-      if (z->img_comp[i].h > h_max) h_max = z->img_comp[i].h;
-      if (z->img_comp[i].v > v_max) v_max = z->img_comp[i].v;
-   }
-
-   // compute interleaved mcu info
-   z->img_h_max = h_max;
-   z->img_v_max = v_max;
-   z->img_mcu_w = h_max * 8;
-   z->img_mcu_h = v_max * 8;
-   z->img_mcu_x = (s->img_x + z->img_mcu_w-1) / z->img_mcu_w;
-   z->img_mcu_y = (s->img_y + z->img_mcu_h-1) / z->img_mcu_h;
-
-   for (i=0; i < s->img_n; ++i) {
-      // number of effective pixels (e.g. for non-interleaved MCU)
-      z->img_comp[i].x = (s->img_x * z->img_comp[i].h + h_max-1) / h_max;
-      z->img_comp[i].y = (s->img_y * z->img_comp[i].v + v_max-1) / v_max;
-      // to simplify generation, we'll allocate enough memory to decode
-      // the bogus oversized data from using interleaved MCUs and their
-      // big blocks (e.g. a 16x16 iMCU on an image of width 33); we won't
-      // discard the extra data until colorspace conversion
-      z->img_comp[i].w2 = z->img_mcu_x * z->img_comp[i].h * 8;
-      z->img_comp[i].h2 = z->img_mcu_y * z->img_comp[i].v * 8;
-      z->img_comp[i].raw_data = malloc(z->img_comp[i].w2 * z->img_comp[i].h2+15);
-      if (z->img_comp[i].raw_data == NULL) {
-         for(--i; i >= 0; --i) {
-            free(z->img_comp[i].raw_data);
-            z->img_comp[i].data = NULL;
-         }
-         return e("outofmem", "Out of memory");
-      }
-      // align blocks for installable-idct using mmx/sse
-      z->img_comp[i].data = (uint8*) (((size_t) z->img_comp[i].raw_data + 15) & ~15);
-      z->img_comp[i].linebuf = NULL;
-   }
-
-   return 1;
-}
-
-// use comparisons since in some cases we handle more than one case (e.g. SOF)
-#define DNL(x)         ((x) == 0xdc)
-#define SOI(x)         ((x) == 0xd8)
-#define EOI(x)         ((x) == 0xd9)
-#define SOF(x)         ((x) == 0xc0 || (x) == 0xc1)
-#define SOS(x)         ((x) == 0xda)
-
-static int decode_jpeg_header(jpeg *z, int scan)
-{
-   int m;
-   z->marker = MARKER_none; // initialize cached marker to empty
-   m = get_marker(z);
-   if (!SOI(m)) return e("no SOI","Corrupt JPEG");
-   if (scan == SCAN_type) return 1;
-   m = get_marker(z);
-   while (!SOF(m)) {
-      if (!process_marker(z,m)) return 0;
-      m = get_marker(z);
-      while (m == MARKER_none) {
-         // some files have extra padding after their blocks, so ok, we'll scan
-         if (at_eof(z->s)) return e("no SOF", "Corrupt JPEG");
-         m = get_marker(z);
-      }
-   }
-   if (!process_frame_header(z, scan)) return 0;
-   return 1;
-}
-
-static int decode_jpeg_image(jpeg *j)
-{
-   int m;
-   j->restart_interval = 0;
-   if (!decode_jpeg_header(j, SCAN_load)) return 0;
-   m = get_marker(j);
-   while (!EOI(m)) {
-      if (SOS(m)) {
-         if (!process_scan_header(j)) return 0;
-         if (!parse_entropy_coded_data(j)) return 0;
-         if (j->marker == MARKER_none ) {
-            // handle 0s at the end of image data from IP Kamera 9060
-            while (!at_eof(j->s)) {
-               int x = get8(j->s);
-               if (x == 255) {
-                  j->marker = get8u(j->s);
-                  break;
-               } else if (x != 0) {
-                  return 0;
-               }
-            }
-            // if we reach eof without hitting a marker, get_marker() below will fail and we'll eventually return 0
-         }
-      } else {
-         if (!process_marker(j, m)) return 0;
-      }
-      m = get_marker(j);
-   }
-   return 1;
-}
-
-// static jfif-centered resampling (across block boundaries)
-
-typedef uint8 *(*resample_row_func)(uint8 *out, uint8 *in0, uint8 *in1,
-                                    int w, int hs);
-
-#define div4(x) ((uint8) ((x) >> 2))
-
-static uint8 *resample_row_1(uint8 *out, uint8 *in_near, uint8 *in_far, int w, int hs)
-{
-   STBI_NOTUSED(out);
-   STBI_NOTUSED(in_far);
-   STBI_NOTUSED(w);
-   STBI_NOTUSED(hs);
-   return in_near;
-}
-
-static uint8* resample_row_v_2(uint8 *out, uint8 *in_near, uint8 *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] = div4(3*in_near[i] + in_far[i] + 2);
-   return out;
-}
-
-static uint8*  resample_row_h_2(uint8 *out, uint8 *in_near, uint8 *in_far, int w, int hs)
-{
-   // need to generate two samples horizontally for every one in input
-   int i;
-   uint8 *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] = div4(input[0]*3 + input[1] + 2);
-   for (i=1; i < w-1; ++i) {
-      int n = 3*input[i]+2;
-      out[i*2+0] = div4(n+input[i-1]);
-      out[i*2+1] = div4(n+input[i+1]);
-   }
-   out[i*2+0] = 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 div16(x) ((uint8) ((x) >> 4))
-
-static uint8 *resample_row_hv_2(uint8 *out, uint8 *in_near, uint8 *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] = div4(3*in_near[0] + in_far[0] + 2);
-      return out;
-   }
-
-   t1 = 3*in_near[0] + in_far[0];
-   out[0] = div4(t1+2);
-   for (i=1; i < w; ++i) {
-      t0 = t1;
-      t1 = 3*in_near[i]+in_far[i];
-      out[i*2-1] = div16(3*t0 + t1 + 8);
-      out[i*2  ] = div16(3*t1 + t0 + 8);
-   }
-   out[w*2-1] = div4(t1+2);
-
-   STBI_NOTUSED(hs);
-
-   return out;
-}
-
-static uint8 *resample_row_generic(uint8 *out, uint8 *in_near, uint8 *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;
-}
-
-#define float2fixed(x)  ((int) ((x) * 65536 + 0.5))
-
-// 0.38 seconds on 3*anemones.jpg   (0.25 with processor = Pro)
-// VC6 without processor=Pro is generating multiple LEAs per multiply!
-static void YCbCr_to_RGB_row(uint8 *out, const uint8 *y, const uint8 *pcb, const uint8 *pcr, int count, int step)
-{
-   int i;
-   for (i=0; i < count; ++i) {
-      int y_fixed = (y[i] << 16) + 32768; // rounding
-      int r,g,b;
-      int cr = pcr[i] - 128;
-      int cb = pcb[i] - 128;
-      r = y_fixed + cr*float2fixed(1.40200f);
-      g = y_fixed - cr*float2fixed(0.71414f) - cb*float2fixed(0.34414f);
-      b = y_fixed                            + cb*float2fixed(1.77200f);
-      r >>= 16;
-      g >>= 16;
-      b >>= 16;
-      if ((unsigned) r > 255) { if (r < 0) r = 0; else r = 255; }
-      if ((unsigned) g > 255) { if (g < 0) g = 0; else g = 255; }
-      if ((unsigned) b > 255) { if (b < 0) b = 0; else b = 255; }
-      out[0] = (uint8)r;
-      out[1] = (uint8)g;
-      out[2] = (uint8)b;
-      out[3] = 255;
-      out += step;
-   }
-}
-
-#ifdef STBI_SIMD
-static stbi_YCbCr_to_RGB_run stbi_YCbCr_installed = YCbCr_to_RGB_row;
-
-void stbi_install_YCbCr_to_RGB(stbi_YCbCr_to_RGB_run func)
-{
-   stbi_YCbCr_installed = func;
-}
-#endif
-
-
-// clean up the temporary component buffers
-static void cleanup_jpeg(jpeg *j)
-{
-   int i;
-   for (i=0; i < j->s->img_n; ++i) {
-      if (j->img_comp[i].data) {
-         free(j->img_comp[i].raw_data);
-         j->img_comp[i].data = NULL;
-      }
-      if (j->img_comp[i].linebuf) {
-         free(j->img_comp[i].linebuf);
-         j->img_comp[i].linebuf = NULL;
-      }
-   }
-}
-
-typedef struct
-{
-   resample_row_func resample;
-   uint8 *line0,*line1;
-   int hs,vs;   // expansion factor in each axis
-   int w_lores; // horizontal pixels pre-expansion
-   int ystep;   // how far through vertical expansion we are
-   int ypos;    // which pre-expansion row we're on
-} stbi_resample;
-
-static uint8 *load_jpeg_image(jpeg *z, int *out_x, int *out_y, int *comp, int req_comp)
-{
-   int n, decode_n;
-   // validate req_comp
-   if (req_comp < 0 || req_comp > 4) return epuc("bad req_comp", "Internal error");
-   z->s->img_n = 0;
-
-   // load a jpeg image from whichever source
-   if (!decode_jpeg_image(z)) { cleanup_jpeg(z); return NULL; }
-
-   // determine actual number of components to generate
-   n = req_comp ? req_comp : z->s->img_n;
-
-   if (z->s->img_n == 3 && n < 3)
-      decode_n = 1;
-   else
-      decode_n = z->s->img_n;
-
-   // resample and color-convert
-   {
-      int k;
-      uint i,j;
-      uint8 *output;
-      uint8 *coutput[4];
-
-      stbi_resample res_comp[4];
-
-      for (k=0; k < decode_n; ++k) {
-         stbi_resample *r = &res_comp[k];
-
-         // allocate line buffer big enough for upsampling off the edges
-         // with upsample factor of 4
-         z->img_comp[k].linebuf = (uint8 *) malloc(z->s->img_x + 3);
-         if (!z->img_comp[k].linebuf) { cleanup_jpeg(z); return epuc("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 = resample_row_v_2;
-         else if (r->hs == 2 && r->vs == 1) r->resample = resample_row_h_2;
-         else if (r->hs == 2 && r->vs == 2) r->resample = resample_row_hv_2;
-         else                               r->resample = resample_row_generic;
-      }
-
-      // can't error after this so, this is safe
-      output = (uint8 *) malloc(n * z->s->img_x * z->s->img_y + 1);
-      if (!output) { cleanup_jpeg(z); return epuc("outofmem", "Out of memory"); }
-
-      // now go ahead and resample
-      for (j=0; j < z->s->img_y; ++j) {
-         uint8 *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) {
-            uint8 *y = coutput[0];
-            if (z->s->img_n == 3) {
-               #ifdef STBI_SIMD
-               stbi_YCbCr_installed(out, y, coutput[1], coutput[2], z->s.img_x, n);
-               #else
-               YCbCr_to_RGB_row(out, y, coutput[1], coutput[2], z->s->img_x, n);
-               #endif
-            } 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 {
-            uint8 *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;
-         }
-      }
-      cleanup_jpeg(z);
-      *out_x = z->s->img_x;
-      *out_y = z->s->img_y;
-      if (comp) *comp  = z->s->img_n; // report original components, not output
-      return output;
-   }
-}
-
-static unsigned char *stbi_jpeg_load(stbi *s, int *x, int *y, int *comp, int req_comp)
-{
-   jpeg j;
-   j.s = s;
-   return load_jpeg_image(&j, x,y,comp,req_comp);
-}
-
-static int stbi_jpeg_test(stbi *s)
-{
-   int r;
-   jpeg j;
-   j.s = s;
-   r = decode_jpeg_header(&j, SCAN_type);
-   stbi_rewind(s);
-   return r;
-}
-
-static int stbi_jpeg_info_raw(jpeg *j, int *x, int *y, int *comp)
-{
-   if (!decode_jpeg_header(j, SCAN_header)) {
-      stbi_rewind( j->s );
-      return 0;
-   }
-   if (x) *x = j->s->img_x;
-   if (y) *y = j->s->img_y;
-   if (comp) *comp = j->s->img_n;
-   return 1;
-}
-
-static int stbi_jpeg_info(stbi *s, int *x, int *y, int *comp)
-{
-   jpeg j;
-   j.s = s;
-   return stbi_jpeg_info_raw(&j, x, y, comp);
-}
-
-// 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
-
-// fast-way is faster to check than jpeg huffman, but slow way is slower
-#define ZFAST_BITS  9 // accelerate all cases in default tables
-#define ZFAST_MASK  ((1 << ZFAST_BITS) - 1)
-
-// zlib-style huffman encoding
-// (jpegs packs from left, zlib from right, so can't share code)
-typedef struct
-{
-   uint16 fast[1 << ZFAST_BITS];
-   uint16 firstcode[16];
-   int maxcode[17];
-   uint16 firstsymbol[16];
-   uint8  size[288];
-   uint16 value[288];
-} zhuffman;
-
-stbi_inline static int 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 bit_reverse(int v, int bits)
-{
-   assert(bits <= 16);
-   // to bit reverse n bits, reverse 16 and shift
-   // e.g. 11 bits, bit reverse and shift away 5
-   return bitreverse16(v) >> (16-bits);
-}
-
-static int zbuild_huffman(zhuffman *z, uint8 *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, 255, sizeof(z->fast));
-   for (i=0; i < num; ++i)
-      ++sizes[sizelist[i]];
-   sizes[0] = 0;
-   for (i=1; i < 16; ++i)
-      assert(sizes[i] <= (1 << i));
-   code = 0;
-   for (i=1; i < 16; ++i) {
-      next_code[i] = code;
-      z->firstcode[i] = (uint16) code;
-      z->firstsymbol[i] = (uint16) k;
-      code = (code + sizes[i]);
-      if (sizes[i])
-         if (code-1 >= (1 << i)) return e("bad codelengths","Corrupt JPEG");
-      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];
-         z->size[c] = (uint8)s;
-         z->value[c] = (uint16)i;
-         if (s <= ZFAST_BITS) {
-            int k = bit_reverse(next_code[s],s);
-            while (k < (1 << ZFAST_BITS)) {
-               z->fast[k] = (uint16) c;
-               k += (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
-{
-   uint8 *zbuffer, *zbuffer_end;
-   int num_bits;
-   uint32 code_buffer;
-
-   char *zout;
-   char *zout_start;
-   char *zout_end;
-   int   z_expandable;
-
-   zhuffman z_length, z_distance;
-} zbuf;
-
-stbi_inline static int zget8(zbuf *z)
-{
-   if (z->zbuffer >= z->zbuffer_end) return 0;
-   return *z->zbuffer++;
-}
-
-static void fill_bits(zbuf *z)
-{
-   do {
-      assert(z->code_buffer < (1U << z->num_bits));
-      z->code_buffer |= zget8(z) << z->num_bits;
-      z->num_bits += 8;
-   } while (z->num_bits <= 24);
-}
-
-stbi_inline static unsigned int zreceive(zbuf *z, int n)
-{
-   unsigned int k;
-   if (z->num_bits < n) fill_bits(z);
-   k = z->code_buffer & ((1 << n) - 1);
-   z->code_buffer >>= n;
-   z->num_bits -= n;
-   return k;
-}
-
-stbi_inline static int zhuffman_decode(zbuf *a, zhuffman *z)
-{
-   int b,s,k;
-   if (a->num_bits < 16) fill_bits(a);
-   b = z->fast[a->code_buffer & ZFAST_MASK];
-   if (b < 0xffff) {
-      s = z->size[b];
-      a->code_buffer >>= s;
-      a->num_bits -= s;
-      return z->value[b];
-   }
-
-   // not resolved by fast table, so compute it the slow way
-   // use jpeg approach, which requires MSbits at top
-   k = bit_reverse(a->code_buffer, 16);
-   for (s=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];
-   assert(z->size[b] == s);
-   a->code_buffer >>= s;
-   a->num_bits -= s;
-   return z->value[b];
-}
-
-static int expand(zbuf *z, int n)  // need to make room for n bytes
-{
-   char *q;
-   int cur, limit;
-   if (!z->z_expandable) return e("output buffer limit","Corrupt PNG");
-   cur   = (int) (z->zout     - z->zout_start);
-   limit = (int) (z->zout_end - z->zout_start);
-   while (cur + n > limit)
-      limit *= 2;
-   q = (char *) realloc(z->zout_start, limit);
-   if (q == NULL) return e("outofmem", "Out of memory");
-   z->zout_start = q;
-   z->zout       = q + cur;
-   z->zout_end   = q + limit;
-   return 1;
-}
-
-static int length_base[31] = {
-   3,4,5,6,7,8,9,10,11,13,
-   15,17,19,23,27,31,35,43,51,59,
-   67,83,99,115,131,163,195,227,258,0,0 };
-
-static int length_extra[31]=
-{ 0,0,0,0,0,0,0,0,1,1,1,1,2,2,2,2,3,3,3,3,4,4,4,4,5,5,5,5,0,0,0 };
-
-static int dist_base[32] = { 1,2,3,4,5,7,9,13,17,25,33,49,65,97,129,193,
-257,385,513,769,1025,1537,2049,3073,4097,6145,8193,12289,16385,24577,0,0};
-
-static int dist_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 parse_huffman_block(zbuf *a)
-{
-   for(;;) {
-      int z = zhuffman_decode(a, &a->z_length);
-      if (z < 256) {
-         if (z < 0) return e("bad huffman code","Corrupt PNG"); // error in huffman codes
-         if (a->zout >= a->zout_end) if (!expand(a, 1)) return 0;
-         *a->zout++ = (char) z;
-      } else {
-         uint8 *p;
-         int len,dist;
-         if (z == 256) return 1;
-         z -= 257;
-         len = length_base[z];
-         if (length_extra[z]) len += zreceive(a, length_extra[z]);
-         z = zhuffman_decode(a, &a->z_distance);
-         if (z < 0) return e("bad huffman code","Corrupt PNG");
-         dist = dist_base[z];
-         if (dist_extra[z]) dist += zreceive(a, dist_extra[z]);
-         if (a->zout - a->zout_start < dist) return e("bad dist","Corrupt PNG");
-         if (a->zout + len > a->zout_end) if (!expand(a, len)) return 0;
-         p = (uint8 *) (a->zout - dist);
-         while (len--)
-            *a->zout++ = *p++;
-      }
-   }
-}
-
-static int compute_huffman_codes(zbuf *a)
-{
-   static uint8 length_dezigzag[19] = { 16,17,18,0,8,7,9,6,10,5,11,4,12,3,13,2,14,1,15 };
-   zhuffman z_codelength;
-   uint8 lencodes[286+32+137];//padding for maximum single op
-   uint8 codelength_sizes[19];
-   int i,n;
-
-   int hlit  = zreceive(a,5) + 257;
-   int hdist = zreceive(a,5) + 1;
-   int hclen = zreceive(a,4) + 4;
-
-   memset(codelength_sizes, 0, sizeof(codelength_sizes));
-   for (i=0; i < hclen; ++i) {
-      int s = zreceive(a,3);
-      codelength_sizes[length_dezigzag[i]] = (uint8) s;
-   }
-   if (!zbuild_huffman(&z_codelength, codelength_sizes, 19)) return 0;
-
-   n = 0;
-   while (n < hlit + hdist) {
-      int c = zhuffman_decode(a, &z_codelength);
-      assert(c >= 0 && c < 19);
-      if (c < 16)
-         lencodes[n++] = (uint8) c;
-      else if (c == 16) {
-         c = zreceive(a,2)+3;
-         memset(lencodes+n, lencodes[n-1], c);
-         n += c;
-      } else if (c == 17) {
-         c = zreceive(a,3)+3;
-         memset(lencodes+n, 0, c);
-         n += c;
-      } else {
-         assert(c == 18);
-         c = zreceive(a,7)+11;
-         memset(lencodes+n, 0, c);
-         n += c;
-      }
-   }
-   if (n != hlit+hdist) return e("bad codelengths","Corrupt PNG");
-   if (!zbuild_huffman(&a->z_length, lencodes, hlit)) return 0;
-   if (!zbuild_huffman(&a->z_distance, lencodes+hlit, hdist)) return 0;
-   return 1;
-}
-
-static int parse_uncompressed_block(zbuf *a)
-{
-   uint8 header[4];
-   int len,nlen,k;
-   if (a->num_bits & 7)
-      zreceive(a, a->num_bits & 7); // discard
-   // drain the bit-packed data into header
-   k = 0;
-   while (a->num_bits > 0) {
-      header[k++] = (uint8) (a->code_buffer & 255); // wtf this warns?
-      a->code_buffer >>= 8;
-      a->num_bits -= 8;
-   }
-   assert(a->num_bits == 0);
-   // now fill header the normal way
-   while (k < 4)
-      header[k++] = (uint8) zget8(a);
-   len  = header[1] * 256 + header[0];
-   nlen = header[3] * 256 + header[2];
-   if (nlen != (len ^ 0xffff)) return e("zlib corrupt","Corrupt PNG");
-   if (a->zbuffer + len > a->zbuffer_end) return e("read past buffer","Corrupt PNG");
-   if (a->zout + len > a->zout_end)
-      if (!expand(a, len)) return 0;
-   memcpy(a->zout, a->zbuffer, len);
-   a->zbuffer += len;
-   a->zout += len;
-   return 1;
-}
-
-static int parse_zlib_header(zbuf *a)
-{
-   int cmf   = zget8(a);
-   int cm    = cmf & 15;
-   /* int cinfo = cmf >> 4; */
-   int flg   = zget8(a);
-   if ((cmf*256+flg) % 31 != 0) return e("bad zlib header","Corrupt PNG"); // zlib spec
-   if (flg & 32) return e("no preset dict","Corrupt PNG"); // preset dictionary not allowed in png
-   if (cm != 8) return e("bad compression","Corrupt PNG"); // DEFLATE required for png
-   // window = 1 << (8 + cinfo)... but who cares, we fully buffer output
-   return 1;
-}
-
-// @TODO: should statically initialize these for optimal thread safety
-static uint8 default_length[288], default_distance[32];
-static void init_defaults(void)
-{
-   int i;   // use <= to match clearly with spec
-   for (i=0; i <= 143; ++i)     default_length[i]   = 8;
-   for (   ; i <= 255; ++i)     default_length[i]   = 9;
-   for (   ; i <= 279; ++i)     default_length[i]   = 7;
-   for (   ; i <= 287; ++i)     default_length[i]   = 8;
-
-   for (i=0; i <=  31; ++i)     default_distance[i] = 5;
-}
-
-int stbi_png_partial; // a quick hack to only allow decoding some of a PNG... I should implement real streaming support instead
-static int parse_zlib(zbuf *a, int parse_header)
-{
-   int final_, type;
-   if (parse_header)
-      if (!parse_zlib_header(a)) return 0;
-   a->num_bits = 0;
-   a->code_buffer = 0;
-   do {
-      final_ = zreceive(a,1);
-      type = zreceive(a,2);
-      if (type == 0) {
-         if (!parse_uncompressed_block(a)) return 0;
-      } else if (type == 3) {
-         return 0;
-      } else {
-         if (type == 1) {
-            // use fixed code lengths
-            if (!default_distance[31]) init_defaults();
-            if (!zbuild_huffman(&a->z_length  , default_length  , 288)) return 0;
-            if (!zbuild_huffman(&a->z_distance, default_distance,  32)) return 0;
-         } else {
-            if (!compute_huffman_codes(a)) return 0;
-         }
-         if (!parse_huffman_block(a)) return 0;
-      }
-      if (stbi_png_partial && a->zout - a->zout_start > 65536)
-         break;
-   } while (!final_);
-   return 1;
-}
-
-static int do_zlib(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 parse_zlib(a, parse_header);
-}
-
-char *stbi_zlib_decode_malloc_guesssize(const char *buffer, int len, int initial_size, int *outlen)
-{
-   zbuf a;
-   char *p = (char *) malloc(initial_size);
-   if (p == NULL) return NULL;
-   a.zbuffer = (uint8 *) buffer;
-   a.zbuffer_end = (uint8 *) buffer + len;
-   if (do_zlib(&a, p, initial_size, 1, 1)) {
-      if (outlen) *outlen = (int) (a.zout - a.zout_start);
-      return a.zout_start;
-   } else {
-      free(a.zout_start);
-      return NULL;
-   }
-}
-
-char *stbi_zlib_decode_malloc(char const *buffer, int len, int *outlen)
-{
-   return stbi_zlib_decode_malloc_guesssize(buffer, len, 16384, outlen);
-}
-
-char *stbi_zlib_decode_malloc_guesssize_headerflag(const char *buffer, int len, int initial_size, int *outlen, int parse_header)
-{
-   zbuf a;
-   char *p = (char *) malloc(initial_size);
-   if (p == NULL) return NULL;
-   a.zbuffer = (uint8 *) buffer;
-   a.zbuffer_end = (uint8 *) buffer + len;
-   if (do_zlib(&a, p, initial_size, 1, parse_header)) {
-      if (outlen) *outlen = (int) (a.zout - a.zout_start);
-      return a.zout_start;
-   } else {
-      free(a.zout_start);
-      return NULL;
-   }
-}
-
-int stbi_zlib_decode_buffer(char *obuffer, int olen, char const *ibuffer, int ilen)
-{
-   zbuf a;
-   a.zbuffer = (uint8 *) ibuffer;
-   a.zbuffer_end = (uint8 *) ibuffer + ilen;
-   if (do_zlib(&a, obuffer, olen, 0, 1))
-      return (int) (a.zout - a.zout_start);
-   else
-      return -1;
-}
-
-char *stbi_zlib_decode_noheader_malloc(char const *buffer, int len, int *outlen)
-{
-   zbuf a;
-   char *p = (char *) malloc(16384);
-   if (p == NULL) return NULL;
-   a.zbuffer = (uint8 *) buffer;
-   a.zbuffer_end = (uint8 *) buffer+len;
-   if (do_zlib(&a, p, 16384, 1, 0)) {
-      if (outlen) *outlen = (int) (a.zout - a.zout_start);
-      return a.zout_start;
-   } else {
-      free(a.zout_start);
-      return NULL;
-   }
-}
-
-int stbi_zlib_decode_noheader_buffer(char *obuffer, int olen, const char *ibuffer, int ilen)
-{
-   zbuf a;
-   a.zbuffer = (uint8 *) ibuffer;
-   a.zbuffer_end = (uint8 *) ibuffer + ilen;
-   if (do_zlib(&a, obuffer, olen, 0, 0))
-      return (int) (a.zout - a.zout_start);
-   else
-      return -1;
-}
-
-// 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
-
-
-typedef struct
-{
-   uint32 length;
-   uint32 type;
-} chunk;
-
-#define PNG_TYPE(a,b,c,d)  (((a) << 24) + ((b) << 16) + ((c) << 8) + (d))
-
-static chunk get_chunk_header(stbi *s)
-{
-   chunk c;
-   c.length = get32(s);
-   c.type   = get32(s);
-   return c;
-}
-
-static int check_png_header(stbi *s)
-{
-   static uint8 png_sig[8] = { 137,80,78,71,13,10,26,10 };
-   int i;
-   for (i=0; i < 8; ++i)
-      if (get8u(s) != png_sig[i]) return e("bad png sig","Not a PNG");
-   return 1;
-}
-
-typedef struct
-{
-   stbi *s;
-   uint8 *idata, *expanded, *out;
-} png;
-
-
-enum {
-   F_none=0, F_sub=1, F_up=2, F_avg=3, F_paeth=4,
-   F_avg_first, F_paeth_first
-};
-
-static uint8 first_row_filter[5] =
-{
-   F_none, F_sub, F_none, F_avg_first, F_paeth_first
-};
-
-static int 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;
-}
-
-// create the png data from post-deflated data
-static int create_png_image_raw(png *a, uint8 *raw, uint32 raw_len, int out_n, uint32 x, uint32 y)
-{
-   stbi *s = a->s;
-   uint32 i,j,stride = x*out_n;
-   int k;
-   int img_n = s->img_n; // copy it into a local for later
-   assert(out_n == s->img_n || out_n == s->img_n+1);
-   if (stbi_png_partial) y = 1;
-   a->out = (uint8 *) malloc(x * y * out_n);
-   if (!a->out) return e("outofmem", "Out of memory");
-   if (!stbi_png_partial) {
-      if (s->img_x == x && s->img_y == y) {
-         if (raw_len != (img_n * x + 1) * y) return e("not enough pixels","Corrupt PNG");
-      } else { // interlaced:
-         if (raw_len < (img_n * x + 1) * y) return e("not enough pixels","Corrupt PNG");
-      }
-   }
-   for (j=0; j < y; ++j) {
-      uint8 *cur = a->out + stride*j;
-      uint8 *prior = cur - stride;
-      int filter = *raw++;
-      if (filter > 4) return e("invalid filter","Corrupt PNG");
-      // if first row, use special filter that doesn't sample previous row
-      if (j == 0) filter = first_row_filter[filter];
-      // handle first pixel explicitly
-      for (k=0; k < img_n; ++k) {
-         switch (filter) {
-            case F_none       : cur[k] = raw[k]; break;
-            case F_sub        : cur[k] = raw[k]; break;
-            case F_up         : cur[k] = raw[k] + prior[k]; break;
-            case F_avg        : cur[k] = raw[k] + (prior[k]>>1); break;
-            case F_paeth      : cur[k] = (uint8) (raw[k] + paeth(0,prior[k],0)); break;
-            case F_avg_first  : cur[k] = raw[k]; break;
-            case F_paeth_first: cur[k] = raw[k]; break;
-         }
-      }
-      if (img_n != out_n) cur[img_n] = 255;
-      raw += img_n;
-      cur += out_n;
-      prior += out_n;
-      // this is a little gross, so that we don't switch per-pixel or per-component
-      if (img_n == out_n) {
-         #define CASE(f) \
-             case f:     \
-                for (i=x-1; i >= 1; --i, raw+=img_n,cur+=img_n,prior+=img_n) \
-                   for (k=0; k < img_n; ++k)
-         switch (filter) {
-            CASE(F_none)  cur[k] = raw[k]; break;
-            CASE(F_sub)   cur[k] = raw[k] + cur[k-img_n]; break;
-            CASE(F_up)    cur[k] = raw[k] + prior[k]; break;
-            CASE(F_avg)   cur[k] = raw[k] + ((prior[k] + cur[k-img_n])>>1); break;
-            CASE(F_paeth)  cur[k] = (uint8) (raw[k] + paeth(cur[k-img_n],prior[k],prior[k-img_n])); break;
-            CASE(F_avg_first)    cur[k] = raw[k] + (cur[k-img_n] >> 1); break;
-            CASE(F_paeth_first)  cur[k] = (uint8) (raw[k] + paeth(cur[k-img_n],0,0)); break;
-         }
-         #undef CASE
-      } else {
-         assert(img_n+1 == out_n);
-         #define CASE(f) \
-             case f:     \
-                for (i=x-1; i >= 1; --i, cur[img_n]=255,raw+=img_n,cur+=out_n,prior+=out_n) \
-                   for (k=0; k < img_n; ++k)
-         switch (filter) {
-            CASE(F_none)  cur[k] = raw[k]; break;
-            CASE(F_sub)   cur[k] = raw[k] + cur[k-out_n]; break;
-            CASE(F_up)    cur[k] = raw[k] + prior[k]; break;
-            CASE(F_avg)   cur[k] = raw[k] + ((prior[k] + cur[k-out_n])>>1); break;
-            CASE(F_paeth)  cur[k] = (uint8) (raw[k] + paeth(cur[k-out_n],prior[k],prior[k-out_n])); break;
-            CASE(F_avg_first)    cur[k] = raw[k] + (cur[k-out_n] >> 1); break;
-            CASE(F_paeth_first)  cur[k] = (uint8) (raw[k] + paeth(cur[k-out_n],0,0)); break;
-         }
-         #undef CASE
-      }
-   }
-   return 1;
-}
-
-static int create_png_image(png *a, uint8 *raw, uint32 raw_len, int out_n, int interlaced)
-{
-   uint8 *final_;
-   int p;
-   int save;
-   if (!interlaced)
-      return create_png_image_raw(a, raw, raw_len, out_n, a->s->img_x, a->s->img_y);
-   save = stbi_png_partial;
-   stbi_png_partial = 0;
-
-   // de-interlacing
-   final_ = (uint8 *) malloc(a->s->img_x * a->s->img_y * out_n);
-   for (p=0; p < 7; ++p) {
-      int xorig[] = { 0,4,0,2,0,1,0 };
-      int yorig[] = { 0,0,4,0,2,0,1 };
-      int xspc[]  = { 8,8,4,4,2,2,1 };
-      int yspc[]  = { 8,8,8,4,4,2,2 };
-      int i,j,x,y;
-      // pass1_x[4] = 0, pass1_x[5] = 1, pass1_x[12] = 1
-      x = (a->s->img_x - xorig[p] + xspc[p]-1) / xspc[p];
-      y = (a->s->img_y - yorig[p] + yspc[p]-1) / yspc[p];
-      if (x && y) {
-         if (!create_png_image_raw(a, raw, raw_len, out_n, x, y)) {
-            free(final_);
-            return 0;
-         }
-         for (j=0; j < y; ++j)
-            for (i=0; i < x; ++i)
-               memcpy(final_ + (j*yspc[p]+yorig[p])*a->s->img_x*out_n + (i*xspc[p]+xorig[p])*out_n,
-                      a->out + (j*x+i)*out_n, out_n);
-         free(a->out);
-         raw += (x*out_n+1)*y;
-         raw_len -= (x*out_n+1)*y;
-      }
-   }
-   a->out = final_;
-
-   stbi_png_partial = save;
-   return 1;
-}
-
-static int compute_transparency(png *z, uint8 tc[3], int out_n)
-{
-   stbi *s = z->s;
-   uint32 i, pixel_count = s->img_x * s->img_y;
-   uint8 *p = z->out;
-
-   // compute color-based transparency, assuming we've
-   // already got 255 as the alpha value in the output
-   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 expand_palette(png *a, uint8 *palette, int len, int pal_img_n)
-{
-   uint32 i, pixel_count = a->s->img_x * a->s->img_y;
-   uint8 *p, *temp_out, *orig = a->out;
-
-   p = (uint8 *) malloc(pixel_count * pal_img_n);
-   if (p == NULL) return e("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;
-      }
-   }
-   free(a->out);
-   a->out = temp_out;
-
-   STBI_NOTUSED(len);
-
-   return 1;
-}
-
-static int stbi_unpremultiply_on_load = 0;
-static int stbi_de_iphone_flag = 0;
-
-void stbi_set_unpremultiply_on_load(int flag_true_if_should_unpremultiply)
-{
-   stbi_unpremultiply_on_load = flag_true_if_should_unpremultiply;
-}
-void stbi_convert_iphone_png_to_rgb(int flag_true_if_should_convert)
-{
-   stbi_de_iphone_flag = flag_true_if_should_convert;
-}
-
-static void stbi_de_iphone(png *z)
-{
-   stbi *s = z->s;
-   uint32 i, pixel_count = s->img_x * s->img_y;
-   uint8 *p = z->out;
-
-   if (s->img_out_n == 3) {  // convert bgr to rgb
-      for (i=0; i < pixel_count; ++i) {
-         uint8 t = p[0];
-         p[0] = p[2];
-         p[2] = t;
-         p += 3;
-      }
-   } else {
-      assert(s->img_out_n == 4);
-      if (stbi_unpremultiply_on_load) {
-         // convert bgr to rgb and unpremultiply
-         for (i=0; i < pixel_count; ++i) {
-            uint8 a = p[3];
-            uint8 t = p[0];
-            if (a) {
-               p[0] = p[2] * 255 / a;
-               p[1] = p[1] * 255 / a;
-               p[2] =  t   * 255 / a;
-            } else {
-               p[0] = p[2];
-               p[2] = t;
-            }
-            p += 4;
-         }
-      } else {
-         // convert bgr to rgb
-         for (i=0; i < pixel_count; ++i) {
-            uint8 t = p[0];
-            p[0] = p[2];
-            p[2] = t;
-            p += 4;
-         }
-      }
-   }
-}
-
-static int parse_png_file(png *z, int scan, int req_comp)
-{
-   uint8 palette[1024], pal_img_n=0;
-   uint8 has_trans=0, tc[3];
-   uint32 ioff=0, idata_limit=0, i, pal_len=0;
-   int first=1,k,interlace=0, iphone=0;
-   stbi *s = z->s;
-
-   z->expanded = NULL;
-   z->idata = NULL;
-   z->out = NULL;
-
-   if (!check_png_header(s)) return 0;
-
-   if (scan == SCAN_type) return 1;
-
-   for (;;) {
-      chunk c = get_chunk_header(s);
-      switch (c.type) {
-         case PNG_TYPE('C','g','B','I'):
-            iphone = stbi_de_iphone_flag;
-            skip(s, c.length);
-            break;
-         case PNG_TYPE('I','H','D','R'): {
-            int depth,color,comp,filter;
-            if (!first) return e("multiple IHDR","Corrupt PNG");
-            first = 0;
-            if (c.length != 13) return e("bad IHDR len","Corrupt PNG");
-            s->img_x = get32(s); if (s->img_x > (1 << 24)) return e("too large","Very large image (corrupt?)");
-            s->img_y = get32(s); if (s->img_y > (1 << 24)) return e("too large","Very large image (corrupt?)");
-            depth = get8(s);  if (depth != 8)        return e("8bit only","PNG not supported: 8-bit only");
-            color = get8(s);  if (color > 6)         return e("bad ctype","Corrupt PNG");
-            if (color == 3) pal_img_n = 3; else if (color & 1) return e("bad ctype","Corrupt PNG");
-            comp  = get8(s);  if (comp) return e("bad comp method","Corrupt PNG");
-            filter= get8(s);  if (filter) return e("bad filter method","Corrupt PNG");
-            interlace = get8(s); if (interlace>1) return e("bad interlace method","Corrupt PNG");
-            if (!s->img_x || !s->img_y) return e("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 e("too large", "Image too large to decode");
-               if (scan == SCAN_header) return 1;
-            } else {
-               // if paletted, then pal_n is our final components, and
-               // img_n is # components to decompress/filter.
-               s->img_n = 1;
-               if ((1 << 30) / s->img_x / 4 < s->img_y) return e("too large","Corrupt PNG");
-               // if SCAN_header, have to scan to see if we have a tRNS
-            }
-            break;
-         }
-
-         case PNG_TYPE('P','L','T','E'):  {
-            if (first) return e("first not IHDR", "Corrupt PNG");
-            if (c.length > 256*3) return e("invalid PLTE","Corrupt PNG");
-            pal_len = c.length / 3;
-            if (pal_len * 3 != c.length) return e("invalid PLTE","Corrupt PNG");
-            for (i=0; i < pal_len; ++i) {
-               palette[i*4+0] = get8u(s);
-               palette[i*4+1] = get8u(s);
-               palette[i*4+2] = get8u(s);
-               palette[i*4+3] = 255;
-            }
-            break;
-         }
-
-         case PNG_TYPE('t','R','N','S'): {
-            if (first) return e("first not IHDR", "Corrupt PNG");
-            if (z->idata) return e("tRNS after IDAT","Corrupt PNG");
-            if (pal_img_n) {
-               if (scan == SCAN_header) { s->img_n = 4; return 1; }
-               if (pal_len == 0) return e("tRNS before PLTE","Corrupt PNG");
-               if (c.length > pal_len) return e("bad tRNS len","Corrupt PNG");
-               pal_img_n = 4;
-               for (i=0; i < c.length; ++i)
-                  palette[i*4+3] = get8u(s);
-            } else {
-               if (!(s->img_n & 1)) return e("tRNS with alpha","Corrupt PNG");
-               if (c.length != (uint32) s->img_n*2) return e("bad tRNS len","Corrupt PNG");
-               has_trans = 1;
-               for (k=0; k < s->img_n; ++k)
-                  tc[k] = (uint8) get16(s); // non 8-bit images will be larger
-            }
-            break;
-         }
-
-         case PNG_TYPE('I','D','A','T'): {
-            if (first) return e("first not IHDR", "Corrupt PNG");
-            if (pal_img_n && !pal_len) return e("no PLTE","Corrupt PNG");
-            if (scan == SCAN_header) { s->img_n = pal_img_n; return 1; }
-            if (ioff + c.length > idata_limit) {
-               uint8 *p;
-               if (idata_limit == 0) idata_limit = c.length > 4096 ? c.length : 4096;
-               while (ioff + c.length > idata_limit)
-                  idata_limit *= 2;
-               p = (uint8 *) realloc(z->idata, idata_limit); if (p == NULL) return e("outofmem", "Out of memory");
-               z->idata = p;
-            }
-            if (!getn(s, z->idata+ioff,c.length)) return e("outofdata","Corrupt PNG");
-            ioff += c.length;
-            break;
-         }
-
-         case PNG_TYPE('I','E','N','D'): {
-            uint32 raw_len;
-            if (first) return e("first not IHDR", "Corrupt PNG");
-            if (scan != SCAN_load) return 1;
-            if (z->idata == NULL) return e("no IDAT","Corrupt PNG");
-            z->expanded = (uint8 *) stbi_zlib_decode_malloc_guesssize_headerflag((char *) z->idata, ioff, 16384, (int *) &raw_len, !iphone);
-            if (z->expanded == NULL) return 0; // zlib should set error
-            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 (!create_png_image(z, z->expanded, raw_len, s->img_out_n, interlace)) return 0;
-            if (has_trans)
-               if (!compute_transparency(z, tc, s->img_out_n)) return 0;
-            if (iphone && 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 (!expand_palette(z, palette, pal_len, s->img_out_n))
-                  return 0;
-            }
-            free(z->expanded); z->expanded = NULL;
-            return 1;
-         }
-
-         default:
-            // if critical, fail
-            if (first) return e("first not IHDR", "Corrupt PNG");
-            if ((c.type & (1 << 29)) == 0) {
-               #ifndef STBI_NO_FAILURE_STRINGS
-               // not threadsafe
-               static char invalid_chunk[] = "XXXX chunk not known";
-               invalid_chunk[0] = (uint8) (c.type >> 24);
-               invalid_chunk[1] = (uint8) (c.type >> 16);
-               invalid_chunk[2] = (uint8) (c.type >>  8);
-               invalid_chunk[3] = (uint8) (c.type >>  0);
-               #endif
-               return e(invalid_chunk, "PNG not supported: unknown chunk type");
-            }
-            skip(s, c.length);
-            break;
-      }
-      // end of chunk, read and skip CRC
-      get32(s);
-   }
-}
-
-static unsigned char *do_png(png *p, int *x, int *y, int *n, int req_comp)
-{
-   unsigned char *result=NULL;
-   if (req_comp < 0 || req_comp > 4) return epuc("bad req_comp", "Internal error");
-   if (parse_png_file(p, SCAN_load, req_comp)) {
-      result = p->out;
-      p->out = NULL;
-      if (req_comp && req_comp != p->s->img_out_n) {
-         result = convert_format(result, p->s->img_out_n, req_comp, p->s->img_x, p->s->img_y);
-         p->s->img_out_n = req_comp;
-         if (result == NULL) return result;
-      }
-      *x = p->s->img_x;
-      *y = p->s->img_y;
-      if (n) *n = p->s->img_n;
-   }
-   free(p->out);      p->out      = NULL;
-   free(p->expanded); p->expanded = NULL;
-   free(p->idata);    p->idata    = NULL;
-
-   return result;
-}
-
-static unsigned char *stbi_png_load(stbi *s, int *x, int *y, int *comp, int req_comp)
-{
-   png p;
-   p.s = s;
-   return do_png(&p, x,y,comp,req_comp);
-}
-
-static int stbi_png_test(stbi *s)
-{
-   int r;
-   r = check_png_header(s);
-   stbi_rewind(s);
-   return r;
-}
-
-static int stbi_png_info_raw(png *p, int *x, int *y, int *comp)
-{
-   if (!parse_png_file(p, 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 *s, int *x, int *y, int *comp)
-{
-   png p;
-   p.s = s;
-   return stbi_png_info_raw(&p, x, y, comp);
-}
-
-// Microsoft/Windows BMP image
-
-static int bmp_test(stbi *s)
-{
-   int sz;
-   if (get8(s) != 'B') return 0;
-   if (get8(s) != 'M') return 0;
-   get32le(s); // discard filesize
-   get16le(s); // discard reserved
-   get16le(s); // discard reserved
-   get32le(s); // discard data offset
-   sz = get32le(s);
-   if (sz == 12 || sz == 40 || sz == 56 || sz == 108) return 1;
-   return 0;
-}
-
-static int stbi_bmp_test(stbi *s)
-{
-   int r = bmp_test(s);
-   stbi_rewind(s);
-   return r;
-}
-
-
-// returns 0..31 for the highest set bit
-static int high_bit(unsigned int z)
-{
-   int n=0;
-   if (z == 0) return -1;
-   if (z >= 0x10000) n += 16, z >>= 16;
-   if (z >= 0x00100) n +=  8, z >>=  8;
-   if (z >= 0x00010) n +=  4, z >>=  4;
-   if (z >= 0x00004) n +=  2, z >>=  2;
-   if (z >= 0x00002) n +=  1, z >>=  1;
-   return n;
-}
-
-static int bitcount(unsigned int a)
-{
-   a = (a & 0x55555555) + ((a >>  1) & 0x55555555); // max 2
-   a = (a & 0x33333333) + ((a >>  2) & 0x33333333); // max 4
-   a = (a + (a >> 4)) & 0x0f0f0f0f; // max 8 per 4, now 8 bits
-   a = (a + (a >> 8)); // max 16 per 8 bits
-   a = (a + (a >> 16)); // max 32 per 8 bits
-   return a & 0xff;
-}
-
-static int shiftsigned(int v, int shift, int bits)
-{
-   int result;
-   int z=0;
-
-   if (shift < 0) v <<= -shift;
-   else v >>= shift;
-   result = v;
-
-   z = bits;
-   while (z < 8) {
-      result += v >> z;
-      z += bits;
-   }
-   return result;
-}
-
-static stbi_uc *bmp_load(stbi *s, int *x, int *y, int *comp, int req_comp)
-{
-   uint8 *out;
-   unsigned int mr=0,mg=0,mb=0,ma=0, fake_a=0;
-   stbi_uc pal[256][4];
-   int psize=0,i,j,compress=0,width;
-   int bpp, flip_vertically, pad, target, offset, hsz;
-   if (get8(s) != 'B' || get8(s) != 'M') return epuc("not BMP", "Corrupt BMP");
-   get32le(s); // discard filesize
-   get16le(s); // discard reserved
-   get16le(s); // discard reserved
-   offset = get32le(s);
-   hsz = get32le(s);
-   if (hsz != 12 && hsz != 40 && hsz != 56 && hsz != 108) return epuc("unknown BMP", "BMP type not supported: unknown");
-   if (hsz == 12) {
-      s->img_x = get16le(s);
-      s->img_y = get16le(s);
-   } else {
-      s->img_x = get32le(s);
-      s->img_y = get32le(s);
-   }
-   if (get16le(s) != 1) return epuc("bad BMP", "bad BMP");
-   bpp = get16le(s);
-   if (bpp == 1) return epuc("monochrome", "BMP type not supported: 1-bit");
-   flip_vertically = ((int) s->img_y) > 0;
-   s->img_y = abs((int) s->img_y);
-   if (hsz == 12) {
-      if (bpp < 24)
-         psize = (offset - 14 - 24) / 3;
-   } else {
-      compress = get32le(s);
-      if (compress == 1 || compress == 2) return epuc("BMP RLE", "BMP type not supported: RLE");
-      get32le(s); // discard sizeof
-      get32le(s); // discard hres
-      get32le(s); // discard vres
-      get32le(s); // discard colorsused
-      get32le(s); // discard max important
-      if (hsz == 40 || hsz == 56) {
-         if (hsz == 56) {
-            get32le(s);
-            get32le(s);
-            get32le(s);
-            get32le(s);
-         }
-         if (bpp == 16 || bpp == 32) {
-            mr = mg = mb = 0;
-            if (compress == 0) {
-               if (bpp == 32) {
-                  mr = 0xffu << 16;
-                  mg = 0xffu <<  8;
-                  mb = 0xffu <<  0;
-                  ma = 0xffu << 24;
-                  fake_a = 1; // @TODO: check for cases like alpha value is all 0 and switch it to 255
-               } else {
-                  mr = 31u << 10;
-                  mg = 31u <<  5;
-                  mb = 31u <<  0;
-               }
-            } else if (compress == 3) {
-               mr = get32le(s);
-               mg = get32le(s);
-               mb = get32le(s);
-               // not documented, but generated by photoshop and handled by mspaint
-               if (mr == mg && mg == mb) {
-                  // ?!?!?
-                  return epuc("bad BMP", "bad BMP");
-               }
-            } else
-               return epuc("bad BMP", "bad BMP");
-         }
-      } else {
-         assert(hsz == 108);
-         mr = get32le(s);
-         mg = get32le(s);
-         mb = get32le(s);
-         ma = get32le(s);
-         get32le(s); // discard color space
-         for (i=0; i < 12; ++i)
-            get32le(s); // discard color space parameters
-      }
-      if (bpp < 16)
-         psize = (offset - 14 - hsz) >> 2;
-   }
-   s->img_n = ma ? 4 : 3;
-   if (req_comp && req_comp >= 3) // we can directly decode 3 or 4
-      target = req_comp;
-   else
-      target = s->img_n; // if they want monochrome, we'll post-convert
-   out = (stbi_uc *) malloc(target * s->img_x * s->img_y);
-   if (!out) return epuc("outofmem", "Out of memory");
-   if (bpp < 16) {
-      int z=0;
-      if (psize == 0 || psize > 256) { free(out); return epuc("invalid", "Corrupt BMP"); }
-      for (i=0; i < psize; ++i) {
-         pal[i][2] = get8u(s);
-         pal[i][1] = get8u(s);
-         pal[i][0] = get8u(s);
-         if (hsz != 12) get8(s);
-         pal[i][3] = 255;
-      }
-      skip(s, offset - 14 - hsz - psize * (hsz == 12 ? 3 : 4));
-      if (bpp == 4) width = (s->img_x + 1) >> 1;
-      else if (bpp == 8) width = s->img_x;
-      else { free(out); return epuc("bad bpp", "Corrupt BMP"); }
-      pad = (-width)&3;
-      for (j=0; j < (int) s->img_y; ++j) {
-         for (i=0; i < (int) s->img_x; i += 2) {
-            int v=get8(s),v2=0;
-            if (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 = (bpp == 8) ? get8(s) : v2;
-            out[z++] = pal[v][0];
-            out[z++] = pal[v][1];
-            out[z++] = pal[v][2];
-            if (target == 4) out[z++] = 255;
-         }
-         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;
-      skip(s, offset - 14 - hsz);
-      if (bpp == 24) width = 3 * s->img_x;
-      else if (bpp == 16) width = 2*s->img_x;
-      else /* bpp = 32 and pad = 0 */ width=0;
-      pad = (-width) & 3;
-      if (bpp == 24) {
-         easy = 1;
-      } else if (bpp == 32) {
-         if (mb == 0xff && mg == 0xff00 && mr == 0x00ff0000 && ma == 0xff000000)
-            easy = 2;
-      }
-      if (!easy) {
-         if (!mr || !mg || !mb) { free(out); return epuc("bad masks", "Corrupt BMP"); }
-         // right shift amt to put high bit in position #7
-         rshift = high_bit(mr)-7; rcount = bitcount(mr);
-         gshift = high_bit(mg)-7; gcount = bitcount(mr);
-         bshift = high_bit(mb)-7; bcount = bitcount(mr);
-         ashift = high_bit(ma)-7; acount = bitcount(mr);
-      }
-      for (j=0; j < (int) s->img_y; ++j) {
-         if (easy) {
-            for (i=0; i < (int) s->img_x; ++i) {
-               int a;
-               out[z+2] = get8u(s);
-               out[z+1] = get8u(s);
-               out[z+0] = get8u(s);
-               z += 3;
-               a = (easy == 2 ? get8(s) : 255);
-               if (target == 4) out[z++] = (uint8) a;
-            }
-         } else {
-            for (i=0; i < (int) s->img_x; ++i) {
-               uint32 v = (bpp == 16 ? get16le(s) : get32le(s));
-               int a;
-               out[z++] = (uint8) shiftsigned(v & mr, rshift, rcount);
-               out[z++] = (uint8) shiftsigned(v & mg, gshift, gcount);
-               out[z++] = (uint8) shiftsigned(v & mb, bshift, bcount);
-               a = (ma ? shiftsigned(v & ma, ashift, acount) : 255);
-               if (target == 4) out[z++] = (uint8) a;
-            }
-         }
-         skip(s, pad);
-      }
-   }
-   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 = convert_format(out, target, req_comp, s->img_x, s->img_y);
-      if (out == NULL) return out; // convert_format frees input on failure
-   }
-
-   *x = s->img_x;
-   *y = s->img_y;
-   if (comp) *comp = s->img_n;
-   return out;
-
-   NVG_NOTUSED(fake_a);
-}
-
-static stbi_uc *stbi_bmp_load(stbi *s,int *x, int *y, int *comp, int req_comp)
-{
-   return bmp_load(s, x,y,comp,req_comp);
-}
-
-
-// Targa Truevision - TGA
-// by Jonathan Dummer
-
-static int tga_info(stbi *s, int *x, int *y, int *comp)
-{
-    int tga_w, tga_h, tga_comp;
-    int sz;
-    get8u(s);                   // discard Offset
-    sz = get8u(s);              // color type
-    if( sz > 1 ) {
-        stbi_rewind(s);
-        return 0;      // only RGB or indexed allowed
-    }
-    sz = get8u(s);              // image type
-    // only RGB or grey allowed, +/- RLE
-    if ((sz != 1) && (sz != 2) && (sz != 3) && (sz != 9) && (sz != 10) && (sz != 11)) return 0;
-    skip(s,9);
-    tga_w = get16le(s);
-    if( tga_w < 1 ) {
-        stbi_rewind(s);
-        return 0;   // test width
-    }
-    tga_h = get16le(s);
-    if( tga_h < 1 ) {
-        stbi_rewind(s);
-        return 0;   // test height
-    }
-    sz = get8(s);               // bits per pixel
-    // only RGB or RGBA or grey allowed
-    if ((sz != 8) && (sz != 16) && (sz != 24) && (sz != 32)) {
-        stbi_rewind(s);
-        return 0;
-    }
-    tga_comp = sz;
-    if (x) *x = tga_w;
-    if (y) *y = tga_h;
-    if (comp) *comp = tga_comp / 8;
-    return 1;                   // seems to have passed everything
-}
-
-int stbi_tga_info(stbi *s, int *x, int *y, int *comp)
-{
-    return tga_info(s, x, y, comp);
-}
-
-static int tga_test(stbi *s)
-{
-   int sz;
-   get8u(s);      //   discard Offset
-   sz = get8u(s);   //   color type
-   if ( sz > 1 ) return 0;   //   only RGB or indexed allowed
-   sz = get8u(s);   //   image type
-   if ( (sz != 1) && (sz != 2) && (sz != 3) && (sz != 9) && (sz != 10) && (sz != 11) ) return 0;   //   only RGB or grey allowed, +/- RLE
-   get16(s);      //   discard palette start
-   get16(s);      //   discard palette length
-   get8(s);         //   discard bits per palette color entry
-   get16(s);      //   discard x origin
-   get16(s);      //   discard y origin
-   if ( get16(s) < 1 ) return 0;      //   test width
-   if ( get16(s) < 1 ) return 0;      //   test height
-   sz = get8(s);   //   bits per pixel
-   if ( (sz != 8) && (sz != 16) && (sz != 24) && (sz != 32) ) return 0;   //   only RGB or RGBA or grey allowed
-   return 1;      //   seems to have passed everything
-}
-
-static int stbi_tga_test(stbi *s)
-{
-   int res = tga_test(s);
-   stbi_rewind(s);
-   return res;
-}
-
-static stbi_uc *tga_load(stbi *s, int *x, int *y, int *comp, int req_comp)
-{
-   //   read in the TGA header stuff
-   int tga_offset = get8u(s);
-   int tga_indexed = get8u(s);
-   int tga_image_type = get8u(s);
-   int tga_is_RLE = 0;
-   int tga_palette_start = get16le(s);
-   int tga_palette_len = get16le(s);
-   int tga_palette_bits = get8u(s);
-   int tga_x_origin = get16le(s);
-   int tga_y_origin = get16le(s);
-   int tga_width = get16le(s);
-   int tga_height = get16le(s);
-   int tga_bits_per_pixel = get8u(s);
-   int tga_inverted = get8u(s);
-   //   image data
-   unsigned char *tga_data;
-   unsigned char *tga_palette = NULL;
-   int i, j;
-   unsigned char raw_data[4];
-   unsigned char trans_data[4];
-   int RLE_count = 0;
-   int RLE_repeating = 0;
-   int read_next_pixel = 1;
-
-   //   do a tiny bit of precessing
-   if ( tga_image_type >= 8 )
-   {
-      tga_image_type -= 8;
-      tga_is_RLE = 1;
-   }
-   /* int tga_alpha_bits = tga_inverted & 15; */
-   tga_inverted = 1 - ((tga_inverted >> 5) & 1);
-
-   //   error check
-   if ( //(tga_indexed) ||
-      (tga_width < 1) || (tga_height < 1) ||
-      (tga_image_type < 1) || (tga_image_type > 3) ||
-      ((tga_bits_per_pixel != 8) && (tga_bits_per_pixel != 16) &&
-      (tga_bits_per_pixel != 24) && (tga_bits_per_pixel != 32))
-      )
-   {
-      return NULL; // we don't report this as a bad TGA because we don't even know if it's TGA
-   }
-
-   //   If I'm paletted, then I'll use the number of bits from the palette
-   if ( tga_indexed )
-   {
-      tga_bits_per_pixel = tga_palette_bits;
-   }
-
-   //   tga info
-   *x = tga_width;
-   *y = tga_height;
-   if ( (req_comp < 1) || (req_comp > 4) )
-   {
-      //   just use whatever the file was
-      req_comp = tga_bits_per_pixel / 8;
-      *comp = req_comp;
-   } else
-   {
-      //   force a new number of components
-      *comp = tga_bits_per_pixel/8;
-   }
-   tga_data = (unsigned char*)malloc( tga_width * tga_height * req_comp );
-   if (!tga_data) return epuc("outofmem", "Out of memory");
-
-   //   skip to the data's starting position (offset usually = 0)
-   skip(s, tga_offset );
-   //   do I need to load a palette?
-   if ( tga_indexed )
-   {
-      //   any data to skip? (offset usually = 0)
-      skip(s, tga_palette_start );
-      //   load the palette
-      tga_palette = (unsigned char*)malloc( tga_palette_len * tga_palette_bits / 8 );
-      if (!tga_palette) return epuc("outofmem", "Out of memory");
-      if (!getn(s, tga_palette, tga_palette_len * tga_palette_bits / 8 )) {
-         free(tga_data);
-         free(tga_palette);
-         return epuc("bad palette", "Corrupt TGA");
-      }
-   }
-   //   load the data
-   trans_data[0] = trans_data[1] = trans_data[2] = trans_data[3] = 0;
-   for (i=0; i < tga_width * tga_height; ++i)
-   {
-      //   if I'm in RLE mode, do I need to get a RLE chunk?
-      if ( tga_is_RLE )
-      {
-         if ( RLE_count == 0 )
-         {
-            //   yep, get the next byte as a RLE command
-            int RLE_cmd = get8u(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 1 byte, then perform the lookup
-            int pal_idx = get8u(s);
-            if ( pal_idx >= tga_palette_len )
-            {
-               //   invalid index
-               pal_idx = 0;
-            }
-            pal_idx *= tga_bits_per_pixel / 8;
-            for (j = 0; j*8 < tga_bits_per_pixel; ++j)
-            {
-               raw_data[j] = tga_palette[pal_idx+j];
-            }
-         } else
-         {
-            //   read in the data raw
-            for (j = 0; j*8 < tga_bits_per_pixel; ++j)
-            {
-               raw_data[j] = get8u(s);
-            }
-         }
-         //   convert raw to the intermediate format
-         switch (tga_bits_per_pixel)
-         {
-         case 8:
-            //   Luminous => RGBA
-            trans_data[0] = raw_data[0];
-            trans_data[1] = raw_data[0];
-            trans_data[2] = raw_data[0];
-            trans_data[3] = 255;
-            break;
-         case 16:
-            //   Luminous,Alpha => RGBA
-            trans_data[0] = raw_data[0];
-            trans_data[1] = raw_data[0];
-            trans_data[2] = raw_data[0];
-            trans_data[3] = raw_data[1];
-            break;
-         case 24:
-            //   BGR => RGBA
-            trans_data[0] = raw_data[2];
-            trans_data[1] = raw_data[1];
-            trans_data[2] = raw_data[0];
-            trans_data[3] = 255;
-            break;
-         case 32:
-            //   BGRA => RGBA
-            trans_data[0] = raw_data[2];
-            trans_data[1] = raw_data[1];
-            trans_data[2] = raw_data[0];
-            trans_data[3] = raw_data[3];
-            break;
-         }
-         //   clear the reading flag for the next pixel
-         read_next_pixel = 0;
-      } // end of reading a pixel
-      //   convert to final format
-      switch (req_comp)
-      {
-      case 1:
-         //   RGBA => Luminance
-         tga_data[i*req_comp+0] = compute_y(trans_data[0],trans_data[1],trans_data[2]);
-         break;
-      case 2:
-         //   RGBA => Luminance,Alpha
-         tga_data[i*req_comp+0] = compute_y(trans_data[0],trans_data[1],trans_data[2]);
-         tga_data[i*req_comp+1] = trans_data[3];
-         break;
-      case 3:
-         //   RGBA => RGB
-         tga_data[i*req_comp+0] = trans_data[0];
-         tga_data[i*req_comp+1] = trans_data[1];
-         tga_data[i*req_comp+2] = trans_data[2];
-         break;
-      case 4:
-         //   RGBA => RGBA
-         tga_data[i*req_comp+0] = trans_data[0];
-         tga_data[i*req_comp+1] = trans_data[1];
-         tga_data[i*req_comp+2] = trans_data[2];
-         tga_data[i*req_comp+3] = trans_data[3];
-         break;
-      }
-      //   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 * req_comp;
-         int index2 = (tga_height - 1 - j) * tga_width * req_comp;
-         for (i = tga_width * req_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 )
-   {
-      free( tga_palette );
-   }
-   //   the things I do to get rid of an error message, and yet keep
-   //   Microsoft's C compilers happy... [8^(
-   tga_palette_start = tga_palette_len = tga_palette_bits =
-         tga_x_origin = tga_y_origin = 0;
-   //   OK, done
-   return tga_data;
-}
-
-static stbi_uc *stbi_tga_load(stbi *s, int *x, int *y, int *comp, int req_comp)
-{
-   return tga_load(s,x,y,comp,req_comp);
-}
-
-
-// *************************************************************************************************
-// Photoshop PSD loader -- PD by Thatcher Ulrich, integration by Nicolas Schulz, tweaked by STB
-
-static int psd_test(stbi *s)
-{
-   if (get32(s) != 0x38425053) return 0;   // "8BPS"
-   else return 1;
-}
-
-static int stbi_psd_test(stbi *s)
-{
-   int r = psd_test(s);
-   stbi_rewind(s);
-   return r;
-}
-
-static stbi_uc *psd_load(stbi *s, int *x, int *y, int *comp, int req_comp)
-{
-   int   pixelCount;
-   int channelCount, compression;
-   int channel, i, count, len;
-   int w,h;
-   uint8 *out;
-
-   // Check identifier
-   if (get32(s) != 0x38425053)   // "8BPS"
-      return epuc("not PSD", "Corrupt PSD image");
-
-   // Check file type version.
-   if (get16(s) != 1)
-      return epuc("wrong version", "Unsupported version of PSD image");
-
-   // Skip 6 reserved bytes.
-   skip(s, 6 );
-
-   // Read the number of channels (R, G, B, A, etc).
-   channelCount = get16(s);
-   if (channelCount < 0 || channelCount > 16)
-      return epuc("wrong channel count", "Unsupported number of channels in PSD image");
-
-   // Read the rows and columns of the image.
-   h = get32(s);
-   w = get32(s);
-
-   // Make sure the depth is 8 bits.
-   if (get16(s) != 8)
-      return epuc("unsupported bit depth", "PSD bit depth is not 8 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 (get16(s) != 3)
-      return epuc("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.)
-   skip(s,get32(s) );
-
-   // Skip the image resources.  (resolution, pen tool paths, etc)
-   skip(s, get32(s) );
-
-   // Skip the reserved data.
-   skip(s, get32(s) );
-
-   // Find out if the data is compressed.
-   // Known values:
-   //   0: no compression
-   //   1: RLE compressed
-   compression = get16(s);
-   if (compression > 1)
-      return epuc("bad compression", "PSD has an unknown compression format");
-
-   // Create the destination image.
-   out = (stbi_uc *) malloc(4 * w*h);
-   if (!out) return epuc("outofmem", "Out of memory");
-   pixelCount = w*h;
-
-   // Initialize the data to zero.
-   //memset( out, 0, pixelCount * 4 );
-
-   // Finally, the image data.
-   if (compression) {
-      // RLE as used by .PSD and .TIFF
-      // Loop until you get the number of unpacked bytes you are expecting:
-      //     Read the next source byte into n.
-      //     If n is between 0 and 127 inclusive, copy the next n+1 bytes literally.
-      //     Else if n is between -127 and -1 inclusive, copy the next byte -n+1 times.
-      //     Else if n is 128, noop.
-      // Endloop
-
-      // The RLE-compressed data is preceeded by a 2-byte data count for each row in the data,
-      // which we're going to just skip.
-      skip(s, h * channelCount * 2 );
-
-      // Read the RLE data by channel.
-      for (channel = 0; channel < 4; channel++) {
-         uint8 *p;
-
-         p = out+channel;
-         if (channel >= channelCount) {
-            // Fill this channel with default data.
-            for (i = 0; i < pixelCount; i++) *p = (channel == 3 ? 255 : 0), p += 4;
-         } else {
-            // Read the RLE data.
-            count = 0;
-            while (count < pixelCount) {
-               len = get8(s);
-               if (len == 128) {
-                  // No-op.
-               } else if (len < 128) {
-                  // Copy next len+1 bytes literally.
-                  len++;
-                  count += len;
-                  while (len) {
-                     *p = get8u(s);
-                     p += 4;
-                     len--;
-                  }
-               } else if (len > 128) {
-                  uint8   val;
-                  // Next -len+1 bytes in the dest are replicated from next source byte.
-                  // (Interpret len as a negative 8-bit int.)
-                  len ^= 0x0FF;
-                  len += 2;
-                  val = get8u(s);
-                  count += len;
-                  while (len) {
-                     *p = val;
-                     p += 4;
-                     len--;
-                  }
-               }
-            }
-         }
-      }
-
-   } else {
-      // We're at the raw image data.  It's each channel in order (Red, Green, Blue, Alpha, ...)
-      // where each channel consists of an 8-bit value for each pixel in the image.
-
-      // Read the data by channel.
-      for (channel = 0; channel < 4; channel++) {
-         uint8 *p;
-
-         p = out + channel;
-         if (channel > channelCount) {
-            // Fill this channel with default data.
-            for (i = 0; i < pixelCount; i++) *p = channel == 3 ? 255 : 0, p += 4;
-         } else {
-            // Read the data.
-            for (i = 0; i < pixelCount; i++)
-               *p = get8u(s), p += 4;
-         }
-      }
-   }
-
-   if (req_comp && req_comp != 4) {
-      out = convert_format(out, 4, req_comp, w, h);
-      if (out == NULL) return out; // convert_format frees input on failure
-   }
-
-   if (comp) *comp = channelCount;
-   *y = h;
-   *x = w;
-
-   return out;
-}
-
-static stbi_uc *stbi_psd_load(stbi *s, int *x, int *y, int *comp, int req_comp)
-{
-   return psd_load(s,x,y,comp,req_comp);
-}
-
-// *************************************************************************************************
-// 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/
-
-static int pic_is4(stbi *s,const char *str)
-{
-   int i;
-   for (i=0; i<4; ++i)
-      if (get8(s) != (stbi_uc)str[i])
-         return 0;
-
-   return 1;
-}
-
-static int pic_test(stbi *s)
-{
-   int i;
-
-   if (!pic_is4(s,"\x53\x80\xF6\x34"))
-      return 0;
-
-   for(i=0;i<84;++i)
-      get8(s);
-
-   if (!pic_is4(s,"PICT"))
-      return 0;
-
-   return 1;
-}
-
-typedef struct
-{
-   stbi_uc size,type,channel;
-} pic_packet_t;
-
-static stbi_uc *pic_readval(stbi *s, int channel, stbi_uc *dest)
-{
-   int mask=0x80, i;
-
-   for (i=0; i<4; ++i, mask>>=1) {
-      if (channel & mask) {
-         if (at_eof(s)) return epuc("bad file","PIC file too short");
-         dest[i]=get8u(s);
-      }
-   }
-
-   return dest;
-}
-
-static void pic_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 *pic_load2(stbi *s,int width,int height,int *comp, stbi_uc *result)
-{
-   int act_comp=0,num_packets=0,y,chained;
-   pic_packet_t packets[10];
-
-   // this will (should...) cater for even some bizarre stuff like having data
-    // for the same channel in multiple packets.
-   do {
-      pic_packet_t *packet;
-
-      if (num_packets==sizeof(packets)/sizeof(packets[0]))
-         return epuc("bad format","too many packets");
-
-      packet = &packets[num_packets++];
-
-      chained = get8(s);
-      packet->size    = get8u(s);
-      packet->type    = get8u(s);
-      packet->channel = get8u(s);
-
-      act_comp |= packet->channel;
-
-      if (at_eof(s))          return epuc("bad file","file too short (reading packets)");
-      if (packet->size != 8)  return epuc("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) {
-         pic_packet_t *packet = &packets[packet_idx];
-         stbi_uc *dest = result+y*width*4;
-
-         switch (packet->type) {
-            default:
-               return epuc("bad format","packet has bad compression type");
-
-            case 0: {//uncompressed
-               int x;
-
-               for(x=0;x<width;++x, dest+=4)
-                  if (!pic_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=get8u(s);
-                     if (at_eof(s))   return epuc("bad file","file too short (pure read count)");
-
-                     if (count > left)
-                        count = (uint8) left;
-
-                     if (!pic_readval(s,packet->channel,value))  return 0;
-
-                     for(i=0; i<count; ++i,dest+=4)
-                        pic_copyval(packet->channel,dest,value);
-                     left -= count;
-                  }
-               }
-               break;
-
-            case 2: {//Mixed RLE
-               int left=width;
-               while (left>0) {
-                  int count = get8(s), i;
-                  if (at_eof(s))  return epuc("bad file","file too short (mixed read count)");
-
-                  if (count >= 128) { // Repeated
-                     stbi_uc value[4];
-                     int i;
-
-                     if (count==128)
-                        count = get16(s);
-                     else
-                        count -= 127;
-                     if (count > left)
-                        return epuc("bad file","scanline overrun");
-
-                     if (!pic_readval(s,packet->channel,value))
-                        return 0;
-
-                     for(i=0;i<count;++i, dest += 4)
-                        pic_copyval(packet->channel,dest,value);
-                  } else { // Raw
-                     ++count;
-                     if (count>left) return epuc("bad file","scanline overrun");
-
-                     for(i=0;i<count;++i, dest+=4)
-                        if (!pic_readval(s,packet->channel,dest))
-                           return 0;
-                  }
-                  left-=count;
-               }
-               break;
-            }
-         }
-      }
-   }
-
-   return result;
-}
-
-static stbi_uc *pic_load(stbi *s,int *px,int *py,int *comp,int req_comp)
-{
-   stbi_uc *result;
-   int i, x,y;
-
-   for (i=0; i<92; ++i)
-      get8(s);
-
-   x = get16(s);
-   y = get16(s);
-   if (at_eof(s))  return epuc("bad file","file too short (pic header)");
-   if ((1 << 28) / x < y) return epuc("too large", "Image too large to decode");
-
-   get32(s); //skip `ratio'
-   get16(s); //skip `fields'
-   get16(s); //skip `pad'
-
-   // intermediate buffer is RGBA
-   result = (stbi_uc *) malloc(x*y*4);
-   memset(result, 0xff, x*y*4);
-
-   if (!pic_load2(s,x,y,comp, result)) {
-      free(result);
-      result=0;
-   }
-   *px = x;
-   *py = y;
-   if (req_comp == 0) req_comp = *comp;
-   result=convert_format(result,4,req_comp,x,y);
-
-   return result;
-}
-
-static int stbi_pic_test(stbi *s)
-{
-   int r = pic_test(s);
-   stbi_rewind(s);
-   return r;
-}
-
-static stbi_uc *stbi_pic_load(stbi *s, int *x, int *y, int *comp, int req_comp)
-{
-   return pic_load(s,x,y,comp,req_comp);
-}
-
-// *************************************************************************************************
-// GIF loader -- public domain by Jean-Marc Lienher -- simplified/shrunk by stb
-typedef struct stbi_gif_lzw_struct {
-   int16 prefix;
-   uint8 first;
-   uint8 suffix;
-} stbi_gif_lzw;
-
-typedef struct stbi_gif_struct
-{
-   int w,h;
-   stbi_uc *out;                 // output buffer (always 4 components)
-   int flags, bgindex, ratio, transparent, eflags;
-   uint8  pal[256][4];
-   uint8 lpal[256][4];
-   stbi_gif_lzw codes[4096];
-   uint8 *color_table;
-   int parse, step;
-   int lflags;
-   int start_x, start_y;
-   int max_x, max_y;
-   int cur_x, cur_y;
-   int line_size;
-} stbi_gif;
-
-static int gif_test(stbi *s)
-{
-   int sz;
-   if (get8(s) != 'G' || get8(s) != 'I' || get8(s) != 'F' || get8(s) != '8') return 0;
-   sz = get8(s);
-   if (sz != '9' && sz != '7') return 0;
-   if (get8(s) != 'a') return 0;
-   return 1;
-}
-
-static int stbi_gif_test(stbi *s)
-{
-   int r = gif_test(s);
-   stbi_rewind(s);
-   return r;
-}
-
-static void stbi_gif_parse_colortable(stbi *s, uint8 pal[256][4], int num_entries, int transp)
-{
-   int i;
-   for (i=0; i < num_entries; ++i) {
-      pal[i][2] = get8u(s);
-      pal[i][1] = get8u(s);
-      pal[i][0] = get8u(s);
-      pal[i][3] = transp ? 0 : 255;
-   }
-}
-
-static int stbi_gif_header(stbi *s, stbi_gif *g, int *comp, int is_info)
-{
-   uint8 version;
-   if (get8(s) != 'G' || get8(s) != 'I' || get8(s) != 'F' || get8(s) != '8')
-      return e("not GIF", "Corrupt GIF");
-
-   version = get8u(s);
-   if (version != '7' && version != '9')    return e("not GIF", "Corrupt GIF");
-   if (get8(s) != 'a')                      return e("not GIF", "Corrupt GIF");
-
-   failure_reason = "";
-   g->w = get16le(s);
-   g->h = get16le(s);
-   g->flags = get8(s);
-   g->bgindex = get8(s);
-   g->ratio = get8(s);
-   g->transparent = -1;
-
-   if (comp != 0) *comp = 4;  // can't actually tell whether it's 3 or 4 until we parse the comments
-
-   if (is_info) return 1;
-
-   if (g->flags & 0x80)
-      stbi_gif_parse_colortable(s,g->pal, 2 << (g->flags & 7), -1);
-
-   return 1;
-}
-
-static int stbi_gif_info_raw(stbi *s, int *x, int *y, int *comp)
-{
-   stbi_gif g;
-   if (!stbi_gif_header(s, &g, comp, 1)) {
-      stbi_rewind( s );
-      return 0;
-   }
-   if (x) *x = g.w;
-   if (y) *y = g.h;
-   return 1;
-}
-
-static void stbi_out_gif_code(stbi_gif *g, uint16 code)
-{
-   uint8 *p, *c;
-
-   // recurse to decode the prefixes, since the linked-list is backwards,
-   // and working backwards through an interleaved image would be nasty
-   if (g->codes[code].prefix >= 0)
-      stbi_out_gif_code(g, g->codes[code].prefix);
-
-   if (g->cur_y >= g->max_y) return;
-
-   p = &g->out[g->cur_x + g->cur_y];
-   c = &g->color_table[g->codes[code].suffix * 4];
-
-   if (c[3] >= 128) {
-      p[0] = c[2];
-      p[1] = c[1];
-      p[2] = c[0];
-      p[3] = c[3];
-   }
-   g->cur_x += 4;
-
-   if (g->cur_x >= g->max_x) {
-      g->cur_x = g->start_x;
-      g->cur_y += g->step;
-
-      while (g->cur_y >= g->max_y && g->parse > 0) {
-         g->step = (1 << g->parse) * g->line_size;
-         g->cur_y = g->start_y + (g->step >> 1);
-         --g->parse;
-      }
-   }
-}
-
-static uint8 *stbi_process_gif_raster(stbi *s, stbi_gif *g)
-{
-   uint8 lzw_cs;
-   int32 len, code;
-   uint32 first;
-   int32 codesize, codemask, avail, oldcode, bits, valid_bits, clear;
-   stbi_gif_lzw *p;
-
-   lzw_cs = get8u(s);
-   clear = 1 << lzw_cs;
-   first = 1;
-   codesize = lzw_cs + 1;
-   codemask = (1 << codesize) - 1;
-   bits = 0;
-   valid_bits = 0;
-   for (code = 0; code < clear; code++) {
-      g->codes[code].prefix = -1;
-      g->codes[code].first = (uint8) code;
-      g->codes[code].suffix = (uint8) code;
-   }
-
-   // support no starting clear code
-   avail = clear+2;
-   oldcode = -1;
-
-   len = 0;
-   for(;;) {
-      if (valid_bits < codesize) {
-         if (len == 0) {
-            len = get8(s); // start new block
-            if (len == 0)
-               return g->out;
-         }
-         --len;
-         bits |= (int32) get8(s) << valid_bits;
-         valid_bits += 8;
-      } else {
-         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
-            skip(s, len);
-            while ((len = get8(s)) > 0)
-               skip(s,len);
-            return g->out;
-         } else if (code <= avail) {
-            if (first) return epuc("no clear code", "Corrupt GIF");
-
-            if (oldcode >= 0) {
-               p = &g->codes[avail++];
-               if (avail > 4096)        return epuc("too many codes", "Corrupt GIF");
-               p->prefix = (int16) oldcode;
-               p->first = g->codes[oldcode].first;
-               p->suffix = (code == avail) ? p->first : g->codes[code].first;
-            } else if (code == avail)
-               return epuc("illegal code in raster", "Corrupt GIF");
-
-            stbi_out_gif_code(g, (uint16) code);
-
-            if ((avail & codemask) == 0 && avail <= 0x0FFF) {
-               codesize++;
-               codemask = (1 << codesize) - 1;
-            }
-
-            oldcode = code;
-         } else {
-            return epuc("illegal code in raster", "Corrupt GIF");
-         }
-      }
-   }
-}
-
-static void stbi_fill_gif_background(stbi_gif *g)
-{
-   int i;
-   uint8 *c = g->pal[g->bgindex];
-   // @OPTIMIZE: write a dword at a time
-   for (i = 0; i < g->w * g->h * 4; i += 4) {
-      uint8 *p  = &g->out[i];
-      p[0] = c[2];
-      p[1] = c[1];
-      p[2] = c[0];
-      p[3] = c[3];
-   }
-}
-
-// this function is designed to support animated gifs, although stb_image doesn't support it
-static uint8 *stbi_gif_load_next(stbi *s, stbi_gif *g, int *comp, int req_comp)
-{
-   int i;
-   uint8 *old_out = 0;
-
-   if (g->out == 0) {
-      if (!stbi_gif_header(s, g, comp,0))     return 0; // failure_reason set by stbi_gif_header
-      g->out = (uint8 *) malloc(4 * g->w * g->h);
-      if (g->out == 0)                      return epuc("outofmem", "Out of memory");
-      stbi_fill_gif_background(g);
-   } else {
-      // animated-gif-only path
-      if (((g->eflags & 0x1C) >> 2) == 3) {
-         old_out = g->out;
-         g->out = (uint8 *) malloc(4 * g->w * g->h);
-         if (g->out == 0)                   return epuc("outofmem", "Out of memory");
-         memcpy(g->out, old_out, g->w*g->h*4);
-      }
-   }
-
-   for (;;) {
-      switch (get8(s)) {
-         case 0x2C: /* Image Descriptor */
-         {
-            int32 x, y, w, h;
-            uint8 *o;
-
-            x = get16le(s);
-            y = get16le(s);
-            w = get16le(s);
-            h = get16le(s);
-            if (((x + w) > (g->w)) || ((y + h) > (g->h)))
-               return epuc("bad Image Descriptor", "Corrupt GIF");
-
-            g->line_size = g->w * 4;
-            g->start_x = x * 4;
-            g->start_y = y * g->line_size;
-            g->max_x   = g->start_x + w * 4;
-            g->max_y   = g->start_y + h * g->line_size;
-            g->cur_x   = g->start_x;
-            g->cur_y   = g->start_y;
-
-            g->lflags = 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 = (uint8 *) g->lpal;
-            } else if (g->flags & 0x80) {
-               for (i=0; i < 256; ++i)  // @OPTIMIZE: reset only the previous transparent
-                  g->pal[i][3] = 255;
-               if (g->transparent >= 0 && (g->eflags & 0x01))
-                  g->pal[g->transparent][3] = 0;
-               g->color_table = (uint8 *) g->pal;
-            } else
-               return epuc("missing color table", "Corrupt GIF");
-
-            o = stbi_process_gif_raster(s, g);
-            if (o == NULL) return NULL;
-
-            if (req_comp && req_comp != 4)
-               o = convert_format(o, 4, req_comp, g->w, g->h);
-            return o;
-         }
-
-         case 0x21: // Comment Extension.
-         {
-            int len;
-            if (get8(s) == 0xF9) { // Graphic Control Extension.
-               len = get8(s);
-               if (len == 4) {
-                  g->eflags = get8(s);
-                  get16le(s); // delay
-                  g->transparent = get8(s);
-               } else {
-                  skip(s, len);
-                  break;
-               }
-            }
-            while ((len = get8(s)) != 0)
-               skip(s, len);
-            break;
-         }
-
-         case 0x3B: // gif stream termination code
-            return (uint8 *) 1;
-
-         default:
-            return epuc("unknown code", "Corrupt GIF");
-      }
-   }
-}
-
-static stbi_uc *stbi_gif_load(stbi *s, int *x, int *y, int *comp, int req_comp)
-{
-   uint8 *u = 0;
-   stbi_gif g;
-   memset(&g, 0, sizeof(stbi_gif));
-
-   u = stbi_gif_load_next(s, &g, comp, req_comp);
-   if (u == (void *) 1) u = 0;  // end of animated gif marker
-   if (u) {
-      *x = g.w;
-      *y = g.h;
-   }
-
-   return u;
-}
-
-static int stbi_gif_info(stbi *s, int *x, int *y, int *comp)
-{
-   return stbi_gif_info_raw(s,x,y,comp);
-}
-
-
-// *************************************************************************************************
-// Radiance RGBE HDR loader
-// originally by Nicolas Schulz
-#ifndef STBI_NO_HDR
-static int hdr_test(stbi *s)
-{
-   const char *signature = "#?RADIANCE\n";
-   int i;
-   for (i=0; signature[i]; ++i)
-      if (get8(s) != signature[i])
-         return 0;
-   return 1;
-}
-
-static int stbi_hdr_test(stbi* s)
-{
-   int r = hdr_test(s);
-   stbi_rewind(s);
-   return r;
-}
-
-#define HDR_BUFLEN  1024
-static char *hdr_gettoken(stbi *z, char *buffer)
-{
-   int len=0;
-   char c = '\0';
-
-   c = (char) get8(z);
-
-   while (!at_eof(z) && c != '\n') {
-      buffer[len++] = c;
-      if (len == HDR_BUFLEN-1) {
-         // flush to end of line
-         while (!at_eof(z) && get8(z) != '\n')
-            ;
-         break;
-      }
-      c = (char) get8(z);
-   }
-
-   buffer[len] = 0;
-   return buffer;
-}
-
-static void 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 *hdr_load(stbi *s, int *x, int *y, int *comp, int req_comp)
-{
-   char buffer[HDR_BUFLEN];
-   char *token;
-   int valid = 0;
-   int width, height;
-   stbi_uc *scanline;
-   float *hdr_data;
-   int len;
-   unsigned char count, value;
-   int i, j, k, c1,c2, z;
-
-
-   // Check identifier
-   if (strcmp(hdr_gettoken(s,buffer), "#?RADIANCE") != 0)
-      return epf("not HDR", "Corrupt HDR image");
-
-   // Parse header
-   for(;;) {
-      token = hdr_gettoken(s,buffer);
-      if (token[0] == 0) break;
-      if (strcmp(token, "FORMAT=32-bit_rle_rgbe") == 0) valid = 1;
-   }
-
-   if (!valid)    return epf("unsupported format", "Unsupported HDR format");
-
-   // Parse width and height
-   // can't use sscanf() if we're not using stdio!
-   token = hdr_gettoken(s,buffer);
-   if (strncmp(token, "-Y ", 3))  return epf("unsupported data layout", "Unsupported HDR format");
-   token += 3;
-   height = strtol(token, &token, 10);
-   while (*token == ' ') ++token;
-   if (strncmp(token, "+X ", 3))  return epf("unsupported data layout", "Unsupported HDR format");
-   token += 3;
-   width = strtol(token, NULL, 10);
-
-   *x = width;
-   *y = height;
-
-   *comp = 3;
-   if (req_comp == 0) req_comp = 3;
-
-   // Read data
-   hdr_data = (float *) malloc(height * width * req_comp * sizeof(float));
-
-   // Load image data
-   // image data is stored as some number of sca
-   if ( width < 8 || width >= 32768) {
-      // Read flat data
-      for (j=0; j < height; ++j) {
-         for (i=0; i < width; ++i) {
-            stbi_uc rgbe[4];
-           main_decode_loop:
-            getn(s, rgbe, 4);
-            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 = get8(s);
-         c2 = get8(s);
-         len = 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)
-            uint8 rgbe[4];
-            rgbe[0] = (uint8) c1;
-            rgbe[1] = (uint8) c2;
-            rgbe[2] = (uint8) len;
-            rgbe[3] = (uint8) get8u(s);
-            hdr_convert(hdr_data, rgbe, req_comp);
-            i = 1;
-            j = 0;
-            free(scanline);
-            goto main_decode_loop; // yes, this makes no sense
-         }
-         len <<= 8;
-         len |= get8(s);
-         if (len != width) { free(hdr_data); free(scanline); return epf("invalid decoded scanline length", "corrupt HDR"); }
-         if (scanline == NULL) scanline = (stbi_uc *) malloc(width * 4);
-
-         for (k = 0; k < 4; ++k) {
-            i = 0;
-            while (i < width) {
-               count = get8u(s);
-               if (count > 128) {
-                  // Run
-                  value = get8u(s);
-                  count -= 128;
-                  for (z = 0; z < count; ++z)
-                     scanline[i++ * 4 + k] = value;
-               } else {
-                  // Dump
-                  for (z = 0; z < count; ++z)
-                     scanline[i++ * 4 + k] = get8u(s);
-               }
-            }
-         }
-         for (i=0; i < width; ++i)
-            hdr_convert(hdr_data+(j*width + i)*req_comp, scanline + i*4, req_comp);
-      }
-      free(scanline);
-   }
-
-   return hdr_data;
-}
-
-static float *stbi_hdr_load(stbi *s, int *x, int *y, int *comp, int req_comp)
-{
-   return hdr_load(s,x,y,comp,req_comp);
-}
-
-static int stbi_hdr_info(stbi *s, int *x, int *y, int *comp)
-{
-   char buffer[HDR_BUFLEN];
-   char *token;
-   int valid = 0;
-
-   if (strcmp(hdr_gettoken(s,buffer), "#?RADIANCE") != 0) {
-       stbi_rewind( s );
-       return 0;
-   }
-
-   for(;;) {
-      token = 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 = hdr_gettoken(s,buffer);
-   if (strncmp(token, "-Y ", 3)) {
-       stbi_rewind( s );
-       return 0;
-   }
-   token += 3;
-   *y = strtol(token, &token, 10);
-   while (*token == ' ') ++token;
-   if (strncmp(token, "+X ", 3)) {
-       stbi_rewind( s );
-       return 0;
-   }
-   token += 3;
-   *x = strtol(token, NULL, 10);
-   *comp = 3;
-   return 1;
-}
-#endif // STBI_NO_HDR
-
-static int stbi_bmp_info(stbi *s, int *x, int *y, int *comp)
-{
-   int hsz;
-   if (get8(s) != 'B' || get8(s) != 'M') {
-       stbi_rewind( s );
-       return 0;
-   }
-   skip(s,12);
-   hsz = get32le(s);
-   if (hsz != 12 && hsz != 40 && hsz != 56 && hsz != 108) {
-       stbi_rewind( s );
-       return 0;
-   }
-   if (hsz == 12) {
-      *x = get16le(s);
-      *y = get16le(s);
-   } else {
-      *x = get32le(s);
-      *y = get32le(s);
-   }
-   if (get16le(s) != 1) {
-       stbi_rewind( s );
-       return 0;
-   }
-   *comp = get16le(s) / 8;
-   return 1;
-}
-
-static int stbi_psd_info(stbi *s, int *x, int *y, int *comp)
-{
-   int channelCount;
-   if (get32(s) != 0x38425053) {
-       stbi_rewind( s );
-       return 0;
-   }
-   if (get16(s) != 1) {
-       stbi_rewind( s );
-       return 0;
-   }
-   skip(s, 6);
-   channelCount = get16(s);
-   if (channelCount < 0 || channelCount > 16) {
-       stbi_rewind( s );
-       return 0;
-   }
-   *y = get32(s);
-   *x = get32(s);
-   if (get16(s) != 8) {
-       stbi_rewind( s );
-       return 0;
-   }
-   if (get16(s) != 3) {
-       stbi_rewind( s );
-       return 0;
-   }
-   *comp = 4;
-   return 1;
-}
-
-static int stbi_pic_info(stbi *s, int *x, int *y, int *comp)
-{
-   int act_comp=0,num_packets=0,chained;
-   pic_packet_t packets[10];
-
-   skip(s, 92);
-
-   *x = get16(s);
-   *y = get16(s);
-   if (at_eof(s))  return 0;
-   if ( (*x) != 0 && (1 << 28) / (*x) < (*y)) {
-       stbi_rewind( s );
-       return 0;
-   }
-
-   skip(s, 8);
-
-   do {
-      pic_packet_t *packet;
-
-      if (num_packets==sizeof(packets)/sizeof(packets[0]))
-         return 0;
-
-      packet = &packets[num_packets++];
-      chained = get8(s);
-      packet->size    = get8u(s);
-      packet->type    = get8u(s);
-      packet->channel = get8u(s);
-      act_comp |= packet->channel;
-
-      if (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;
-}
-
-static int stbi_info_main(stbi *s, int *x, int *y, int *comp)
-{
-   if (stbi_jpeg_info(s, x, y, comp))
-       return 1;
-   if (stbi_png_info(s, x, y, comp))
-       return 1;
-   if (stbi_gif_info(s, x, y, comp))
-       return 1;
-   if (stbi_bmp_info(s, x, y, comp))
-       return 1;
-   if (stbi_psd_info(s, x, y, comp))
-       return 1;
-   if (stbi_pic_info(s, x, y, comp))
-       return 1;
-   #ifndef STBI_NO_HDR
-   if (stbi_hdr_info(s, x, y, comp))
-       return 1;
-   #endif
-   // test tga last because it's a crappy test!
-   if (stbi_tga_info(s, x, y, comp))
-       return 1;
-   return e("unknown image type", "Image not of any known type, or corrupt");
-}
-
-#ifndef STBI_NO_STDIO
-int stbi_info(char const *filename, int *x, int *y, int *comp)
-{
-    FILE *f = fopen(filename, "rb");
-    int result;
-    if (!f) return e("can't fopen", "Unable to open file");
-    result = stbi_info_from_file(f, x, y, comp);
-    fclose(f);
-    return result;
-}
-
-int stbi_info_from_file(FILE *f, int *x, int *y, int *comp)
-{
-   int r;
-   stbi s;
-   long pos = ftell(f);
-   start_file(&s, f);
-   r = stbi_info_main(&s,x,y,comp);
-   fseek(f,pos,SEEK_SET);
-   return r;
-}
-#endif // !STBI_NO_STDIO
-
-int stbi_info_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp)
-{
-   stbi s;
-   start_mem(&s,buffer,len);
-   return stbi_info_main(&s,x,y,comp);
-}
-
-int stbi_info_from_callbacks(stbi_io_callbacks const *c, void *user, int *x, int *y, int *comp)
-{
-   stbi s;
-   start_callbacks(&s, (stbi_io_callbacks *) c, user);
-   return stbi_info_main(&s,x,y,comp);
-}
-
-#endif // STBI_HEADER_FILE_ONLY
-
-/*
-   revision history:
-      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-uint8 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.e. Janez (U+017D)emva)
-      1.21   fix use of 'uint8' 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 - 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   first released version
-*/
+/* stb_image - v1.39 - public domain JPEG/PNG reader - http://nothings.org/stb_image.c
+   when you control the images you're loading
+                                     no warranty implied; use at your own risk
+
+   Do this:
+      #define STB_IMAGE_IMPLEMENTATION
+   before you include this file in *one* C or C++ file to create the implementation.
+
+   QUICK NOTES:
+      Primarily of interest to game developers and other people who can
+          avoid problematic images and only need the trivial interface
+
+      JPEG baseline (no JPEG progressive)
+      PNG 8-bit-per-channel only
+
+      TGA (not sure what subset, if a subset)
+      BMP non-1bpp, non-RLE
+      PSD (composited view only, no extra channels)
+
+      GIF (*comp always reports as 4-channel)
+      HDR (radiance rgbE format)
+      PIC (Softimage PIC)
+
+      - stbi__jpeg_huff_decode from memory or through FILE (define STBI_NO_STDIO to remove code)
+      - stbi__jpeg_huff_decode from arbitrary I/O callbacks
+      - overridable dequantizing-IDCT, YCbCr-to-RGB conversion (define STBI_SIMD)
+
+   Latest revisions:
+      1.39 (2014-06-15) TGA optimization fix, multiple BMP fixes
+      1.38 (2014-06-06) suppress MSVC run-time warnings, fix accidental rename of 'skip'
+      1.37 (2014-06-04) remove duplicate typedef
+      1.36 (2014-06-03) converted to header file, allow reading incorrect iphoned-images without iphone flag
+      1.35 (2014-05-27) warnings, bugfixes, TGA optimization, etc
+      1.34 (unknown   ) warning fix
+      1.33 (2011-07-14) minor fixes suggested by Dave Moore
+
+   See end of file for full revision history.
+
+   TODO:
+      stbi_info support for BMP,PSD,HDR,PIC
+
+
+ ============================    Contributors    =========================
+              
+ Image formats                                Bug fixes & warning fixes
+    Sean Barrett (jpeg, png, bmp)                Marc LeBlanc
+    Nicolas Schulz (hdr, psd)                    Christpher Lloyd
+    Jonathan Dummer (tga)                        Dave Moore
+    Jean-Marc Lienher (gif)                      Won Chun
+    Tom Seddon (pic)                             the Horde3D community
+    Thatcher Ulrich (psd)                        Janez Zemva
+                                                 Jonathan Blow
+                                                 Laurent Gomila
+ Extensions, features                            Aruelien Pocheville
+    Jetro Lauha (stbi_info)                      Ryamond Barbiero
+    James "moose2000" Brown (iPhone PNG)         David Woo
+    Ben "Disch" Wenger (io callbacks)            Roy Eltham
+    Martin "SpartanJ" Golini                     Luke Graham
+                                                 Thomas Ruf
+                                                 John Bartholomew
+ Optimizations & bugfixes                        Ken Hamada
+    Fabian "ryg" Giesen                          Cort Stratton
+    Arseny Kapoulkine                            Blazej Dariusz Roszkowski
+                                                 Thibault Reuille
+ If your name should be here but                 Paul Du Bois
+ isn't, let Sean know.                           Guillaume George
+                                                 Jerry Jansson
+                                                 Hayaki Saito
+*/
+
+#ifndef STBI_INCLUDE_STB_IMAGE_H
+#define STBI_INCLUDE_STB_IMAGE_H
+
+// Limitations:
+//    - no jpeg progressive support
+//    - non-HDR formats support 8-bit samples only (jpeg, png)
+//    - no delayed line count (jpeg) -- IJG doesn't support either
+//    - no 1-bit BMP
+//    - GIF always returns *comp=4
+//
+// Basic usage (see HDR discussion below):
+//    int x,y,n;
+//    unsigned char *data = stbi_load(filename, &x, &y, &n, 0);
+//    // ... process data if not NULL ... 
+//    // ... x = width, y = height, n = # 8-bit components per pixel ...
+//    // ... replace '0' with '1'..'4' to force that many components per pixel
+//    // ... but 'n' will always be the number that it would have been if you said 0
+//    stbi_image_free(data)
+//
+// Standard parameters:
+//    int *x       -- outputs image width in pixels
+//    int *y       -- outputs image height in pixels
+//    int *comp    -- outputs # of image components in image file
+//    int req_comp -- if non-zero, # of image components requested in result
+//
+// The return value from an image loader is an 'unsigned char *' which points
+// to the pixel data. The pixel data consists of *y scanlines of *x pixels,
+// with each pixel consisting of N interleaved 8-bit components; the first
+// pixel pointed to is top-left-most in the image. There is no padding between
+// image scanlines or between pixels, regardless of format. The number of
+// components N is 'req_comp' if req_comp is non-zero, or *comp otherwise.
+// If req_comp is non-zero, *comp has the number of components that _would_
+// have been output otherwise. E.g. if you set req_comp to 4, you will always
+// get RGBA output, but you can check *comp to easily see if it's opaque.
+//
+// An output image with N components has the following components interleaved
+// in this order in each pixel:
+//
+//     N=#comp     components
+//       1           grey
+//       2           grey, alpha
+//       3           red, green, blue
+//       4           red, green, blue, alpha
+//
+// If image loading fails for any reason, the return value will be NULL,
+// and *x, *y, *comp will be unchanged. The function stbi_failure_reason()
+// can be queried for an extremely brief, end-user unfriendly explanation
+// of why the load failed. Define STBI_NO_FAILURE_STRINGS to avoid
+// compiling these strings at all, and STBI_FAILURE_USERMSG to get slightly
+// more user-friendly ones.
+//
+// Paletted PNG, BMP, GIF, and PIC images are automatically depalettized.
+//
+// ===========================================================================
+//
+// iPhone PNG support:
+//
+// By default we convert iphone-formatted PNGs back to RGB; nominally they
+// would silently load as BGR, except the existing code should have just
+// failed on such iPhone PNGs. But you can disable this conversion by
+// by calling stbi_convert_iphone_png_to_rgb(0), in which case
+// you will always just get the native iphone "format" through.
+//
+// Call stbi_set_unpremultiply_on_load(1) as well to force a divide per
+// pixel to remove any premultiplied alpha *only* if the image file explicitly
+// says there's premultiplied data (currently only happens in iPhone images,
+// and only if iPhone convert-to-rgb processing is on).
+//
+// ===========================================================================
+//
+// HDR image support   (disable by defining STBI_NO_HDR)
+//
+// stb_image now supports loading HDR images in general, and currently
+// the Radiance .HDR file format, although the support is provided
+// generically. You can still load any file through the existing interface;
+// if you attempt to load an HDR file, it will be automatically remapped to
+// LDR, assuming gamma 2.2 and an arbitrary scale factor defaulting to 1;
+// both of these constants can be reconfigured through this interface:
+//
+//     stbi_hdr_to_ldr_gamma(2.2f);
+//     stbi_hdr_to_ldr_scale(1.0f);
+//
+// (note, do not use _inverse_ constants; stbi_image will invert them
+// appropriately).
+//
+// Additionally, there is a new, parallel interface for loading files as
+// (linear) floats to preserve the full dynamic range:
+//
+//    float *data = stbi_loadf(filename, &x, &y, &n, 0);
+// 
+// If you load LDR images through this interface, those images will
+// be promoted to floating point values, run through the inverse of
+// constants corresponding to the above:
+//
+//     stbi_ldr_to_hdr_scale(1.0f);
+//     stbi_ldr_to_hdr_gamma(2.2f);
+//
+// Finally, given a filename (or an open file or memory block--see header
+// file for details) containing image data, you can query for the "most
+// appropriate" interface to use (that is, whether the image is HDR or
+// not), using:
+//
+//     stbi_is_hdr(char *filename);
+//
+// ===========================================================================
+//
+// 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).
+
+
+#ifndef STBI_NO_STDIO
+
+#if defined(_MSC_VER) && _MSC_VER >= 1400
+#define _CRT_SECURE_NO_WARNINGS // suppress warnings about fopen()
+#pragma warning(push)
+#pragma warning(disable:4996)   // suppress even more warnings about fopen()
+#endif
+#include <stdio.h>
+#endif // STBI_NO_STDIO
+
+#define STBI_VERSION 1
+
+enum
+{
+   STBI_default = 0, // only used for req_comp
+
+   STBI_grey       = 1,
+   STBI_grey_alpha = 2,
+   STBI_rgb        = 3,
+   STBI_rgb_alpha  = 4
+};
+
+typedef unsigned char stbi_uc;
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#ifdef STB_IMAGE_STATIC
+#define STBIDEF static
+#else
+#define STBIDEF extern
+#endif
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// PRIMARY API - works on images of any type
+//
+
+//
+// load image by filename, open file, or memory buffer
+//
+
+STBIDEF stbi_uc *stbi_load_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp);
+
+#ifndef STBI_NO_STDIO
+STBIDEF stbi_uc *stbi_load            (char const *filename,     int *x, int *y, int *comp, int req_comp);
+STBIDEF stbi_uc *stbi_load_from_file  (FILE *f,                  int *x, int *y, int *comp, int req_comp);
+// for stbi_load_from_file, file pointer is left pointing immediately after image
+#endif
+
+typedef struct
+{
+   int      (*read)  (void *user,char *data,int size);   // fill 'data' with 'size' bytes.  return number of bytes actually read 
+   void     (*skip)  (void *user,int n);                 // skip the next 'n' bytes, or 'unget' the last -n bytes if negative
+   int      (*eof)   (void *user);                       // returns nonzero if we are at end of file/data
+} stbi_io_callbacks;
+
+STBIDEF stbi_uc *stbi_load_from_callbacks  (stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp, int req_comp);
+
+#ifndef STBI_NO_HDR
+   STBIDEF float *stbi_loadf_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp);
+
+   #ifndef STBI_NO_STDIO
+   STBIDEF float *stbi_loadf            (char const *filename,   int *x, int *y, int *comp, int req_comp);
+   STBIDEF float *stbi_loadf_from_file  (FILE *f,                int *x, int *y, int *comp, int req_comp);
+   #endif
+   
+   STBIDEF float *stbi_loadf_from_callbacks  (stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp, int req_comp);
+
+   STBIDEF void   stbi_hdr_to_ldr_gamma(float gamma);
+   STBIDEF void   stbi_hdr_to_ldr_scale(float scale);
+
+   STBIDEF void   stbi_ldr_to_hdr_gamma(float gamma);
+   STBIDEF void   stbi_ldr_to_hdr_scale(float scale);
+#endif // STBI_NO_HDR
+
+// stbi_is_hdr is always defined
+STBIDEF int    stbi_is_hdr_from_callbacks(stbi_io_callbacks const *clbk, void *user);
+STBIDEF int    stbi_is_hdr_from_memory(stbi_uc const *buffer, int len);
+#ifndef STBI_NO_STDIO
+STBIDEF int      stbi_is_hdr          (char const *filename);
+STBIDEF int      stbi_is_hdr_from_file(FILE *f);
+#endif // STBI_NO_STDIO
+
+
+// get a VERY brief reason for failure
+// NOT THREADSAFE
+STBIDEF const char *stbi_failure_reason  (void); 
+
+// free the loaded image -- this is just free()
+STBIDEF void     stbi_image_free      (void *retval_from_stbi_load);
+
+// get image dimensions & components without fully decoding
+STBIDEF int      stbi_info_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp);
+STBIDEF int      stbi_info_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp);
+
+#ifndef STBI_NO_STDIO
+STBIDEF int      stbi_info            (char const *filename,     int *x, int *y, int *comp);
+STBIDEF int      stbi_info_from_file  (FILE *f,                  int *x, int *y, int *comp);
+
+#endif
+
+
+
+// for image formats that explicitly notate that they have premultiplied alpha,
+// we just return the colors as stored in the file. set this flag to force
+// unpremultiplication. results are undefined if the unpremultiply overflow.
+STBIDEF void stbi_set_unpremultiply_on_load(int flag_true_if_should_unpremultiply);
+
+// indicate whether we should process iphone images back to canonical format,
+// or just pass them through "as-is"
+STBIDEF void stbi_convert_iphone_png_to_rgb(int flag_true_if_should_convert);
+
+
+// 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);
+
+
+// define faster low-level operations (typically SIMD support)
+#ifdef STBI_SIMD
+typedef void (*stbi_idct_8x8)(stbi_uc *out, int out_stride, short data[64], unsigned short *dequantize);
+// compute an integer IDCT on "input"
+//     input[x] = data[x] * dequantize[x]
+//     write results to 'out': 64 samples, each run of 8 spaced by 'out_stride'
+//                             CLAMP results to 0..255
+typedef void (*stbi_YCbCr_to_RGB_run)(stbi_uc *output, stbi_uc const  *y, stbi_uc const *cb, stbi_uc const *cr, int count, int step);
+// compute a conversion from YCbCr to RGB
+//     'count' pixels
+//     write pixels to 'output'; each pixel is 'step' bytes (either 3 or 4; if 4, write '255' as 4th), order R,G,B
+//     y: Y input channel
+//     cb: Cb input channel; scale/biased to be 0..255
+//     cr: Cr input channel; scale/biased to be 0..255
+
+STBIDEF void stbi_install_idct(stbi_idct_8x8 func);
+STBIDEF void stbi_install_YCbCr_to_RGB(stbi_YCbCr_to_RGB_run func);
+#endif // STBI_SIMD
+
+
+#ifdef __cplusplus
+}
+#endif
+
+//
+//
+////   end header file   /////////////////////////////////////////////////////
+#endif // STBI_INCLUDE_STB_IMAGE_H
+
+#ifdef STB_IMAGE_IMPLEMENTATION
+
+#ifndef STBI_NO_HDR
+#include <math.h>  // ldexp
+#include <string.h> // strcmp, strtok
+#endif
+
+#ifndef STBI_NO_STDIO
+#include <stdio.h>
+#endif
+#include <stdlib.h>
+#include <memory.h>
+#include <assert.h>
+#include <stdarg.h>
+#include <stddef.h> // ptrdiff_t on osx
+
+#ifndef _MSC_VER
+   #ifdef __cplusplus
+   #define stbi_inline inline
+   #else
+   #define stbi_inline
+   #endif
+#else
+   #define stbi_inline __forceinline
+#endif
+
+
+#ifdef _MSC_VER
+typedef unsigned short stbi__uint16;
+typedef   signed short stbi__int16;
+typedef unsigned int   stbi__uint32;
+typedef   signed int   stbi__int32;
+#else
+#include <stdint.h>
+typedef uint16_t stbi__uint16;
+typedef int16_t  stbi__int16;
+typedef uint32_t stbi__uint32;
+typedef int32_t  stbi__int32;
+#endif
+
+// should produce compiler error if size is wrong
+typedef unsigned char validate_uint32[sizeof(stbi__uint32)==4 ? 1 : -1];
+
+#ifdef _MSC_VER
+#define STBI_NOTUSED(v)  (void)(v)
+#else
+#define STBI_NOTUSED(v)  (void)sizeof(v)
+#endif
+
+#ifdef _MSC_VER
+#define STBI_HAS_LROTL
+#endif
+
+#ifdef STBI_HAS_LROTL
+   #define stbi_lrot(x,y)  _lrotl(x,y)
+#else
+   #define stbi_lrot(x,y)  (((x) << (y)) | ((x) >> (32 - (y))))
+#endif
+
+///////////////////////////////////////////////
+//
+//  stbi__context struct and start_xxx functions
+
+// stbi__context structure is our basic context used by all images, so it
+// contains all the IO context, plus some basic image information
+typedef struct
+{
+   stbi__uint32 img_x, img_y;
+   int img_n, img_out_n;
+   
+   stbi_io_callbacks io;
+   void *io_user_data;
+
+   int read_from_callbacks;
+   int buflen;
+   stbi_uc buffer_start[128];
+
+   stbi_uc *img_buffer, *img_buffer_end;
+   stbi_uc *img_buffer_original;
+} stbi__context;
+
+
+static void stbi__refill_buffer(stbi__context *s);
+
+// initialize a memory-stbi__jpeg_huff_decode context
+static void stbi__start_mem(stbi__context *s, stbi_uc const *buffer, int len)
+{
+   s->io.read = NULL;
+   s->read_from_callbacks = 0;
+   s->img_buffer = s->img_buffer_original = (stbi_uc *) buffer;
+   s->img_buffer_end = (stbi_uc *) buffer+len;
+}
+
+// initialize a callback-based context
+static void stbi__start_callbacks(stbi__context *s, stbi_io_callbacks *c, void *user)
+{
+   s->io = *c;
+   s->io_user_data = user;
+   s->buflen = sizeof(s->buffer_start);
+   s->read_from_callbacks = 1;
+   s->img_buffer_original = s->buffer_start;
+   stbi__refill_buffer(s);
+}
+
+#ifndef STBI_NO_STDIO
+
+static int stbi__stdio_read(void *user, char *data, int size)
+{
+   return (int) fread(data,1,size,(FILE*) user);
+}
+
+static void stbi__stdio_skip(void *user, int n)
+{
+   fseek((FILE*) user, n, SEEK_CUR);
+}
+
+static int stbi__stdio_eof(void *user)
+{
+   return feof((FILE*) user);
+}
+
+static stbi_io_callbacks stbi__stdio_callbacks =
+{
+   stbi__stdio_read,
+   stbi__stdio_skip,
+   stbi__stdio_eof,
+};
+
+static void stbi__start_file(stbi__context *s, FILE *f)
+{
+   stbi__start_callbacks(s, &stbi__stdio_callbacks, (void *) f);
+}
+
+//static void stop_file(stbi__context *s) { }
+
+#endif // !STBI_NO_STDIO
+
+static void stbi__rewind(stbi__context *s)
+{
+   // conceptually rewind SHOULD rewind to the beginning of the stream,
+   // but we just rewind to the beginning of the initial buffer, because
+   // we only use it after doing 'test', which only ever looks at at most 92 bytes
+   s->img_buffer = s->img_buffer_original;
+}
+
+static int      stbi__jpeg_test(stbi__context *s);
+static stbi_uc *stbi__jpeg_load(stbi__context *s, int *x, int *y, int *comp, int req_comp);
+static int      stbi__jpeg_info(stbi__context *s, int *x, int *y, int *comp);
+static int      stbi__png_test(stbi__context *s);
+static stbi_uc *stbi__png_load(stbi__context *s, int *x, int *y, int *comp, int req_comp);
+static int      stbi__png_info(stbi__context *s, int *x, int *y, int *comp);
+static int      stbi__bmp_test(stbi__context *s);
+static stbi_uc *stbi__bmp_load(stbi__context *s, int *x, int *y, int *comp, int req_comp);
+static int      stbi__tga_test(stbi__context *s);
+static stbi_uc *stbi__tga_load(stbi__context *s, int *x, int *y, int *comp, int req_comp);
+static int      stbi__tga_info(stbi__context *s, int *x, int *y, int *comp);
+static int      stbi__psd_test(stbi__context *s);
+static stbi_uc *stbi__psd_load(stbi__context *s, int *x, int *y, int *comp, int req_comp);
+#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);
+#endif
+static int      stbi__pic_test(stbi__context *s);
+static stbi_uc *stbi__pic_load(stbi__context *s, int *x, int *y, int *comp, int req_comp);
+static int      stbi__gif_test(stbi__context *s);
+static stbi_uc *stbi__gif_load(stbi__context *s, int *x, int *y, int *comp, int req_comp);
+static int      stbi__gif_info(stbi__context *s, int *x, int *y, int *comp);
+
+
+// this is not threadsafe
+static const char *stbi__g_failure_reason;
+
+STBIDEF const char *stbi_failure_reason(void)
+{
+   return stbi__g_failure_reason;
+}
+
+static int stbi__err(const char *str)
+{
+   stbi__g_failure_reason = str;
+   return 0;
+}
+
+// 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 *) (stbi__err(x,y)?NULL:NULL))
+#define stbi__errpuc(x,y)  ((unsigned char *) (stbi__err(x,y)?NULL:NULL))
+
+STBIDEF void stbi_image_free(void *retval_from_stbi_load)
+{
+   free(retval_from_stbi_load);
+}
+
+#ifndef STBI_NO_HDR
+static float   *stbi__ldr_to_hdr(stbi_uc *data, int x, int y, int comp);
+static stbi_uc *stbi__hdr_to_ldr(float   *data, int x, int y, int comp);
+#endif
+
+static unsigned char *stbi_load_main(stbi__context *s, int *x, int *y, int *comp, int req_comp)
+{
+   if (stbi__jpeg_test(s)) return stbi__jpeg_load(s,x,y,comp,req_comp);
+   if (stbi__png_test(s))  return stbi__png_load(s,x,y,comp,req_comp);
+   if (stbi__bmp_test(s))  return stbi__bmp_load(s,x,y,comp,req_comp);
+   if (stbi__gif_test(s))  return stbi__gif_load(s,x,y,comp,req_comp);
+   if (stbi__psd_test(s))  return stbi__psd_load(s,x,y,comp,req_comp);
+   if (stbi__pic_test(s))  return stbi__pic_load(s,x,y,comp,req_comp);
+
+   #ifndef STBI_NO_HDR
+   if (stbi__hdr_test(s)) {
+      float *hdr = stbi__hdr_load(s, x,y,comp,req_comp);
+      return stbi__hdr_to_ldr(hdr, *x, *y, req_comp ? req_comp : *comp);
+   }
+   #endif
+
+   // test tga last because it's a crappy test!
+   if (stbi__tga_test(s))
+      return stbi__tga_load(s,x,y,comp,req_comp);
+   return stbi__errpuc("unknown image type", "Image not of any known type, or corrupt");
+}
+
+#ifndef STBI_NO_STDIO
+STBIDEF unsigned char *stbi_load(char const *filename, int *x, int *y, int *comp, int req_comp)
+{
+   FILE *f = 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 unsigned char *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_main(&s,x,y,comp,req_comp);
+   if (result) {
+      // need to 'unget' all the characters in the IO buffer
+      fseek(f, - (int) (s.img_buffer_end - s.img_buffer), SEEK_CUR);
+   }
+   return result;
+}
+#endif //!STBI_NO_STDIO
+
+STBIDEF unsigned char *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_main(&s,x,y,comp,req_comp);
+}
+
+unsigned char *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_main(&s,x,y,comp,req_comp);
+}
+
+#ifndef STBI_NO_HDR
+
+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))
+      return stbi__hdr_load(s,x,y,comp,req_comp);
+   #endif
+   data = stbi_load_main(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");
+}
+
+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);
+}
+
+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
+float *stbi_loadf(char const *filename, int *x, int *y, int *comp, int req_comp)
+{
+   FILE *f = fopen(filename, "rb");
+   float *result;
+   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;
+}
+
+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_HDR
+
+// these is-hdr-or-not is defined independent of whether STBI_NO_HDR is
+// defined, for API simplicity; if STBI_NO_HDR is defined, it always
+// reports false!
+
+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 = fopen(filename, "rb");
+   int result=0;
+   if (f) {
+      result = stbi_is_hdr_from_file(f);
+      fclose(f);
+   }
+   return result;
+}
+
+STBIDEF int      stbi_is_hdr_from_file(FILE *f)
+{
+   #ifndef STBI_NO_HDR
+   stbi__context s;
+   stbi__start_file(&s,f);
+   return stbi__hdr_test(&s);
+   #else
+   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
+   return 0;
+   #endif
+}
+
+#ifndef STBI_NO_HDR
+static float stbi__h2l_gamma_i=1.0f/2.2f, stbi__h2l_scale_i=1.0f;
+static float stbi__l2h_gamma=2.2f, stbi__l2h_scale=1.0f;
+
+void   stbi_hdr_to_ldr_gamma(float gamma) { stbi__h2l_gamma_i = 1/gamma; }
+void   stbi_hdr_to_ldr_scale(float scale) { stbi__h2l_scale_i = 1/scale; }
+
+void   stbi_ldr_to_hdr_gamma(float gamma) { stbi__l2h_gamma = gamma; }
+void   stbi_ldr_to_hdr_scale(float scale) { stbi__l2h_scale = scale; }
+#endif
+
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// Common code used by all image loaders
+//
+
+enum
+{
+   SCAN_load=0,
+   SCAN_type,
+   SCAN_header
+};
+
+static void stbi__refill_buffer(stbi__context *s)
+{
+   int n = (s->io.read)(s->io_user_data,(char*)s->buffer_start,s->buflen);
+   if (n == 0) {
+      // at end of file, treat same as if from memory, but need to handle case
+      // where s->img_buffer isn't pointing to safe memory, stbi__err.g. 0-byte file
+      s->read_from_callbacks = 0;
+      s->img_buffer = s->buffer_start;
+      s->img_buffer_end = s->buffer_start+1;
+      *s->img_buffer = 0;
+   } else {
+      s->img_buffer = s->buffer_start;
+      s->img_buffer_end = s->buffer_start + n;
+   }
+}
+
+stbi_inline static stbi_uc stbi__get8(stbi__context *s)
+{
+   if (s->img_buffer < s->img_buffer_end)
+      return *s->img_buffer++;
+   if (s->read_from_callbacks) {
+      stbi__refill_buffer(s);
+      return *s->img_buffer++;
+   }
+   return 0;
+}
+
+stbi_inline static int stbi__at_eof(stbi__context *s)
+{
+   if (s->io.read) {
+      if (!(s->io.eof)(s->io_user_data)) return 0;
+      // if feof() is true, check if buffer = end
+      // special case: we've only got the special 0 character at the end
+      if (s->read_from_callbacks == 0) return 1;
+   }
+
+   return s->img_buffer >= s->img_buffer_end;   
+}
+
+static void stbi__skip(stbi__context *s, int n)
+{
+   if (s->io.read) {
+      int blen = (int) (s->img_buffer_end - s->img_buffer);
+      if (blen < n) {
+         s->img_buffer = s->img_buffer_end;
+         (s->io.skip)(s->io_user_data, n - blen);
+         return;
+      }
+   }
+   s->img_buffer += n;
+}
+
+static int stbi__getn(stbi__context *s, stbi_uc *buffer, int n)
+{
+   if (s->io.read) {
+      int blen = (int) (s->img_buffer_end - s->img_buffer);
+      if (blen < n) {
+         int res, count;
+
+         memcpy(buffer, s->img_buffer, blen);
+         
+         count = (s->io.read)(s->io_user_data, (char*) buffer + blen, n - blen);
+         res = (count == (n-blen));
+         s->img_buffer = s->img_buffer_end;
+         return res;
+      }
+   }
+
+   if (s->img_buffer+n <= s->img_buffer_end) {
+      memcpy(buffer, s->img_buffer, n);
+      s->img_buffer += n;
+      return 1;
+   } else
+      return 0;
+}
+
+static int stbi__get16be(stbi__context *s)
+{
+   int z = stbi__get8(s);
+   return (z << 8) + stbi__get8(s);
+}
+
+static stbi__uint32 stbi__get32be(stbi__context *s)
+{
+   stbi__uint32 z = stbi__get16be(s);
+   return (z << 16) + stbi__get16be(s);
+}
+
+static int stbi__get16le(stbi__context *s)
+{
+   int z = stbi__get8(s);
+   return z + (stbi__get8(s) << 8);
+}
+
+static stbi__uint32 stbi__get32le(stbi__context *s)
+{
+   stbi__uint32 z = stbi__get16le(s);
+   return z + (stbi__get16le(s) << 16);
+}
+
+//////////////////////////////////////////////////////////////////////////////
+//
+//  generic converter from built-in img_n to req_comp
+//    individual types do this automatically as much as possible (stbi__err.g. jpeg
+//    does all cases internally since it needs to colorspace convert anyway,
+//    and it never has alpha, so very few cases ). png can automatically
+//    interleave an alpha=255 channel, but falls back to this for other cases
+//
+//  assume data buffer is malloced, so malloc a new one and free that one
+//  only failure mode is malloc failing
+
+static stbi_uc stbi__compute_y(int r, int g, int b)
+{
+   return (stbi_uc) (((r*77) + (g*150) +  (29*b)) >> 8);
+}
+
+static unsigned char *stbi__convert_format(unsigned char *data, int img_n, int req_comp, unsigned int x, unsigned int y)
+{
+   int i,j;
+   unsigned char *good;
+
+   if (req_comp == img_n) return data;
+   assert(req_comp >= 1 && req_comp <= 4);
+
+   good = (unsigned char *) malloc(req_comp * x * y);
+   if (good == NULL) {
+      free(data);
+      return stbi__errpuc("outofmem", "Out of memory");
+   }
+
+   for (j=0; j < (int) y; ++j) {
+      unsigned char *src  = data + j * x * img_n   ;
+      unsigned char *dest = good + j * x * req_comp;
+
+      #define COMBO(a,b)  ((a)*8+(b))
+      #define CASE(a,b)   case COMBO(a,b): for(i=x-1; i >= 0; --i, src += a, dest += b)
+      // convert source image with img_n components to one with req_comp components;
+      // avoid switch per pixel, so use switch per scanline and massive macros
+      switch (COMBO(img_n, req_comp)) {
+         CASE(1,2) dest[0]=src[0], dest[1]=255; break;
+         CASE(1,3) dest[0]=dest[1]=dest[2]=src[0]; break;
+         CASE(1,4) dest[0]=dest[1]=dest[2]=src[0], dest[3]=255; break;
+         CASE(2,1) dest[0]=src[0]; break;
+         CASE(2,3) dest[0]=dest[1]=dest[2]=src[0]; break;
+         CASE(2,4) dest[0]=dest[1]=dest[2]=src[0], dest[3]=src[1]; break;
+         CASE(3,4) dest[0]=src[0],dest[1]=src[1],dest[2]=src[2],dest[3]=255; break;
+         CASE(3,1) dest[0]=stbi__compute_y(src[0],src[1],src[2]); break;
+         CASE(3,2) dest[0]=stbi__compute_y(src[0],src[1],src[2]), dest[1] = 255; break;
+         CASE(4,1) dest[0]=stbi__compute_y(src[0],src[1],src[2]); break;
+         CASE(4,2) dest[0]=stbi__compute_y(src[0],src[1],src[2]), dest[1] = src[3]; break;
+         CASE(4,3) dest[0]=src[0],dest[1]=src[1],dest[2]=src[2]; break;
+         default: assert(0);
+      }
+      #undef CASE
+   }
+
+   free(data);
+   return good;
+}
+
+#ifndef STBI_NO_HDR
+static float   *stbi__ldr_to_hdr(stbi_uc *data, int x, int y, int comp)
+{
+   int i,k,n;
+   float *output = (float *) malloc(x * y * comp * sizeof(float));
+   if (output == NULL) { free(data); return stbi__errpf("outofmem", "Out of memory"); }
+   // compute number of non-alpha components
+   if (comp & 1) n = comp; else n = comp-1;
+   for (i=0; i < x*y; ++i) {
+      for (k=0; k < n; ++k) {
+         output[i*comp + k] = (float) (pow(data[i*comp+k]/255.0f, stbi__l2h_gamma) * stbi__l2h_scale);
+      }
+      if (k < comp) output[i*comp + k] = data[i*comp+k]/255.0f;
+   }
+   free(data);
+   return output;
+}
+
+#define stbi__float2int(x)   ((int) (x))
+static stbi_uc *stbi__hdr_to_ldr(float   *data, int x, int y, int comp)
+{
+   int i,k,n;
+   stbi_uc *output = (stbi_uc *) malloc(x * y * comp);
+   if (output == NULL) { 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);
+      }
+   }
+   free(data);
+   return output;
+}
+#endif
+
+//////////////////////////////////////////////////////////////////////////////
+//
+//  "baseline" JPEG/JFIF decoder (not actually fully baseline implementation)
+//
+//    simple implementation
+//      - channel subsampling of at most 2 in each dimension
+//      - 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
+//      - uses a lot of intermediate memory, could cache poorly
+//      - load http://nothings.org/remote/anemones.jpg 3 times on 2.8Ghz P4
+//          stb_jpeg:   1.34 seconds (MSVC6, default release build)
+//          stb_jpeg:   1.06 seconds (MSVC6, processor = Pentium Pro)
+//          IJL11.dll:  1.08 seconds (compiled by intel)
+//          IJG 1998:   0.98 seconds (MSVC6, makefile provided by IJG)
+//          IJG 1998:   0.95 seconds (MSVC6, makefile + proc=PPro)
+
+// 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
+{
+   #ifdef STBI_SIMD
+   unsigned short dequant2[4][64];
+   #endif
+   stbi__context *s;
+   stbi__huffman huff_dc[4];
+   stbi__huffman huff_ac[4];
+   stbi_uc dequant[4][64];
+
+// 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;
+      stbi_uc *linebuf;
+   } 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 scan_n, order[4];
+   int restart_interval, todo;
+} stbi__jpeg;
+
+static int stbi__build_huffman(stbi__huffman *h, int *count)
+{
+   int i,j,k=0,code;
+   // build size list for each symbol (from JPEG spec)
+   for (i=0; i < 16; ++i)
+      for (j=0; j < count[i]; ++j)
+         h->size[k++] = (stbi_uc) (i+1);
+   h->size[k] = 0;
+
+   // compute actual symbols (from jpeg spec)
+   code = 0;
+   k = 0;
+   for(j=1; j <= 16; ++j) {
+      // compute delta to add to code to compute symbol id
+      h->delta[j] = k - code;
+      if (h->size[k] == j) {
+         while (h->size[k] == j)
+            h->code[k++] = (stbi__uint16) (code++);
+         if (code-1 >= (1 << j)) return stbi__err("bad code lengths","Corrupt JPEG");
+      }
+      // compute largest code + 1 for this size, preshifted as needed later
+      h->maxcode[j] = code << (16-j);
+      code <<= 1;
+   }
+   h->maxcode[j] = 0xffffffff;
+
+   // build non-spec acceleration table; 255 is flag for not-accelerated
+   memset(h->fast, 255, 1 << FAST_BITS);
+   for (i=0; i < k; ++i) {
+      int s = h->size[i];
+      if (s <= FAST_BITS) {
+         int c = h->code[i] << (FAST_BITS-s);
+         int m = 1 << (FAST_BITS-s);
+         for (j=0; j < m; ++j) {
+            h->fast[c+j] = (stbi_uc) i;
+         }
+      }
+   }
+   return 1;
+}
+
+static void stbi__grow_buffer_unsafe(stbi__jpeg *j)
+{
+   do {
+      int b = j->nomore ? 0 : stbi__get8(j->s);
+      if (b == 0xff) {
+         int c = stbi__get8(j->s);
+         if (c != 0) {
+            j->marker = (unsigned char) c;
+            j->nomore = 1;
+            return;
+         }
+      }
+      j->code_buffer |= b << (24 - j->code_bits);
+      j->code_bits += 8;
+   } while (j->code_bits <= 24);
+}
+
+// (1 << n) - 1
+static stbi__uint32 stbi__bmask[17]={0,1,3,7,15,31,63,127,255,511,1023,2047,4095,8191,16383,32767,65535};
+
+// stbi__jpeg_huff_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];
+   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];
+}
+
+// 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 m = 1 << (n-1);
+   unsigned int k;
+   if (j->code_bits < n) stbi__grow_buffer_unsafe(j);
+
+   #if 1
+   k = stbi_lrot(j->code_buffer, n);
+   j->code_buffer = k & ~stbi__bmask[n];
+   k &= stbi__bmask[n];
+   j->code_bits -= n;
+   #else
+   k = (j->code_buffer >> (32 - n)) & stbi__bmask[n];
+   j->code_bits -= n;
+   j->code_buffer <<= n;
+   #endif
+   // the following test is probably a random branch that won't
+   // predict well. I tried to table accelerate it but failed.
+   // maybe it's compiling as a conditional move?
+   if (k < m)
+      return (-1 << n) + k + 1;
+   else
+      return k;
+}
+
+// given a value that's at position X in the zigzag stream,
+// where does it appear in the 8x8 matrix coded as row-major?
+static stbi_uc stbi__jpeg_dezigzag[64+15] =
+{
+    0,  1,  8, 16,  9,  2,  3, 10,
+   17, 24, 32, 25, 18, 11,  4,  5,
+   12, 19, 26, 33, 40, 48, 41, 34,
+   27, 20, 13,  6,  7, 14, 21, 28,
+   35, 42, 49, 56, 57, 50, 43, 36,
+   29, 22, 15, 23, 30, 37, 44, 51,
+   58, 59, 52, 45, 38, 31, 39, 46,
+   53, 60, 61, 54, 47, 55, 62, 63,
+   // let corrupt input sample past end
+   63, 63, 63, 63, 63, 63, 63, 63,
+   63, 63, 63, 63, 63, 63, 63
+};
+
+// stbi__jpeg_huff_decode one 64-entry block--
+static int stbi__jpeg_decode_block(stbi__jpeg *j, short data[64], stbi__huffman *hdc, stbi__huffman *hac, int b)
+{
+   int diff,dc,k;
+   int t = stbi__jpeg_huff_decode(j, hdc);
+   if (t < 0) return stbi__err("bad huffman code","Corrupt JPEG");
+
+   // 0 all the ac values now so we can do it 32-bits at a time
+   memset(data,0,64*sizeof(data[0]));
+
+   diff = t ? stbi__extend_receive(j, t) : 0;
+   dc = j->img_comp[b].dc_pred + diff;
+   j->img_comp[b].dc_pred = dc;
+   data[0] = (short) dc;
+
+   // stbi__jpeg_huff_decode AC components, see JPEG spec
+   k = 1;
+   do {
+      int r,s;
+      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;
+         // stbi__jpeg_huff_decode into unzigzag'd location
+         data[stbi__jpeg_dezigzag[k++]] = (short) stbi__extend_receive(j,s);
+      }
+   } while (k < 64);
+   return 1;
+}
+
+// take a -128..127 value and stbi__clamp it and convert to 0..255
+stbi_inline static stbi_uc stbi__clamp(int x)
+{
+   // trick to use a single test to catch both cases
+   if ((unsigned int) x > 255) {
+      if (x < 0) return 0;
+      if (x > 255) return 255;
+   }
+   return (stbi_uc) x;
+}
+
+#define stbi__f2f(x)  (int) (((x) * 4096 + 0.5))
+#define stbi__fsh(x)  ((x) << 12)
+
+// derived from jidctint -- DCT_ISLOW
+#define STBI__IDCT_1D(s0,s1,s2,s3,s4,s5,s6,s7)       \
+   int t0,t1,t2,t3,p1,p2,p3,p4,p5,x0,x1,x2,x3; \
+   p2 = s2;                                    \
+   p3 = s6;                                    \
+   p1 = (p2+p3) * stbi__f2f(0.5411961f);             \
+   t2 = p1 + p3*stbi__f2f(-1.847759065f);            \
+   t3 = p1 + p2*stbi__f2f( 0.765366865f);            \
+   p2 = s0;                                    \
+   p3 = s4;                                    \
+   t0 = stbi__fsh(p2+p3);                            \
+   t1 = stbi__fsh(p2-p3);                            \
+   x0 = t0+t3;                                 \
+   x3 = t0-t3;                                 \
+   x1 = t1+t2;                                 \
+   x2 = t1-t2;                                 \
+   t0 = s7;                                    \
+   t1 = s5;                                    \
+   t2 = s3;                                    \
+   t3 = s1;                                    \
+   p3 = t0+t2;                                 \
+   p4 = t1+t3;                                 \
+   p1 = t0+t3;                                 \
+   p2 = t1+t2;                                 \
+   p5 = (p3+p4)*stbi__f2f( 1.175875602f);            \
+   t0 = t0*stbi__f2f( 0.298631336f);                 \
+   t1 = t1*stbi__f2f( 2.053119869f);                 \
+   t2 = t2*stbi__f2f( 3.072711026f);                 \
+   t3 = t3*stbi__f2f( 1.501321110f);                 \
+   p1 = p5 + p1*stbi__f2f(-0.899976223f);            \
+   p2 = p5 + p2*stbi__f2f(-2.562915447f);            \
+   p3 = p3*stbi__f2f(-1.961570560f);                 \
+   p4 = p4*stbi__f2f(-0.390180644f);                 \
+   t3 += p1+p4;                                \
+   t2 += p2+p3;                                \
+   t1 += p2+p4;                                \
+   t0 += p1+p3;
+
+#ifdef STBI_SIMD
+typedef unsigned short stbi_dequantize_t;
+#else
+typedef stbi_uc stbi_dequantize_t;
+#endif
+
+// .344 seconds on 3*anemones.jpg
+static void stbi__idct_block(stbi_uc *out, int out_stride, short data[64], stbi_dequantize_t *dequantize)
+{
+   int i,val[64],*v=val;
+   stbi_dequantize_t *dq = dequantize;
+   stbi_uc *o;
+   short *d = data;
+
+   // columns
+   for (i=0; i < 8; ++i,++d,++dq, ++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] * dq[0] << 2;
+         v[0] = v[8] = v[16] = v[24] = v[32] = v[40] = v[48] = v[56] = dcterm;
+      } else {
+         STBI__IDCT_1D(d[ 0]*dq[ 0],d[ 8]*dq[ 8],d[16]*dq[16],d[24]*dq[24],
+                 d[32]*dq[32],d[40]*dq[40],d[48]*dq[48],d[56]*dq[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_SIMD
+static stbi_idct_8x8 stbi__idct_installed = stbi__idct_block;
+
+STBIDEF void stbi_install_idct(stbi_idct_8x8 func)
+{
+   stbi__idct_installed = func;
+}
+#endif
+
+#define STBI__MARKER_none  0xff
+// if there's a pending marker from the entropy stream, return that
+// otherwise, fetch from the stream and get a marker. if there's no
+// marker, return 0xff, which is never a valid marker value
+static stbi_uc stbi__get_marker(stbi__jpeg *j)
+{
+   stbi_uc x;
+   if (j->marker != STBI__MARKER_none) { x = j->marker; j->marker = STBI__MARKER_none; return x; }
+   x = stbi__get8(j->s);
+   if (x != 0xff) return STBI__MARKER_none;
+   while (x == 0xff)
+      x = stbi__get8(j->s);
+   return x;
+}
+
+// in each scan, we'll have scan_n components, and the order
+// of the components is specified by order[]
+#define STBI__RESTART(x)     ((x) >= 0xd0 && (x) <= 0xd7)
+
+// after a restart interval, stbi__jpeg_reset the entropy decoder and
+// the dc prediction
+static void stbi__jpeg_reset(stbi__jpeg *j)
+{
+   j->code_bits = 0;
+   j->code_buffer = 0;
+   j->nomore = 0;
+   j->img_comp[0].dc_pred = j->img_comp[1].dc_pred = j->img_comp[2].dc_pred = 0;
+   j->marker = STBI__MARKER_none;
+   j->todo = j->restart_interval ? j->restart_interval : 0x7fffffff;
+   // 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->scan_n == 1) {
+      int i,j;
+      #ifdef STBI_SIMD
+      __declspec(align(16))
+      #endif
+      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) {
+            if (!stbi__jpeg_decode_block(z, data, z->huff_dc+z->img_comp[n].hd, z->huff_ac+z->img_comp[n].ha, n)) return 0;
+            #ifdef STBI_SIMD
+            stbi__idct_installed(z->img_comp[n].data+z->img_comp[n].w2*j*8+i*8, z->img_comp[n].w2, data, z->dequant2[z->img_comp[n].tq]);
+            #else
+            stbi__idct_block(z->img_comp[n].data+z->img_comp[n].w2*j*8+i*8, z->img_comp[n].w2, data, z->dequant[z->img_comp[n].tq]);
+            #endif
+            // 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);
+            }
+         }
+      }
+   } else { // interleaved!
+      int i,j,k,x,y;
+      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;
+                     if (!stbi__jpeg_decode_block(z, data, z->huff_dc+z->img_comp[n].hd, z->huff_ac+z->img_comp[n].ha, n)) return 0;
+                     #ifdef STBI_SIMD
+                     stbi__idct_installed(z->img_comp[n].data+z->img_comp[n].w2*y2+x2, z->img_comp[n].w2, data, z->dequant2[z->img_comp[n].tq]);
+                     #else
+                     stbi__idct_block(z->img_comp[n].data+z->img_comp[n].w2*y2+x2, z->img_comp[n].w2, data, z->dequant[z->img_comp[n].tq]);
+                     #endif
+                  }
+               }
+            }
+            // 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 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;
+}
+
+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 0xC2: // stbi__SOF - progressive
+         return stbi__err("progressive jpeg","JPEG format not supported (progressive)");
+
+      case 0xDD: // DRI - specify restart interval
+         if (stbi__get16be(z->s) != 4) return stbi__err("bad DRI len","Corrupt JPEG");
+         z->restart_interval = stbi__get16be(z->s);
+         return 1;
+
+      case 0xDB: // DQT - define quantization table
+         L = stbi__get16be(z->s)-2;
+         while (L > 0) {
+            int q = stbi__get8(z->s);
+            int p = q >> 4;
+            int t = q & 15,i;
+            if (p != 0) return stbi__err("bad DQT type","Corrupt JPEG");
+            if (t > 3) return stbi__err("bad DQT table","Corrupt JPEG");
+            for (i=0; i < 64; ++i)
+               z->dequant[t][stbi__jpeg_dezigzag[i]] = stbi__get8(z->s);
+            #ifdef STBI_SIMD
+            for (i=0; i < 64; ++i)
+               z->dequant2[t][i] = z->dequant[t][i];
+            #endif
+            L -= 65;
+         }
+         return L==0;
+
+      case 0xC4: // DHT - define huffman table
+         L = stbi__get16be(z->s)-2;
+         while (L > 0) {
+            stbi_uc *v;
+            int sizes[16],i,n=0;
+            int q = stbi__get8(z->s);
+            int tc = q >> 4;
+            int th = q & 15;
+            if (tc > 1 || th > 3) return stbi__err("bad DHT header","Corrupt JPEG");
+            for (i=0; i < 16; ++i) {
+               sizes[i] = stbi__get8(z->s);
+               n += sizes[i];
+            }
+            L -= 17;
+            if (tc == 0) {
+               if (!stbi__build_huffman(z->huff_dc+th, sizes)) return 0;
+               v = z->huff_dc[th].values;
+            } else {
+               if (!stbi__build_huffman(z->huff_ac+th, sizes)) return 0;
+               v = z->huff_ac[th].values;
+            }
+            for (i=0; i < n; ++i)
+               v[i] = stbi__get8(z->s);
+            L -= n;
+         }
+         return L==0;
+   }
+   // check for comment block or APP blocks
+   if ((m >= 0xE0 && m <= 0xEF) || m == 0xFE) {
+      stbi__skip(z->s, stbi__get16be(z->s)-2);
+      return 1;
+   }
+   return 0;
+}
+
+// after we see stbi__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 stbi__SOS component count","Corrupt JPEG");
+   if (Ls != 6+2*z->scan_n) return stbi__err("bad stbi__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;
+      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;
+   }
+   if (stbi__get8(z->s) != 0) return stbi__err("bad stbi__SOS","Corrupt JPEG");
+   stbi__get8(z->s); // should be 63, but might be 0
+   if (stbi__get8(z->s) != 0) return stbi__err("bad stbi__SOS","Corrupt JPEG");
+
+   return 1;
+}
+
+static int stbi__process_frame_header(stbi__jpeg *z, int scan)
+{
+   stbi__context *s = z->s;
+   int Lf,p,i,q, h_max=1,v_max=1,c;
+   Lf = stbi__get16be(s);         if (Lf < 11) return stbi__err("bad stbi__SOF len","Corrupt JPEG"); // JPEG
+   p  = stbi__get8(s);          if (p != 8) return stbi__err("only 8-bit","JPEG format not supported: 8-bit only"); // JPEG baseline
+   s->img_y = stbi__get16be(s);   if (s->img_y == 0) return stbi__err("no header height", "JPEG format not supported: delayed height"); // Legal, but we don't handle it--but neither does IJG
+   s->img_x = stbi__get16be(s);   if (s->img_x == 0) return stbi__err("0 width","Corrupt JPEG"); // JPEG requires
+   c = stbi__get8(s);
+   if (c != 3 && c != 1) return stbi__err("bad component count","Corrupt JPEG");    // JFIF requires
+   s->img_n = c;
+   for (i=0; i < c; ++i) {
+      z->img_comp[i].data = NULL;
+      z->img_comp[i].linebuf = NULL;
+   }
+
+   if (Lf != 8+3*s->img_n) return stbi__err("bad stbi__SOF len","Corrupt JPEG");
+
+   for (i=0; i < s->img_n; ++i) {
+      z->img_comp[i].id = stbi__get8(s);
+      if (z->img_comp[i].id != i+1)   // JFIF requires
+         if (z->img_comp[i].id != i)  // some version of jpegtran outputs non-JFIF-compliant files!
+            return stbi__err("bad component ID","Corrupt JPEG");
+      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 != SCAN_load) return 1;
+
+   if ((1 << 30) / s->img_x / s->img_n < s->img_y) return stbi__err("too large", "Image too large to stbi__jpeg_huff_decode");
+
+   for (i=0; i < s->img_n; ++i) {
+      if (z->img_comp[i].h > h_max) h_max = z->img_comp[i].h;
+      if (z->img_comp[i].v > v_max) v_max = z->img_comp[i].v;
+   }
+
+   // compute interleaved mcu info
+   z->img_h_max = h_max;
+   z->img_v_max = v_max;
+   z->img_mcu_w = h_max * 8;
+   z->img_mcu_h = v_max * 8;
+   z->img_mcu_x = (s->img_x + z->img_mcu_w-1) / z->img_mcu_w;
+   z->img_mcu_y = (s->img_y + z->img_mcu_h-1) / z->img_mcu_h;
+
+   for (i=0; i < s->img_n; ++i) {
+      // number of effective pixels (stbi__err.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 stbi__jpeg_huff_decode
+      // the bogus oversized data from using interleaved MCUs and their
+      // big blocks (stbi__err.g. a 16x16 iMCU on an image of width 33); we won't
+      // discard the extra data until colorspace conversion
+      z->img_comp[i].w2 = z->img_mcu_x * z->img_comp[i].h * 8;
+      z->img_comp[i].h2 = z->img_mcu_y * z->img_comp[i].v * 8;
+      z->img_comp[i].raw_data = malloc(z->img_comp[i].w2 * z->img_comp[i].h2+15);
+      if (z->img_comp[i].raw_data == NULL) {
+         for(--i; i >= 0; --i) {
+            free(z->img_comp[i].raw_data);
+            z->img_comp[i].data = NULL;
+         }
+         return stbi__err("outofmem", "Out of memory");
+      }
+      // align blocks for installable-idct using mmx/sse
+      z->img_comp[i].data = (stbi_uc*) (((size_t) z->img_comp[i].raw_data + 15) & ~15);
+      z->img_comp[i].linebuf = NULL;
+   }
+
+   return 1;
+}
+
+// use comparisons since in some cases we handle more than one case (stbi__err.g. stbi__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)
+#define stbi__SOS(x)         ((x) == 0xda)
+
+static int decode_jpeg_header(stbi__jpeg *z, int scan)
+{
+   int m;
+   z->marker = STBI__MARKER_none; // initialize cached marker to empty
+   m = stbi__get_marker(z);
+   if (!stbi__SOI(m)) return stbi__err("no stbi__SOI","Corrupt JPEG");
+   if (scan == 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 stbi__SOF", "Corrupt JPEG");
+         m = stbi__get_marker(z);
+      }
+   }
+   if (!stbi__process_frame_header(z, scan)) return 0;
+   return 1;
+}
+
+static int decode_jpeg_image(stbi__jpeg *j)
+{
+   int m;
+   j->restart_interval = 0;
+   if (!decode_jpeg_header(j, SCAN_load)) return 0;
+   m = stbi__get_marker(j);
+   while (!stbi__EOI(m)) {
+      if (stbi__SOS(m)) {
+         if (!stbi__process_scan_header(j)) return 0;
+         if (!stbi__parse_entropy_coded_data(j)) return 0;
+         if (j->marker == STBI__MARKER_none ) {
+            // handle 0s at the end of image data from IP Kamera 9060
+            while (!stbi__at_eof(j->s)) {
+               int x = stbi__get8(j->s);
+               if (x == 255) {
+                  j->marker = stbi__get8(j->s);
+                  break;
+               } else if (x != 0) {
+                  return 0;
+               }
+            }
+            // if we reach eof without hitting a marker, stbi__get_marker() below will fail and we'll eventually return 0
+         }
+      } else {
+         if (!stbi__process_marker(j, m)) return 0;
+      }
+      m = stbi__get_marker(j);
+   }
+   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;
+}
+
+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;
+}
+
+#define float2fixed(x)  ((int) ((x) * 65536 + 0.5))
+
+// 0.38 seconds on 3*anemones.jpg   (0.25 with processor = Pro)
+// VC6 without processor=Pro is generating multiple LEAs per multiply!
+static void stbi__YCbCr_to_RGB_row(stbi_uc *out, const stbi_uc *y, const stbi_uc *pcb, const stbi_uc *pcr, int count, int step)
+{
+   int i;
+   for (i=0; i < count; ++i) {
+      int y_fixed = (y[i] << 16) + 32768; // rounding
+      int r,g,b;
+      int cr = pcr[i] - 128;
+      int cb = pcb[i] - 128;
+      r = y_fixed + cr*float2fixed(1.40200f);
+      g = y_fixed - cr*float2fixed(0.71414f) - cb*float2fixed(0.34414f);
+      b = y_fixed                            + cb*float2fixed(1.77200f);
+      r >>= 16;
+      g >>= 16;
+      b >>= 16;
+      if ((unsigned) r > 255) { if (r < 0) r = 0; else r = 255; }
+      if ((unsigned) g > 255) { if (g < 0) g = 0; else g = 255; }
+      if ((unsigned) b > 255) { if (b < 0) b = 0; else b = 255; }
+      out[0] = (stbi_uc)r;
+      out[1] = (stbi_uc)g;
+      out[2] = (stbi_uc)b;
+      out[3] = 255;
+      out += step;
+   }
+}
+
+#ifdef STBI_SIMD
+static stbi_YCbCr_to_RGB_run stbi__YCbCr_installed = stbi__YCbCr_to_RGB_row;
+
+STBIDEF void stbi_install_YCbCr_to_RGB(stbi_YCbCr_to_RGB_run func)
+{
+   stbi__YCbCr_installed = func;
+}
+#endif
+
+
+// clean up the temporary component buffers
+static void stbi__cleanup_jpeg(stbi__jpeg *j)
+{
+   int i;
+   for (i=0; i < j->s->img_n; ++i) {
+      if (j->img_comp[i].data) {
+         free(j->img_comp[i].raw_data);
+         j->img_comp[i].data = NULL;
+      }
+      if (j->img_comp[i].linebuf) {
+         free(j->img_comp[i].linebuf);
+         j->img_comp[i].linebuf = NULL;
+      }
+   }
+}
+
+typedef struct
+{
+   resample_row_func resample;
+   stbi_uc *line0,*line1;
+   int hs,vs;   // expansion factor in each axis
+   int w_lores; // horizontal pixels pre-expansion 
+   int ystep;   // how far through vertical expansion we are
+   int ypos;    // which pre-expansion row we're on
+} stbi__resample;
+
+static stbi_uc *load_jpeg_image(stbi__jpeg *z, int *out_x, int *out_y, int *comp, int req_comp)
+{
+   int n, decode_n;
+   // validate req_comp
+   if (req_comp < 0 || req_comp > 4) return stbi__errpuc("bad req_comp", "Internal error");
+   z->s->img_n = 0;
+
+   // load a jpeg image from whichever source
+   if (!decode_jpeg_image(z)) { stbi__cleanup_jpeg(z); return NULL; }
+
+   // determine actual number of components to generate
+   n = req_comp ? req_comp : z->s->img_n;
+
+   if (z->s->img_n == 3 && n < 3)
+      decode_n = 1;
+   else
+      decode_n = z->s->img_n;
+
+   // resample and color-convert
+   {
+      int k;
+      unsigned int i,j;
+      stbi_uc *output;
+      stbi_uc *coutput[4];
+
+      stbi__resample res_comp[4];
+
+      for (k=0; k < decode_n; ++k) {
+         stbi__resample *r = &res_comp[k];
+
+         // allocate line buffer big enough for upsampling off the edges
+         // with upsample factor of 4
+         z->img_comp[k].linebuf = (stbi_uc *) 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 = stbi__resample_row_hv_2;
+         else                               r->resample = stbi__resample_row_generic;
+      }
+
+      // can't error after this so, this is safe
+      output = (stbi_uc *) malloc(n * z->s->img_x * z->s->img_y + 1);
+      if (!output) { stbi__cleanup_jpeg(z); return stbi__errpuc("outofmem", "Out of memory"); }
+
+      // now go ahead and resample
+      for (j=0; j < z->s->img_y; ++j) {
+         stbi_uc *out = output + n * z->s->img_x * j;
+         for (k=0; k < decode_n; ++k) {
+            stbi__resample *r = &res_comp[k];
+            int y_bot = r->ystep >= (r->vs >> 1);
+            coutput[k] = r->resample(z->img_comp[k].linebuf,
+                                     y_bot ? r->line1 : r->line0,
+                                     y_bot ? r->line0 : r->line1,
+                                     r->w_lores, r->hs);
+            if (++r->ystep >= r->vs) {
+               r->ystep = 0;
+               r->line0 = r->line1;
+               if (++r->ypos < z->img_comp[k].y)
+                  r->line1 += z->img_comp[k].w2;
+            }
+         }
+         if (n >= 3) {
+            stbi_uc *y = coutput[0];
+            if (z->s->img_n == 3) {
+               #ifdef STBI_SIMD
+               stbi__YCbCr_installed(out, y, coutput[1], coutput[2], z->s->img_x, n);
+               #else
+               stbi__YCbCr_to_RGB_row(out, y, coutput[1], coutput[2], z->s->img_x, n);
+               #endif
+            } else
+               for (i=0; i < z->s->img_x; ++i) {
+                  out[0] = out[1] = out[2] = y[i];
+                  out[3] = 255; // not used if n==3
+                  out += n;
+               }
+         } else {
+            stbi_uc *y = coutput[0];
+            if (n == 1)
+               for (i=0; i < z->s->img_x; ++i) out[i] = y[i];
+            else
+               for (i=0; i < z->s->img_x; ++i) *out++ = y[i], *out++ = 255;
+         }
+      }
+      stbi__cleanup_jpeg(z);
+      *out_x = z->s->img_x;
+      *out_y = z->s->img_y;
+      if (comp) *comp  = z->s->img_n; // report original components, not output
+      return output;
+   }
+}
+
+static unsigned char *stbi__jpeg_load(stbi__context *s, int *x, int *y, int *comp, int req_comp)
+{
+   stbi__jpeg j;
+   j.s = s;
+   return load_jpeg_image(&j, x,y,comp,req_comp);
+}
+
+static int stbi__jpeg_test(stbi__context *s)
+{
+   int r;
+   stbi__jpeg j;
+   j.s = s;
+   r = decode_jpeg_header(&j, SCAN_type);
+   stbi__rewind(s);
+   return r;
+}
+
+static int stbi__jpeg_info_raw(stbi__jpeg *j, int *x, int *y, int *comp)
+{
+   if (!decode_jpeg_header(j, SCAN_header)) {
+      stbi__rewind( j->s );
+      return 0;
+   }
+   if (x) *x = j->s->img_x;
+   if (y) *y = j->s->img_y;
+   if (comp) *comp = j->s->img_n;
+   return 1;
+}
+
+static int stbi__jpeg_info(stbi__context *s, int *x, int *y, int *comp)
+{
+   stbi__jpeg j;
+   j.s = s;
+   return stbi__jpeg_info_raw(&j, x, y, comp);
+}
+
+// public domain zlib stbi__jpeg_huff_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
+
+// fast-way is faster to check than jpeg huffman, but slow way is slower
+#define STBI__ZFAST_BITS  9 // accelerate all cases in default tables
+#define STBI__ZFAST_MASK  ((1 << STBI__ZFAST_BITS) - 1)
+
+// zlib-style huffman encoding
+// (jpegs packs from left, zlib from right, so can't share code)
+typedef struct
+{
+   stbi__uint16 fast[1 << STBI__ZFAST_BITS];
+   stbi__uint16 firstcode[16];
+   int maxcode[17];
+   stbi__uint16 firstsymbol[16];
+   stbi_uc  size[288];
+   stbi__uint16 value[288]; 
+} stbi__zhuffman;
+
+stbi_inline static int stbi__bitreverse16(int n)
+{
+  n = ((n & 0xAAAA) >>  1) | ((n & 0x5555) << 1);
+  n = ((n & 0xCCCC) >>  2) | ((n & 0x3333) << 2);
+  n = ((n & 0xF0F0) >>  4) | ((n & 0x0F0F) << 4);
+  n = ((n & 0xFF00) >>  8) | ((n & 0x00FF) << 8);
+  return n;
+}
+
+stbi_inline static int stbi__bit_reverse(int v, int bits)
+{
+   assert(bits <= 16);
+   // to bit reverse n bits, reverse 16 and shift
+   // stbi__err.g. 11 bits, bit reverse and shift away 5
+   return stbi__bitreverse16(v) >> (16-bits);
+}
+
+static int stbi__zbuild_huffman(stbi__zhuffman *z, stbi_uc *sizelist, int num)
+{
+   int i,k=0;
+   int code, next_code[16], sizes[17];
+
+   // DEFLATE spec for generating codes
+   memset(sizes, 0, sizeof(sizes));
+   memset(z->fast, 255, sizeof(z->fast));
+   for (i=0; i < num; ++i) 
+      ++sizes[sizelist[i]];
+   sizes[0] = 0;
+   for (i=1; i < 16; ++i)
+      assert(sizes[i] <= (1 << i));
+   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 JPEG");
+      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];
+         z->size [c] = (stbi_uc     ) s;
+         z->value[c] = (stbi__uint16) i;
+         if (s <= STBI__ZFAST_BITS) {
+            int k = stbi__bit_reverse(next_code[s],s);
+            while (k < (1 << STBI__ZFAST_BITS)) {
+               z->fast[k] = (stbi__uint16) c;
+               k += (1 << s);
+            }
+         }
+         ++next_code[s];
+      }
+   }
+   return 1;
+}
+
+// zlib-from-memory implementation for PNG reading
+//    because PNG allows splitting the zlib stream arbitrarily,
+//    and it's annoying structurally to have PNG call ZLIB call PNG,
+//    we require PNG read all the IDATs and combine them into a single
+//    memory buffer
+
+typedef struct
+{
+   stbi_uc *zbuffer, *zbuffer_end;
+   int num_bits;
+   stbi__uint32 code_buffer;
+
+   char *zout;
+   char *zout_start;
+   char *zout_end;
+   int   z_expandable;
+
+   stbi__zhuffman z_length, z_distance;
+} stbi__zbuf;
+
+stbi_inline static stbi_uc stbi__zget8(stbi__zbuf *z)
+{
+   if (z->zbuffer >= z->zbuffer_end) return 0;
+   return *z->zbuffer++;
+}
+
+static void stbi__fill_bits(stbi__zbuf *z)
+{
+   do {
+      assert(z->code_buffer < (1U << z->num_bits));
+      z->code_buffer |= 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;   
+}
+
+stbi_inline static int stbi__zhuffman_decode(stbi__zbuf *a, stbi__zhuffman *z)
+{
+   int b,s,k;
+   if (a->num_bits < 16) stbi__fill_bits(a);
+   b = z->fast[a->code_buffer & STBI__ZFAST_MASK];
+   if (b < 0xffff) {
+      s = z->size[b];
+      a->code_buffer >>= s;
+      a->num_bits -= s;
+      return z->value[b];
+   }
+
+   // 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];
+   assert(z->size[b] == s);
+   a->code_buffer >>= s;
+   a->num_bits -= s;
+   return z->value[b];
+}
+
+static int stbi__zexpand(stbi__zbuf *z, int n)  // need to make room for n bytes
+{
+   char *q;
+   int cur, limit;
+   if (!z->z_expandable) return stbi__err("output buffer limit","Corrupt PNG");
+   cur   = (int) (z->zout     - z->zout_start);
+   limit = (int) (z->zout_end - z->zout_start);
+   while (cur + n > limit)
+      limit *= 2;
+   q = (char *) realloc(z->zout_start, limit);
+   if (q == NULL) return stbi__err("outofmem", "Out of memory");
+   z->zout_start = q;
+   z->zout       = q + cur;
+   z->zout_end   = q + limit;
+   return 1;
+}
+
+static int stbi__zlength_base[31] = {
+   3,4,5,6,7,8,9,10,11,13,
+   15,17,19,23,27,31,35,43,51,59,
+   67,83,99,115,131,163,195,227,258,0,0 };
+
+static int stbi__zlength_extra[31]= 
+{ 0,0,0,0,0,0,0,0,1,1,1,1,2,2,2,2,3,3,3,3,4,4,4,4,5,5,5,5,0,0,0 };
+
+static int stbi__zdist_base[32] = { 1,2,3,4,5,7,9,13,17,25,33,49,65,97,129,193,
+257,385,513,769,1025,1537,2049,3073,4097,6145,8193,12289,16385,24577,0,0};
+
+static int stbi__zdist_extra[32] =
+{ 0,0,0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,8,8,9,9,10,10,11,11,12,12,13,13};
+
+static int stbi__parse_huffman_block(stbi__zbuf *a)
+{
+   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 (a->zout >= a->zout_end) if (!stbi__zexpand(a, 1)) return 0;
+         *a->zout++ = (char) z;
+      } else {
+         stbi_uc *p;
+         int len,dist;
+         if (z == 256) return 1;
+         z -= 257;
+         len = stbi__zlength_base[z];
+         if (stbi__zlength_extra[z]) len += stbi__zreceive(a, stbi__zlength_extra[z]);
+         z = stbi__zhuffman_decode(a, &a->z_distance);
+         if (z < 0) return stbi__err("bad huffman code","Corrupt PNG");
+         dist = stbi__zdist_base[z];
+         if (stbi__zdist_extra[z]) dist += stbi__zreceive(a, stbi__zdist_extra[z]);
+         if (a->zout - a->zout_start < dist) return stbi__err("bad dist","Corrupt PNG");
+         if (a->zout + len > a->zout_end) if (!stbi__zexpand(a, len)) return 0;
+         p = (stbi_uc *) (a->zout - dist);
+         while (len--)
+            *a->zout++ = *p++;
+      }
+   }
+}
+
+static int stbi__compute_huffman_codes(stbi__zbuf *a)
+{
+   static stbi_uc length_dezigzag[19] = { 16,17,18,0,8,7,9,6,10,5,11,4,12,3,13,2,14,1,15 };
+   stbi__zhuffman z_codelength;
+   stbi_uc lencodes[286+32+137];//padding for maximum single op
+   stbi_uc codelength_sizes[19];
+   int i,n;
+
+   int hlit  = stbi__zreceive(a,5) + 257;
+   int hdist = stbi__zreceive(a,5) + 1;
+   int hclen = stbi__zreceive(a,4) + 4;
+
+   memset(codelength_sizes, 0, sizeof(codelength_sizes));
+   for (i=0; i < hclen; ++i) {
+      int s = stbi__zreceive(a,3);
+      codelength_sizes[length_dezigzag[i]] = (stbi_uc) s;
+   }
+   if (!stbi__zbuild_huffman(&z_codelength, codelength_sizes, 19)) return 0;
+
+   n = 0;
+   while (n < hlit + hdist) {
+      int c = stbi__zhuffman_decode(a, &z_codelength);
+      assert(c >= 0 && c < 19);
+      if (c < 16)
+         lencodes[n++] = (stbi_uc) c;
+      else if (c == 16) {
+         c = stbi__zreceive(a,2)+3;
+         memset(lencodes+n, lencodes[n-1], c);
+         n += c;
+      } else if (c == 17) {
+         c = stbi__zreceive(a,3)+3;
+         memset(lencodes+n, 0, c);
+         n += c;
+      } else {
+         assert(c == 18);
+         c = stbi__zreceive(a,7)+11;
+         memset(lencodes+n, 0, c);
+         n += c;
+      }
+   }
+   if (n != hlit+hdist) return stbi__err("bad codelengths","Corrupt PNG");
+   if (!stbi__zbuild_huffman(&a->z_length, lencodes, hlit)) return 0;
+   if (!stbi__zbuild_huffman(&a->z_distance, lencodes+hlit, hdist)) return 0;
+   return 1;
+}
+
+static int stbi__parse_uncomperssed_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;
+   }
+   assert(a->num_bits == 0);
+   // now fill header the normal way
+   while (k < 4)
+      header[k++] = stbi__zget8(a);
+   len  = header[1] * 256 + header[0];
+   nlen = header[3] * 256 + header[2];
+   if (nlen != (len ^ 0xffff)) return stbi__err("zlib corrupt","Corrupt PNG");
+   if (a->zbuffer + len > a->zbuffer_end) return stbi__err("read past buffer","Corrupt PNG");
+   if (a->zout + len > a->zout_end)
+      if (!stbi__zexpand(a, len)) return 0;
+   memcpy(a->zout, a->zbuffer, len);
+   a->zbuffer += len;
+   a->zout += len;
+   return 1;
+}
+
+static int stbi__parse_zlib_header(stbi__zbuf *a)
+{
+   int cmf   = stbi__zget8(a);
+   int cm    = cmf & 15;
+   /* int cinfo = cmf >> 4; */
+   int flg   = stbi__zget8(a);
+   if ((cmf*256+flg) % 31 != 0) return stbi__err("bad zlib header","Corrupt PNG"); // zlib spec
+   if (flg & 32) return stbi__err("no preset dict","Corrupt PNG"); // preset dictionary not allowed in png
+   if (cm != 8) return stbi__err("bad compression","Corrupt PNG"); // DEFLATE required for png
+   // window = 1 << (8 + cinfo)... but who cares, we fully buffer output
+   return 1;
+}
+
+// @TODO: should statically initialize these for optimal thread safety
+static stbi_uc stbi__zdefault_length[288], stbi__zdefault_distance[32];
+static void stbi__init_zdefaults(void)
+{
+   int i;   // use <= to match clearly with spec
+   for (i=0; i <= 143; ++i)     stbi__zdefault_length[i]   = 8;
+   for (   ; i <= 255; ++i)     stbi__zdefault_length[i]   = 9;
+   for (   ; i <= 279; ++i)     stbi__zdefault_length[i]   = 7;
+   for (   ; i <= 287; ++i)     stbi__zdefault_length[i]   = 8;
+
+   for (i=0; i <=  31; ++i)     stbi__zdefault_distance[i] = 5;
+}
+
+static int stbi__parse_zlib(stbi__zbuf *a, int parse_header)
+{
+   int final, type;
+   if (parse_header)
+      if (!stbi__parse_zlib_header(a)) return 0;
+   a->num_bits = 0;
+   a->code_buffer = 0;
+   do {
+      final = stbi__zreceive(a,1);
+      type = stbi__zreceive(a,2);
+      if (type == 0) {
+         if (!stbi__parse_uncomperssed_block(a)) return 0;
+      } else if (type == 3) {
+         return 0;
+      } else {
+         if (type == 1) {
+            // use fixed code lengths
+            if (!stbi__zdefault_distance[31]) stbi__init_zdefaults();
+            if (!stbi__zbuild_huffman(&a->z_length  , stbi__zdefault_length  , 288)) return 0;
+            if (!stbi__zbuild_huffman(&a->z_distance, stbi__zdefault_distance,  32)) return 0;
+         } else {
+            if (!stbi__compute_huffman_codes(a)) return 0;
+         }
+         if (!stbi__parse_huffman_block(a)) return 0;
+      }
+   } while (!final);
+   return 1;
+}
+
+static int stbi__do_zlib(stbi__zbuf *a, char *obuf, int olen, int exp, int parse_header)
+{
+   a->zout_start = obuf;
+   a->zout       = obuf;
+   a->zout_end   = obuf + olen;
+   a->z_expandable = exp;
+
+   return stbi__parse_zlib(a, parse_header);
+}
+
+STBIDEF char *stbi_zlib_decode_malloc_guesssize(const char *buffer, int len, int initial_size, int *outlen)
+{
+   stbi__zbuf a;
+   char *p = (char *) 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 {
+      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 *) 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 {
+      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 *) 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 {
+      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;
+}
+
+// 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
+
+
+typedef struct
+{
+   stbi__uint32 length;
+   stbi__uint32 type;
+} stbi__pngchunk;
+
+#define PNG_TYPE(a,b,c,d)  (((a) << 24) + ((b) << 16) + ((c) << 8) + (d))
+
+static stbi__pngchunk stbi__get_chunk_header(stbi__context *s)
+{
+   stbi__pngchunk c;
+   c.length = stbi__get32be(s);
+   c.type   = stbi__get32be(s);
+   return c;
+}
+
+static int stbi__check_png_header(stbi__context *s)
+{
+   static stbi_uc png_sig[8] = { 137,80,78,71,13,10,26,10 };
+   int i;
+   for (i=0; i < 8; ++i)
+      if (stbi__get8(s) != png_sig[i]) return stbi__err("bad png sig","Not a PNG");
+   return 1;
+}
+
+typedef struct
+{
+   stbi__context *s;
+   stbi_uc *idata, *expanded, *out;
+} stbi__png;
+
+
+enum {
+   STBI__F_none=0, STBI__F_sub=1, STBI__F_up=2, STBI__F_avg=3, STBI__F_paeth=4,
+   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;
+}
+
+#define STBI__BYTECAST(x)  ((stbi_uc) ((x) & 255))  // truncate int to byte without warnings
+
+// 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)
+{
+   stbi__context *s = a->s;
+   stbi__uint32 i,j,stride = x*out_n;
+   int k;
+   int img_n = s->img_n; // copy it into a local for later
+   assert(out_n == s->img_n || out_n == s->img_n+1);
+   a->out = (stbi_uc *) malloc(x * y * out_n);
+   if (!a->out) return stbi__err("outofmem", "Out of memory");
+   if (s->img_x == x && s->img_y == y) {
+      if (raw_len != (img_n * x + 1) * y) return stbi__err("not enough pixels","Corrupt PNG");
+   } else { // interlaced:
+      if (raw_len < (img_n * x + 1) * y) return stbi__err("not enough pixels","Corrupt PNG");
+   }
+   for (j=0; j < y; ++j) {
+      stbi_uc *cur = a->out + stride*j;
+      stbi_uc *prior = cur - stride;
+      int filter = *raw++;
+      if (filter > 4) return stbi__err("invalid filter","Corrupt PNG");
+      // if first row, use special filter that doesn't sample previous row
+      if (j == 0) filter = first_row_filter[filter];
+      // handle first pixel explicitly
+      for (k=0; k < img_n; ++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 (img_n != out_n) cur[img_n] = 255;
+      raw += img_n;
+      cur += out_n;
+      prior += out_n;
+      // this is a little gross, so that we don't switch per-pixel or per-component
+      if (img_n == out_n) {
+         #define CASE(f) \
+             case f:     \
+                for (i=x-1; i >= 1; --i, raw+=img_n,cur+=img_n,prior+=img_n) \
+                   for (k=0; k < img_n; ++k)
+         switch (filter) {
+            CASE(STBI__F_none)         cur[k] = raw[k]; break;
+            CASE(STBI__F_sub)          cur[k] = STBI__BYTECAST(raw[k] + cur[k-img_n]); break;
+            CASE(STBI__F_up)           cur[k] = STBI__BYTECAST(raw[k] + prior[k]); break;
+            CASE(STBI__F_avg)          cur[k] = STBI__BYTECAST(raw[k] + ((prior[k] + cur[k-img_n])>>1)); break;
+            CASE(STBI__F_paeth)        cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k-img_n],prior[k],prior[k-img_n])); break;
+            CASE(STBI__F_avg_first)    cur[k] = STBI__BYTECAST(raw[k] + (cur[k-img_n] >> 1)); break;
+            CASE(STBI__F_paeth_first)  cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k-img_n],0,0)); break;
+         }
+         #undef CASE
+      } else {
+         assert(img_n+1 == out_n);
+         #define CASE(f) \
+             case f:     \
+                for (i=x-1; i >= 1; --i, cur[img_n]=255,raw+=img_n,cur+=out_n,prior+=out_n) \
+                   for (k=0; k < img_n; ++k)
+         switch (filter) {
+            CASE(STBI__F_none)         cur[k] = raw[k]; break;
+            CASE(STBI__F_sub)          cur[k] = STBI__BYTECAST(raw[k] + cur[k-out_n]); break;
+            CASE(STBI__F_up)           cur[k] = STBI__BYTECAST(raw[k] + prior[k]); break;
+            CASE(STBI__F_avg)          cur[k] = STBI__BYTECAST(raw[k] + ((prior[k] + cur[k-out_n])>>1)); break;
+            CASE(STBI__F_paeth)        cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k-out_n],prior[k],prior[k-out_n])); break;
+            CASE(STBI__F_avg_first)    cur[k] = STBI__BYTECAST(raw[k] + (cur[k-out_n] >> 1)); break;
+            CASE(STBI__F_paeth_first)  cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k-out_n],0,0)); break;
+         }
+         #undef CASE
+      }
+   }
+   return 1;
+}
+
+static int stbi__create_png_image(stbi__png *a, stbi_uc *raw, stbi__uint32 raw_len, int out_n, int interlaced)
+{
+   stbi_uc *final;
+   int p;
+   if (!interlaced)
+      return stbi__create_png_image_raw(a, raw, raw_len, out_n, a->s->img_x, a->s->img_y);
+
+   // de-interlacing
+   final = (stbi_uc *) malloc(a->s->img_x * a->s->img_y * out_n);
+   for (p=0; p < 7; ++p) {
+      int xorig[] = { 0,4,0,2,0,1,0 };
+      int yorig[] = { 0,0,4,0,2,0,1 };
+      int xspc[]  = { 8,8,4,4,2,2,1 };
+      int yspc[]  = { 8,8,8,4,4,2,2 };
+      int i,j,x,y;
+      // pass1_x[4] = 0, pass1_x[5] = 1, pass1_x[12] = 1
+      x = (a->s->img_x - xorig[p] + xspc[p]-1) / xspc[p];
+      y = (a->s->img_y - yorig[p] + yspc[p]-1) / yspc[p];
+      if (x && y) {
+         if (!stbi__create_png_image_raw(a, raw, raw_len, out_n, x, y)) {
+            free(final);
+            return 0;
+         }
+         for (j=0; j < y; ++j)
+            for (i=0; i < x; ++i)
+               memcpy(final + (j*yspc[p]+yorig[p])*a->s->img_x*out_n + (i*xspc[p]+xorig[p])*out_n,
+                      a->out + (j*x+i)*out_n, out_n);
+         free(a->out);
+         raw += (x*out_n+1)*y;
+         raw_len -= (x*out_n+1)*y;
+      }
+   }
+   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
+   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__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 *) malloc(pixel_count * pal_img_n);
+   if (p == NULL) return stbi__err("outofmem", "Out of memory");
+
+   // between here and free(out) below, exitting would leak
+   temp_out = p;
+
+   if (pal_img_n == 3) {
+      for (i=0; i < pixel_count; ++i) {
+         int n = orig[i]*4;
+         p[0] = palette[n  ];
+         p[1] = palette[n+1];
+         p[2] = palette[n+2];
+         p += 3;
+      }
+   } else {
+      for (i=0; i < pixel_count; ++i) {
+         int n = orig[i]*4;
+         p[0] = palette[n  ];
+         p[1] = palette[n+1];
+         p[2] = palette[n+2];
+         p[3] = palette[n+3];
+         p += 4;
+      }
+   }
+   free(a->out);
+   a->out = temp_out;
+
+   STBI_NOTUSED(len);
+
+   return 1;
+}
+
+static int stbi__unpremultiply_on_load = 0;
+static int stbi__de_iphone_flag = 0;
+
+STBIDEF void stbi_set_unpremultiply_on_load(int flag_true_if_should_unpremultiply)
+{
+   stbi__unpremultiply_on_load = flag_true_if_should_unpremultiply;
+}
+
+STBIDEF void stbi_convert_iphone_png_to_rgb(int flag_true_if_should_convert)
+{
+   stbi__de_iphone_flag = flag_true_if_should_convert;
+}
+
+static void stbi__de_iphone(stbi__png *z)
+{
+   stbi__context *s = z->s;
+   stbi__uint32 i, pixel_count = s->img_x * s->img_y;
+   stbi_uc *p = z->out;
+
+   if (s->img_out_n == 3) {  // convert bgr to rgb
+      for (i=0; i < pixel_count; ++i) {
+         stbi_uc t = p[0];
+         p[0] = p[2];
+         p[2] = t;
+         p += 3;
+      }
+   } else {
+      assert(s->img_out_n == 4);
+      if (stbi__unpremultiply_on_load) {
+         // convert bgr to rgb and unpremultiply
+         for (i=0; i < pixel_count; ++i) {
+            stbi_uc a = p[3];
+            stbi_uc t = p[0];
+            if (a) {
+               p[0] = p[2] * 255 / a;
+               p[1] = p[1] * 255 / a;
+               p[2] =  t   * 255 / a;
+            } else {
+               p[0] = p[2];
+               p[2] = t;
+            } 
+            p += 4;
+         }
+      } else {
+         // convert bgr to rgb
+         for (i=0; i < pixel_count; ++i) {
+            stbi_uc t = p[0];
+            p[0] = p[2];
+            p[2] = t;
+            p += 4;
+         }
+      }
+   }
+}
+
+static int stbi__parse_png_file(stbi__png *z, int scan, int req_comp)
+{
+   stbi_uc palette[1024], pal_img_n=0;
+   stbi_uc has_trans=0, tc[3];
+   stbi__uint32 ioff=0, idata_limit=0, i, pal_len=0;
+   int first=1,k,interlace=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 == SCAN_type) return 1;
+
+   for (;;) {
+      stbi__pngchunk c = stbi__get_chunk_header(s);
+      switch (c.type) {
+         case PNG_TYPE('C','g','B','I'):
+            is_iphone = 1;
+            stbi__skip(s, c.length);
+            break;
+         case PNG_TYPE('I','H','D','R'): {
+            int depth,color,comp,filter;
+            if (!first) return stbi__err("multiple IHDR","Corrupt PNG");
+            first = 0;
+            if (c.length != 13) return stbi__err("bad IHDR len","Corrupt PNG");
+            s->img_x = stbi__get32be(s); if (s->img_x > (1 << 24)) return stbi__err("too large","Very large image (corrupt?)");
+            s->img_y = stbi__get32be(s); if (s->img_y > (1 << 24)) return stbi__err("too large","Very large image (corrupt?)");
+            depth = stbi__get8(s);  if (depth != 8)        return stbi__err("8bit only","PNG not supported: 8-bit only");
+            color = stbi__get8(s);  if (color > 6)         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 stbi__jpeg_huff_decode");
+               if (scan == SCAN_header) return 1;
+            } else {
+               // if paletted, then pal_n is our final components, and
+               // img_n is # components to decompress/filter.
+               s->img_n = 1;
+               if ((1 << 30) / s->img_x / 4 < s->img_y) return stbi__err("too large","Corrupt PNG");
+               // if SCAN_header, have to scan to see if we have a tRNS
+            }
+            break;
+         }
+
+         case 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 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 == 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;
+               for (k=0; k < s->img_n; ++k)
+                  tc[k] = (stbi_uc) (stbi__get16be(s) & 255); // non 8-bit images will be larger
+            }
+            break;
+         }
+
+         case 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 == SCAN_header) { s->img_n = pal_img_n; return 1; }
+            if (ioff + c.length > 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;
+               p = (stbi_uc *) realloc(z->idata, 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 PNG_TYPE('I','E','N','D'): {
+            stbi__uint32 raw_len;
+            if (first) return stbi__err("first not IHDR", "Corrupt PNG");
+            if (scan != SCAN_load) return 1;
+            if (z->idata == NULL) return stbi__err("no IDAT","Corrupt PNG");
+            z->expanded = (stbi_uc *) stbi_zlib_decode_malloc_guesssize_headerflag((char *) z->idata, ioff, 16384, (int *) &raw_len, !is_iphone);
+            if (z->expanded == NULL) return 0; // zlib should set error
+            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, interlace)) return 0;
+            if (has_trans)
+               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;
+            }
+            free(z->expanded); z->expanded = NULL;
+            return 1;
+         }
+
+         default:
+            // if critical, fail
+            if (first) return stbi__err("first not IHDR", "Corrupt PNG");
+            if ((c.type & (1 << 29)) == 0) {
+               #ifndef STBI_NO_FAILURE_STRINGS
+               // not threadsafe
+               static char invalid_chunk[] = "XXXX PNG chunk not known";
+               invalid_chunk[0] = STBI__BYTECAST(c.type >> 24);
+               invalid_chunk[1] = STBI__BYTECAST(c.type >> 16);
+               invalid_chunk[2] = STBI__BYTECAST(c.type >>  8);
+               invalid_chunk[3] = STBI__BYTECAST(c.type >>  0);
+               #endif
+               return stbi__err(invalid_chunk, "PNG not supported: unknown PNG chunk type");
+            }
+            stbi__skip(s, c.length);
+            break;
+      }
+      // end of PNG chunk, read and skip CRC
+      stbi__get32be(s);
+   }
+}
+
+static unsigned char *stbi__do_png(stbi__png *p, int *x, int *y, int *n, int req_comp)
+{
+   unsigned char *result=NULL;
+   if (req_comp < 0 || req_comp > 4) return stbi__errpuc("bad req_comp", "Internal error");
+   if (stbi__parse_png_file(p, SCAN_load, req_comp)) {
+      result = p->out;
+      p->out = NULL;
+      if (req_comp && req_comp != p->s->img_out_n) {
+         result = stbi__convert_format(result, p->s->img_out_n, req_comp, p->s->img_x, p->s->img_y);
+         p->s->img_out_n = req_comp;
+         if (result == NULL) return result;
+      }
+      *x = p->s->img_x;
+      *y = p->s->img_y;
+      if (n) *n = p->s->img_n;
+   }
+   free(p->out);      p->out      = NULL;
+   free(p->expanded); p->expanded = NULL;
+   free(p->idata);    p->idata    = NULL;
+
+   return result;
+}
+
+static unsigned char *stbi__png_load(stbi__context *s, int *x, int *y, int *comp, int req_comp)
+{
+   stbi__png p;
+   p.s = s;
+   return stbi__do_png(&p, x,y,comp,req_comp);
+}
+
+static int stbi__png_test(stbi__context *s)
+{
+   int r;
+   r = stbi__check_png_header(s);
+   stbi__rewind(s);
+   return r;
+}
+
+static int stbi__png_info_raw(stbi__png *p, int *x, int *y, int *comp)
+{
+   if (!stbi__parse_png_file(p, 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);
+}
+
+// Microsoft/Windows BMP image
+static int stbi__bmp_test_raw(stbi__context *s)
+{
+   int r;
+   int sz;
+   if (stbi__get8(s) != 'B') return 0;
+   if (stbi__get8(s) != 'M') return 0;
+   stbi__get32le(s); // discard filesize
+   stbi__get16le(s); // discard reserved
+   stbi__get16le(s); // discard reserved
+   stbi__get32le(s); // discard data offset
+   sz = stbi__get32le(s);
+   r = (sz == 12 || sz == 40 || sz == 56 || sz == 108 || sz == 124);
+   return r;
+}
+
+static int stbi__bmp_test(stbi__context *s)
+{
+   int r = stbi__bmp_test_raw(s);
+   stbi__rewind(s);
+   return r;
+}
+
+
+// returns 0..31 for the highest set bit
+static int stbi__high_bit(unsigned int z)
+{
+   int n=0;
+   if (z == 0) return -1;
+   if (z >= 0x10000) n += 16, z >>= 16;
+   if (z >= 0x00100) n +=  8, z >>=  8;
+   if (z >= 0x00010) n +=  4, z >>=  4;
+   if (z >= 0x00004) n +=  2, z >>=  2;
+   if (z >= 0x00002) n +=  1, z >>=  1;
+   return n;
+}
+
+static int stbi__bitcount(unsigned int a)
+{
+   a = (a & 0x55555555) + ((a >>  1) & 0x55555555); // max 2
+   a = (a & 0x33333333) + ((a >>  2) & 0x33333333); // max 4
+   a = (a + (a >> 4)) & 0x0f0f0f0f; // max 8 per 4, now 8 bits
+   a = (a + (a >> 8)); // max 16 per 8 bits
+   a = (a + (a >> 16)); // max 32 per 8 bits
+   return a & 0xff;
+}
+
+static int stbi__shiftsigned(int v, int shift, int bits)
+{
+   int result;
+   int z=0;
+
+   if (shift < 0) v <<= -shift;
+   else v >>= shift;
+   result = v;
+
+   z = bits;
+   while (z < 8) {
+      result += v >> z;
+      z += bits;
+   }
+   return result;
+}
+
+static stbi_uc *stbi__bmp_load(stbi__context *s, int *x, int *y, int *comp, int req_comp)
+{
+   stbi_uc *out;
+   unsigned int mr=0,mg=0,mb=0,ma=0, fake_a=0;
+   stbi_uc pal[256][4];
+   int psize=0,i,j,compress=0,width;
+   int bpp, flip_vertically, pad, target, offset, 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
+   offset = stbi__get32le(s);
+   hsz = stbi__get32le(s);
+   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");
+   bpp = stbi__get16le(s);
+   if (bpp == 1) return stbi__errpuc("monochrome", "BMP type not supported: 1-bit");
+   flip_vertically = ((int) s->img_y) > 0;
+   s->img_y = abs((int) s->img_y);
+   if (hsz == 12) {
+      if (bpp < 24)
+         psize = (offset - 14 - 24) / 3;
+   } else {
+      compress = stbi__get32le(s);
+      if (compress == 1 || compress == 2) return stbi__errpuc("BMP RLE", "BMP type not supported: RLE");
+      stbi__get32le(s); // discard sizeof
+      stbi__get32le(s); // discard hres
+      stbi__get32le(s); // discard vres
+      stbi__get32le(s); // discard colorsused
+      stbi__get32le(s); // discard max important
+      if (hsz == 40 || hsz == 56) {
+         if (hsz == 56) {
+            stbi__get32le(s);
+            stbi__get32le(s);
+            stbi__get32le(s);
+            stbi__get32le(s);
+         }
+         if (bpp == 16 || bpp == 32) {
+            mr = mg = mb = 0;
+            if (compress == 0) {
+               if (bpp == 32) {
+                  mr = 0xffu << 16;
+                  mg = 0xffu <<  8;
+                  mb = 0xffu <<  0;
+                  ma = 0xffu << 24;
+                  fake_a = 1; // @TODO: check for cases like alpha value is all 0 and switch it to 255
+                  STBI_NOTUSED(fake_a);
+               } else {
+                  mr = 31u << 10;
+                  mg = 31u <<  5;
+                  mb = 31u <<  0;
+               }
+            } else if (compress == 3) {
+               mr = stbi__get32le(s);
+               mg = stbi__get32le(s);
+               mb = stbi__get32le(s);
+               // not documented, but generated by photoshop and handled by mspaint
+               if (mr == mg && mg == mb) {
+                  // ?!?!?
+                  return stbi__errpuc("bad BMP", "bad BMP");
+               }
+            } else
+               return stbi__errpuc("bad BMP", "bad BMP");
+         }
+      } else {
+         assert(hsz == 108 || hsz == 124);
+         mr = stbi__get32le(s);
+         mg = stbi__get32le(s);
+         mb = stbi__get32le(s);
+         ma = stbi__get32le(s);
+         stbi__get32le(s); // discard color space
+         for (i=0; i < 12; ++i)
+            stbi__get32le(s); // discard color space parameters
+         if (hsz == 124) {
+            stbi__get32le(s); // discard rendering intent
+            stbi__get32le(s); // discard offset of profile data
+            stbi__get32le(s); // discard size of profile data
+            stbi__get32le(s); // discard reserved
+         }
+      }
+      if (bpp < 16)
+         psize = (offset - 14 - hsz) >> 2;
+   }
+   s->img_n = ma ? 4 : 3;
+   if (req_comp && req_comp >= 3) // we can directly stbi__jpeg_huff_decode 3 or 4
+      target = req_comp;
+   else
+      target = s->img_n; // if they want monochrome, we'll post-convert
+   out = (stbi_uc *) malloc(target * s->img_x * s->img_y);
+   if (!out) return stbi__errpuc("outofmem", "Out of memory");
+   if (bpp < 16) {
+      int z=0;
+      if (psize == 0 || psize > 256) { 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 (hsz != 12) stbi__get8(s);
+         pal[i][3] = 255;
+      }
+      stbi__skip(s, offset - 14 - hsz - psize * (hsz == 12 ? 3 : 4));
+      if (bpp == 4) width = (s->img_x + 1) >> 1;
+      else if (bpp == 8) width = s->img_x;
+      else { free(out); return stbi__errpuc("bad bpp", "Corrupt BMP"); }
+      pad = (-width)&3;
+      for (j=0; j < (int) s->img_y; ++j) {
+         for (i=0; i < (int) s->img_x; i += 2) {
+            int v=stbi__get8(s),v2=0;
+            if (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 = (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, offset - 14 - hsz);
+      if (bpp == 24) width = 3 * s->img_x;
+      else if (bpp == 16) width = 2*s->img_x;
+      else /* bpp = 32 and pad = 0 */ width=0;
+      pad = (-width) & 3;
+      if (bpp == 24) {
+         easy = 1;
+      } else if (bpp == 32) {
+         if (mb == 0xff && mg == 0xff00 && mr == 0x00ff0000 && ma == 0xff000000)
+            easy = 2;
+      }
+      if (!easy) {
+         if (!mr || !mg || !mb) { free(out); return stbi__errpuc("bad masks", "Corrupt BMP"); }
+         // right shift amt to put high bit in position #7
+         rshift = stbi__high_bit(mr)-7; rcount = stbi__bitcount(mr);
+         gshift = stbi__high_bit(mg)-7; gcount = stbi__bitcount(mg);
+         bshift = stbi__high_bit(mb)-7; bcount = stbi__bitcount(mb);
+         ashift = stbi__high_bit(ma)-7; acount = stbi__bitcount(ma);
+      }
+      for (j=0; j < (int) s->img_y; ++j) {
+         if (easy) {
+            for (i=0; i < (int) s->img_x; ++i) {
+               unsigned char a;
+               out[z+2] = stbi__get8(s);
+               out[z+1] = stbi__get8(s);
+               out[z+0] = stbi__get8(s);
+               z += 3;
+               a = (easy == 2 ? stbi__get8(s) : 255);
+               if (target == 4) out[z++] = a;
+            }
+         } else {
+            for (i=0; i < (int) s->img_x; ++i) {
+               stbi__uint32 v = (stbi__uint32) (bpp == 16 ? stbi__get16le(s) : stbi__get32le(s));
+               int a;
+               out[z++] = STBI__BYTECAST(stbi__shiftsigned(v & mr, rshift, rcount));
+               out[z++] = STBI__BYTECAST(stbi__shiftsigned(v & mg, gshift, gcount));
+               out[z++] = STBI__BYTECAST(stbi__shiftsigned(v & mb, bshift, bcount));
+               a = (ma ? stbi__shiftsigned(v & ma, ashift, acount) : 255);
+               if (target == 4) out[z++] = STBI__BYTECAST(a); 
+            }
+         }
+         stbi__skip(s, pad);
+      }
+   }
+   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;
+}
+
+// Targa Truevision - TGA
+// by Jonathan Dummer
+
+static int stbi__tga_info(stbi__context *s, int *x, int *y, int *comp)
+{
+    int tga_w, tga_h, tga_comp;
+    int sz;
+    stbi__get8(s);                   // discard Offset
+    sz = stbi__get8(s);              // color type
+    if( sz > 1 ) {
+        stbi__rewind(s);
+        return 0;      // only RGB or indexed allowed
+    }
+    sz = stbi__get8(s);              // image type
+    // only RGB or grey allowed, +/- RLE
+    if ((sz != 1) && (sz != 2) && (sz != 3) && (sz != 9) && (sz != 10) && (sz != 11)) return 0;
+    stbi__skip(s,9);
+    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
+    }
+    sz = stbi__get8(s);               // bits per pixel
+    // only RGB or RGBA or grey allowed
+    if ((sz != 8) && (sz != 16) && (sz != 24) && (sz != 32)) {
+        stbi__rewind(s);
+        return 0;
+    }
+    tga_comp = sz;
+    if (x) *x = tga_w;
+    if (y) *y = tga_h;
+    if (comp) *comp = tga_comp / 8;
+    return 1;                   // seems to have passed everything
+}
+
+static int stbi__tga_test(stbi__context *s)
+{
+   int res;
+   int sz;
+   stbi__get8(s);      //   discard Offset
+   sz = stbi__get8(s);   //   color type
+   if ( sz > 1 ) return 0;   //   only RGB or indexed allowed
+   sz = stbi__get8(s);   //   image type
+   if ( (sz != 1) && (sz != 2) && (sz != 3) && (sz != 9) && (sz != 10) && (sz != 11) ) return 0;   //   only RGB or grey allowed, +/- RLE
+   stbi__get16be(s);      //   discard palette start
+   stbi__get16be(s);      //   discard palette length
+   stbi__get8(s);         //   discard bits per palette color entry
+   stbi__get16be(s);      //   discard x origin
+   stbi__get16be(s);      //   discard y origin
+   if ( stbi__get16be(s) < 1 ) return 0;      //   test width
+   if ( stbi__get16be(s) < 1 ) return 0;      //   test height
+   sz = stbi__get8(s);   //   bits per pixel
+   if ( (sz != 8) && (sz != 16) && (sz != 24) && (sz != 32) )
+      res = 0;
+   else
+      res = 1;
+   stbi__rewind(s);
+   return res;
+}
+
+static stbi_uc *stbi__tga_load(stbi__context *s, int *x, int *y, int *comp, int req_comp)
+{
+   //   read in the TGA header stuff
+   int tga_offset = stbi__get8(s);
+   int tga_indexed = stbi__get8(s);
+   int tga_image_type = stbi__get8(s);
+   int tga_is_RLE = 0;
+   int tga_palette_start = stbi__get16le(s);
+   int tga_palette_len = stbi__get16le(s);
+   int tga_palette_bits = stbi__get8(s);
+   int tga_x_origin = stbi__get16le(s);
+   int tga_y_origin = stbi__get16le(s);
+   int tga_width = stbi__get16le(s);
+   int tga_height = stbi__get16le(s);
+   int tga_bits_per_pixel = stbi__get8(s);
+   int tga_comp = tga_bits_per_pixel / 8;
+   int tga_inverted = stbi__get8(s);
+   //   image data
+   unsigned char *tga_data;
+   unsigned char *tga_palette = NULL;
+   int i, j;
+   unsigned char raw_data[4];
+   int RLE_count = 0;
+   int RLE_repeating = 0;
+   int read_next_pixel = 1;
+
+   //   do a tiny bit of precessing
+   if ( tga_image_type >= 8 )
+   {
+      tga_image_type -= 8;
+      tga_is_RLE = 1;
+   }
+   /* int tga_alpha_bits = tga_inverted & 15; */
+   tga_inverted = 1 - ((tga_inverted >> 5) & 1);
+
+   //   error check
+   if ( //(tga_indexed) ||
+      (tga_width < 1) || (tga_height < 1) ||
+      (tga_image_type < 1) || (tga_image_type > 3) ||
+      ((tga_bits_per_pixel != 8) && (tga_bits_per_pixel != 16) &&
+      (tga_bits_per_pixel != 24) && (tga_bits_per_pixel != 32))
+      )
+   {
+      return NULL; // we don't report this as a bad TGA because we don't even know if it's TGA
+   }
+
+   //   If I'm paletted, then I'll use the number of bits from the palette
+   if ( tga_indexed )
+   {
+      tga_comp = tga_palette_bits / 8;
+   }
+
+   //   tga info
+   *x = tga_width;
+   *y = tga_height;
+   if (comp) *comp = tga_comp;
+
+   tga_data = (unsigned char*)malloc( tga_width * tga_height * tga_comp );
+   if (!tga_data) return stbi__errpuc("outofmem", "Out of memory");
+
+   // skip to the data's starting position (offset usually = 0)
+   stbi__skip(s, tga_offset );
+
+   if ( !tga_indexed && !tga_is_RLE) {
+      for (i=0; i < tga_height; ++i) {
+         int y = tga_inverted ? tga_height -i - 1 : i;
+         stbi_uc *tga_row = tga_data + y*tga_width*tga_comp;
+         stbi__getn(s, tga_row, tga_width * tga_comp);
+      }
+   } else  {
+      //   do I need to load a palette?
+      if ( tga_indexed)
+      {
+         //   any data to skip? (offset usually = 0)
+         stbi__skip(s, tga_palette_start );
+         //   load the palette
+         tga_palette = (unsigned char*)malloc( tga_palette_len * tga_palette_bits / 8 );
+         if (!tga_palette) {
+            free(tga_data);
+            return stbi__errpuc("outofmem", "Out of memory");
+         }
+         if (!stbi__getn(s, tga_palette, tga_palette_len * tga_palette_bits / 8 )) {
+            free(tga_data);
+            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 1 byte, then perform the lookup
+               int pal_idx = stbi__get8(s);
+               if ( pal_idx >= tga_palette_len )
+               {
+                  //   invalid index
+                  pal_idx = 0;
+               }
+               pal_idx *= tga_bits_per_pixel / 8;
+               for (j = 0; j*8 < tga_bits_per_pixel; ++j)
+               {
+                  raw_data[j] = tga_palette[pal_idx+j];
+               }
+            } else
+            {
+               //   read in the data raw
+               for (j = 0; j*8 < tga_bits_per_pixel; ++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 )
+      {
+         free( tga_palette );
+      }
+   }
+
+   // swap RGB
+   if (tga_comp >= 3)
+   {
+      unsigned char* tga_pixel = tga_data;
+      for (i=0; i < tga_width * tga_height; ++i)
+      {
+         unsigned char temp = tga_pixel[0];
+         tga_pixel[0] = tga_pixel[2];
+         tga_pixel[2] = temp;
+         tga_pixel += tga_comp;
+      }
+   }
+
+   // convert to target component count
+   if (req_comp && req_comp != tga_comp)
+      tga_data = stbi__convert_format(tga_data, tga_comp, req_comp, tga_width, tga_height);
+
+   //   the things I do to get rid of an error message, and yet keep
+   //   Microsoft's C compilers happy... [8^(
+   tga_palette_start = tga_palette_len = tga_palette_bits =
+         tga_x_origin = tga_y_origin = 0;
+   //   OK, done
+   return tga_data;
+}
+
+// *************************************************************************************************
+// Photoshop PSD loader -- PD by Thatcher Ulrich, integration by Nicolas Schulz, tweaked by STB
+
+static int stbi__psd_test(stbi__context *s)
+{
+   int r = (stbi__get32be(s) == 0x38425053);
+   stbi__rewind(s);
+   return r;
+}
+
+static stbi_uc *stbi__psd_load(stbi__context *s, int *x, int *y, int *comp, int req_comp)
+{
+   int   pixelCount;
+   int channelCount, compression;
+   int channel, i, count, len;
+   int w,h;
+   stbi_uc *out;
+
+   // Check identifier
+   if (stbi__get32be(s) != 0x38425053)   // "8BPS"
+      return stbi__errpuc("not PSD", "Corrupt PSD image");
+
+   // Check file type version.
+   if (stbi__get16be(s) != 1)
+      return stbi__errpuc("wrong version", "Unsupported version of PSD image");
+
+   // Skip 6 reserved bytes.
+   stbi__skip(s, 6 );
+
+   // Read the number of channels (R, G, B, A, etc).
+   channelCount = stbi__get16be(s);
+   if (channelCount < 0 || channelCount > 16)
+      return stbi__errpuc("wrong channel count", "Unsupported number of channels in PSD image");
+
+   // Read the rows and columns of the image.
+   h = stbi__get32be(s);
+   w = stbi__get32be(s);
+   
+   // Make sure the depth is 8 bits.
+   if (stbi__get16be(s) != 8)
+      return stbi__errpuc("unsupported bit depth", "PSD bit depth is not 8 bit");
+
+   // Make sure the color mode is RGB.
+   // Valid options are:
+   //   0: Bitmap
+   //   1: Grayscale
+   //   2: Indexed color
+   //   3: RGB color
+   //   4: CMYK color
+   //   7: Multichannel
+   //   8: Duotone
+   //   9: Lab color
+   if (stbi__get16be(s) != 3)
+      return stbi__errpuc("wrong color format", "PSD is not in RGB color format");
+
+   // Skip the Mode Data.  (It's the palette for indexed color; other info for other modes.)
+   stbi__skip(s,stbi__get32be(s) );
+
+   // Skip the image resources.  (resolution, pen tool paths, etc)
+   stbi__skip(s, stbi__get32be(s) );
+
+   // Skip the reserved data.
+   stbi__skip(s, stbi__get32be(s) );
+
+   // Find out if the data is compressed.
+   // Known values:
+   //   0: no compression
+   //   1: RLE compressed
+   compression = stbi__get16be(s);
+   if (compression > 1)
+      return stbi__errpuc("bad compression", "PSD has an unknown compression format");
+
+   // Create the destination image.
+   out = (stbi_uc *) malloc(4 * w*h);
+   if (!out) return stbi__errpuc("outofmem", "Out of memory");
+   pixelCount = w*h;
+
+   // Initialize the data to zero.
+   //memset( out, 0, pixelCount * 4 );
+   
+   // Finally, the image data.
+   if (compression) {
+      // RLE as used by .PSD and .TIFF
+      // Loop until you get the number of unpacked bytes you are expecting:
+      //     Read the next source byte into n.
+      //     If n is between 0 and 127 inclusive, copy the next n+1 bytes literally.
+      //     Else if n is between -127 and -1 inclusive, copy the next byte -n+1 times.
+      //     Else if n is 128, noop.
+      // Endloop
+
+      // The RLE-compressed data is preceeded by a 2-byte data count for each row in the data,
+      // which we're going to just skip.
+      stbi__skip(s, h * channelCount * 2 );
+
+      // Read the RLE data by channel.
+      for (channel = 0; channel < 4; channel++) {
+         stbi_uc *p;
+         
+         p = out+channel;
+         if (channel >= channelCount) {
+            // Fill this channel with default data.
+            for (i = 0; i < pixelCount; i++) *p = (channel == 3 ? 255 : 0), p += 4;
+         } else {
+            // Read the RLE data.
+            count = 0;
+            while (count < pixelCount) {
+               len = stbi__get8(s);
+               if (len == 128) {
+                  // No-op.
+               } else if (len < 128) {
+                  // Copy next len+1 bytes literally.
+                  len++;
+                  count += len;
+                  while (len) {
+                     *p = stbi__get8(s);
+                     p += 4;
+                     len--;
+                  }
+               } else if (len > 128) {
+                  stbi_uc   val;
+                  // Next -len+1 bytes in the dest are replicated from next source byte.
+                  // (Interpret len as a negative 8-bit int.)
+                  len ^= 0x0FF;
+                  len += 2;
+                  val = stbi__get8(s);
+                  count += len;
+                  while (len) {
+                     *p = val;
+                     p += 4;
+                     len--;
+                  }
+               }
+            }
+         }
+      }
+      
+   } else {
+      // We're at the raw image data.  It's each channel in order (Red, Green, Blue, Alpha, ...)
+      // where each channel consists of an 8-bit value for each pixel in the image.
+      
+      // Read the data by channel.
+      for (channel = 0; channel < 4; channel++) {
+         stbi_uc *p;
+         
+         p = out + channel;
+         if (channel > channelCount) {
+            // Fill this channel with default data.
+            for (i = 0; i < pixelCount; i++) *p = channel == 3 ? 255 : 0, p += 4;
+         } else {
+            // Read the data.
+            for (i = 0; i < pixelCount; i++)
+               *p = stbi__get8(s), p += 4;
+         }
+      }
+   }
+
+   if (req_comp && req_comp != 4) {
+      out = stbi__convert_format(out, 4, req_comp, w, h);
+      if (out == NULL) return out; // stbi__convert_format frees input on failure
+   }
+
+   if (comp) *comp = channelCount;
+   *y = h;
+   *x = w;
+   
+   return out;
+}
+
+// *************************************************************************************************
+// 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/
+
+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];
+                     int i;
+
+                     if (count==128)
+                        count = stbi__get16be(s);
+                     else
+                        count -= 127;
+                     if (count > left)
+                        return stbi__errpuc("bad file","scanline overrun");
+
+                     if (!stbi__readval(s,packet->channel,value))
+                        return 0;
+
+                     for(i=0;i<count;++i, dest += 4)
+                        stbi__copyval(packet->channel,dest,value);
+                  } else { // Raw
+                     ++count;
+                     if (count>left) return stbi__errpuc("bad file","scanline overrun");
+
+                     for(i=0;i<count;++i, dest+=4)
+                        if (!stbi__readval(s,packet->channel,dest))
+                           return 0;
+                  }
+                  left-=count;
+               }
+               break;
+            }
+         }
+      }
+   }
+
+   return result;
+}
+
+static stbi_uc *stbi__pic_load(stbi__context *s,int *px,int *py,int *comp,int req_comp)
+{
+   stbi_uc *result;
+   int i, x,y;
+
+   for (i=0; i<92; ++i)
+      stbi__get8(s);
+
+   x = stbi__get16be(s);
+   y = stbi__get16be(s);
+   if (stbi__at_eof(s))  return stbi__errpuc("bad file","file too short (pic header)");
+   if ((1 << 28) / x < y) return stbi__errpuc("too large", "Image too large to stbi__jpeg_huff_decode");
+
+   stbi__get32be(s); //skip `ratio'
+   stbi__get16be(s); //skip `fields'
+   stbi__get16be(s); //skip `pad'
+
+   // intermediate buffer is RGBA
+   result = (stbi_uc *) malloc(x*y*4);
+   memset(result, 0xff, x*y*4);
+
+   if (!stbi__pic_load_core(s,x,y,comp, result)) {
+      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;
+}
+
+// *************************************************************************************************
+// GIF loader -- public domain by Jean-Marc Lienher -- simplified/shrunk by stb
+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)
+   int flags, bgindex, ratio, transparent, eflags;
+   stbi_uc  pal[256][4];
+   stbi_uc lpal[256][4];
+   stbi__gif_lzw codes[4096];
+   stbi_uc *color_table;
+   int parse, step;
+   int lflags;
+   int start_x, start_y;
+   int max_x, max_y;
+   int cur_x, cur_y;
+   int line_size;
+} stbi__gif;
+
+static int stbi__gif_test_raw(stbi__context *s)
+{
+   int sz;
+   if (stbi__get8(s) != 'G' || stbi__get8(s) != 'I' || stbi__get8(s) != 'F' || stbi__get8(s) != '8') return 0;
+   sz = stbi__get8(s);
+   if (sz != '9' && sz != '7') return 0;
+   if (stbi__get8(s) != 'a') return 0;
+   return 1;
+}
+
+static int stbi__gif_test(stbi__context *s)
+{
+   int r = stbi__gif_test_raw(s);
+   stbi__rewind(s);
+   return r;
+}
+
+static void stbi__gif_parse_colortable(stbi__context *s, stbi_uc pal[256][4], int num_entries, int transp)
+{
+   int i;
+   for (i=0; i < num_entries; ++i) {
+      pal[i][2] = stbi__get8(s);
+      pal[i][1] = stbi__get8(s);
+      pal[i][0] = stbi__get8(s);
+      pal[i][3] = transp ? 0 : 255;
+   }   
+}
+
+static int stbi__gif_header(stbi__context *s, stbi__gif *g, int *comp, int is_info)
+{
+   stbi_uc version;
+   if (stbi__get8(s) != 'G' || stbi__get8(s) != 'I' || stbi__get8(s) != 'F' || stbi__get8(s) != '8')
+      return stbi__err("not GIF", "Corrupt GIF");
+
+   version = stbi__get8(s);
+   if (version != '7' && version != '9')    return stbi__err("not GIF", "Corrupt GIF");
+   if (stbi__get8(s) != 'a')                      return stbi__err("not GIF", "Corrupt GIF");
+ 
+   stbi__g_failure_reason = "";
+   g->w = stbi__get16le(s);
+   g->h = stbi__get16le(s);
+   g->flags = stbi__get8(s);
+   g->bgindex = stbi__get8(s);
+   g->ratio = stbi__get8(s);
+   g->transparent = -1;
+
+   if (comp != 0) *comp = 4;  // can't actually tell whether it's 3 or 4 until we parse the comments
+
+   if (is_info) return 1;
+
+   if (g->flags & 0x80)
+      stbi__gif_parse_colortable(s,g->pal, 2 << (g->flags & 7), -1);
+
+   return 1;
+}
+
+static int stbi__gif_info_raw(stbi__context *s, int *x, int *y, int *comp)
+{
+   stbi__gif g;   
+   if (!stbi__gif_header(s, &g, comp, 1)) {
+      stbi__rewind( s );
+      return 0;
+   }
+   if (x) *x = g.w;
+   if (y) *y = g.h;
+   return 1;
+}
+
+static void stbi__out_gif_code(stbi__gif *g, stbi__uint16 code)
+{
+   stbi_uc *p, *c;
+
+   // recurse to stbi__jpeg_huff_decode the prefixes, since the linked-list is backwards,
+   // and working backwards through an interleaved image would be nasty
+   if (g->codes[code].prefix >= 0)
+      stbi__out_gif_code(g, g->codes[code].prefix);
+
+   if (g->cur_y >= g->max_y) return;
+  
+   p = &g->out[g->cur_x + g->cur_y];
+   c = &g->color_table[g->codes[code].suffix * 4];
+
+   if (c[3] >= 128) {
+      p[0] = c[2];
+      p[1] = c[1];
+      p[2] = c[0];
+      p[3] = c[3];
+   }
+   g->cur_x += 4;
+
+   if (g->cur_x >= g->max_x) {
+      g->cur_x = g->start_x;
+      g->cur_y += g->step;
+
+      while (g->cur_y >= g->max_y && g->parse > 0) {
+         g->step = (1 << g->parse) * g->line_size;
+         g->cur_y = g->start_y + (g->step >> 1);
+         --g->parse;
+      }
+   }
+}
+
+static stbi_uc *stbi__process_gif_raster(stbi__context *s, stbi__gif *g)
+{
+   stbi_uc lzw_cs;
+   stbi__int32 len, code;
+   stbi__uint32 first;
+   stbi__int32 codesize, codemask, avail, oldcode, bits, valid_bits, clear;
+   stbi__gif_lzw *p;
+
+   lzw_cs = stbi__get8(s);
+   clear = 1 << lzw_cs;
+   first = 1;
+   codesize = lzw_cs + 1;
+   codemask = (1 << codesize) - 1;
+   bits = 0;
+   valid_bits = 0;
+   for (code = 0; code < clear; code++) {
+      g->codes[code].prefix = -1;
+      g->codes[code].first = (stbi_uc) code;
+      g->codes[code].suffix = (stbi_uc) code;
+   }
+
+   // support no starting clear code
+   avail = clear+2;
+   oldcode = -1;
+
+   len = 0;
+   for(;;) {
+      if (valid_bits < codesize) {
+         if (len == 0) {
+            len = stbi__get8(s); // start new block
+            if (len == 0) 
+               return g->out;
+         }
+         --len;
+         bits |= (stbi__int32) stbi__get8(s) << valid_bits;
+         valid_bits += 8;
+      } else {
+         stbi__int32 code = bits & codemask;
+         bits >>= codesize;
+         valid_bits -= codesize;
+         // @OPTIMIZE: is there some way we can accelerate the non-clear path?
+         if (code == clear) {  // clear code
+            codesize = lzw_cs + 1;
+            codemask = (1 << codesize) - 1;
+            avail = clear + 2;
+            oldcode = -1;
+            first = 0;
+         } else if (code == clear + 1) { // end of stream code
+            stbi__skip(s, len);
+            while ((len = stbi__get8(s)) > 0)
+               stbi__skip(s,len);
+            return g->out;
+         } else if (code <= avail) {
+            if (first) return stbi__errpuc("no clear code", "Corrupt GIF");
+
+            if (oldcode >= 0) {
+               p = &g->codes[avail++];
+               if (avail > 4096)        return stbi__errpuc("too many codes", "Corrupt GIF");
+               p->prefix = (stbi__int16) oldcode;
+               p->first = g->codes[oldcode].first;
+               p->suffix = (code == avail) ? p->first : g->codes[code].first;
+            } else if (code == avail)
+               return stbi__errpuc("illegal code in raster", "Corrupt GIF");
+
+            stbi__out_gif_code(g, (stbi__uint16) code);
+
+            if ((avail & codemask) == 0 && avail <= 0x0FFF) {
+               codesize++;
+               codemask = (1 << codesize) - 1;
+            }
+
+            oldcode = code;
+         } else {
+            return stbi__errpuc("illegal code in raster", "Corrupt GIF");
+         }
+      } 
+   }
+}
+
+static void stbi__fill_gif_background(stbi__gif *g)
+{
+   int i;
+   stbi_uc *c = g->pal[g->bgindex];
+   // @OPTIMIZE: write a dword at a time
+   for (i = 0; i < g->w * g->h * 4; i += 4) {
+      stbi_uc *p  = &g->out[i];
+      p[0] = c[2];
+      p[1] = c[1];
+      p[2] = c[0];
+      p[3] = c[3];
+   }
+}
+
+// this function is designed to support animated gifs, although stb_image doesn't support it
+static stbi_uc *stbi__gif_load_next(stbi__context *s, stbi__gif *g, int *comp, int req_comp)
+{
+   int i;
+   stbi_uc *old_out = 0;
+
+   if (g->out == 0) {
+      if (!stbi__gif_header(s, g, comp,0))     return 0; // stbi__g_failure_reason set by stbi__gif_header
+      g->out = (stbi_uc *) malloc(4 * g->w * g->h);
+      if (g->out == 0)                      return stbi__errpuc("outofmem", "Out of memory");
+      stbi__fill_gif_background(g);
+   } else {
+      // animated-gif-only path
+      if (((g->eflags & 0x1C) >> 2) == 3) {
+         old_out = g->out;
+         g->out = (stbi_uc *) malloc(4 * g->w * g->h);
+         if (g->out == 0)                   return stbi__errpuc("outofmem", "Out of memory");
+         memcpy(g->out, old_out, g->w*g->h*4);
+      }
+   }
+    
+   for (;;) {
+      switch (stbi__get8(s)) {
+         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;
+
+            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) {
+               for (i=0; i < 256; ++i)  // @OPTIMIZE: stbi__jpeg_reset only the previous transparent
+                  g->pal[i][3] = 255; 
+               if (g->transparent >= 0 && (g->eflags & 0x01))
+                  g->pal[g->transparent][3] = 0;
+               g->color_table = (stbi_uc *) g->pal;
+            } else
+               return stbi__errpuc("missing color table", "Corrupt GIF");
+   
+            o = stbi__process_gif_raster(s, g);
+            if (o == NULL) return NULL;
+
+            if (req_comp && req_comp != 4)
+               o = stbi__convert_format(o, 4, req_comp, g->w, g->h);
+            return o;
+         }
+
+         case 0x21: // Comment Extension.
+         {
+            int len;
+            if (stbi__get8(s) == 0xF9) { // Graphic Control Extension.
+               len = stbi__get8(s);
+               if (len == 4) {
+                  g->eflags = stbi__get8(s);
+                  stbi__get16le(s); // delay
+                  g->transparent = stbi__get8(s);
+               } else {
+                  stbi__skip(s, len);
+                  break;
+               }
+            }
+            while ((len = stbi__get8(s)) != 0)
+               stbi__skip(s, len);
+            break;
+         }
+
+         case 0x3B: // gif stream termination code
+            return (stbi_uc *) 1;
+
+         default:
+            return stbi__errpuc("unknown code", "Corrupt GIF");
+      }
+   }
+}
+
+static stbi_uc *stbi__gif_load(stbi__context *s, int *x, int *y, int *comp, int req_comp)
+{
+   stbi_uc *u = 0;
+   stbi__gif g={0};
+
+   u = stbi__gif_load_next(s, &g, comp, req_comp);
+   if (u == (void *) 1) u = 0;  // end of animated gif marker
+   if (u) {
+      *x = g.w;
+      *y = g.h;
+   }
+
+   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);
+}
+
+
+// *************************************************************************************************
+// Radiance RGBE HDR loader
+// originally by Nicolas Schulz
+#ifndef STBI_NO_HDR
+static int stbi__hdr_test_core(stbi__context *s)
+{
+   const char *signature = "#?RADIANCE\n";
+   int i;
+   for (i=0; signature[i]; ++i)
+      if (stbi__get8(s) != signature[i])
+         return 0;
+   return 1;
+}
+
+static int stbi__hdr_test(stbi__context* s)
+{
+   int r = stbi__hdr_test_core(s);
+   stbi__rewind(s);
+   return r;
+}
+
+#define STBI__HDR_BUFLEN  1024
+static char *stbi__hdr_gettoken(stbi__context *z, char *buffer)
+{
+   int len=0;
+   char c = '\0';
+
+   c = (char) stbi__get8(z);
+
+   while (!stbi__at_eof(z) && c != '\n') {
+      buffer[len++] = c;
+      if (len == STBI__HDR_BUFLEN-1) {
+         // flush to end of line
+         while (!stbi__at_eof(z) && stbi__get8(z) != '\n')
+            ;
+         break;
+      }
+      c = (char) stbi__get8(z);
+   }
+
+   buffer[len] = 0;
+   return buffer;
+}
+
+static void stbi__hdr_convert(float *output, stbi_uc *input, int req_comp)
+{
+   if ( input[3] != 0 ) {
+      float f1;
+      // Exponent
+      f1 = (float) ldexp(1.0f, input[3] - (int)(128 + 8));
+      if (req_comp <= 2)
+         output[0] = (input[0] + input[1] + input[2]) * f1 / 3;
+      else {
+         output[0] = input[0] * f1;
+         output[1] = input[1] * f1;
+         output[2] = input[2] * f1;
+      }
+      if (req_comp == 2) output[1] = 1;
+      if (req_comp == 4) output[3] = 1;
+   } else {
+      switch (req_comp) {
+         case 4: output[3] = 1; /* fallthrough */
+         case 3: output[0] = output[1] = output[2] = 0;
+                 break;
+         case 2: output[1] = 1; /* fallthrough */
+         case 1: output[0] = 0;
+                 break;
+      }
+   }
+}
+
+static float *stbi__hdr_load(stbi__context *s, int *x, int *y, int *comp, int req_comp)
+{
+   char buffer[STBI__HDR_BUFLEN];
+   char *token;
+   int valid = 0;
+   int width, height;
+   stbi_uc *scanline;
+   float *hdr_data;
+   int len;
+   unsigned char count, value;
+   int i, j, k, c1,c2, z;
+
+
+   // Check identifier
+   if (strcmp(stbi__hdr_gettoken(s,buffer), "#?RADIANCE") != 0)
+      return stbi__errpf("not HDR", "Corrupt HDR image");
+   
+   // Parse header
+   for(;;) {
+      token = stbi__hdr_gettoken(s,buffer);
+      if (token[0] == 0) break;
+      if (strcmp(token, "FORMAT=32-bit_rle_rgbe") == 0) valid = 1;
+   }
+
+   if (!valid)    return stbi__errpf("unsupported format", "Unsupported HDR format");
+
+   // Parse width and height
+   // can't use sscanf() if we're not using stdio!
+   token = stbi__hdr_gettoken(s,buffer);
+   if (strncmp(token, "-Y ", 3))  return stbi__errpf("unsupported data layout", "Unsupported HDR format");
+   token += 3;
+   height = (int) strtol(token, &token, 10);
+   while (*token == ' ') ++token;
+   if (strncmp(token, "+X ", 3))  return stbi__errpf("unsupported data layout", "Unsupported HDR format");
+   token += 3;
+   width = (int) strtol(token, NULL, 10);
+
+   *x = width;
+   *y = height;
+
+   if (comp) *comp = 3;
+   if (req_comp == 0) req_comp = 3;
+
+   // Read data
+   hdr_data = (float *) malloc(height * width * req_comp * sizeof(float));
+
+   // Load image data
+   // image data is stored as some number of sca
+   if ( width < 8 || width >= 32768) {
+      // Read flat data
+      for (j=0; j < height; ++j) {
+         for (i=0; i < width; ++i) {
+            stbi_uc rgbe[4];
+           main_decode_loop:
+            stbi__getn(s, rgbe, 4);
+            stbi__hdr_convert(hdr_data + j * width * req_comp + i * req_comp, rgbe, req_comp);
+         }
+      }
+   } else {
+      // Read RLE-encoded data
+      scanline = NULL;
+
+      for (j = 0; j < height; ++j) {
+         c1 = stbi__get8(s);
+         c2 = stbi__get8(s);
+         len = stbi__get8(s);
+         if (c1 != 2 || c2 != 2 || (len & 0x80)) {
+            // not run-length encoded, so we have to actually use THIS data as a decoded
+            // pixel (note this can't be a valid pixel--one of RGB must be >= 128)
+            stbi_uc rgbe[4];
+            rgbe[0] = (stbi_uc) c1;
+            rgbe[1] = (stbi_uc) c2;
+            rgbe[2] = (stbi_uc) len;
+            rgbe[3] = (stbi_uc) stbi__get8(s);
+            stbi__hdr_convert(hdr_data, rgbe, req_comp);
+            i = 1;
+            j = 0;
+            free(scanline);
+            goto main_decode_loop; // yes, this makes no sense
+         }
+         len <<= 8;
+         len |= stbi__get8(s);
+         if (len != width) { free(hdr_data); free(scanline); return stbi__errpf("invalid decoded scanline length", "corrupt HDR"); }
+         if (scanline == NULL) scanline = (stbi_uc *) malloc(width * 4);
+            
+         for (k = 0; k < 4; ++k) {
+            i = 0;
+            while (i < width) {
+               count = stbi__get8(s);
+               if (count > 128) {
+                  // Run
+                  value = stbi__get8(s);
+                  count -= 128;
+                  for (z = 0; z < count; ++z)
+                     scanline[i++ * 4 + k] = value;
+               } else {
+                  // Dump
+                  for (z = 0; z < count; ++z)
+                     scanline[i++ * 4 + k] = stbi__get8(s);
+               }
+            }
+         }
+         for (i=0; i < width; ++i)
+            stbi__hdr_convert(hdr_data+(j*width + i)*req_comp, scanline + i*4, req_comp);
+      }
+      free(scanline);
+   }
+
+   return hdr_data;
+}
+
+static int stbi__hdr_info(stbi__context *s, int *x, int *y, int *comp)
+{
+   char buffer[STBI__HDR_BUFLEN];
+   char *token;
+   int valid = 0;
+
+   if (strcmp(stbi__hdr_gettoken(s,buffer), "#?RADIANCE") != 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
+
+static int stbi__bmp_info(stbi__context *s, int *x, int *y, int *comp)
+{
+   int hsz;
+   if (stbi__get8(s) != 'B' || stbi__get8(s) != 'M') {
+       stbi__rewind( s );
+       return 0;
+   }
+   stbi__skip(s,12);
+   hsz = stbi__get32le(s);
+   if (hsz != 12 && hsz != 40 && hsz != 56 && hsz != 108 && hsz != 124) {
+       stbi__rewind( s );
+       return 0;
+   }
+   if (hsz == 12) {
+      *x = stbi__get16le(s);
+      *y = stbi__get16le(s);
+   } else {
+      *x = stbi__get32le(s);
+      *y = stbi__get32le(s);
+   }
+   if (stbi__get16le(s) != 1) {
+       stbi__rewind( s );
+       return 0;
+   }
+   *comp = stbi__get16le(s) / 8;
+   return 1;
+}
+
+static int stbi__psd_info(stbi__context *s, int *x, int *y, int *comp)
+{
+   int channelCount;
+   if (stbi__get32be(s) != 0x38425053) {
+       stbi__rewind( s );
+       return 0;
+   }
+   if (stbi__get16be(s) != 1) {
+       stbi__rewind( s );
+       return 0;
+   }
+   stbi__skip(s, 6);
+   channelCount = stbi__get16be(s);
+   if (channelCount < 0 || channelCount > 16) {
+       stbi__rewind( s );
+       return 0;
+   }
+   *y = stbi__get32be(s);
+   *x = stbi__get32be(s);
+   if (stbi__get16be(s) != 8) {
+       stbi__rewind( s );
+       return 0;
+   }
+   if (stbi__get16be(s) != 3) {
+       stbi__rewind( s );
+       return 0;
+   }
+   *comp = 4;
+   return 1;
+}
+
+static int stbi__pic_info(stbi__context *s, int *x, int *y, int *comp)
+{
+   int act_comp=0,num_packets=0,chained;
+   stbi__pic_packet packets[10];
+
+   stbi__skip(s, 92);
+
+   *x = stbi__get16be(s);
+   *y = stbi__get16be(s);
+   if (stbi__at_eof(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;
+}
+
+static int stbi__info_main(stbi__context *s, int *x, int *y, int *comp)
+{
+   if (stbi__jpeg_info(s, x, y, comp))
+       return 1;
+   if (stbi__png_info(s, x, y, comp))
+       return 1;
+   if (stbi__gif_info(s, x, y, comp))
+       return 1;
+   if (stbi__bmp_info(s, x, y, comp))
+       return 1;
+   if (stbi__psd_info(s, x, y, comp))
+       return 1;
+   if (stbi__pic_info(s, x, y, comp))
+       return 1;
+   #ifndef STBI_NO_HDR
+   if (stbi__hdr_info(s, x, y, comp))
+       return 1;
+   #endif
+   // test tga last because it's a crappy test!
+   if (stbi__tga_info(s, x, y, comp))
+       return 1;
+   return stbi__err("unknown image type", "Image not of any known type, or corrupt");
+}
+
+#ifndef STBI_NO_STDIO
+STBIDEF int stbi_info(char const *filename, int *x, int *y, int *comp)
+{
+    FILE *f = fopen(filename, "rb");
+    int result;
+    if (!f) return stbi__err("can't fopen", "Unable to open file");
+    result = stbi_info_from_file(f, x, y, comp);
+    fclose(f);
+    return result;
+}
+
+STBIDEF int stbi_info_from_file(FILE *f, int *x, int *y, int *comp)
+{
+   int r;
+   stbi__context s;
+   long pos = ftell(f);
+   stbi__start_file(&s, f);
+   r = stbi__info_main(&s,x,y,comp);
+   fseek(f,pos,SEEK_SET);
+   return r;
+}
+#endif // !STBI_NO_STDIO
+
+STBIDEF int stbi_info_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp)
+{
+   stbi__context s;
+   stbi__start_mem(&s,buffer,len);
+   return stbi__info_main(&s,x,y,comp);
+}
+
+STBIDEF int stbi_info_from_callbacks(stbi_io_callbacks const *c, void *user, int *x, int *y, int *comp)
+{
+   stbi__context s;
+   stbi__start_callbacks(&s, (stbi_io_callbacks *) c, user);
+   return stbi__info_main(&s,x,y,comp);
+}
+
+#endif // STB_IMAGE_IMPLEMENTATION
+
+#if !defined(STBI_NO_STDIO) && defined(_MSC_VER) && _MSC_VER >= 1400
+#pragma warning(pop)
+#endif
+
+
+/*
+   revision history:
+      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   first released version
+*/
diff --git a/dgl/src/nanovg2/LICENSE.txt b/dgl/src/nanovg2/LICENSE.txt
new file mode 100644
index 00000000..2a03a1a6
--- /dev/null
+++ b/dgl/src/nanovg2/LICENSE.txt
@@ -0,0 +1,18 @@
+Copyright (c) 2013 Mikko Mononen memon@inside.org
+
+This software is provided 'as-is', without any express or implied
+warranty.  In no event will the authors be held liable for any damages
+arising from the use of this software.
+
+Permission is granted to anyone to use this software for any purpose,
+including commercial applications, and to alter it and redistribute it
+freely, subject to the following restrictions:
+
+1. The origin of this software must not be misrepresented; you must not
+claim that you wrote the original software. If you use this software
+in a product, an acknowledgment in the product documentation would be
+appreciated but is not required.
+2. Altered source versions must be plainly marked as such, and must not be
+misrepresented as being the original software.
+3. This notice may not be removed or altered from any source distribution.
+
diff --git a/dgl/src/nanovg2/fontstash.h b/dgl/src/nanovg2/fontstash.h
new file mode 100644
index 00000000..71377096
--- /dev/null
+++ b/dgl/src/nanovg2/fontstash.h
@@ -0,0 +1,1688 @@
+//
+// Copyright (c) 2009-2013 Mikko Mononen memon@inside.org
+//
+// This software is provided 'as-is', without any express or implied
+// warranty.  In no event will the authors be held liable for any damages
+// arising from the use of this software.
+// Permission is granted to anyone to use this software for any purpose,
+// including commercial applications, and to alter it and redistribute it
+// freely, subject to the following restrictions:
+// 1. The origin of this software must not be misrepresented; you must not
+//    claim that you wrote the original software. If you use this software
+//    in a product, an acknowledgment in the product documentation would be
+//    appreciated but is not required.
+// 2. Altered source versions must be plainly marked as such, and must not be
+//    misrepresented as being the original software.
+// 3. This notice may not be removed or altered from any source distribution.
+//
+
+#ifndef FONS_H
+#define FONS_H
+
+#define FONS_INVALID -1
+
+enum FONSflags {
+	FONS_ZERO_TOPLEFT = 1,
+	FONS_ZERO_BOTTOMLEFT = 2,
+};
+
+enum FONSalign {
+	// Horizontal align
+	FONS_ALIGN_LEFT 	= 1<<0,	// Default
+	FONS_ALIGN_CENTER 	= 1<<1,
+	FONS_ALIGN_RIGHT 	= 1<<2,
+	// Vertical align
+	FONS_ALIGN_TOP 		= 1<<3,
+	FONS_ALIGN_MIDDLE	= 1<<4,
+	FONS_ALIGN_BOTTOM	= 1<<5,
+	FONS_ALIGN_BASELINE	= 1<<6, // Default
+};
+
+enum FONSerrorCode {
+	// Font atlas is full.
+	FONS_ATLAS_FULL = 1,
+	// Scratch memory used to render glyphs is full, requested size reported in 'val', you may need to bump up FONS_SCRATCH_BUF_SIZE.		
+	FONS_SCRATCH_FULL = 2,
+	// Calls to fonsPushState has created too large stack, if you need deep state stack bump up FONS_MAX_STATES.
+	FONS_STATES_OVERFLOW = 3,
+	// Trying to pop too many states fonsPopState().
+	FONS_STATES_UNDERFLOW = 4,
+};
+
+struct FONSparams {
+	int width, height;
+	unsigned char flags;
+	void* userPtr;
+	int (*renderCreate)(void* uptr, int width, int height);
+	int (*renderResize)(void* uptr, int width, int height);
+	void (*renderUpdate)(void* uptr, int* rect, const unsigned char* data);
+	void (*renderDraw)(void* uptr, const float* verts, const float* tcoords, const unsigned int* colors, int nverts);
+	void (*renderDelete)(void* uptr);
+};
+typedef struct FONSparams FONSparams;
+
+struct FONSquad
+{
+	float x0,y0,s0,t0;
+	float x1,y1,s1,t1;
+};
+typedef struct FONSquad FONSquad;
+
+struct FONStextIter {
+	float x, y, nextx, nexty, scale, spacing;
+	unsigned int codepoint;
+	short isize, iblur;
+	struct FONSfont* font;
+	int prevGlyphIndex;
+	const char* str;
+	const char* next;
+	const char* end;
+	unsigned int utf8state;
+};
+typedef struct FONStextIter FONStextIter;
+
+typedef struct FONScontext FONScontext;
+
+// Constructor and destructor.
+FONScontext* fonsCreateInternal(FONSparams* params);
+void fonsDeleteInternal(FONScontext* s);
+
+void fonsSetErrorCallback(FONScontext* s, void (*callback)(void* uptr, int error, int val), void* uptr);
+// Returns current atlas size.
+void fonsGetAtlasSize(FONScontext* s, int* width, int* height);
+// Expands the atlas size. 
+int fonsExpandAtlas(FONScontext* s, int width, int height);
+// Resets the whole stash.
+int fonsResetAtlas(FONScontext* stash, int width, int height);
+
+// Add fonts
+int fonsAddFont(FONScontext* s, const char* name, const char* path);
+int fonsAddFontMem(FONScontext* s, const char* name, unsigned char* data, int ndata, int freeData);
+int fonsGetFontByName(FONScontext* s, const char* name);
+
+// State handling
+void fonsPushState(FONScontext* s);
+void fonsPopState(FONScontext* s);
+void fonsClearState(FONScontext* s);
+
+// State setting
+void fonsSetSize(FONScontext* s, float size);
+void fonsSetColor(FONScontext* s, unsigned int color);
+void fonsSetSpacing(FONScontext* s, float spacing);
+void fonsSetBlur(FONScontext* s, float blur);
+void fonsSetAlign(FONScontext* s, int align);
+void fonsSetFont(FONScontext* s, int font);
+
+// Draw text
+float fonsDrawText(FONScontext* s, float x, float y, const char* string, const char* end);
+
+// Measure text
+float fonsTextBounds(FONScontext* s, float x, float y, const char* string, const char* end, float* bounds);
+void fonsLineBounds(FONScontext* s, float y, float* miny, float* maxy);
+void fonsVertMetrics(FONScontext* s, float* ascender, float* descender, float* lineh);
+
+// Text iterator
+int fonsTextIterInit(FONScontext* stash, FONStextIter* iter, float x, float y, const char* str, const char* end);
+int fonsTextIterNext(FONScontext* stash, FONStextIter* iter, struct FONSquad* quad);
+
+// Pull texture changes
+const unsigned char* fonsGetTextureData(FONScontext* stash, int* width, int* height);
+int fonsValidateTexture(FONScontext* s, int* dirty);
+
+// Draws the stash texture for debugging
+void fonsDrawDebug(FONScontext* s, float x, float y);
+
+#endif // FONTSTASH_H
+
+
+#ifdef FONTSTASH_IMPLEMENTATION
+
+#define FONS_NOTUSED(v)  (void)sizeof(v)
+
+#ifdef FONS_USE_FREETYPE
+
+#include <ft2build.h>
+#include FT_FREETYPE_H
+#include FT_ADVANCES_H
+#include <math.h>
+
+struct FONSttFontImpl {
+	FT_Face font;
+};
+typedef struct FONSttFontImpl FONSttFontImpl;
+
+static FT_Library ftLibrary;
+
+int fons__tt_init(FONScontext *context)
+{ 
+	FT_Error ftError;
+        FONS_NOTUSED(context);
+	ftError = FT_Init_FreeType(&ftLibrary);
+	return ftError == 0;
+}
+
+int fons__tt_loadFont(FONScontext *context, FONSttFontImpl *font, unsigned char *data, int dataSize)
+{
+	FT_Error ftError;
+	FONS_NOTUSED(context);
+
+	//font->font.userdata = stash;
+	ftError = FT_New_Memory_Face(ftLibrary, (const FT_Byte*)data, dataSize, 0, &font->font);
+	return ftError == 0;
+}
+
+void fons__tt_getFontVMetrics(FONSttFontImpl *font, int *ascent, int *descent, int *lineGap)
+{
+	*ascent = font->font->ascender;
+	*descent = font->font->descender;
+	*lineGap = font->font->height - (*ascent - *descent);
+}
+
+float fons__tt_getPixelHeightScale(FONSttFontImpl *font, float size)
+{
+	return size / (font->font->ascender - font->font->descender);
+}
+
+int fons__tt_getGlyphIndex(FONSttFontImpl *font, int codepoint)
+{
+	return FT_Get_Char_Index(font->font, codepoint);
+}
+
+int fons__tt_buildGlyphBitmap(FONSttFontImpl *font, int glyph, float size, float scale,
+							  int *advance, int *lsb, int *x0, int *y0, int *x1, int *y1)
+{
+	FT_Error ftError;
+	FT_GlyphSlot ftGlyph;
+	FONS_NOTUSED(scale);
+
+	ftError = FT_Set_Pixel_Sizes(font->font, 0, (FT_UInt)(size * (float)font->font->units_per_EM / (float)(font->font->ascender - font->font->descender)));
+	if (ftError) return 0;
+	ftError = FT_Load_Glyph(font->font, glyph, FT_LOAD_RENDER);
+	if (ftError) return 0;
+	ftError = FT_Get_Advance(font->font, glyph, FT_LOAD_NO_SCALE, (FT_Fixed*)advance);
+	if (ftError) return 0;
+	ftGlyph = font->font->glyph;
+	*lsb = ftGlyph->metrics.horiBearingX;
+	*x0 = ftGlyph->bitmap_left;
+	*x1 = *x0 + ftGlyph->bitmap.width;
+	*y0 = -ftGlyph->bitmap_top;
+	*y1 = *y0 + ftGlyph->bitmap.rows;
+	return 1;
+}
+
+void fons__tt_renderGlyphBitmap(FONSttFontImpl *font, unsigned char *output, int outWidth, int outHeight, int outStride,
+								float scaleX, float scaleY, int glyph)
+{
+	FT_GlyphSlot ftGlyph = font->font->glyph;
+	int ftGlyphOffset = 0;
+	int x, y;
+	FONS_NOTUSED(outWidth);
+	FONS_NOTUSED(outHeight);
+	FONS_NOTUSED(scaleX);
+	FONS_NOTUSED(scaleY);
+	FONS_NOTUSED(glyph);	// glyph has already been loaded by fons__tt_buildGlyphBitmap
+
+	for ( y = 0; y < ftGlyph->bitmap.rows; y++ ) {
+		for ( x = 0; x < ftGlyph->bitmap.width; x++ ) {
+			output[(y * outStride) + x] = ftGlyph->bitmap.buffer[ftGlyphOffset++];
+		}
+	}
+}
+
+int fons__tt_getGlyphKernAdvance(FONSttFontImpl *font, int glyph1, int glyph2)
+{
+	FT_Vector ftKerning;
+	FT_Get_Kerning(font->font, glyph1, glyph2, FT_KERNING_DEFAULT, &ftKerning);
+	return ftKerning.x;
+}
+
+#else
+
+#define STB_TRUETYPE_IMPLEMENTATION
+static void* fons__tmpalloc(size_t size, void* up);
+static void fons__tmpfree(void* ptr, void* up);
+#define STBTT_malloc(x,u)    fons__tmpalloc(x,u)
+#define STBTT_free(x,u)      fons__tmpfree(x,u)
+#include "stb_truetype.h"
+
+struct FONSttFontImpl {
+	stbtt_fontinfo font;
+};
+typedef struct FONSttFontImpl FONSttFontImpl;
+
+int fons__tt_init(FONScontext *context)
+{
+	FONS_NOTUSED(context);
+	return 1;
+}
+
+int fons__tt_loadFont(FONScontext *context, FONSttFontImpl *font, unsigned char *data, int dataSize)
+{
+	int stbError;
+	FONS_NOTUSED(dataSize);
+
+	font->font.userdata = context;
+	stbError = stbtt_InitFont(&font->font, data, 0);
+	return stbError;
+}
+
+void fons__tt_getFontVMetrics(FONSttFontImpl *font, int *ascent, int *descent, int *lineGap)
+{
+	stbtt_GetFontVMetrics(&font->font, ascent, descent, lineGap);
+}
+
+float fons__tt_getPixelHeightScale(FONSttFontImpl *font, float size)
+{
+	return stbtt_ScaleForPixelHeight(&font->font, size);
+}
+
+int fons__tt_getGlyphIndex(FONSttFontImpl *font, int codepoint)
+{
+	return stbtt_FindGlyphIndex(&font->font, codepoint);
+}
+
+int fons__tt_buildGlyphBitmap(FONSttFontImpl *font, int glyph, float size, float scale,
+							  int *advance, int *lsb, int *x0, int *y0, int *x1, int *y1)
+{
+	FONS_NOTUSED(size);
+	stbtt_GetGlyphHMetrics(&font->font, glyph, advance, lsb);
+	stbtt_GetGlyphBitmapBox(&font->font, glyph, scale, scale, x0, y0, x1, y1);
+	return 1;
+}
+
+void fons__tt_renderGlyphBitmap(FONSttFontImpl *font, unsigned char *output, int outWidth, int outHeight, int outStride,
+								float scaleX, float scaleY, int glyph)
+{
+	stbtt_MakeGlyphBitmap(&font->font, output, outWidth, outHeight, outStride, scaleX, scaleY, glyph);
+}
+
+int fons__tt_getGlyphKernAdvance(FONSttFontImpl *font, int glyph1, int glyph2)
+{
+	return stbtt_GetGlyphKernAdvance(&font->font, glyph1, glyph2);
+}
+
+#endif
+
+#ifndef FONS_SCRATCH_BUF_SIZE
+#	define FONS_SCRATCH_BUF_SIZE 16000
+#endif
+#ifndef FONS_HASH_LUT_SIZE
+#	define FONS_HASH_LUT_SIZE 256
+#endif
+#ifndef FONS_INIT_FONTS
+#	define FONS_INIT_FONTS 4
+#endif
+#ifndef FONS_INIT_GLYPHS
+#	define FONS_INIT_GLYPHS 256
+#endif
+#ifndef FONS_INIT_ATLAS_NODES
+#	define FONS_INIT_ATLAS_NODES 256
+#endif
+#ifndef FONS_VERTEX_COUNT
+#	define FONS_VERTEX_COUNT 1024
+#endif
+#ifndef FONS_MAX_STATES
+#	define FONS_MAX_STATES 20
+#endif
+
+static unsigned int fons__hashint(unsigned int a)
+{
+	a += ~(a<<15);
+	a ^=  (a>>10);
+	a +=  (a<<3);
+	a ^=  (a>>6);
+	a += ~(a<<11);
+	a ^=  (a>>16);
+	return a;
+}
+
+static int fons__mini(int a, int b)
+{
+	return a < b ? a : b;
+}
+
+static int fons__maxi(int a, int b)
+{
+	return a > b ? a : b;
+}
+
+struct FONSglyph
+{
+	unsigned int codepoint;
+	int index;
+	int next;
+	short size, blur;
+	short x0,y0,x1,y1;
+	short xadv,xoff,yoff;
+};
+typedef struct FONSglyph FONSglyph;
+
+struct FONSfont
+{
+	FONSttFontImpl font;
+	char name[64];
+	unsigned char* data;
+	int dataSize;
+	unsigned char freeData;
+	float ascender;
+	float descender;
+	float lineh;
+	FONSglyph* glyphs;
+	int cglyphs;
+	int nglyphs;
+	int lut[FONS_HASH_LUT_SIZE];
+};
+typedef struct FONSfont FONSfont;
+
+struct FONSstate
+{
+	int font;
+	int align;
+	float size;
+	unsigned int color;
+	float blur;
+	float spacing;
+};
+typedef struct FONSstate FONSstate;
+
+struct FONSatlasNode {
+    short x, y, width;
+};
+typedef struct FONSatlasNode FONSatlasNode;
+
+struct FONSatlas
+{
+	int width, height;
+	FONSatlasNode* nodes;
+	int nnodes;
+	int cnodes;
+};
+typedef struct FONSatlas FONSatlas;
+
+struct FONScontext
+{
+	FONSparams params;
+	float itw,ith;
+	unsigned char* texData;
+	int dirtyRect[4];
+	FONSfont** fonts;
+	FONSatlas* atlas;
+	int cfonts;
+	int nfonts;
+	float verts[FONS_VERTEX_COUNT*2];
+	float tcoords[FONS_VERTEX_COUNT*2];
+	unsigned int colors[FONS_VERTEX_COUNT];
+	int nverts;
+	unsigned char* scratch;
+	int nscratch;
+	FONSstate states[FONS_MAX_STATES];
+	int nstates;
+	void (*handleError)(void* uptr, int error, int val);
+	void* errorUptr;
+};
+
+static void* fons__tmpalloc(size_t size, void* up)
+{
+	unsigned char* ptr;
+	FONScontext* stash = (FONScontext*)up;
+
+	// 16-byte align the returned pointer
+	size = (size + 0xf) & ~0xf;
+
+	if (stash->nscratch+(int)size > FONS_SCRATCH_BUF_SIZE) {
+		if (stash->handleError)
+			stash->handleError(stash->errorUptr, FONS_SCRATCH_FULL, stash->nscratch+(int)size);
+		return NULL;
+	}
+	ptr = stash->scratch + stash->nscratch;
+	stash->nscratch += (int)size;
+	return ptr;
+}
+
+static void fons__tmpfree(void* ptr, void* up)
+{
+	(void)ptr;
+	(void)up;
+	// empty
+}
+
+// Copyright (c) 2008-2010 Bjoern Hoehrmann <bjoern@hoehrmann.de>
+// See http://bjoern.hoehrmann.de/utf-8/decoder/dfa/ for details.
+
+#define FONS_UTF8_ACCEPT 0
+#define FONS_UTF8_REJECT 12
+
+static unsigned int fons__decutf8(unsigned int* state, unsigned int* codep, unsigned int byte)
+{
+	static const unsigned char utf8d[] = {
+		// The first part of the table maps bytes to character classes that
+		// to reduce the size of the transition table and create bitmasks.
+		0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,  0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
+		0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,  0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
+		0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,  0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
+		0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,  0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
+		1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,  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,  7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,
+		8,8,2,2,2,2,2,2,2,2,2,2,2,2,2,2,  2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,
+		10,3,3,3,3,3,3,3,3,3,3,3,3,4,3,3, 11,6,6,6,5,8,8,8,8,8,8,8,8,8,8,8,
+
+		// The second part is a transition table that maps a combination
+		// of a state of the automaton and a character class to a state.
+		0,12,24,36,60,96,84,12,12,12,48,72, 12,12,12,12,12,12,12,12,12,12,12,12,
+		12, 0,12,12,12,12,12, 0,12, 0,12,12, 12,24,12,12,12,12,12,24,12,24,12,12,
+		12,12,12,12,12,12,12,24,12,12,12,12, 12,24,12,12,12,12,12,12,12,24,12,12,
+		12,12,12,12,12,12,12,36,12,36,12,12, 12,36,12,12,12,12,12,36,12,36,12,12,
+		12,36,12,12,12,12,12,12,12,12,12,12,
+    };
+
+	unsigned int type = utf8d[byte];
+
+    *codep = (*state != FONS_UTF8_ACCEPT) ?
+		(byte & 0x3fu) | (*codep << 6) :
+		(0xff >> type) & (byte);
+
+	*state = utf8d[256 + *state + type];
+	return *state;
+}
+
+// Atlas based on Skyline Bin Packer by Jukka Jylänki
+
+static void fons__deleteAtlas(FONSatlas* atlas)
+{
+	if (atlas == NULL) return;
+	if (atlas->nodes != NULL) free(atlas->nodes);
+	free(atlas);
+}
+
+static FONSatlas* fons__allocAtlas(int w, int h, int nnodes)
+{
+	FONSatlas* atlas = NULL;
+
+	// Allocate memory for the font stash.
+	atlas = (FONSatlas*)malloc(sizeof(FONSatlas));
+	if (atlas == NULL) goto error;
+	memset(atlas, 0, sizeof(FONSatlas));
+
+	atlas->width = w;
+	atlas->height = h;
+
+	// Allocate space for skyline nodes
+	atlas->nodes = (FONSatlasNode*)malloc(sizeof(FONSatlasNode) * nnodes);
+	if (atlas->nodes == NULL) goto error;
+	memset(atlas->nodes, 0, sizeof(FONSatlasNode) * nnodes);
+	atlas->nnodes = 0;
+	atlas->cnodes = nnodes;
+
+	// Init root node.
+	atlas->nodes[0].x = 0;
+	atlas->nodes[0].y = 0;
+	atlas->nodes[0].width = (short)w;
+	atlas->nnodes++;
+
+	return atlas;
+
+error:
+	if (atlas) fons__deleteAtlas(atlas);
+	return NULL;
+}
+
+static int fons__atlasInsertNode(FONSatlas* atlas, int idx, int x, int y, int w)
+{
+	int i;
+	// Insert node
+	if (atlas->nnodes+1 > atlas->cnodes) {
+		atlas->cnodes = atlas->cnodes == 0 ? 8 : atlas->cnodes * 2;
+		atlas->nodes = (FONSatlasNode*)realloc(atlas->nodes, sizeof(FONSatlasNode) * atlas->cnodes);
+		if (atlas->nodes == NULL)
+			return 0;
+	}
+	for (i = atlas->nnodes; i > idx; i--)
+		atlas->nodes[i] = atlas->nodes[i-1];
+	atlas->nodes[idx].x = (short)x;
+	atlas->nodes[idx].y = (short)y;
+	atlas->nodes[idx].width = (short)w;
+	atlas->nnodes++;
+
+	return 1;
+}
+
+static void fons__atlasRemoveNode(FONSatlas* atlas, int idx)
+{
+	int i;
+	if (atlas->nnodes == 0) return;
+	for (i = idx; i < atlas->nnodes-1; i++)
+		atlas->nodes[i] = atlas->nodes[i+1];
+	atlas->nnodes--;
+}
+
+static void fons__atlasExpand(FONSatlas* atlas, int w, int h)
+{
+	// Insert node for empty space
+	if (w > atlas->width)
+		fons__atlasInsertNode(atlas, atlas->nnodes, atlas->width, 0, w - atlas->width);
+	atlas->width = w;
+	atlas->height = h;
+}
+
+static void fons__atlasReset(FONSatlas* atlas, int w, int h)
+{
+	atlas->width = w;
+	atlas->height = h;
+	atlas->nnodes = 0;
+
+	// Init root node.
+	atlas->nodes[0].x = 0;
+	atlas->nodes[0].y = 0;
+	atlas->nodes[0].width = (short)w;
+	atlas->nnodes++;
+}
+
+static int fons__atlasAddSkylineLevel(FONSatlas* atlas, int idx, int x, int y, int w, int h)
+{
+	int i;
+
+	// Insert new node
+	if (fons__atlasInsertNode(atlas, idx, x, y+h, w) == 0)
+		return 0;
+
+	// Delete skyline segments that fall under the shadow of the new segment.
+	for (i = idx+1; i < atlas->nnodes; i++) {
+		if (atlas->nodes[i].x < atlas->nodes[i-1].x + atlas->nodes[i-1].width) {
+			int shrink = atlas->nodes[i-1].x + atlas->nodes[i-1].width - atlas->nodes[i].x;
+			atlas->nodes[i].x += (short)shrink;
+			atlas->nodes[i].width -= (short)shrink;
+			if (atlas->nodes[i].width <= 0) {
+				fons__atlasRemoveNode(atlas, i);
+				i--;
+			} else {
+				break;
+			}
+		} else {
+			break;
+		}
+	}
+
+	// Merge same height skyline segments that are next to each other.
+	for (i = 0; i < atlas->nnodes-1; i++) {
+		if (atlas->nodes[i].y == atlas->nodes[i+1].y) {
+			atlas->nodes[i].width += atlas->nodes[i+1].width;
+			fons__atlasRemoveNode(atlas, i+1);
+			i--;
+		}
+	}
+
+	return 1;
+}
+
+static int fons__atlasRectFits(FONSatlas* atlas, int i, int w, int h)
+{
+	// Checks if there is enough space at the location of skyline span 'i',
+	// and return the max height of all skyline spans under that at that location,
+	// (think tetris block being dropped at that position). Or -1 if no space found.
+	int x = atlas->nodes[i].x;
+	int y = atlas->nodes[i].y;
+	int spaceLeft;
+	if (x + w > atlas->width)
+		return -1;
+	spaceLeft = w;
+	while (spaceLeft > 0) {
+		if (i == atlas->nnodes) return -1;
+		y = fons__maxi(y, atlas->nodes[i].y);
+		if (y + h > atlas->height) return -1;
+		spaceLeft -= atlas->nodes[i].width;
+		++i;
+	}
+	return y;
+}
+
+static int fons__atlasAddRect(FONSatlas* atlas, int rw, int rh, int* rx, int* ry)
+{
+	int besth = atlas->height, bestw = atlas->width, besti = -1;
+	int bestx = -1, besty = -1, i;
+
+	// Bottom left fit heuristic.
+	for (i = 0; i < atlas->nnodes; i++) {
+		int y = fons__atlasRectFits(atlas, i, rw, rh);
+		if (y != -1) {
+			if (y + rh < besth || (y + rh == besth && atlas->nodes[i].width < bestw)) {
+				besti = i;
+				bestw = atlas->nodes[i].width;
+				besth = y + rh;
+				bestx = atlas->nodes[i].x;
+				besty = y;
+			}
+		}
+	}
+
+	if (besti == -1)
+		return 0;
+
+	// Perform the actual packing.
+	if (fons__atlasAddSkylineLevel(atlas, besti, bestx, besty, rw, rh) == 0)
+		return 0;
+
+	*rx = bestx;
+	*ry = besty;
+
+	return 1;
+}
+
+static void fons__addWhiteRect(FONScontext* stash, int w, int h)
+{
+	int x, y, gx, gy;
+	unsigned char* dst;
+	if (fons__atlasAddRect(stash->atlas, w, h, &gx, &gy) == 0)
+		return;
+
+	// Rasterize
+	dst = &stash->texData[gx + gy * stash->params.width];
+	for (y = 0; y < h; y++) {
+		for (x = 0; x < w; x++)
+			dst[x] = 0xff;
+		dst += stash->params.width;
+	}
+
+	stash->dirtyRect[0] = fons__mini(stash->dirtyRect[0], gx);
+	stash->dirtyRect[1] = fons__mini(stash->dirtyRect[1], gy);
+	stash->dirtyRect[2] = fons__maxi(stash->dirtyRect[2], gx+w);
+	stash->dirtyRect[3] = fons__maxi(stash->dirtyRect[3], gy+h);
+}
+
+FONScontext* fonsCreateInternal(FONSparams* params)
+{
+	FONScontext* stash = NULL;
+
+	// Allocate memory for the font stash.
+	stash = (FONScontext*)malloc(sizeof(FONScontext));
+	if (stash == NULL) goto error;
+	memset(stash, 0, sizeof(FONScontext));
+
+	stash->params = *params;
+
+	// Allocate scratch buffer.
+	stash->scratch = (unsigned char*)malloc(FONS_SCRATCH_BUF_SIZE);
+	if (stash->scratch == NULL) goto error;
+
+	// Initialize implementation library
+	if (!fons__tt_init(stash)) goto error;
+
+	if (stash->params.renderCreate != NULL) {
+		if (stash->params.renderCreate(stash->params.userPtr, stash->params.width, stash->params.height) == 0)
+			goto error;
+	}
+
+	stash->atlas = fons__allocAtlas(stash->params.width, stash->params.height, FONS_INIT_ATLAS_NODES);
+	if (stash->atlas == NULL) goto error;
+
+	// Allocate space for fonts.
+	stash->fonts = (FONSfont**)malloc(sizeof(FONSfont*) * FONS_INIT_FONTS);
+	if (stash->fonts == NULL) goto error;
+	memset(stash->fonts, 0, sizeof(FONSfont*) * FONS_INIT_FONTS);
+	stash->cfonts = FONS_INIT_FONTS;
+	stash->nfonts = 0;
+
+	// Create texture for the cache.
+	stash->itw = 1.0f/stash->params.width;
+	stash->ith = 1.0f/stash->params.height;
+	stash->texData = (unsigned char*)malloc(stash->params.width * stash->params.height);
+	if (stash->texData == NULL) goto error;
+	memset(stash->texData, 0, stash->params.width * stash->params.height);
+
+	stash->dirtyRect[0] = stash->params.width;
+	stash->dirtyRect[1] = stash->params.height;
+	stash->dirtyRect[2] = 0;
+	stash->dirtyRect[3] = 0;
+
+	// Add white rect at 0,0 for debug drawing.
+	fons__addWhiteRect(stash, 2,2);
+
+	fonsPushState(stash);
+	fonsClearState(stash);
+
+	return stash;
+
+error:
+	fonsDeleteInternal(stash);
+	return NULL;
+}
+
+static FONSstate* fons__getState(FONScontext* stash)
+{
+	return &stash->states[stash->nstates-1];
+}
+
+void fonsSetSize(FONScontext* stash, float size)
+{
+	fons__getState(stash)->size = size;
+}
+
+void fonsSetColor(FONScontext* stash, unsigned int color)
+{
+	fons__getState(stash)->color = color;
+}
+
+void fonsSetSpacing(FONScontext* stash, float spacing)
+{
+	fons__getState(stash)->spacing = spacing;
+}
+
+void fonsSetBlur(FONScontext* stash, float blur)
+{
+	fons__getState(stash)->blur = blur;
+}
+
+void fonsSetAlign(FONScontext* stash, int align)
+{
+	fons__getState(stash)->align = align;
+}
+
+void fonsSetFont(FONScontext* stash, int font)
+{
+	fons__getState(stash)->font = font;
+}
+
+void fonsPushState(FONScontext* stash)
+{
+	if (stash->nstates >= FONS_MAX_STATES) {
+		if (stash->handleError)
+			stash->handleError(stash->errorUptr, FONS_STATES_OVERFLOW, 0);
+		return;
+	}
+	if (stash->nstates > 0)
+		memcpy(&stash->states[stash->nstates], &stash->states[stash->nstates-1], sizeof(FONSstate));
+	stash->nstates++;
+}
+
+void fonsPopState(FONScontext* stash)
+{
+	if (stash->nstates <= 1) {
+		if (stash->handleError)
+			stash->handleError(stash->errorUptr, FONS_STATES_UNDERFLOW, 0);
+		return;
+	}
+	stash->nstates--;
+}
+
+void fonsClearState(FONScontext* stash)
+{
+	FONSstate* state = fons__getState(stash);
+	state->size = 12.0f;
+	state->color = 0xffffffff;
+	state->font = 0;
+	state->blur = 0;
+	state->spacing = 0;
+	state->align = FONS_ALIGN_LEFT | FONS_ALIGN_BASELINE;
+}
+
+static void fons__freeFont(FONSfont* font)
+{
+	if (font == NULL) return;
+	if (font->glyphs) free(font->glyphs);
+	if (font->freeData && font->data) free(font->data);
+	free(font);
+}
+
+static int fons__allocFont(FONScontext* stash)
+{
+	FONSfont* font = NULL;
+	if (stash->nfonts+1 > stash->cfonts) {
+		stash->cfonts = stash->cfonts == 0 ? 8 : stash->cfonts * 2;
+		stash->fonts = (FONSfont**)realloc(stash->fonts, sizeof(FONSfont*) * stash->cfonts);
+		if (stash->fonts == NULL)
+			return -1;
+	}
+	font = (FONSfont*)malloc(sizeof(FONSfont));
+	if (font == NULL) goto error;
+	memset(font, 0, sizeof(FONSfont));
+
+	font->glyphs = (FONSglyph*)malloc(sizeof(FONSglyph) * FONS_INIT_GLYPHS);
+	if (font->glyphs == NULL) goto error;
+	font->cglyphs = FONS_INIT_GLYPHS;
+	font->nglyphs = 0;
+
+	stash->fonts[stash->nfonts++] = font;
+	return stash->nfonts-1;
+
+error:
+	fons__freeFont(font);
+
+	return FONS_INVALID;
+}
+
+int fonsAddFont(FONScontext* stash, const char* name, const char* path)
+{
+	FILE* fp = 0;
+	int dataSize = 0;
+	unsigned char* data = NULL;
+
+	// Read in the font data.
+	fp = fopen(path, "rb");
+	if (fp == NULL) goto error;
+	fseek(fp,0,SEEK_END);
+	dataSize = (int)ftell(fp);
+	fseek(fp,0,SEEK_SET);
+	data = (unsigned char*)malloc(dataSize);
+	if (data == NULL) goto error;
+	fread(data, 1, dataSize, fp);
+	fclose(fp);
+	fp = 0;
+
+	return fonsAddFontMem(stash, name, data, dataSize, 1);
+
+error:
+	if (data) free(data);
+	if (fp) fclose(fp);
+	return FONS_INVALID;
+}
+
+int fonsAddFontMem(FONScontext* stash, const char* name, unsigned char* data, int dataSize, int freeData)
+{
+	int i, ascent, descent, fh, lineGap;
+	FONSfont* font;
+
+	int idx = fons__allocFont(stash);
+	if (idx == FONS_INVALID)
+		return FONS_INVALID;
+
+	font = stash->fonts[idx];
+
+	strncpy(font->name, name, sizeof(font->name));
+	font->name[sizeof(font->name)-1] = '\0';
+
+	// Init hash lookup.
+	for (i = 0; i < FONS_HASH_LUT_SIZE; ++i)
+		font->lut[i] = -1;
+
+	// Read in the font data.
+	font->dataSize = dataSize;
+	font->data = data;
+	font->freeData = (unsigned char)freeData;
+
+	// Init font
+	stash->nscratch = 0;
+	if (!fons__tt_loadFont(stash, &font->font, data, dataSize)) goto error;
+
+	// Store normalized line height. The real line height is got
+	// by multiplying the lineh by font size.
+	fons__tt_getFontVMetrics( &font->font, &ascent, &descent, &lineGap);
+	fh = ascent - descent;
+	font->ascender = (float)ascent / (float)fh;
+	font->descender = (float)descent / (float)fh;
+	font->lineh = (float)(fh + lineGap) / (float)fh;
+
+	return idx;
+
+error:
+	fons__freeFont(font);
+	stash->nfonts--;
+	return FONS_INVALID;
+}
+
+int fonsGetFontByName(FONScontext* s, const char* name)
+{
+	int i;
+	for (i = 0; i < s->nfonts; i++) {
+		if (strcmp(s->fonts[i]->name, name) == 0)
+			return i;
+	}
+	return FONS_INVALID;
+}
+
+
+static FONSglyph* fons__allocGlyph(FONSfont* font)
+{
+	if (font->nglyphs+1 > font->cglyphs) {
+		font->cglyphs = font->cglyphs == 0 ? 8 : font->cglyphs * 2;
+		font->glyphs = (FONSglyph*)realloc(font->glyphs, sizeof(FONSglyph) * font->cglyphs);
+		if (font->glyphs == NULL) return NULL;
+	}
+	font->nglyphs++;
+	return &font->glyphs[font->nglyphs-1];
+}
+
+
+// Based on Exponential blur, Jani Huhtanen, 2006
+
+#define APREC 16
+#define ZPREC 7
+
+static void fons__blurCols(unsigned char* dst, int w, int h, int dstStride, int alpha)
+{
+	int x, y;
+	for (y = 0; y < h; y++) {
+		int z = 0; // force zero border
+		for (x = 1; x < w; x++) {
+			z += (alpha * (((int)(dst[x]) << ZPREC) - z)) >> APREC;
+			dst[x] = (unsigned char)(z >> ZPREC);
+		}
+		dst[w-1] = 0; // force zero border
+		z = 0;
+		for (x = w-2; x >= 0; x--) {
+			z += (alpha * (((int)(dst[x]) << ZPREC) - z)) >> APREC;
+			dst[x] = (unsigned char)(z >> ZPREC);
+		}
+		dst[0] = 0; // force zero border
+		dst += dstStride;
+	}
+}
+
+static void fons__blurRows(unsigned char* dst, int w, int h, int dstStride, int alpha)
+{
+	int x, y;
+	for (x = 0; x < w; x++) {
+		int z = 0; // force zero border
+		for (y = dstStride; y < h*dstStride; y += dstStride) {
+			z += (alpha * (((int)(dst[y]) << ZPREC) - z)) >> APREC;
+			dst[y] = (unsigned char)(z >> ZPREC);
+		}
+		dst[(h-1)*dstStride] = 0; // force zero border
+		z = 0;
+		for (y = (h-2)*dstStride; y >= 0; y -= dstStride) {
+			z += (alpha * (((int)(dst[y]) << ZPREC) - z)) >> APREC;
+			dst[y] = (unsigned char)(z >> ZPREC);
+		}
+		dst[0] = 0; // force zero border
+		dst++;
+	}
+}
+
+
+static void fons__blur(FONScontext* stash, unsigned char* dst, int w, int h, int dstStride, int blur)
+{
+	int alpha;
+	float sigma;
+	(void)stash;
+
+	if (blur < 1)
+		return;
+	// Calculate the alpha such that 90% of the kernel is within the radius. (Kernel extends to infinity)
+	sigma = (float)blur * 0.57735f; // 1 / sqrt(3)
+	alpha = (int)((1<<APREC) * (1.0f - expf(-2.3f / (sigma+1.0f))));
+	fons__blurRows(dst, w, h, dstStride, alpha);
+	fons__blurCols(dst, w, h, dstStride, alpha);
+	fons__blurRows(dst, w, h, dstStride, alpha);
+	fons__blurCols(dst, w, h, dstStride, alpha);
+//	fons__blurrows(dst, w, h, dstStride, alpha);
+//	fons__blurcols(dst, w, h, dstStride, alpha);
+}
+
+static FONSglyph* fons__getGlyph(FONScontext* stash, FONSfont* font, unsigned int codepoint,
+								 short isize, short iblur)
+{
+	int i, g, advance, lsb, x0, y0, x1, y1, gw, gh, gx, gy, x, y;
+	float scale;
+	FONSglyph* glyph = NULL;
+	unsigned int h;
+	float size = isize/10.0f;
+	int pad, added;
+	unsigned char* bdst;
+	unsigned char* dst;
+
+	if (isize < 2) return NULL;
+	if (iblur > 20) iblur = 20;
+	pad = iblur+2;
+
+	// Reset allocator.
+	stash->nscratch = 0;
+
+	// Find code point and size.
+	h = fons__hashint(codepoint) & (FONS_HASH_LUT_SIZE-1);
+	i = font->lut[h];
+	while (i != -1) {
+		if (font->glyphs[i].codepoint == codepoint && font->glyphs[i].size == isize && font->glyphs[i].blur == iblur)
+			return &font->glyphs[i];
+		i = font->glyphs[i].next;
+	}
+
+	// Could not find glyph, create it.
+	scale = fons__tt_getPixelHeightScale(&font->font, size);
+	g = fons__tt_getGlyphIndex(&font->font, codepoint);
+	fons__tt_buildGlyphBitmap(&font->font, g, size, scale, &advance, &lsb, &x0, &y0, &x1, &y1);
+	gw = x1-x0 + pad*2;
+	gh = y1-y0 + pad*2;
+
+	// Find free spot for the rect in the atlas
+	added = fons__atlasAddRect(stash->atlas, gw, gh, &gx, &gy);
+	if (added == 0 && stash->handleError != NULL) {
+		// Atlas is full, let the user to resize the atlas (or not), and try again.
+		stash->handleError(stash->errorUptr, FONS_ATLAS_FULL, 0);
+		added = fons__atlasAddRect(stash->atlas, gw, gh, &gx, &gy);
+	}
+	if (added == 0) return NULL;
+
+	// Init glyph.
+	glyph = fons__allocGlyph(font);
+	glyph->codepoint = codepoint;
+	glyph->size = isize;
+	glyph->blur = iblur;
+	glyph->index = g;
+	glyph->x0 = (short)gx;
+	glyph->y0 = (short)gy;
+	glyph->x1 = (short)(glyph->x0+gw);
+	glyph->y1 = (short)(glyph->y0+gh);
+	glyph->xadv = (short)(scale * advance * 10.0f);
+	glyph->xoff = (short)(x0 - pad);
+	glyph->yoff = (short)(y0 - pad);
+	glyph->next = 0;
+
+	// Insert char to hash lookup.
+	glyph->next = font->lut[h];
+	font->lut[h] = font->nglyphs-1;
+
+	// Rasterize
+	dst = &stash->texData[(glyph->x0+pad) + (glyph->y0+pad) * stash->params.width];
+	fons__tt_renderGlyphBitmap(&font->font, dst, gw-pad*2,gh-pad*2, stash->params.width, scale,scale, g);
+
+	// Make sure there is one pixel empty border.
+	dst = &stash->texData[glyph->x0 + glyph->y0 * stash->params.width];
+	for (y = 0; y < gh; y++) {
+		dst[y*stash->params.width] = 0;
+		dst[gw-1 + y*stash->params.width] = 0;
+	}
+	for (x = 0; x < gw; x++) {
+		dst[x] = 0;
+		dst[x + (gh-1)*stash->params.width] = 0;
+	}
+
+	// Debug code to color the glyph background
+/*	unsigned char* fdst = &stash->texData[glyph->x0 + glyph->y0 * stash->params.width];
+	for (y = 0; y < gh; y++) {
+		for (x = 0; x < gw; x++) {
+			int a = (int)fdst[x+y*stash->params.width] + 20;
+			if (a > 255) a = 255;
+			fdst[x+y*stash->params.width] = a;
+		}
+	}*/
+
+	// Blur
+	if (iblur > 0) {
+		stash->nscratch = 0;
+		bdst = &stash->texData[glyph->x0 + glyph->y0 * stash->params.width];
+		fons__blur(stash, bdst, gw,gh, stash->params.width, iblur);
+	}
+
+	stash->dirtyRect[0] = fons__mini(stash->dirtyRect[0], glyph->x0);
+	stash->dirtyRect[1] = fons__mini(stash->dirtyRect[1], glyph->y0);
+	stash->dirtyRect[2] = fons__maxi(stash->dirtyRect[2], glyph->x1);
+	stash->dirtyRect[3] = fons__maxi(stash->dirtyRect[3], glyph->y1);
+
+	return glyph;
+}
+
+static void fons__getQuad(FONScontext* stash, FONSfont* font,
+						   int prevGlyphIndex, FONSglyph* glyph,
+						   float scale, float spacing, float* x, float* y, FONSquad* q)
+{
+	float rx,ry,xoff,yoff,x0,y0,x1,y1;
+
+	if (prevGlyphIndex != -1) {
+		float adv = fons__tt_getGlyphKernAdvance(&font->font, prevGlyphIndex, glyph->index) * scale;
+		*x += (int)(adv + spacing + 0.5f);
+	}
+
+	// Each glyph has 2px border to allow good interpolation,
+	// one pixel to prevent leaking, and one to allow good interpolation for rendering.
+	// Inset the texture region by one pixel for correct interpolation.
+	xoff = (short)(glyph->xoff+1);
+	yoff = (short)(glyph->yoff+1);
+	x0 = (float)(glyph->x0+1);
+	y0 = (float)(glyph->y0+1);
+	x1 = (float)(glyph->x1-1);
+	y1 = (float)(glyph->y1-1);
+
+	if (stash->params.flags & FONS_ZERO_TOPLEFT) {
+		rx = (float)(int)(*x + xoff);
+		ry = (float)(int)(*y + yoff);
+
+		q->x0 = rx;
+		q->y0 = ry;
+		q->x1 = rx + x1 - x0;
+		q->y1 = ry + y1 - y0;
+
+		q->s0 = x0 * stash->itw;
+		q->t0 = y0 * stash->ith;
+		q->s1 = x1 * stash->itw;
+		q->t1 = y1 * stash->ith;
+	} else {
+		rx = (float)(int)(*x + xoff);
+		ry = (float)(int)(*y - yoff);
+
+		q->x0 = rx;
+		q->y0 = ry;
+		q->x1 = rx + x1 - x0;
+		q->y1 = ry - y1 + y0;
+
+		q->s0 = x0 * stash->itw;
+		q->t0 = y0 * stash->ith;
+		q->s1 = x1 * stash->itw;
+		q->t1 = y1 * stash->ith;
+	}
+
+	*x += (int)(glyph->xadv / 10.0f + 0.5f);
+}
+
+static void fons__flush(FONScontext* stash)
+{
+	// Flush texture
+	if (stash->dirtyRect[0] < stash->dirtyRect[2] && stash->dirtyRect[1] < stash->dirtyRect[3]) {
+		if (stash->params.renderUpdate != NULL)
+			stash->params.renderUpdate(stash->params.userPtr, stash->dirtyRect, stash->texData);
+		// Reset dirty rect
+		stash->dirtyRect[0] = stash->params.width;
+		stash->dirtyRect[1] = stash->params.height;
+		stash->dirtyRect[2] = 0;
+		stash->dirtyRect[3] = 0;
+	}
+
+	// Flush triangles
+	if (stash->nverts > 0) {
+		if (stash->params.renderDraw != NULL)
+			stash->params.renderDraw(stash->params.userPtr, stash->verts, stash->tcoords, stash->colors, stash->nverts);
+		stash->nverts = 0;
+	}
+}
+
+static __inline void fons__vertex(FONScontext* stash, float x, float y, float s, float t, unsigned int c)
+{
+	stash->verts[stash->nverts*2+0] = x;
+	stash->verts[stash->nverts*2+1] = y;
+	stash->tcoords[stash->nverts*2+0] = s;
+	stash->tcoords[stash->nverts*2+1] = t;
+	stash->colors[stash->nverts] = c;
+	stash->nverts++;
+}
+
+static float fons__getVertAlign(FONScontext* stash, FONSfont* font, int align, short isize)
+{
+	if (stash->params.flags & FONS_ZERO_TOPLEFT) {
+		if (align & FONS_ALIGN_TOP) {
+			return font->ascender * (float)isize/10.0f;
+		} else if (align & FONS_ALIGN_MIDDLE) {
+			return (font->ascender + font->descender) / 2.0f * (float)isize/10.0f;
+		} else if (align & FONS_ALIGN_BASELINE) {
+			return 0.0f;
+		} else if (align & FONS_ALIGN_BOTTOM) {
+			return font->descender * (float)isize/10.0f;
+		}
+	} else {
+		if (align & FONS_ALIGN_TOP) {
+			return -font->ascender * (float)isize/10.0f;
+		} else if (align & FONS_ALIGN_MIDDLE) {
+			return -(font->ascender + font->descender) / 2.0f * (float)isize/10.0f;
+		} else if (align & FONS_ALIGN_BASELINE) {
+			return 0.0f;
+		} else if (align & FONS_ALIGN_BOTTOM) {
+			return -font->descender * (float)isize/10.0f;
+		}
+	}
+	return 0.0;
+}
+
+float fonsDrawText(FONScontext* stash,
+				   float x, float y,
+				   const char* str, const char* end)
+{
+	FONSstate* state = fons__getState(stash);
+	unsigned int codepoint;
+	unsigned int utf8state = 0;
+	FONSglyph* glyph = NULL;
+	FONSquad q;
+	int prevGlyphIndex = -1;
+	short isize = (short)(state->size*10.0f);
+	short iblur = (short)state->blur;
+	float scale;
+	FONSfont* font;
+	float width;
+
+	if (stash == NULL) return x;
+	if (state->font < 0 || state->font >= stash->nfonts) return x;
+	font = stash->fonts[state->font];
+	if (font->data == NULL) return x;
+
+	scale = fons__tt_getPixelHeightScale(&font->font, (float)isize/10.0f);
+
+	if (end == NULL)
+		end = str + strlen(str);
+
+	// Align horizontally
+	if (state->align & FONS_ALIGN_LEFT) {
+		// empty
+	} else if (state->align & FONS_ALIGN_RIGHT) {
+		width = fonsTextBounds(stash, x,y, str, end, NULL);
+		x -= width;
+	} else if (state->align & FONS_ALIGN_CENTER) {
+		width = fonsTextBounds(stash, x,y, str, end, NULL);
+		x -= width * 0.5f;
+	}
+	// Align vertically.
+	y += fons__getVertAlign(stash, font, state->align, isize);
+
+	for (; str != end; ++str) {
+		if (fons__decutf8(&utf8state, &codepoint, *(const unsigned char*)str))
+			continue;
+		glyph = fons__getGlyph(stash, font, codepoint, isize, iblur);
+		if (glyph != NULL) {
+			fons__getQuad(stash, font, prevGlyphIndex, glyph, scale, state->spacing, &x, &y, &q);
+
+			if (stash->nverts+6 > FONS_VERTEX_COUNT)
+				fons__flush(stash);
+
+			fons__vertex(stash, q.x0, q.y0, q.s0, q.t0, state->color);
+			fons__vertex(stash, q.x1, q.y1, q.s1, q.t1, state->color);
+			fons__vertex(stash, q.x1, q.y0, q.s1, q.t0, state->color);
+
+			fons__vertex(stash, q.x0, q.y0, q.s0, q.t0, state->color);
+			fons__vertex(stash, q.x0, q.y1, q.s0, q.t1, state->color);
+			fons__vertex(stash, q.x1, q.y1, q.s1, q.t1, state->color);
+		}
+		prevGlyphIndex = glyph != NULL ? glyph->index : -1;
+	}
+	fons__flush(stash);
+
+	return x;
+}
+
+int fonsTextIterInit(FONScontext* stash, FONStextIter* iter,
+					 float x, float y, const char* str, const char* end)
+{
+	FONSstate* state = fons__getState(stash);
+	float width;
+
+	memset(iter, 0, sizeof(*iter));
+
+	if (stash == NULL) return 0;
+	if (state->font < 0 || state->font >= stash->nfonts) return 0;
+	iter->font = stash->fonts[state->font];
+	if (iter->font->data == NULL) return 0;
+
+	iter->isize = (short)(state->size*10.0f);
+	iter->iblur = (short)state->blur;
+	iter->scale = fons__tt_getPixelHeightScale(&iter->font->font, (float)iter->isize/10.0f);
+
+	// Align horizontally
+	if (state->align & FONS_ALIGN_LEFT) {
+		// empty
+	} else if (state->align & FONS_ALIGN_RIGHT) {
+		width = fonsTextBounds(stash, x,y, str, end, NULL);
+		x -= width;
+	} else if (state->align & FONS_ALIGN_CENTER) {
+		width = fonsTextBounds(stash, x,y, str, end, NULL);
+		x -= width * 0.5f;
+	}
+	// Align vertically.
+	y += fons__getVertAlign(stash, iter->font, state->align, iter->isize);
+
+	if (end == NULL)
+		end = str + strlen(str);
+
+	iter->x = iter->nextx = x;
+	iter->y = iter->nexty = y;
+	iter->spacing = state->spacing;
+	iter->str = str;
+	iter->next = str;
+	iter->end = end;
+	iter->codepoint = 0;
+	iter->prevGlyphIndex = -1;
+
+	return 1;
+}
+
+int fonsTextIterNext(FONScontext* stash, FONStextIter* iter, FONSquad* quad)
+{
+	FONSglyph* glyph = NULL;
+	const char* str = iter->next;
+	iter->str = iter->next;
+
+	if (str == iter->end)
+		return 0;
+
+	for (; str != iter->end; str++) {
+		if (fons__decutf8(&iter->utf8state, &iter->codepoint, *(const unsigned char*)str))
+			continue;
+		str++;
+		// Get glyph and quad
+		iter->x = iter->nextx;
+		iter->y = iter->nexty;
+		glyph = fons__getGlyph(stash, iter->font, iter->codepoint, iter->isize, iter->iblur);
+		if (glyph != NULL)
+			fons__getQuad(stash, iter->font, iter->prevGlyphIndex, glyph, iter->scale, iter->spacing, &iter->nextx, &iter->nexty, quad);
+		iter->prevGlyphIndex = glyph != NULL ? glyph->index : -1;
+		break;
+	}
+	iter->next = str;
+
+	return 1;
+}
+
+void fonsDrawDebug(FONScontext* stash, float x, float y)
+{
+	int i;
+	int w = stash->params.width;
+	int h = stash->params.height;
+	float u = w == 0 ? 0 : (1.0f / w);
+	float v = h == 0 ? 0 : (1.0f / h);
+
+	if (stash->nverts+6+6 > FONS_VERTEX_COUNT)
+		fons__flush(stash);
+
+	// Draw background
+	fons__vertex(stash, x+0, y+0, u, v, 0x0fffffff);
+	fons__vertex(stash, x+w, y+h, u, v, 0x0fffffff);
+	fons__vertex(stash, x+w, y+0, u, v, 0x0fffffff);
+
+	fons__vertex(stash, x+0, y+0, u, v, 0x0fffffff);
+	fons__vertex(stash, x+0, y+h, u, v, 0x0fffffff);
+	fons__vertex(stash, x+w, y+h, u, v, 0x0fffffff);
+
+	// Draw texture
+	fons__vertex(stash, x+0, y+0, 0, 0, 0xffffffff);
+	fons__vertex(stash, x+w, y+h, 1, 1, 0xffffffff);
+	fons__vertex(stash, x+w, y+0, 1, 0, 0xffffffff);
+
+	fons__vertex(stash, x+0, y+0, 0, 0, 0xffffffff);
+	fons__vertex(stash, x+0, y+h, 0, 1, 0xffffffff);
+	fons__vertex(stash, x+w, y+h, 1, 1, 0xffffffff);
+
+	// Drawbug draw atlas
+	for (i = 0; i < stash->atlas->nnodes; i++) {
+		FONSatlasNode* n = &stash->atlas->nodes[i];
+
+		if (stash->nverts+6 > FONS_VERTEX_COUNT)
+			fons__flush(stash);
+
+		fons__vertex(stash, x+n->x+0, y+n->y+0, u, v, 0xc00000ff);
+		fons__vertex(stash, x+n->x+n->width, y+n->y+1, u, v, 0xc00000ff);
+		fons__vertex(stash, x+n->x+n->width, y+n->y+0, u, v, 0xc00000ff);
+
+		fons__vertex(stash, x+n->x+0, y+n->y+0, u, v, 0xc00000ff);
+		fons__vertex(stash, x+n->x+0, y+n->y+1, u, v, 0xc00000ff);
+		fons__vertex(stash, x+n->x+n->width, y+n->y+1, u, v, 0xc00000ff);
+	}
+
+	fons__flush(stash);
+}
+
+float fonsTextBounds(FONScontext* stash,
+					 float x, float y, 
+					 const char* str, const char* end,
+					 float* bounds)
+{
+	FONSstate* state = fons__getState(stash);
+	unsigned int codepoint;
+	unsigned int utf8state = 0;
+	FONSquad q;
+	FONSglyph* glyph = NULL;
+	int prevGlyphIndex = -1;
+	short isize = (short)(state->size*10.0f);
+	short iblur = (short)state->blur;
+	float scale;
+	FONSfont* font;
+	float startx, advance;
+	float minx, miny, maxx, maxy;
+
+	if (stash == NULL) return 0;
+	if (state->font < 0 || state->font >= stash->nfonts) return 0;
+	font = stash->fonts[state->font];
+	if (font->data == NULL) return 0;
+
+	scale = fons__tt_getPixelHeightScale(&font->font, (float)isize/10.0f);
+
+	// Align vertically.
+	y += fons__getVertAlign(stash, font, state->align, isize);
+
+	minx = maxx = x;
+	miny = maxy = y;
+	startx = x;
+
+	if (end == NULL)
+		end = str + strlen(str);
+
+	for (; str != end; ++str) {
+		if (fons__decutf8(&utf8state, &codepoint, *(const unsigned char*)str))
+			continue;
+		glyph = fons__getGlyph(stash, font, codepoint, isize, iblur);
+		if (glyph != NULL) {
+			fons__getQuad(stash, font, prevGlyphIndex, glyph, scale, state->spacing, &x, &y, &q);
+			if (q.x0 < minx) minx = q.x0;
+			if (q.x1 > maxx) maxx = q.x1;
+			if (stash->params.flags & FONS_ZERO_TOPLEFT) {
+				if (q.y0 < miny) miny = q.y0;
+				if (q.y1 > maxy) maxy = q.y1;
+			} else {
+				if (q.y1 < miny) miny = q.y1;
+				if (q.y0 > maxy) maxy = q.y0;
+			}
+		}
+		prevGlyphIndex = glyph != NULL ? glyph->index : -1;
+	}
+
+	advance = x - startx;
+
+	// Align horizontally
+	if (state->align & FONS_ALIGN_LEFT) {
+		// empty
+	} else if (state->align & FONS_ALIGN_RIGHT) {
+		minx -= advance;
+		maxx -= advance;
+	} else if (state->align & FONS_ALIGN_CENTER) {
+		minx -= advance * 0.5f;
+		maxx -= advance * 0.5f;
+	}
+
+	if (bounds) {
+		bounds[0] = minx;
+		bounds[1] = miny;
+		bounds[2] = maxx;
+		bounds[3] = maxy;
+	}
+
+	return advance;
+}
+
+void fonsVertMetrics(FONScontext* stash,
+					 float* ascender, float* descender, float* lineh)
+{
+	FONSfont* font;
+	FONSstate* state = fons__getState(stash);
+	short isize;
+
+	if (stash == NULL) return;
+	if (state->font < 0 || state->font >= stash->nfonts) return;
+	font = stash->fonts[state->font];
+	isize = (short)(state->size*10.0f);
+	if (font->data == NULL) return;
+
+	if (ascender)
+		*ascender = font->ascender*isize/10.0f;
+	if (descender)
+		*descender = font->descender*isize/10.0f;
+	if (lineh)
+		*lineh = font->lineh*isize/10.0f;
+}
+
+void fonsLineBounds(FONScontext* stash, float y, float* miny, float* maxy)
+{
+	FONSfont* font;
+	FONSstate* state = fons__getState(stash);
+	short isize;
+
+	if (stash == NULL) return;
+	if (state->font < 0 || state->font >= stash->nfonts) return;
+	font = stash->fonts[state->font];
+	isize = (short)(state->size*10.0f);
+	if (font->data == NULL) return;
+
+	y += fons__getVertAlign(stash, font, state->align, isize);
+
+	if (stash->params.flags & FONS_ZERO_TOPLEFT) {
+		*miny = y - font->ascender * (float)isize/10.0f;
+		*maxy = *miny + font->lineh*isize/10.0f;
+	} else {
+		*maxy = y + font->descender * (float)isize/10.0f;
+		*miny = *maxy - font->lineh*isize/10.0f;
+	}
+}
+
+const unsigned char* fonsGetTextureData(FONScontext* stash, int* width, int* height)
+{
+	if (width != NULL)
+		*width = stash->params.width;
+	if (height != NULL)
+		*height = stash->params.height;
+	return stash->texData;
+}
+
+int fonsValidateTexture(FONScontext* stash, int* dirty)
+{
+	if (stash->dirtyRect[0] < stash->dirtyRect[2] && stash->dirtyRect[1] < stash->dirtyRect[3]) {
+		dirty[0] = stash->dirtyRect[0];
+		dirty[1] = stash->dirtyRect[1];
+		dirty[2] = stash->dirtyRect[2];
+		dirty[3] = stash->dirtyRect[3];
+		// Reset dirty rect
+		stash->dirtyRect[0] = stash->params.width;
+		stash->dirtyRect[1] = stash->params.height;
+		stash->dirtyRect[2] = 0;
+		stash->dirtyRect[3] = 0;
+		return 1;
+	}
+	return 0;
+}
+
+void fonsDeleteInternal(FONScontext* stash)
+{
+	int i;
+	if (stash == NULL) return;
+
+	if (stash->params.renderDelete)
+		stash->params.renderDelete(stash->params.userPtr);
+
+	for (i = 0; i < stash->nfonts; ++i)
+		fons__freeFont(stash->fonts[i]);
+
+	if (stash->atlas) fons__deleteAtlas(stash->atlas);
+	if (stash->fonts) free(stash->fonts);
+	if (stash->texData) free(stash->texData);
+	if (stash->scratch) free(stash->scratch);
+	free(stash);
+}
+
+void fonsSetErrorCallback(FONScontext* stash, void (*callback)(void* uptr, int error, int val), void* uptr)
+{
+	if (stash == NULL) return;
+	stash->handleError = callback;
+	stash->errorUptr = uptr;
+}
+
+void fonsGetAtlasSize(FONScontext* stash, int* width, int* height)
+{
+	if (stash == NULL) return;
+	*width = stash->params.width;
+	*height = stash->params.height;
+}
+
+int fonsExpandAtlas(FONScontext* stash, int width, int height)
+{
+	int i, maxy = 0;
+	unsigned char* data = NULL;
+	if (stash == NULL) return 0;
+
+	width = fons__maxi(width, stash->params.width);
+	height = fons__maxi(height, stash->params.height);
+
+	if (width == stash->params.width && height == stash->params.height)
+		return 1;	
+
+	// Flush pending glyphs.
+	fons__flush(stash);
+
+	// Create new texture
+	if (stash->params.renderResize != NULL) {
+		if (stash->params.renderResize(stash->params.userPtr, width, height) == 0)
+			return 0;
+	}
+	// Copy old texture data over.
+	data = (unsigned char*)malloc(width * height);
+	if (data == NULL)
+		return 0;
+	for (i = 0; i < stash->params.height; i++) {
+		unsigned char* dst = &data[i*width];
+		unsigned char* src = &stash->texData[i*stash->params.width];
+		memcpy(dst, src, stash->params.width);
+		if (width > stash->params.width)
+			memset(dst+stash->params.width, 0, width - stash->params.width);
+	}
+	if (height > stash->params.height)
+		memset(&data[stash->params.height * width], 0, (height - stash->params.height) * width);
+
+	free(stash->texData);
+	stash->texData = data;
+
+	// Increase atlas size
+	fons__atlasExpand(stash->atlas, width, height);
+
+	// Add existing data as dirty.
+	for (i = 0; i < stash->atlas->nnodes; i++)
+		maxy = fons__maxi(maxy, stash->atlas->nodes[i].y);
+	stash->dirtyRect[0] = 0;
+	stash->dirtyRect[1] = 0;
+	stash->dirtyRect[2] = stash->params.width;
+	stash->dirtyRect[3] = maxy;
+
+	stash->params.width = width;
+	stash->params.height = height;
+	stash->itw = 1.0f/stash->params.width;
+	stash->ith = 1.0f/stash->params.height;
+
+	return 1;
+}
+
+int fonsResetAtlas(FONScontext* stash, int width, int height)
+{
+	int i, j;
+	if (stash == NULL) return 0;
+
+	// Flush pending glyphs.
+	fons__flush(stash);
+
+	// Create new texture
+	if (stash->params.renderResize != NULL) {
+		if (stash->params.renderResize(stash->params.userPtr, width, height) == 0)
+			return 0;
+	}
+
+	// Reset atlas
+	fons__atlasReset(stash->atlas, width, height);
+
+	// Clear texture data.
+	stash->texData = (unsigned char*)realloc(stash->texData, width * height);
+	if (stash->texData == NULL) return 0;
+	memset(stash->texData, 0, width * height);
+
+	// Reset dirty rect
+	stash->dirtyRect[0] = width;
+	stash->dirtyRect[1] = height;
+	stash->dirtyRect[2] = 0;
+	stash->dirtyRect[3] = 0;
+
+	// Reset cached glyphs
+	for (i = 0; i < stash->nfonts; i++) {
+		FONSfont* font = stash->fonts[i];
+		font->nglyphs = 0;
+		for (j = 0; j < FONS_HASH_LUT_SIZE; j++)
+			font->lut[j] = -1;
+	}
+
+	stash->params.width = width;
+	stash->params.height = height;
+	stash->itw = 1.0f/stash->params.width;
+	stash->ith = 1.0f/stash->params.height;
+
+	// Add white rect at 0,0 for debug drawing.
+	fons__addWhiteRect(stash, 2,2);
+
+	return 1;
+}
+
+
+#endif
diff --git a/dgl/src/nanovg2/nanovg.c b/dgl/src/nanovg2/nanovg.c
new file mode 100644
index 00000000..44cf25f8
--- /dev/null
+++ b/dgl/src/nanovg2/nanovg.c
@@ -0,0 +1,2752 @@
+//
+// Copyright (c) 2013 Mikko Mononen memon@inside.org
+//
+// This software is provided 'as-is', without any express or implied
+// warranty.  In no event will the authors be held liable for any damages
+// arising from the use of this software.
+// Permission is granted to anyone to use this software for any purpose,
+// including commercial applications, and to alter it and redistribute it
+// freely, subject to the following restrictions:
+// 1. The origin of this software must not be misrepresented; you must not
+//    claim that you wrote the original software. If you use this software
+//    in a product, an acknowledgment in the product documentation would be
+//    appreciated but is not required.
+// 2. Altered source versions must be plainly marked as such, and must not be
+//    misrepresented as being the original software.
+// 3. This notice may not be removed or altered from any source distribution.
+//
+
+#include <stdio.h>
+#include <math.h>
+#include "nanovg.h"
+#define FONTSTASH_IMPLEMENTATION
+#include "fontstash.h"
+#define STB_IMAGE_IMPLEMENTATION
+#include "stb_image.h"
+
+#ifdef _MSC_VER
+#pragma warning(disable: 4100)  // unreferenced formal parameter
+#pragma warning(disable: 4127)  // conditional expression is constant
+#pragma warning(disable: 4204)  // nonstandard extension used : non-constant aggregate initializer
+#pragma warning(disable: 4706)  // assignment within conditional expression
+#endif
+
+#define NVG_INIT_FONTIMAGE_SIZE  512
+#define NVG_MAX_FONTIMAGE_SIZE   2048
+#define NVG_MAX_FONTIMAGES       4
+
+#define NVG_INIT_COMMANDS_SIZE 256
+#define NVG_INIT_POINTS_SIZE 128
+#define NVG_INIT_PATHS_SIZE 16
+#define NVG_INIT_VERTS_SIZE 256
+#define NVG_MAX_STATES 32
+
+#define NVG_KAPPA90 0.5522847493f	// Length proportional to radius of a cubic bezier handle for 90deg arcs.
+
+#define NVG_COUNTOF(arr) (sizeof(arr) / sizeof(0[arr]))
+
+
+enum NVGcommands {
+	NVG_MOVETO = 0,
+	NVG_LINETO = 1,
+	NVG_BEZIERTO = 2,
+	NVG_CLOSE = 3,
+	NVG_WINDING = 4,
+};
+
+enum NVGpointFlags
+{
+	NVG_PT_CORNER = 0x01,
+	NVG_PT_LEFT = 0x02,
+	NVG_PT_BEVEL = 0x04,
+	NVG_PR_INNERBEVEL = 0x08,
+};
+
+struct NVGstate {
+	NVGpaint fill;
+	NVGpaint stroke;
+	float strokeWidth;
+	float miterLimit;
+	int lineJoin;
+	int lineCap;
+	float alpha;
+	float xform[6];
+	NVGscissor scissor;
+	float fontSize;
+	float letterSpacing;
+	float lineHeight;
+	float fontBlur;
+	int textAlign;
+	int fontId;
+};
+typedef struct NVGstate NVGstate;
+
+struct NVGpoint {
+	float x,y;
+	float dx, dy;
+	float len;
+	float dmx, dmy;
+	unsigned char flags;
+};
+typedef struct NVGpoint NVGpoint;
+
+struct NVGpathCache {
+	NVGpoint* points;
+	int npoints;
+	int cpoints;
+	NVGpath* paths;
+	int npaths;
+	int cpaths;
+	NVGvertex* verts;
+	int nverts;
+	int cverts;
+	float bounds[4];
+};
+typedef struct NVGpathCache NVGpathCache;
+
+struct NVGcontext {
+	NVGparams params;
+	float* commands;
+	int ccommands;
+	int ncommands;
+	float commandx, commandy;
+	NVGstate states[NVG_MAX_STATES];
+	int nstates;
+	NVGpathCache* cache;
+	float tessTol;
+	float distTol;
+	float fringeWidth;
+	float devicePxRatio;
+	struct FONScontext* fs;
+	int fontImages[NVG_MAX_FONTIMAGES];
+	int fontImageIdx;
+	int drawCallCount;
+	int fillTriCount;
+	int strokeTriCount;
+	int textTriCount;
+};
+
+static float nvg__sqrtf(float a) { return sqrtf(a); }
+static float nvg__modf(float a, float b) { return fmodf(a, b); }
+static float nvg__sinf(float a) { return sinf(a); }
+static float nvg__cosf(float a) { return cosf(a); }
+static float nvg__tanf(float a) { return tanf(a); }
+static float nvg__atan2f(float a,float b) { return atan2f(a, b); }
+static float nvg__acosf(float a) { return acosf(a); }
+
+static int nvg__mini(int a, int b) { return a < b ? a : b; }
+static int nvg__maxi(int a, int b) { return a > b ? a : b; }
+static int nvg__clampi(int a, int mn, int mx) { return a < mn ? mn : (a > mx ? mx : a); }
+static float nvg__minf(float a, float b) { return a < b ? a : b; }
+static float nvg__maxf(float a, float b) { return a > b ? a : b; }
+static float nvg__absf(float a) { return a >= 0.0f ? a : -a; }
+static float nvg__signf(float a) { return a >= 0.0f ? 1.0f : -1.0f; }
+static float nvg__clampf(float a, float mn, float mx) { return a < mn ? mn : (a > mx ? mx : a); }
+static float nvg__cross(float dx0, float dy0, float dx1, float dy1) { return dx1*dy0 - dx0*dy1; }
+
+static float nvg__normalize(float *x, float* y)
+{
+	float d = nvg__sqrtf((*x)*(*x) + (*y)*(*y));
+	if (d > 1e-6f) {
+		float id = 1.0f / d;
+		*x *= id;
+		*y *= id;
+	}
+	return d;
+}
+
+
+static void nvg__deletePathCache(NVGpathCache* c)
+{
+	if (c == NULL) return;
+	if (c->points != NULL) free(c->points);
+	if (c->paths != NULL) free(c->paths);
+	if (c->verts != NULL) free(c->verts);
+	free(c);
+}
+
+static NVGpathCache* nvg__allocPathCache(void)
+{
+	NVGpathCache* c = (NVGpathCache*)malloc(sizeof(NVGpathCache));
+	if (c == NULL) goto error;
+	memset(c, 0, sizeof(NVGpathCache));
+
+	c->points = (NVGpoint*)malloc(sizeof(NVGpoint)*NVG_INIT_POINTS_SIZE);
+	if (!c->points) goto error;
+	c->npoints = 0;
+	c->cpoints = NVG_INIT_POINTS_SIZE;
+
+	c->paths = (NVGpath*)malloc(sizeof(NVGpath)*NVG_INIT_PATHS_SIZE);
+	if (!c->paths) goto error;
+	c->npaths = 0;
+	c->cpaths = NVG_INIT_PATHS_SIZE;
+
+	c->verts = (NVGvertex*)malloc(sizeof(NVGvertex)*NVG_INIT_VERTS_SIZE);
+	if (!c->verts) goto error;
+	c->nverts = 0;
+	c->cverts = NVG_INIT_VERTS_SIZE;
+
+	return c;
+error:
+	nvg__deletePathCache(c);
+	return NULL;
+}
+
+static void nvg__setDevicePixelRatio(NVGcontext* ctx, float ratio)
+{
+	ctx->tessTol = 0.25f / ratio;
+	ctx->distTol = 0.01f / ratio;
+	ctx->fringeWidth = 1.0f / ratio;
+	ctx->devicePxRatio = ratio;
+}
+
+NVGcontext* nvgCreateInternal(NVGparams* params)
+{
+	FONSparams fontParams;
+	NVGcontext* ctx = (NVGcontext*)malloc(sizeof(NVGcontext));
+	int i;
+	if (ctx == NULL) goto error;
+	memset(ctx, 0, sizeof(NVGcontext));
+
+	ctx->params = *params;
+	for (i = 0; i < NVG_MAX_FONTIMAGES; i++)
+		ctx->fontImages[i] = 0;
+
+	ctx->commands = (float*)malloc(sizeof(float)*NVG_INIT_COMMANDS_SIZE);
+	if (!ctx->commands) goto error;
+	ctx->ncommands = 0;
+	ctx->ccommands = NVG_INIT_COMMANDS_SIZE;
+
+	ctx->cache = nvg__allocPathCache();
+	if (ctx->cache == NULL) goto error;
+
+	nvgSave(ctx);
+	nvgReset(ctx);
+
+	nvg__setDevicePixelRatio(ctx, 1.0f);
+
+	if (ctx->params.renderCreate(ctx->params.userPtr) == 0) goto error;
+
+	// Init font rendering
+	memset(&fontParams, 0, sizeof(fontParams));
+	fontParams.width = NVG_INIT_FONTIMAGE_SIZE;
+	fontParams.height = NVG_INIT_FONTIMAGE_SIZE;
+	fontParams.flags = FONS_ZERO_TOPLEFT;
+	fontParams.renderCreate = NULL;
+	fontParams.renderUpdate = NULL;
+	fontParams.renderDraw = NULL;
+	fontParams.renderDelete = NULL;
+	fontParams.userPtr = NULL;
+	ctx->fs = fonsCreateInternal(&fontParams);
+	if (ctx->fs == NULL) goto error;
+
+	// Create font texture
+	ctx->fontImages[0] = ctx->params.renderCreateTexture(ctx->params.userPtr, NVG_TEXTURE_ALPHA, fontParams.width, fontParams.height, 0, NULL);
+	if (ctx->fontImages[0] == 0) goto error;
+	ctx->fontImageIdx = 0;
+
+	return ctx;
+
+error:
+	nvgDeleteInternal(ctx);
+	return 0;
+}
+
+NVGparams* nvgInternalParams(NVGcontext* ctx)
+{
+    return &ctx->params;
+}
+
+void nvgDeleteInternal(NVGcontext* ctx)
+{
+	int i;
+	if (ctx == NULL) return;
+	if (ctx->commands != NULL) free(ctx->commands);
+	if (ctx->cache != NULL) nvg__deletePathCache(ctx->cache);
+
+	if (ctx->fs)
+		fonsDeleteInternal(ctx->fs);
+
+	for (i = 0; i < NVG_MAX_FONTIMAGES; i++) {
+		if (ctx->fontImages[i] != 0) {
+			nvgDeleteImage(ctx, ctx->fontImages[i]);
+			ctx->fontImages[i] = 0;
+		}
+	}
+
+	if (ctx->params.renderDelete != NULL)
+		ctx->params.renderDelete(ctx->params.userPtr);
+
+	free(ctx);
+}
+
+void nvgBeginFrame(NVGcontext* ctx, int windowWidth, int windowHeight, float devicePixelRatio)
+{
+/*	printf("Tris: draws:%d  fill:%d  stroke:%d  text:%d  TOT:%d\n",
+		ctx->drawCallCount, ctx->fillTriCount, ctx->strokeTriCount, ctx->textTriCount,
+		ctx->fillTriCount+ctx->strokeTriCount+ctx->textTriCount);*/
+
+	ctx->nstates = 0;
+	nvgSave(ctx);
+	nvgReset(ctx);
+
+	nvg__setDevicePixelRatio(ctx, devicePixelRatio);
+	
+	ctx->params.renderViewport(ctx->params.userPtr, windowWidth, windowHeight);
+
+	ctx->drawCallCount = 0;
+	ctx->fillTriCount = 0;
+	ctx->strokeTriCount = 0;
+	ctx->textTriCount = 0;
+}
+
+void nvgCancelFrame(NVGcontext* ctx)
+{
+	ctx->params.renderCancel(ctx->params.userPtr);
+}
+
+void nvgEndFrame(NVGcontext* ctx)
+{
+	ctx->params.renderFlush(ctx->params.userPtr);
+	if (ctx->fontImageIdx != 0) {
+		int fontImage = ctx->fontImages[ctx->fontImageIdx];
+		int i, j, iw, ih;
+		// delete images that smaller than current one
+		if (fontImage == 0)
+			return;
+		nvgImageSize(ctx, fontImage, &iw, &ih);
+		for (i = j = 0; i < ctx->fontImageIdx; i++) {
+			if (ctx->fontImages[i] != 0) {
+				int nw, nh;
+				nvgImageSize(ctx, ctx->fontImages[i], &nw, &nh);
+				if (nw < iw || nh < ih)
+					nvgDeleteImage(ctx, ctx->fontImages[i]);
+				else
+					ctx->fontImages[j++] = ctx->fontImages[i];
+			}
+		}
+		// make current font image to first
+		ctx->fontImages[j++] = ctx->fontImages[0];
+		ctx->fontImages[0] = fontImage;
+		ctx->fontImageIdx = 0;
+		// clear all images after j
+		for (i = j; i < NVG_MAX_FONTIMAGES; i++)
+			ctx->fontImages[i] = 0;
+	}
+}
+
+NVGcolor nvgRGB(unsigned char r, unsigned char g, unsigned char b)
+{
+	return nvgRGBA(r,g,b,255);
+}
+
+NVGcolor nvgRGBf(float r, float g, float b)
+{
+	return nvgRGBAf(r,g,b,1.0f);
+}
+
+NVGcolor nvgRGBA(unsigned char r, unsigned char g, unsigned char b, unsigned char a)
+{
+	NVGcolor color;
+	// Use longer initialization to suppress warning.
+	color.r = r / 255.0f;
+	color.g = g / 255.0f;
+	color.b = b / 255.0f;
+	color.a = a / 255.0f;
+	return color;
+}
+
+NVGcolor nvgRGBAf(float r, float g, float b, float a)
+{
+	NVGcolor color;
+	// Use longer initialization to suppress warning.
+	color.r = r;
+	color.g = g;
+	color.b = b;
+	color.a = a;
+	return color;
+}
+
+NVGcolor nvgTransRGBA(NVGcolor c, unsigned char a)
+{
+	c.a = a / 255.0f;
+	return c;
+}
+
+NVGcolor nvgTransRGBAf(NVGcolor c, float a)
+{
+	c.a = a;
+	return c;
+}
+
+NVGcolor nvgLerpRGBA(NVGcolor c0, NVGcolor c1, float u)
+{
+	int i;
+	float oneminu;
+	NVGcolor cint;
+
+	u = nvg__clampf(u, 0.0f, 1.0f);
+	oneminu = 1.0f - u;
+	for( i = 0; i <4; i++ )
+	{
+		cint.rgba[i] = c0.rgba[i] * oneminu + c1.rgba[i] * u;
+	}
+	 
+	return cint;
+}
+
+NVGcolor nvgHSL(float h, float s, float l)
+{
+	return nvgHSLA(h,s,l,255);
+}
+
+static float nvg__hue(float h, float m1, float m2)
+{
+	if (h < 0) h += 1;
+	if (h > 1) h -= 1;
+	if (h < 1.0f/6.0f)
+		return m1 + (m2 - m1) * h * 6.0f;
+	else if (h < 3.0f/6.0f)
+		return m2;
+	else if (h < 4.0f/6.0f)
+		return m1 + (m2 - m1) * (2.0f/3.0f - h) * 6.0f;
+	return m1;
+}
+
+NVGcolor nvgHSLA(float h, float s, float l, unsigned char a)
+{
+	float m1, m2;
+	NVGcolor col;
+	h = nvg__modf(h, 1.0f);
+	if (h < 0.0f) h += 1.0f;
+	s = nvg__clampf(s, 0.0f, 1.0f);
+	l = nvg__clampf(l, 0.0f, 1.0f);
+	m2 = l <= 0.5f ? (l * (1 + s)) : (l + s - l * s);
+	m1 = 2 * l - m2;
+	col.r = nvg__clampf(nvg__hue(h + 1.0f/3.0f, m1, m2), 0.0f, 1.0f);
+	col.g = nvg__clampf(nvg__hue(h, m1, m2), 0.0f, 1.0f);
+	col.b = nvg__clampf(nvg__hue(h - 1.0f/3.0f, m1, m2), 0.0f, 1.0f);
+	col.a = a/255.0f;
+	return col;
+}
+
+
+static NVGstate* nvg__getState(NVGcontext* ctx)
+{
+	return &ctx->states[ctx->nstates-1];
+}
+
+void nvgTransformIdentity(float* t)
+{
+	t[0] = 1.0f; t[1] = 0.0f;
+	t[2] = 0.0f; t[3] = 1.0f;
+	t[4] = 0.0f; t[5] = 0.0f;
+}
+
+void nvgTransformTranslate(float* t, float tx, float ty)
+{
+	t[0] = 1.0f; t[1] = 0.0f;
+	t[2] = 0.0f; t[3] = 1.0f;
+	t[4] = tx; t[5] = ty;
+}
+
+void nvgTransformScale(float* t, float sx, float sy)
+{
+	t[0] = sx; t[1] = 0.0f;
+	t[2] = 0.0f; t[3] = sy;
+	t[4] = 0.0f; t[5] = 0.0f;
+}
+
+void nvgTransformRotate(float* t, float a)
+{
+	float cs = nvg__cosf(a), sn = nvg__sinf(a);
+	t[0] = cs; t[1] = sn;
+	t[2] = -sn; t[3] = cs;
+	t[4] = 0.0f; t[5] = 0.0f;
+}
+
+void nvgTransformSkewX(float* t, float a)
+{
+	t[0] = 1.0f; t[1] = 0.0f;
+	t[2] = nvg__tanf(a); t[3] = 1.0f;
+	t[4] = 0.0f; t[5] = 0.0f;
+}
+
+void nvgTransformSkewY(float* t, float a)
+{
+	t[0] = 1.0f; t[1] = nvg__tanf(a);
+	t[2] = 0.0f; t[3] = 1.0f;
+	t[4] = 0.0f; t[5] = 0.0f;
+}
+
+void nvgTransformMultiply(float* t, const float* s)
+{
+	float t0 = t[0] * s[0] + t[1] * s[2];
+	float t2 = t[2] * s[0] + t[3] * s[2];
+	float t4 = t[4] * s[0] + t[5] * s[2] + s[4];
+	t[1] = t[0] * s[1] + t[1] * s[3];
+	t[3] = t[2] * s[1] + t[3] * s[3];
+	t[5] = t[4] * s[1] + t[5] * s[3] + s[5];
+	t[0] = t0;
+	t[2] = t2;
+	t[4] = t4;
+}
+
+void nvgTransformPremultiply(float* t, const float* s)
+{
+	float s2[6];
+	memcpy(s2, s, sizeof(float)*6);
+	nvgTransformMultiply(s2, t);
+	memcpy(t, s2, sizeof(float)*6);
+}
+
+int nvgTransformInverse(float* inv, const float* t)
+{
+	double invdet, det = (double)t[0] * t[3] - (double)t[2] * t[1];
+	if (det > -1e-6 && det < 1e-6) {
+		nvgTransformIdentity(inv);
+		return 0;
+	}
+	invdet = 1.0 / det;
+	inv[0] = (float)(t[3] * invdet);
+	inv[2] = (float)(-t[2] * invdet);
+	inv[4] = (float)(((double)t[2] * t[5] - (double)t[3] * t[4]) * invdet);
+	inv[1] = (float)(-t[1] * invdet);
+	inv[3] = (float)(t[0] * invdet);
+	inv[5] = (float)(((double)t[1] * t[4] - (double)t[0] * t[5]) * invdet);
+	return 1;
+}
+
+void nvgTransformPoint(float* dx, float* dy, const float* t, float sx, float sy)
+{
+	*dx = sx*t[0] + sy*t[2] + t[4];
+	*dy = sx*t[1] + sy*t[3] + t[5];
+}
+
+float nvgDegToRad(float deg)
+{
+	return deg / 180.0f * NVG_PI;
+}
+
+float nvgRadToDeg(float rad)
+{
+	return rad / NVG_PI * 180.0f;
+}
+
+static void nvg__setPaintColor(NVGpaint* p, NVGcolor color)
+{
+	memset(p, 0, sizeof(*p));
+	nvgTransformIdentity(p->xform);
+	p->radius = 0.0f;
+	p->feather = 1.0f;
+	p->innerColor = color;
+	p->outerColor = color;
+}
+
+
+// State handling
+void nvgSave(NVGcontext* ctx)
+{
+	if (ctx->nstates >= NVG_MAX_STATES)
+		return;
+	if (ctx->nstates > 0)
+		memcpy(&ctx->states[ctx->nstates], &ctx->states[ctx->nstates-1], sizeof(NVGstate));
+	ctx->nstates++;
+}
+
+void nvgRestore(NVGcontext* ctx)
+{
+	if (ctx->nstates <= 1)
+		return;
+	ctx->nstates--;
+}
+
+void nvgReset(NVGcontext* ctx)
+{
+	NVGstate* state = nvg__getState(ctx);
+	memset(state, 0, sizeof(*state));
+
+	nvg__setPaintColor(&state->fill, nvgRGBA(255,255,255,255));
+	nvg__setPaintColor(&state->stroke, nvgRGBA(0,0,0,255));
+	state->strokeWidth = 1.0f;
+	state->miterLimit = 10.0f;
+	state->lineCap = NVG_BUTT;
+	state->lineJoin = NVG_MITER;
+	state->alpha = 1.0f;
+	nvgTransformIdentity(state->xform);
+
+	state->scissor.extent[0] = -1.0f;
+	state->scissor.extent[1] = -1.0f;
+
+	state->fontSize = 16.0f;
+	state->letterSpacing = 0.0f;
+	state->lineHeight = 1.0f;
+	state->fontBlur = 0.0f;
+	state->textAlign = NVG_ALIGN_LEFT | NVG_ALIGN_BASELINE;
+	state->fontId = 0;
+}
+
+// State setting
+void nvgStrokeWidth(NVGcontext* ctx, float width)
+{
+	NVGstate* state = nvg__getState(ctx);
+	state->strokeWidth = width;
+}
+
+void nvgMiterLimit(NVGcontext* ctx, float limit)
+{
+	NVGstate* state = nvg__getState(ctx);
+	state->miterLimit = limit;
+}
+
+void nvgLineCap(NVGcontext* ctx, int cap)
+{
+	NVGstate* state = nvg__getState(ctx);
+	state->lineCap = cap;
+}
+
+void nvgLineJoin(NVGcontext* ctx, int join)
+{
+	NVGstate* state = nvg__getState(ctx);
+	state->lineJoin = join;
+}
+
+void nvgGlobalAlpha(NVGcontext* ctx, float alpha)
+{
+	NVGstate* state = nvg__getState(ctx);
+	state->alpha = alpha;
+}
+
+void nvgTransform(NVGcontext* ctx, float a, float b, float c, float d, float e, float f)
+{
+	NVGstate* state = nvg__getState(ctx);
+	float t[6] = { a, b, c, d, e, f };
+	nvgTransformPremultiply(state->xform, t);
+}
+
+void nvgResetTransform(NVGcontext* ctx)
+{
+	NVGstate* state = nvg__getState(ctx);
+	nvgTransformIdentity(state->xform);
+}
+
+void nvgTranslate(NVGcontext* ctx, float x, float y)
+{
+	NVGstate* state = nvg__getState(ctx);
+	float t[6];
+	nvgTransformTranslate(t, x,y);
+	nvgTransformPremultiply(state->xform, t);
+}
+
+void nvgRotate(NVGcontext* ctx, float angle)
+{
+	NVGstate* state = nvg__getState(ctx);
+	float t[6];
+	nvgTransformRotate(t, angle);
+	nvgTransformPremultiply(state->xform, t);
+}
+
+void nvgSkewX(NVGcontext* ctx, float angle)
+{
+	NVGstate* state = nvg__getState(ctx);
+	float t[6];
+	nvgTransformSkewX(t, angle);
+	nvgTransformPremultiply(state->xform, t);
+}
+
+void nvgSkewY(NVGcontext* ctx, float angle)
+{
+	NVGstate* state = nvg__getState(ctx);
+	float t[6];
+	nvgTransformSkewY(t, angle);
+	nvgTransformPremultiply(state->xform, t);
+}
+
+void nvgScale(NVGcontext* ctx, float x, float y)
+{
+	NVGstate* state = nvg__getState(ctx);
+	float t[6];
+	nvgTransformScale(t, x,y);
+	nvgTransformPremultiply(state->xform, t);
+}
+
+void nvgCurrentTransform(NVGcontext* ctx, float* xform)
+{
+	NVGstate* state = nvg__getState(ctx);
+	if (xform == NULL) return;
+	memcpy(xform, state->xform, sizeof(float)*6);
+}
+
+void nvgStrokeColor(NVGcontext* ctx, NVGcolor color)
+{
+	NVGstate* state = nvg__getState(ctx);
+	nvg__setPaintColor(&state->stroke, color);
+}
+
+void nvgStrokePaint(NVGcontext* ctx, NVGpaint paint)
+{
+	NVGstate* state = nvg__getState(ctx);
+	state->stroke = paint;
+	nvgTransformMultiply(state->stroke.xform, state->xform);
+}
+
+void nvgFillColor(NVGcontext* ctx, NVGcolor color)
+{
+	NVGstate* state = nvg__getState(ctx);
+	nvg__setPaintColor(&state->fill, color);
+}
+
+void nvgFillPaint(NVGcontext* ctx, NVGpaint paint)
+{
+	NVGstate* state = nvg__getState(ctx);
+	state->fill = paint;
+	nvgTransformMultiply(state->fill.xform, state->xform);
+}
+
+int nvgCreateImage(NVGcontext* ctx, const char* filename, int imageFlags)
+{
+	int w, h, n, image;
+	unsigned char* img;
+	stbi_set_unpremultiply_on_load(1);
+	stbi_convert_iphone_png_to_rgb(1);
+	img = stbi_load(filename, &w, &h, &n, 4);
+	if (img == NULL) {
+//		printf("Failed to load %s - %s\n", filename, stbi_failure_reason());
+		return 0;
+	}
+	image = nvgCreateImageRGBA(ctx, w, h, imageFlags, img);
+	stbi_image_free(img);
+	return image;
+}
+
+int nvgCreateImageMem(NVGcontext* ctx, int imageFlags, unsigned char* data, int ndata)
+{
+	int w, h, n, image;
+	unsigned char* img = stbi_load_from_memory(data, ndata, &w, &h, &n, 4);
+	if (img == NULL) {
+//		printf("Failed to load %s - %s\n", filename, stbi_failure_reason());
+		return 0;
+	}
+	image = nvgCreateImageRGBA(ctx, w, h, imageFlags, img);
+	stbi_image_free(img);
+	return image;
+}
+
+int nvgCreateImageRGBA(NVGcontext* ctx, int w, int h, int imageFlags, const unsigned char* data)
+{
+	return ctx->params.renderCreateTexture(ctx->params.userPtr, NVG_TEXTURE_RGBA, w, h, imageFlags, data);
+}
+
+void nvgUpdateImage(NVGcontext* ctx, int image, const unsigned char* data)
+{
+	int w, h;
+	ctx->params.renderGetTextureSize(ctx->params.userPtr, image, &w, &h);
+	ctx->params.renderUpdateTexture(ctx->params.userPtr, image, 0,0, w,h, data);
+}
+
+void nvgImageSize(NVGcontext* ctx, int image, int* w, int* h)
+{
+	ctx->params.renderGetTextureSize(ctx->params.userPtr, image, w, h);
+}
+
+void nvgDeleteImage(NVGcontext* ctx, int image)
+{
+	ctx->params.renderDeleteTexture(ctx->params.userPtr, image);
+}
+
+NVGpaint nvgLinearGradient(NVGcontext* ctx,
+								  float sx, float sy, float ex, float ey,
+								  NVGcolor icol, NVGcolor ocol)
+{
+	NVGpaint p;
+	float dx, dy, d;
+	const float large = 1e5;
+	NVG_NOTUSED(ctx);
+	memset(&p, 0, sizeof(p));
+
+	// Calculate transform aligned to the line
+	dx = ex - sx;
+	dy = ey - sy;
+	d = sqrtf(dx*dx + dy*dy);
+	if (d > 0.0001f) {
+		dx /= d;
+		dy /= d;
+	} else {
+		dx = 0;
+		dy = 1;
+	}
+
+	p.xform[0] = dy; p.xform[1] = -dx;
+	p.xform[2] = dx; p.xform[3] = dy;
+	p.xform[4] = sx - dx*large; p.xform[5] = sy - dy*large;
+
+	p.extent[0] = large;
+	p.extent[1] = large + d*0.5f;
+
+	p.radius = 0.0f;
+
+	p.feather = nvg__maxf(1.0f, d);
+
+	p.innerColor = icol;
+	p.outerColor = ocol;
+
+	return p;
+}
+
+NVGpaint nvgRadialGradient(NVGcontext* ctx,
+								  float cx, float cy, float inr, float outr,
+								  NVGcolor icol, NVGcolor ocol)
+{
+	NVGpaint p;
+	float r = (inr+outr)*0.5f;
+	float f = (outr-inr);
+	NVG_NOTUSED(ctx);
+	memset(&p, 0, sizeof(p));
+
+	nvgTransformIdentity(p.xform);
+	p.xform[4] = cx;
+	p.xform[5] = cy;
+
+	p.extent[0] = r;
+	p.extent[1] = r;
+
+	p.radius = r;
+
+	p.feather = nvg__maxf(1.0f, f);
+
+	p.innerColor = icol;
+	p.outerColor = ocol;
+
+	return p;
+}
+
+NVGpaint nvgBoxGradient(NVGcontext* ctx,
+							   float x, float y, float w, float h, float r, float f,
+							   NVGcolor icol, NVGcolor ocol)
+{
+	NVGpaint p;
+	NVG_NOTUSED(ctx);
+	memset(&p, 0, sizeof(p));
+
+	nvgTransformIdentity(p.xform);
+	p.xform[4] = x+w*0.5f;
+	p.xform[5] = y+h*0.5f;
+
+	p.extent[0] = w*0.5f;
+	p.extent[1] = h*0.5f;
+
+	p.radius = r;
+
+	p.feather = nvg__maxf(1.0f, f);
+
+	p.innerColor = icol;
+	p.outerColor = ocol;
+
+	return p;
+}
+
+
+NVGpaint nvgImagePattern(NVGcontext* ctx,
+								float cx, float cy, float w, float h, float angle,
+								int image, float alpha)
+{
+	NVGpaint p;
+	NVG_NOTUSED(ctx);
+	memset(&p, 0, sizeof(p));
+
+	nvgTransformRotate(p.xform, angle);
+	p.xform[4] = cx;
+	p.xform[5] = cy;
+
+	p.extent[0] = w;
+	p.extent[1] = h;
+
+	p.image = image;
+
+	p.innerColor = p.outerColor = nvgRGBAf(1,1,1,alpha);
+
+	return p;
+}
+
+// Scissoring
+void nvgScissor(NVGcontext* ctx, float x, float y, float w, float h)
+{
+	NVGstate* state = nvg__getState(ctx);
+
+	w = nvg__maxf(0.0f, w);
+	h = nvg__maxf(0.0f, h);
+
+	nvgTransformIdentity(state->scissor.xform);
+	state->scissor.xform[4] = x+w*0.5f;
+	state->scissor.xform[5] = y+h*0.5f;
+	nvgTransformMultiply(state->scissor.xform, state->xform);
+
+	state->scissor.extent[0] = w*0.5f;
+	state->scissor.extent[1] = h*0.5f;
+}
+
+static void nvg__isectRects(float* dst,
+							float ax, float ay, float aw, float ah,
+							float bx, float by, float bw, float bh)
+{
+	float minx = nvg__maxf(ax, bx);
+	float miny = nvg__maxf(ay, by);
+	float maxx = nvg__minf(ax+aw, bx+bw);
+	float maxy = nvg__minf(ay+ah, by+bh);
+	dst[0] = minx;
+	dst[1] = miny;
+	dst[2] = nvg__maxf(0.0f, maxx - minx);
+	dst[3] = nvg__maxf(0.0f, maxy - miny);
+}
+
+void nvgIntersectScissor(NVGcontext* ctx, float x, float y, float w, float h)
+{
+	NVGstate* state = nvg__getState(ctx);
+	float pxform[6], invxorm[6];
+	float rect[4];
+	float ex, ey, tex, tey;
+
+	// If no previous scissor has been set, set the scissor as current scissor.
+	if (state->scissor.extent[0] < 0) {
+		nvgScissor(ctx, x, y, w, h);
+		return;
+	}
+
+	// Transform the current scissor rect into current transform space.
+	// If there is difference in rotation, this will be approximation. 
+	memcpy(pxform, state->scissor.xform, sizeof(float)*6);
+	ex = state->scissor.extent[0];
+	ey = state->scissor.extent[1];
+	nvgTransformInverse(invxorm, state->xform);
+	nvgTransformMultiply(pxform, invxorm);
+	tex = ex*nvg__absf(pxform[0]) + ey*nvg__absf(pxform[2]);
+	tey = ex*nvg__absf(pxform[1]) + ey*nvg__absf(pxform[3]);
+
+	// Intersect rects.
+	nvg__isectRects(rect, pxform[4]-tex,pxform[5]-tey,tex*2,tey*2, x,y,w,h);
+
+	nvgScissor(ctx, rect[0], rect[1], rect[2], rect[3]);
+}
+
+void nvgResetScissor(NVGcontext* ctx)
+{
+	NVGstate* state = nvg__getState(ctx);
+	memset(state->scissor.xform, 0, sizeof(state->scissor.xform));
+	state->scissor.extent[0] = -1.0f;
+	state->scissor.extent[1] = -1.0f;
+}
+
+static int nvg__ptEquals(float x1, float y1, float x2, float y2, float tol)
+{
+	float dx = x2 - x1;
+	float dy = y2 - y1;
+	return dx*dx + dy*dy < tol*tol;
+}
+
+static float nvg__distPtSeg(float x, float y, float px, float py, float qx, float qy)
+{
+	float pqx, pqy, dx, dy, d, t;
+	pqx = qx-px;
+	pqy = qy-py;
+	dx = x-px;
+	dy = y-py;
+	d = pqx*pqx + pqy*pqy;
+	t = pqx*dx + pqy*dy;
+	if (d > 0) t /= d;
+	if (t < 0) t = 0;
+	else if (t > 1) t = 1;
+	dx = px + t*pqx - x;
+	dy = py + t*pqy - y;
+	return dx*dx + dy*dy;
+}
+
+static void nvg__appendCommands(NVGcontext* ctx, float* vals, int nvals)
+{
+	NVGstate* state = nvg__getState(ctx);
+	int i;
+
+	if (ctx->ncommands+nvals > ctx->ccommands) {
+		float* commands;
+		int ccommands = ctx->ncommands+nvals + ctx->ccommands/2;
+		commands = (float*)realloc(ctx->commands, sizeof(float)*ccommands);
+		if (commands == NULL) return;
+		ctx->commands = commands;
+		ctx->ccommands = ccommands;
+	}
+
+	if ((int)vals[0] != NVG_CLOSE && (int)vals[0] != NVG_WINDING) {
+		ctx->commandx = vals[nvals-2];
+		ctx->commandy = vals[nvals-1];
+	}
+
+	// transform commands
+	i = 0;
+	while (i < nvals) {
+		int cmd = (int)vals[i];
+		switch (cmd) {
+		case NVG_MOVETO:
+			nvgTransformPoint(&vals[i+1],&vals[i+2], state->xform, vals[i+1],vals[i+2]);
+			i += 3;
+			break;
+		case NVG_LINETO:
+			nvgTransformPoint(&vals[i+1],&vals[i+2], state->xform, vals[i+1],vals[i+2]);
+			i += 3;
+			break;
+		case NVG_BEZIERTO:
+			nvgTransformPoint(&vals[i+1],&vals[i+2], state->xform, vals[i+1],vals[i+2]);
+			nvgTransformPoint(&vals[i+3],&vals[i+4], state->xform, vals[i+3],vals[i+4]);
+			nvgTransformPoint(&vals[i+5],&vals[i+6], state->xform, vals[i+5],vals[i+6]);
+			i += 7;
+			break;
+		case NVG_CLOSE:
+			i++;
+			break;
+		case NVG_WINDING:
+			i += 2;
+			break;
+		default:
+			i++;
+		}
+	}
+
+	memcpy(&ctx->commands[ctx->ncommands], vals, nvals*sizeof(float));
+
+	ctx->ncommands += nvals;
+}
+
+
+static void nvg__clearPathCache(NVGcontext* ctx)
+{
+	ctx->cache->npoints = 0;
+	ctx->cache->npaths = 0;
+}
+
+static NVGpath* nvg__lastPath(NVGcontext* ctx)
+{
+	if (ctx->cache->npaths > 0)
+		return &ctx->cache->paths[ctx->cache->npaths-1];
+	return NULL;
+}
+
+static void nvg__addPath(NVGcontext* ctx)
+{
+	NVGpath* path;
+	if (ctx->cache->npaths+1 > ctx->cache->cpaths) {
+		NVGpath* paths;
+		int cpaths = ctx->cache->npaths+1 + ctx->cache->cpaths/2;
+		paths = (NVGpath*)realloc(ctx->cache->paths, sizeof(NVGpath)*cpaths);
+		if (paths == NULL) return;
+		ctx->cache->paths = paths;
+		ctx->cache->cpaths = cpaths;
+	}
+	path = &ctx->cache->paths[ctx->cache->npaths];
+	memset(path, 0, sizeof(*path));
+	path->first = ctx->cache->npoints;
+	path->winding = NVG_CCW;
+
+	ctx->cache->npaths++;
+}
+
+static NVGpoint* nvg__lastPoint(NVGcontext* ctx)
+{
+	if (ctx->cache->npoints > 0)
+		return &ctx->cache->points[ctx->cache->npoints-1];
+	return NULL;
+}
+
+static void nvg__addPoint(NVGcontext* ctx, float x, float y, int flags)
+{
+	NVGpath* path = nvg__lastPath(ctx);
+	NVGpoint* pt;
+	if (path == NULL) return;
+
+	if (path->count > 0 && ctx->cache->npoints > 0) {
+		pt = nvg__lastPoint(ctx);
+		if (nvg__ptEquals(pt->x,pt->y, x,y, ctx->distTol)) {
+			pt->flags |= flags;
+			return;
+		}
+	}
+
+	if (ctx->cache->npoints+1 > ctx->cache->cpoints) {
+		NVGpoint* points;
+		int cpoints = ctx->cache->npoints+1 + ctx->cache->cpoints/2;
+		points = (NVGpoint*)realloc(ctx->cache->points, sizeof(NVGpoint)*cpoints);
+		if (points == NULL) return;
+		ctx->cache->points = points;
+		ctx->cache->cpoints = cpoints;
+	}
+
+	pt = &ctx->cache->points[ctx->cache->npoints];
+	memset(pt, 0, sizeof(*pt));
+	pt->x = x;
+	pt->y = y;
+	pt->flags = (unsigned char)flags;
+
+	ctx->cache->npoints++;
+	path->count++;
+}
+
+static void nvg__closePath(NVGcontext* ctx)
+{
+	NVGpath* path = nvg__lastPath(ctx);
+	if (path == NULL) return;
+	path->closed = 1;
+}
+
+static void nvg__pathWinding(NVGcontext* ctx, int winding)
+{
+	NVGpath* path = nvg__lastPath(ctx);
+	if (path == NULL) return;
+	path->winding = winding;
+}
+
+static float nvg__getAverageScale(float *t)
+{
+	float sx = sqrtf(t[0]*t[0] + t[2]*t[2]);
+	float sy = sqrtf(t[1]*t[1] + t[3]*t[3]);
+	return (sx + sy) * 0.5f;
+}
+
+static NVGvertex* nvg__allocTempVerts(NVGcontext* ctx, int nverts)
+{
+	if (nverts > ctx->cache->cverts) {
+		NVGvertex* verts;
+		int cverts = (nverts + 0xff) & ~0xff; // Round up to prevent allocations when things change just slightly.
+		verts = (NVGvertex*)realloc(ctx->cache->verts, sizeof(NVGvertex)*cverts);
+		if (verts == NULL) return NULL;
+		ctx->cache->verts = verts;
+		ctx->cache->cverts = cverts;
+	}
+
+	return ctx->cache->verts;
+}
+
+static float nvg__triarea2(float ax, float ay, float bx, float by, float cx, float cy)
+{
+	float abx = bx - ax;
+	float aby = by - ay;
+	float acx = cx - ax;
+	float acy = cy - ay;
+	return acx*aby - abx*acy;
+}
+
+static float nvg__polyArea(NVGpoint* pts, int npts)
+{
+	int i;
+	float area = 0;
+	for (i = 2; i < npts; i++) {
+		NVGpoint* a = &pts[0];
+		NVGpoint* b = &pts[i-1];
+		NVGpoint* c = &pts[i];
+		area += nvg__triarea2(a->x,a->y, b->x,b->y, c->x,c->y);
+	}
+	return area * 0.5f;
+}
+
+static void nvg__polyReverse(NVGpoint* pts, int npts)
+{
+	NVGpoint tmp;
+	int i = 0, j = npts-1;
+	while (i < j) {
+		tmp = pts[i];
+		pts[i] = pts[j];
+		pts[j] = tmp;
+		i++;
+		j--;
+	}
+}
+
+
+static void nvg__vset(NVGvertex* vtx, float x, float y, float u, float v)
+{
+	vtx->x = x;
+	vtx->y = y;
+	vtx->u = u;
+	vtx->v = v;
+}
+
+static void nvg__tesselateBezier(NVGcontext* ctx,
+								 float x1, float y1, float x2, float y2,
+								 float x3, float y3, float x4, float y4,
+								 int level, int type)
+{
+	float x12,y12,x23,y23,x34,y34,x123,y123,x234,y234,x1234,y1234;
+	float dx,dy,d2,d3;
+	
+	if (level > 10) return;
+
+	x12 = (x1+x2)*0.5f;
+	y12 = (y1+y2)*0.5f;
+	x23 = (x2+x3)*0.5f;
+	y23 = (y2+y3)*0.5f;
+	x34 = (x3+x4)*0.5f;
+	y34 = (y3+y4)*0.5f;
+	x123 = (x12+x23)*0.5f;
+	y123 = (y12+y23)*0.5f;
+
+	dx = x4 - x1;
+	dy = y4 - y1;
+	d2 = nvg__absf(((x2 - x4) * dy - (y2 - y4) * dx));
+	d3 = nvg__absf(((x3 - x4) * dy - (y3 - y4) * dx));
+
+	if ((d2 + d3)*(d2 + d3) < ctx->tessTol * (dx*dx + dy*dy)) {
+		nvg__addPoint(ctx, x4, y4, type);
+		return;
+	}
+
+/*	if (nvg__absf(x1+x3-x2-x2) + nvg__absf(y1+y3-y2-y2) + nvg__absf(x2+x4-x3-x3) + nvg__absf(y2+y4-y3-y3) < ctx->tessTol) {
+		nvg__addPoint(ctx, x4, y4, type);
+		return;
+	}*/
+
+	x234 = (x23+x34)*0.5f;
+	y234 = (y23+y34)*0.5f;
+	x1234 = (x123+x234)*0.5f;
+	y1234 = (y123+y234)*0.5f;
+
+	nvg__tesselateBezier(ctx, x1,y1, x12,y12, x123,y123, x1234,y1234, level+1, 0); 
+	nvg__tesselateBezier(ctx, x1234,y1234, x234,y234, x34,y34, x4,y4, level+1, type); 
+}
+
+static void nvg__flattenPaths(NVGcontext* ctx)
+{
+	NVGpathCache* cache = ctx->cache;
+//	NVGstate* state = nvg__getState(ctx);
+	NVGpoint* last;
+	NVGpoint* p0;
+	NVGpoint* p1;
+	NVGpoint* pts;
+	NVGpath* path;
+	int i, j;
+	float* cp1;
+	float* cp2;
+	float* p;
+	float area;
+
+	if (cache->npaths > 0)
+		return;
+
+	// Flatten
+	i = 0;
+	while (i < ctx->ncommands) {
+		int cmd = (int)ctx->commands[i];
+		switch (cmd) {
+		case NVG_MOVETO:
+			nvg__addPath(ctx);
+			p = &ctx->commands[i+1];
+			nvg__addPoint(ctx, p[0], p[1], NVG_PT_CORNER);
+			i += 3;
+			break;
+		case NVG_LINETO:
+			p = &ctx->commands[i+1];
+			nvg__addPoint(ctx, p[0], p[1], NVG_PT_CORNER);
+			i += 3;
+			break;
+		case NVG_BEZIERTO:
+			last = nvg__lastPoint(ctx);
+			if (last != NULL) {
+				cp1 = &ctx->commands[i+1];
+				cp2 = &ctx->commands[i+3];
+				p = &ctx->commands[i+5];
+				nvg__tesselateBezier(ctx, last->x,last->y, cp1[0],cp1[1], cp2[0],cp2[1], p[0],p[1], 0, NVG_PT_CORNER);
+			}
+			i += 7;
+			break;
+		case NVG_CLOSE:
+			nvg__closePath(ctx);
+			i++;
+			break;
+		case NVG_WINDING:
+			nvg__pathWinding(ctx, (int)ctx->commands[i+1]);
+			i += 2;
+			break;
+		default:
+			i++;
+		}
+	}
+
+	cache->bounds[0] = cache->bounds[1] = 1e6f;
+	cache->bounds[2] = cache->bounds[3] = -1e6f;
+
+	// Calculate the direction and length of line segments.
+	for (j = 0; j < cache->npaths; j++) {
+		path = &cache->paths[j];
+		pts = &cache->points[path->first];
+
+		// If the first and last points are the same, remove the last, mark as closed path.
+		p0 = &pts[path->count-1];
+		p1 = &pts[0];
+		if (nvg__ptEquals(p0->x,p0->y, p1->x,p1->y, ctx->distTol)) {
+			path->count--;
+			p0 = &pts[path->count-1];
+			path->closed = 1;
+		}
+
+		// Enforce winding.
+		if (path->count > 2) {
+			area = nvg__polyArea(pts, path->count);
+			if (path->winding == NVG_CCW && area < 0.0f)
+				nvg__polyReverse(pts, path->count);
+			if (path->winding == NVG_CW && area > 0.0f)
+				nvg__polyReverse(pts, path->count);
+		}
+
+		for(i = 0; i < path->count; i++) {
+			// Calculate segment direction and length
+			p0->dx = p1->x - p0->x;
+			p0->dy = p1->y - p0->y;
+			p0->len = nvg__normalize(&p0->dx, &p0->dy);
+			// Update bounds
+			cache->bounds[0] = nvg__minf(cache->bounds[0], p0->x);
+			cache->bounds[1] = nvg__minf(cache->bounds[1], p0->y);
+			cache->bounds[2] = nvg__maxf(cache->bounds[2], p0->x);
+			cache->bounds[3] = nvg__maxf(cache->bounds[3], p0->y);
+			// Advance
+			p0 = p1++;
+		}
+	}
+}
+
+static int nvg__curveDivs(float r, float arc, float tol)
+{
+	float da = acosf(r / (r + tol)) * 2.0f;
+	return nvg__maxi(2, (int)ceilf(arc / da));
+}
+
+static void nvg__chooseBevel(int bevel, NVGpoint* p0, NVGpoint* p1, float w,
+							float* x0, float* y0, float* x1, float* y1)
+{
+	if (bevel) {
+		*x0 = p1->x + p0->dy * w;
+		*y0 = p1->y - p0->dx * w;
+		*x1 = p1->x + p1->dy * w;
+		*y1 = p1->y - p1->dx * w;
+	} else {
+		*x0 = p1->x + p1->dmx * w;
+		*y0 = p1->y + p1->dmy * w;
+		*x1 = p1->x + p1->dmx * w;
+		*y1 = p1->y + p1->dmy * w;
+	}
+}
+
+static NVGvertex* nvg__roundJoin(NVGvertex* dst, NVGpoint* p0, NVGpoint* p1,
+										float lw, float rw, float lu, float ru, int ncap, float fringe)
+{
+	int i, n;
+	float dlx0 = p0->dy;
+	float dly0 = -p0->dx;
+	float dlx1 = p1->dy;
+	float dly1 = -p1->dx;
+	NVG_NOTUSED(fringe);
+
+	if (p1->flags & NVG_PT_LEFT) {
+		float lx0,ly0,lx1,ly1,a0,a1;
+		nvg__chooseBevel(p1->flags & NVG_PR_INNERBEVEL, p0, p1, lw, &lx0,&ly0, &lx1,&ly1);
+		a0 = atan2f(-dly0, -dlx0);
+		a1 = atan2f(-dly1, -dlx1);
+		if (a1 > a0) a1 -= NVG_PI*2;
+
+		nvg__vset(dst, lx0, ly0, lu,1); dst++;
+		nvg__vset(dst, p1->x - dlx0*rw, p1->y - dly0*rw, ru,1); dst++;
+
+		n = nvg__clampi((int)ceilf(((a0 - a1) / NVG_PI) * ncap), 2, ncap);
+		for (i = 0; i < n; i++) {
+			float u = i/(float)(n-1);
+			float a = a0 + u*(a1-a0);
+			float rx = p1->x + cosf(a) * rw;
+			float ry = p1->y + sinf(a) * rw;
+			nvg__vset(dst, p1->x, p1->y, 0.5f,1); dst++;
+			nvg__vset(dst, rx, ry, ru,1); dst++;
+		}
+
+		nvg__vset(dst, lx1, ly1, lu,1); dst++;
+		nvg__vset(dst, p1->x - dlx1*rw, p1->y - dly1*rw, ru,1); dst++;
+
+	} else {
+		float rx0,ry0,rx1,ry1,a0,a1;
+		nvg__chooseBevel(p1->flags & NVG_PR_INNERBEVEL, p0, p1, -rw, &rx0,&ry0, &rx1,&ry1);
+		a0 = atan2f(dly0, dlx0);
+		a1 = atan2f(dly1, dlx1);
+		if (a1 < a0) a1 += NVG_PI*2;
+
+		nvg__vset(dst, p1->x + dlx0*rw, p1->y + dly0*rw, lu,1); dst++;
+		nvg__vset(dst, rx0, ry0, ru,1); dst++;
+
+		n = nvg__clampi((int)ceilf(((a1 - a0) / NVG_PI) * ncap), 2, ncap);
+		for (i = 0; i < n; i++) {
+			float u = i/(float)(n-1);
+			float a = a0 + u*(a1-a0);
+			float lx = p1->x + cosf(a) * lw;
+			float ly = p1->y + sinf(a) * lw;
+			nvg__vset(dst, lx, ly, lu,1); dst++;
+			nvg__vset(dst, p1->x, p1->y, 0.5f,1); dst++;
+		}
+
+		nvg__vset(dst, p1->x + dlx1*rw, p1->y + dly1*rw, lu,1); dst++;
+		nvg__vset(dst, rx1, ry1, ru,1); dst++;
+
+	}
+	return dst;
+}
+
+static NVGvertex* nvg__bevelJoin(NVGvertex* dst, NVGpoint* p0, NVGpoint* p1,
+										float lw, float rw, float lu, float ru, float fringe)
+{
+	float rx0,ry0,rx1,ry1;
+	float lx0,ly0,lx1,ly1;
+	float dlx0 = p0->dy;
+	float dly0 = -p0->dx;
+	float dlx1 = p1->dy;
+	float dly1 = -p1->dx;
+	NVG_NOTUSED(fringe);
+
+	if (p1->flags & NVG_PT_LEFT) {
+		nvg__chooseBevel(p1->flags & NVG_PR_INNERBEVEL, p0, p1, lw, &lx0,&ly0, &lx1,&ly1);
+
+		nvg__vset(dst, lx0, ly0, lu,1); dst++;
+		nvg__vset(dst, p1->x - dlx0*rw, p1->y - dly0*rw, ru,1); dst++;
+
+		if (p1->flags & NVG_PT_BEVEL) {
+			nvg__vset(dst, lx0, ly0, lu,1); dst++;
+			nvg__vset(dst, p1->x - dlx0*rw, p1->y - dly0*rw, ru,1); dst++;
+
+			nvg__vset(dst, lx1, ly1, lu,1); dst++;
+			nvg__vset(dst, p1->x - dlx1*rw, p1->y - dly1*rw, ru,1); dst++;
+		} else {
+			rx0 = p1->x - p1->dmx * rw;
+			ry0 = p1->y - p1->dmy * rw;
+
+			nvg__vset(dst, p1->x, p1->y, 0.5f,1); dst++;
+			nvg__vset(dst, p1->x - dlx0*rw, p1->y - dly0*rw, ru,1); dst++;
+
+			nvg__vset(dst, rx0, ry0, ru,1); dst++;
+			nvg__vset(dst, rx0, ry0, ru,1); dst++;
+
+			nvg__vset(dst, p1->x, p1->y, 0.5f,1); dst++;
+			nvg__vset(dst, p1->x - dlx1*rw, p1->y - dly1*rw, ru,1); dst++;
+		}
+
+		nvg__vset(dst, lx1, ly1, lu,1); dst++;
+		nvg__vset(dst, p1->x - dlx1*rw, p1->y - dly1*rw, ru,1); dst++;
+
+	} else {
+		nvg__chooseBevel(p1->flags & NVG_PR_INNERBEVEL, p0, p1, -rw, &rx0,&ry0, &rx1,&ry1);
+
+		nvg__vset(dst, p1->x + dlx0*lw, p1->y + dly0*lw, lu,1); dst++;
+		nvg__vset(dst, rx0, ry0, ru,1); dst++;
+
+		if (p1->flags & NVG_PT_BEVEL) {
+			nvg__vset(dst, p1->x + dlx0*lw, p1->y + dly0*lw, lu,1); dst++;
+			nvg__vset(dst, rx0, ry0, ru,1); dst++;
+
+			nvg__vset(dst, p1->x + dlx1*lw, p1->y + dly1*lw, lu,1); dst++;
+			nvg__vset(dst, rx1, ry1, ru,1); dst++;
+		} else {
+			lx0 = p1->x + p1->dmx * lw;
+			ly0 = p1->y + p1->dmy * lw;
+
+			nvg__vset(dst, p1->x + dlx0*lw, p1->y + dly0*lw, lu,1); dst++;
+			nvg__vset(dst, p1->x, p1->y, 0.5f,1); dst++;
+
+			nvg__vset(dst, lx0, ly0, lu,1); dst++;
+			nvg__vset(dst, lx0, ly0, lu,1); dst++;
+
+			nvg__vset(dst, p1->x + dlx1*lw, p1->y + dly1*lw, lu,1); dst++;
+			nvg__vset(dst, p1->x, p1->y, 0.5f,1); dst++;
+		}
+
+		nvg__vset(dst, p1->x + dlx1*lw, p1->y + dly1*lw, lu,1); dst++;
+		nvg__vset(dst, rx1, ry1, ru,1); dst++;
+	}
+
+	return dst;
+}
+
+static NVGvertex* nvg__buttCapStart(NVGvertex* dst, NVGpoint* p,
+										   float dx, float dy, float w, float d, float aa)
+{
+	float px = p->x - dx*d;
+	float py = p->y - dy*d;
+	float dlx = dy;
+	float dly = -dx;
+	nvg__vset(dst, px + dlx*w - dx*aa, py + dly*w - dy*aa, 0,0); dst++;
+	nvg__vset(dst, px - dlx*w - dx*aa, py - dly*w - dy*aa, 1,0); dst++;
+	nvg__vset(dst, px + dlx*w, py + dly*w, 0,1); dst++;
+	nvg__vset(dst, px - dlx*w, py - dly*w, 1,1); dst++;
+	return dst;
+}
+
+static NVGvertex* nvg__buttCapEnd(NVGvertex* dst, NVGpoint* p,
+										   float dx, float dy, float w, float d, float aa)
+{
+	float px = p->x + dx*d;
+	float py = p->y + dy*d;
+	float dlx = dy;
+	float dly = -dx;
+	nvg__vset(dst, px + dlx*w, py + dly*w, 0,1); dst++;
+	nvg__vset(dst, px - dlx*w, py - dly*w, 1,1); dst++;
+	nvg__vset(dst, px + dlx*w + dx*aa, py + dly*w + dy*aa, 0,0); dst++;
+	nvg__vset(dst, px - dlx*w + dx*aa, py - dly*w + dy*aa, 1,0); dst++;
+	return dst;
+}
+
+
+static NVGvertex* nvg__roundCapStart(NVGvertex* dst, NVGpoint* p,
+											float dx, float dy, float w, int ncap, float aa)
+{
+	int i;
+	float px = p->x;
+	float py = p->y;
+	float dlx = dy;
+	float dly = -dx;
+	NVG_NOTUSED(aa);
+	for (i = 0; i < ncap; i++) {
+		float a = i/(float)(ncap-1)*NVG_PI;
+		float ax = cosf(a) * w, ay = sinf(a) * w;
+		nvg__vset(dst, px - dlx*ax - dx*ay, py - dly*ax - dy*ay, 0,1); dst++;
+		nvg__vset(dst, px, py, 0.5f,1); dst++;
+	}
+	nvg__vset(dst, px + dlx*w, py + dly*w, 0,1); dst++;
+	nvg__vset(dst, px - dlx*w, py - dly*w, 1,1); dst++;
+	return dst;
+}
+
+static NVGvertex* nvg__roundCapEnd(NVGvertex* dst, NVGpoint* p,
+										  float dx, float dy, float w, int ncap, float aa)
+{
+	int i;
+	float px = p->x;
+	float py = p->y;
+	float dlx = dy;
+	float dly = -dx;
+	NVG_NOTUSED(aa);
+	nvg__vset(dst, px + dlx*w, py + dly*w, 0,1); dst++;
+	nvg__vset(dst, px - dlx*w, py - dly*w, 1,1); dst++;
+	for (i = 0; i < ncap; i++) {
+		float a = i/(float)(ncap-1)*NVG_PI;
+		float ax = cosf(a) * w, ay = sinf(a) * w;
+		nvg__vset(dst, px, py, 0.5f,1); dst++;
+		nvg__vset(dst, px - dlx*ax + dx*ay, py - dly*ax + dy*ay, 0,1); dst++;
+	}
+	return dst;
+}
+
+
+static void nvg__calculateJoins(NVGcontext* ctx, float w, int lineJoin, float miterLimit)
+{
+	NVGpathCache* cache = ctx->cache;
+	int i, j;
+	float iw = 0.0f;
+
+	if (w > 0.0f) iw = 1.0f / w;
+
+	// Calculate which joins needs extra vertices to append, and gather vertex count.
+	for (i = 0; i < cache->npaths; i++) {
+		NVGpath* path = &cache->paths[i];
+		NVGpoint* pts = &cache->points[path->first];
+		NVGpoint* p0 = &pts[path->count-1];
+		NVGpoint* p1 = &pts[0];
+		int nleft = 0;
+
+		path->nbevel = 0;
+
+		for (j = 0; j < path->count; j++) {
+			float dlx0, dly0, dlx1, dly1, dmr2, cross, limit;
+			dlx0 = p0->dy;
+			dly0 = -p0->dx;
+			dlx1 = p1->dy;
+			dly1 = -p1->dx;
+			// Calculate extrusions
+			p1->dmx = (dlx0 + dlx1) * 0.5f;
+			p1->dmy = (dly0 + dly1) * 0.5f;
+			dmr2 = p1->dmx*p1->dmx + p1->dmy*p1->dmy;
+			if (dmr2 > 0.000001f) {
+				float scale = 1.0f / dmr2;
+				if (scale > 600.0f) {
+					scale = 600.0f;
+				}
+				p1->dmx *= scale;
+				p1->dmy *= scale;
+			}
+
+			// Clear flags, but keep the corner.
+			p1->flags = (p1->flags & NVG_PT_CORNER) ? NVG_PT_CORNER : 0;
+
+			// Keep track of left turns.
+			cross = p1->dx * p0->dy - p0->dx * p1->dy;
+			if (cross > 0.0f) {
+				nleft++;
+				p1->flags |= NVG_PT_LEFT;
+			}
+
+			// Calculate if we should use bevel or miter for inner join.
+			limit = nvg__maxf(1.01f, nvg__minf(p0->len, p1->len) * iw);
+			if ((dmr2 * limit*limit) < 1.0f)
+				p1->flags |= NVG_PR_INNERBEVEL;
+
+			// Check to see if the corner needs to be beveled.
+			if (p1->flags & NVG_PT_CORNER) {
+				if ((dmr2 * miterLimit*miterLimit) < 1.0f || lineJoin == NVG_BEVEL || lineJoin == NVG_ROUND) {
+					p1->flags |= NVG_PT_BEVEL;
+				}
+			}
+
+			if ((p1->flags & (NVG_PT_BEVEL | NVG_PR_INNERBEVEL)) != 0)
+				path->nbevel++;
+
+			p0 = p1++;
+		}
+
+		path->convex = (nleft == path->count) ? 1 : 0;
+	}
+}
+
+
+static int nvg__expandStroke(NVGcontext* ctx, float w, int lineCap, int lineJoin, float miterLimit)
+{	
+	NVGpathCache* cache = ctx->cache;
+	NVGvertex* verts;
+	NVGvertex* dst;
+	int cverts, i, j;
+	float aa = ctx->fringeWidth;
+	int ncap = nvg__curveDivs(w, NVG_PI, ctx->tessTol);	// Calculate divisions per half circle.
+
+	nvg__calculateJoins(ctx, w, lineJoin, miterLimit);
+
+	// Calculate max vertex usage.
+	cverts = 0;
+	for (i = 0; i < cache->npaths; i++) {
+		NVGpath* path = &cache->paths[i];
+		int loop = (path->closed == 0) ? 0 : 1;
+		if (lineJoin == NVG_ROUND)
+			cverts += (path->count + path->nbevel*(ncap+2) + 1) * 2; // plus one for loop
+		else
+			cverts += (path->count + path->nbevel*5 + 1) * 2; // plus one for loop
+		if (loop == 0) {
+			// space for caps
+			if (lineCap == NVG_ROUND) {
+				cverts += (ncap*2 + 2)*2;
+			} else {
+				cverts += (3+3)*2;
+			}
+		}
+	}
+
+	verts = nvg__allocTempVerts(ctx, cverts);
+	if (verts == NULL) return 0;
+
+	for (i = 0; i < cache->npaths; i++) {
+		NVGpath* path = &cache->paths[i];
+		NVGpoint* pts = &cache->points[path->first];
+		NVGpoint* p0;
+		NVGpoint* p1;
+		int s, e, loop;
+		float dx, dy;
+
+		path->fill = 0;
+		path->nfill = 0;
+
+		// Calculate fringe or stroke
+		loop = (path->closed == 0) ? 0 : 1;
+		dst = verts;
+		path->stroke = dst;
+
+		if (loop) {
+			// Looping
+			p0 = &pts[path->count-1];
+			p1 = &pts[0];
+			s = 0;
+			e = path->count;
+		} else {
+			// Add cap
+			p0 = &pts[0];
+			p1 = &pts[1];
+			s = 1;
+			e = path->count-1;
+		}
+
+		if (loop == 0) {
+			// Add cap
+			dx = p1->x - p0->x;
+			dy = p1->y - p0->y;
+			nvg__normalize(&dx, &dy);
+			if (lineCap == NVG_BUTT)
+				dst = nvg__buttCapStart(dst, p0, dx, dy, w, -aa*0.5f, aa);
+			else if (lineCap == NVG_BUTT || lineCap == NVG_SQUARE)
+				dst = nvg__buttCapStart(dst, p0, dx, dy, w, w-aa, aa);
+			else if (lineCap == NVG_ROUND)
+				dst = nvg__roundCapStart(dst, p0, dx, dy, w, ncap, aa);
+		}
+
+		for (j = s; j < e; ++j) {
+			if ((p1->flags & (NVG_PT_BEVEL | NVG_PR_INNERBEVEL)) != 0) {
+				if (lineJoin == NVG_ROUND) {
+					dst = nvg__roundJoin(dst, p0, p1, w, w, 0, 1, ncap, aa);
+				} else {
+					dst = nvg__bevelJoin(dst, p0, p1, w, w, 0, 1, aa);
+				}
+			} else {
+				nvg__vset(dst, p1->x + (p1->dmx * w), p1->y + (p1->dmy * w), 0,1); dst++;
+				nvg__vset(dst, p1->x - (p1->dmx * w), p1->y - (p1->dmy * w), 1,1); dst++;
+			}
+			p0 = p1++;
+		}
+
+		if (loop) {
+			// Loop it
+			nvg__vset(dst, verts[0].x, verts[0].y, 0,1); dst++;
+			nvg__vset(dst, verts[1].x, verts[1].y, 1,1); dst++;
+		} else {
+			// Add cap
+			dx = p1->x - p0->x;
+			dy = p1->y - p0->y;
+			nvg__normalize(&dx, &dy);
+			if (lineCap == NVG_BUTT)
+				dst = nvg__buttCapEnd(dst, p1, dx, dy, w, -aa*0.5f, aa);
+			else if (lineCap == NVG_BUTT || lineCap == NVG_SQUARE)
+				dst = nvg__buttCapEnd(dst, p1, dx, dy, w, w-aa, aa);
+			else if (lineCap == NVG_ROUND)
+				dst = nvg__roundCapEnd(dst, p1, dx, dy, w, ncap, aa);
+		}
+
+		path->nstroke = (int)(dst - verts);
+
+		verts = dst;
+	}
+
+	return 1;
+}
+
+static int nvg__expandFill(NVGcontext* ctx, float w, int lineJoin, float miterLimit)
+{
+	NVGpathCache* cache = ctx->cache;
+	NVGvertex* verts;
+	NVGvertex* dst;
+	int cverts, convex, i, j;
+	float aa = ctx->fringeWidth;
+	int fringe = w > 0.0f;
+
+	nvg__calculateJoins(ctx, w, lineJoin, miterLimit);
+
+	// Calculate max vertex usage.
+	cverts = 0;
+	for (i = 0; i < cache->npaths; i++) {
+		NVGpath* path = &cache->paths[i];
+		cverts += path->count + path->nbevel + 1;
+		if (fringe)
+			cverts += (path->count + path->nbevel*5 + 1) * 2; // plus one for loop
+	}
+
+	verts = nvg__allocTempVerts(ctx, cverts);
+	if (verts == NULL) return 0;
+
+	convex = cache->npaths == 1 && cache->paths[0].convex;
+
+	for (i = 0; i < cache->npaths; i++) {
+		NVGpath* path = &cache->paths[i];
+		NVGpoint* pts = &cache->points[path->first];
+		NVGpoint* p0;
+		NVGpoint* p1;
+		float rw, lw, woff;
+		float ru, lu;
+
+		// Calculate shape vertices.
+		woff = 0.5f*aa;
+		dst = verts;
+		path->fill = dst;
+
+		if (fringe) {
+			// Looping
+			p0 = &pts[path->count-1];
+			p1 = &pts[0];
+			for (j = 0; j < path->count; ++j) {
+				if (p1->flags & NVG_PT_BEVEL) {
+					float dlx0 = p0->dy;
+					float dly0 = -p0->dx;
+					float dlx1 = p1->dy;
+					float dly1 = -p1->dx;
+					if (p1->flags & NVG_PT_LEFT) {
+						float lx = p1->x + p1->dmx * woff;
+						float ly = p1->y + p1->dmy * woff;
+						nvg__vset(dst, lx, ly, 0.5f,1); dst++;
+					} else {
+						float lx0 = p1->x + dlx0 * woff;
+						float ly0 = p1->y + dly0 * woff;
+						float lx1 = p1->x + dlx1 * woff;
+						float ly1 = p1->y + dly1 * woff;
+						nvg__vset(dst, lx0, ly0, 0.5f,1); dst++;
+						nvg__vset(dst, lx1, ly1, 0.5f,1); dst++;
+					}
+				} else {
+					nvg__vset(dst, p1->x + (p1->dmx * woff), p1->y + (p1->dmy * woff), 0.5f,1); dst++;
+				}
+				p0 = p1++;
+			}
+		} else {
+			for (j = 0; j < path->count; ++j) {
+				nvg__vset(dst, pts[j].x, pts[j].y, 0.5f,1);
+				dst++;
+			}
+		}
+
+		path->nfill = (int)(dst - verts);
+		verts = dst;
+
+		// Calculate fringe
+		if (fringe) {
+			lw = w + woff;
+			rw = w - woff;
+			lu = 0;
+			ru = 1;
+			dst = verts;
+			path->stroke = dst;
+
+			// Create only half a fringe for convex shapes so that
+			// the shape can be rendered without stenciling.
+			if (convex) {
+				lw = woff;	// This should generate the same vertex as fill inset above.
+				lu = 0.5f;	// Set outline fade at middle.
+			}
+
+			// Looping
+			p0 = &pts[path->count-1];
+			p1 = &pts[0];
+
+			for (j = 0; j < path->count; ++j) {
+				if ((p1->flags & (NVG_PT_BEVEL | NVG_PR_INNERBEVEL)) != 0) {
+					dst = nvg__bevelJoin(dst, p0, p1, lw, rw, lu, ru, ctx->fringeWidth);
+				} else {
+					nvg__vset(dst, p1->x + (p1->dmx * lw), p1->y + (p1->dmy * lw), lu,1); dst++;
+					nvg__vset(dst, p1->x - (p1->dmx * rw), p1->y - (p1->dmy * rw), ru,1); dst++;
+				}
+				p0 = p1++;
+			}
+
+			// Loop it
+			nvg__vset(dst, verts[0].x, verts[0].y, lu,1); dst++;
+			nvg__vset(dst, verts[1].x, verts[1].y, ru,1); dst++;
+
+			path->nstroke = (int)(dst - verts);
+			verts = dst;
+		} else {
+			path->stroke = NULL;
+			path->nstroke = 0;
+		}
+	}
+
+	return 1;
+}
+
+
+// Draw
+void nvgBeginPath(NVGcontext* ctx)
+{
+	ctx->ncommands = 0;
+	nvg__clearPathCache(ctx);
+}
+
+void nvgMoveTo(NVGcontext* ctx, float x, float y)
+{
+	float vals[] = { NVG_MOVETO, x, y };
+	nvg__appendCommands(ctx, vals, NVG_COUNTOF(vals));
+}
+
+void nvgLineTo(NVGcontext* ctx, float x, float y)
+{
+	float vals[] = { NVG_LINETO, x, y };
+	nvg__appendCommands(ctx, vals, NVG_COUNTOF(vals));
+}
+
+void nvgBezierTo(NVGcontext* ctx, float c1x, float c1y, float c2x, float c2y, float x, float y)
+{
+	float vals[] = { NVG_BEZIERTO, c1x, c1y, c2x, c2y, x, y };
+	nvg__appendCommands(ctx, vals, NVG_COUNTOF(vals));
+}
+
+void nvgQuadTo(NVGcontext* ctx, float cx, float cy, float x, float y)
+{
+    float x0 = ctx->commandx;
+    float y0 = ctx->commandy;
+    float vals[] = { NVG_BEZIERTO, 
+        x0 + 2.0f/3.0f*(cx - x0), y0 + 2.0f/3.0f*(cy - y0),
+        x + 2.0f/3.0f*(cx - x), y + 2.0f/3.0f*(cy - y),
+        x, y };
+    nvg__appendCommands(ctx, vals, NVG_COUNTOF(vals));
+}
+
+void nvgArcTo(NVGcontext* ctx, float x1, float y1, float x2, float y2, float radius)
+{
+	float x0 = ctx->commandx;
+	float y0 = ctx->commandy;
+	float dx0,dy0, dx1,dy1, a, d, cx,cy, a0,a1;
+	int dir;
+
+	if (ctx->ncommands == 0) {
+		return;
+	}
+
+	// Handle degenerate cases.
+	if (nvg__ptEquals(x0,y0, x1,y1, ctx->distTol) ||
+		nvg__ptEquals(x1,y1, x2,y2, ctx->distTol) ||
+		nvg__distPtSeg(x1,y1, x0,y0, x2,y2) < ctx->distTol*ctx->distTol ||
+		radius < ctx->distTol) {
+		nvgLineTo(ctx, x1,y1);
+		return;
+	}
+
+	// Calculate tangential circle to lines (x0,y0)-(x1,y1) and (x1,y1)-(x2,y2).
+	dx0 = x0-x1;
+	dy0 = y0-y1;
+	dx1 = x2-x1;
+	dy1 = y2-y1;
+	nvg__normalize(&dx0,&dy0);
+	nvg__normalize(&dx1,&dy1);
+	a = nvg__acosf(dx0*dx1 + dy0*dy1);
+	d = radius / nvg__tanf(a/2.0f);
+
+//	printf("a=%f° d=%f\n", a/NVG_PI*180.0f, d);
+
+	if (d > 10000.0f) {
+		nvgLineTo(ctx, x1,y1);
+		return;
+	}
+
+	if (nvg__cross(dx0,dy0, dx1,dy1) > 0.0f) {
+		cx = x1 + dx0*d + dy0*radius;
+		cy = y1 + dy0*d + -dx0*radius;
+		a0 = nvg__atan2f(dx0, -dy0);
+		a1 = nvg__atan2f(-dx1, dy1);
+		dir = NVG_CW;
+//		printf("CW c=(%f, %f) a0=%f° a1=%f°\n", cx, cy, a0/NVG_PI*180.0f, a1/NVG_PI*180.0f);
+	} else {
+		cx = x1 + dx0*d + -dy0*radius;
+		cy = y1 + dy0*d + dx0*radius;
+		a0 = nvg__atan2f(-dx0, dy0);
+		a1 = nvg__atan2f(dx1, -dy1);
+		dir = NVG_CCW;
+//		printf("CCW c=(%f, %f) a0=%f° a1=%f°\n", cx, cy, a0/NVG_PI*180.0f, a1/NVG_PI*180.0f);
+	}
+
+	nvgArc(ctx, cx, cy, radius, a0, a1, dir);
+}
+
+void nvgClosePath(NVGcontext* ctx)
+{
+	float vals[] = { NVG_CLOSE };
+	nvg__appendCommands(ctx, vals, NVG_COUNTOF(vals));
+}
+
+void nvgPathWinding(NVGcontext* ctx, int dir)
+{
+	float vals[] = { NVG_WINDING, (float)dir };
+	nvg__appendCommands(ctx, vals, NVG_COUNTOF(vals));
+}
+
+void nvgArc(NVGcontext* ctx, float cx, float cy, float r, float a0, float a1, int dir)
+{
+	float a = 0, da = 0, hda = 0, kappa = 0;
+	float dx = 0, dy = 0, x = 0, y = 0, tanx = 0, tany = 0;
+	float px = 0, py = 0, ptanx = 0, ptany = 0;
+	float vals[3 + 5*7 + 100];
+	int i, ndivs, nvals;
+	int move = ctx->ncommands > 0 ? NVG_LINETO : NVG_MOVETO; 
+
+	// Clamp angles
+	da = a1 - a0;
+	if (dir == NVG_CW) {
+		if (nvg__absf(da) >= NVG_PI*2) {
+			da = NVG_PI*2;
+		} else {
+			while (da < 0.0f) da += NVG_PI*2;
+		}
+	} else {
+		if (nvg__absf(da) >= NVG_PI*2) {
+			da = -NVG_PI*2;
+		} else {
+			while (da > 0.0f) da -= NVG_PI*2;
+		}
+	}
+
+	// Split arc into max 90 degree segments.
+	ndivs = nvg__maxi(1, nvg__mini((int)(nvg__absf(da) / (NVG_PI*0.5f) + 0.5f), 5));
+	hda = (da / (float)ndivs) / 2.0f;
+	kappa = nvg__absf(4.0f / 3.0f * (1.0f - nvg__cosf(hda)) / nvg__sinf(hda));
+
+	if (dir == NVG_CCW)
+		kappa = -kappa;
+
+	nvals = 0;
+	for (i = 0; i <= ndivs; i++) {
+		a = a0 + da * (i/(float)ndivs);
+		dx = nvg__cosf(a);
+		dy = nvg__sinf(a);
+		x = cx + dx*r;
+		y = cy + dy*r;
+		tanx = -dy*r*kappa;
+		tany = dx*r*kappa;
+
+		if (i == 0) {
+			vals[nvals++] = (float)move;
+			vals[nvals++] = x;
+			vals[nvals++] = y;
+		} else {
+			vals[nvals++] = NVG_BEZIERTO;
+			vals[nvals++] = px+ptanx;
+			vals[nvals++] = py+ptany;
+			vals[nvals++] = x-tanx;
+			vals[nvals++] = y-tany;
+			vals[nvals++] = x;
+			vals[nvals++] = y;
+		}
+		px = x;
+		py = y;
+		ptanx = tanx;
+		ptany = tany;
+	}
+
+	nvg__appendCommands(ctx, vals, nvals);
+}
+
+void nvgRect(NVGcontext* ctx, float x, float y, float w, float h)
+{
+	float vals[] = {
+		NVG_MOVETO, x,y,
+		NVG_LINETO, x,y+h,
+		NVG_LINETO, x+w,y+h,
+		NVG_LINETO, x+w,y,
+		NVG_CLOSE
+	};
+	nvg__appendCommands(ctx, vals, NVG_COUNTOF(vals));
+}
+
+void nvgRoundedRect(NVGcontext* ctx, float x, float y, float w, float h, float r)
+{
+	if (r < 0.1f) {
+		nvgRect(ctx, x,y,w,h);
+		return;
+	}
+	else {
+		float rx = nvg__minf(r, nvg__absf(w)*0.5f) * nvg__signf(w), ry = nvg__minf(r, nvg__absf(h)*0.5f) * nvg__signf(h);
+		float vals[] = {
+			NVG_MOVETO, x, y+ry,
+			NVG_LINETO, x, y+h-ry,
+			NVG_BEZIERTO, x, y+h-ry*(1-NVG_KAPPA90), x+rx*(1-NVG_KAPPA90), y+h, x+rx, y+h,
+			NVG_LINETO, x+w-rx, y+h,
+			NVG_BEZIERTO, x+w-rx*(1-NVG_KAPPA90), y+h, x+w, y+h-ry*(1-NVG_KAPPA90), x+w, y+h-ry,
+			NVG_LINETO, x+w, y+ry,
+			NVG_BEZIERTO, x+w, y+ry*(1-NVG_KAPPA90), x+w-rx*(1-NVG_KAPPA90), y, x+w-rx, y,
+			NVG_LINETO, x+rx, y,
+			NVG_BEZIERTO, x+rx*(1-NVG_KAPPA90), y, x, y+ry*(1-NVG_KAPPA90), x, y+ry,
+			NVG_CLOSE
+		};
+		nvg__appendCommands(ctx, vals, NVG_COUNTOF(vals));
+	}
+}
+
+void nvgEllipse(NVGcontext* ctx, float cx, float cy, float rx, float ry)
+{
+	float vals[] = {
+		NVG_MOVETO, cx-rx, cy,
+		NVG_BEZIERTO, cx-rx, cy+ry*NVG_KAPPA90, cx-rx*NVG_KAPPA90, cy+ry, cx, cy+ry,
+		NVG_BEZIERTO, cx+rx*NVG_KAPPA90, cy+ry, cx+rx, cy+ry*NVG_KAPPA90, cx+rx, cy,
+		NVG_BEZIERTO, cx+rx, cy-ry*NVG_KAPPA90, cx+rx*NVG_KAPPA90, cy-ry, cx, cy-ry,
+		NVG_BEZIERTO, cx-rx*NVG_KAPPA90, cy-ry, cx-rx, cy-ry*NVG_KAPPA90, cx-rx, cy,
+		NVG_CLOSE
+	};
+	nvg__appendCommands(ctx, vals, NVG_COUNTOF(vals));
+}
+
+void nvgCircle(NVGcontext* ctx, float cx, float cy, float r)
+{
+	nvgEllipse(ctx, cx,cy, r,r);
+}
+
+void nvgDebugDumpPathCache(NVGcontext* ctx)
+{
+	const NVGpath* path;
+	int i, j;
+
+	printf("Dumping %d cached paths\n", ctx->cache->npaths);
+	for (i = 0; i < ctx->cache->npaths; i++) {
+		path = &ctx->cache->paths[i];
+		printf(" - Path %d\n", i);
+		if (path->nfill) {
+			printf("   - fill: %d\n", path->nfill);
+			for (j = 0; j < path->nfill; j++)
+				printf("%f\t%f\n", path->fill[j].x, path->fill[j].y);
+		}
+		if (path->nstroke) {
+			printf("   - stroke: %d\n", path->nstroke);
+			for (j = 0; j < path->nstroke; j++)
+				printf("%f\t%f\n", path->stroke[j].x, path->stroke[j].y);
+		}
+	}
+}
+
+void nvgFill(NVGcontext* ctx)
+{
+	NVGstate* state = nvg__getState(ctx);
+	const NVGpath* path;
+	NVGpaint fillPaint = state->fill;
+	int i;
+
+	nvg__flattenPaths(ctx);
+	if (ctx->params.edgeAntiAlias)
+		nvg__expandFill(ctx, ctx->fringeWidth, NVG_MITER, 2.4f);
+	else
+		nvg__expandFill(ctx, 0.0f, NVG_MITER, 2.4f);
+
+	// Apply global alpha
+	fillPaint.innerColor.a *= state->alpha;
+	fillPaint.outerColor.a *= state->alpha;
+
+	ctx->params.renderFill(ctx->params.userPtr, &fillPaint, &state->scissor, ctx->fringeWidth,
+						   ctx->cache->bounds, ctx->cache->paths, ctx->cache->npaths);
+
+	// Count triangles
+	for (i = 0; i < ctx->cache->npaths; i++) {
+		path = &ctx->cache->paths[i];
+		ctx->fillTriCount += path->nfill-2;
+		ctx->fillTriCount += path->nstroke-2;
+		ctx->drawCallCount += 2;
+	}
+}
+
+void nvgStroke(NVGcontext* ctx)
+{
+	NVGstate* state = nvg__getState(ctx);
+	float scale = nvg__getAverageScale(state->xform);
+	float strokeWidth = nvg__clampf(state->strokeWidth * scale, 0.0f, 200.0f);
+	NVGpaint strokePaint = state->stroke;
+	const NVGpath* path;
+	int i;
+
+	if (strokeWidth < ctx->fringeWidth) {
+		// If the stroke width is less than pixel size, use alpha to emulate coverage.
+		// Since coverage is area, scale by alpha*alpha.
+		float alpha = nvg__clampf(strokeWidth / ctx->fringeWidth, 0.0f, 1.0f);
+		strokePaint.innerColor.a *= alpha*alpha;
+		strokePaint.outerColor.a *= alpha*alpha;
+		strokeWidth = ctx->fringeWidth;
+	}
+
+	// Apply global alpha
+	strokePaint.innerColor.a *= state->alpha;
+	strokePaint.outerColor.a *= state->alpha;
+
+	nvg__flattenPaths(ctx);
+
+	if (ctx->params.edgeAntiAlias)
+		nvg__expandStroke(ctx, strokeWidth*0.5f + ctx->fringeWidth*0.5f, state->lineCap, state->lineJoin, state->miterLimit);
+	else
+		nvg__expandStroke(ctx, strokeWidth*0.5f, state->lineCap, state->lineJoin, state->miterLimit);
+
+	ctx->params.renderStroke(ctx->params.userPtr, &strokePaint, &state->scissor, ctx->fringeWidth,
+							 strokeWidth, ctx->cache->paths, ctx->cache->npaths);
+
+	// Count triangles
+	for (i = 0; i < ctx->cache->npaths; i++) {
+		path = &ctx->cache->paths[i];
+		ctx->strokeTriCount += path->nstroke-2;
+		ctx->drawCallCount++;
+	}
+}
+
+// Add fonts
+int nvgCreateFont(NVGcontext* ctx, const char* name, const char* path)
+{
+	return fonsAddFont(ctx->fs, name, path);
+}
+
+int nvgCreateFontMem(NVGcontext* ctx, const char* name, unsigned char* data, int ndata, int freeData)
+{
+	return fonsAddFontMem(ctx->fs, name, data, ndata, freeData);
+}
+
+int nvgFindFont(NVGcontext* ctx, const char* name)
+{
+	if (name == NULL) return -1;
+	return fonsGetFontByName(ctx->fs, name);
+}
+
+// State setting
+void nvgFontSize(NVGcontext* ctx, float size)
+{
+	NVGstate* state = nvg__getState(ctx);
+	state->fontSize = size;
+}
+
+void nvgFontBlur(NVGcontext* ctx, float blur)
+{
+	NVGstate* state = nvg__getState(ctx);
+	state->fontBlur = blur;
+}
+
+void nvgTextLetterSpacing(NVGcontext* ctx, float spacing)
+{
+	NVGstate* state = nvg__getState(ctx);
+	state->letterSpacing = spacing;
+}
+
+void nvgTextLineHeight(NVGcontext* ctx, float lineHeight)
+{
+	NVGstate* state = nvg__getState(ctx);
+	state->lineHeight = lineHeight;
+}
+
+void nvgTextAlign(NVGcontext* ctx, int align)
+{
+	NVGstate* state = nvg__getState(ctx);
+	state->textAlign = align;
+}
+
+void nvgFontFaceId(NVGcontext* ctx, int font)
+{
+	NVGstate* state = nvg__getState(ctx);
+	state->fontId = font;
+}
+
+void nvgFontFace(NVGcontext* ctx, const char* font)
+{
+	NVGstate* state = nvg__getState(ctx);
+	state->fontId = fonsGetFontByName(ctx->fs, font);
+}
+
+static float nvg__quantize(float a, float d)
+{
+	return ((int)(a / d + 0.5f)) * d;
+}
+
+static float nvg__getFontScale(NVGstate* state)
+{
+	return nvg__minf(nvg__quantize(nvg__getAverageScale(state->xform), 0.01f), 4.0f);
+}
+
+static void nvg__flushTextTexture(NVGcontext* ctx)
+{
+	int dirty[4];
+
+	if (fonsValidateTexture(ctx->fs, dirty)) {
+		int fontImage = ctx->fontImages[ctx->fontImageIdx];
+		// Update texture
+		if (fontImage != 0) {
+			int iw, ih;
+			const unsigned char* data = fonsGetTextureData(ctx->fs, &iw, &ih);
+			int x = dirty[0];
+			int y = dirty[1];
+			int w = dirty[2] - dirty[0];
+			int h = dirty[3] - dirty[1];
+			ctx->params.renderUpdateTexture(ctx->params.userPtr, fontImage, x,y, w,h, data);
+		}
+	}
+}
+
+static int nvg__allocTextAtlas(NVGcontext* ctx)
+{
+	int iw, ih;
+	nvg__flushTextTexture(ctx);
+	if (ctx->fontImageIdx >= NVG_MAX_FONTIMAGES-1)
+		return 0;
+	// if next fontImage already have a texture
+	if (ctx->fontImages[ctx->fontImageIdx+1] != 0)
+		nvgImageSize(ctx, ctx->fontImages[ctx->fontImageIdx+1], &iw, &ih);
+	else { // calculate the new font image size and create it.
+		nvgImageSize(ctx, ctx->fontImages[ctx->fontImageIdx], &iw, &ih);
+		if (iw > ih)
+			ih *= 2;
+		else
+			iw *= 2;
+		if (iw > NVG_MAX_FONTIMAGE_SIZE || ih > NVG_MAX_FONTIMAGE_SIZE)
+			iw = ih = NVG_MAX_FONTIMAGE_SIZE;
+		ctx->fontImages[ctx->fontImageIdx+1] = ctx->params.renderCreateTexture(ctx->params.userPtr, NVG_TEXTURE_ALPHA, iw, ih, 0, NULL);
+	}
+	++ctx->fontImageIdx;
+	fonsResetAtlas(ctx->fs, iw, ih);
+	return 1;
+}
+
+static void nvg__renderText(NVGcontext* ctx, NVGvertex* verts, int nverts)
+{
+	NVGstate* state = nvg__getState(ctx);
+	NVGpaint paint = state->fill;
+
+	// Render triangles.
+	paint.image = ctx->fontImages[ctx->fontImageIdx];
+
+	// Apply global alpha
+	paint.innerColor.a *= state->alpha;
+	paint.outerColor.a *= state->alpha;
+
+	ctx->params.renderTriangles(ctx->params.userPtr, &paint, &state->scissor, verts, nverts);
+
+	ctx->drawCallCount++;
+	ctx->textTriCount += nverts/3;
+}
+
+float nvgText(NVGcontext* ctx, float x, float y, const char* string, const char* end)
+{
+	NVGstate* state = nvg__getState(ctx);
+	FONStextIter iter, prevIter;
+	FONSquad q;
+	NVGvertex* verts;
+	float scale = nvg__getFontScale(state) * ctx->devicePxRatio;
+	float invscale = 1.0f / scale;
+	int cverts = 0;
+	int nverts = 0;
+
+	if (end == NULL)
+		end = string + strlen(string);
+
+	if (state->fontId == FONS_INVALID) return x;
+
+	fonsSetSize(ctx->fs, state->fontSize*scale);
+	fonsSetSpacing(ctx->fs, state->letterSpacing*scale);
+	fonsSetBlur(ctx->fs, state->fontBlur*scale);
+	fonsSetAlign(ctx->fs, state->textAlign);
+	fonsSetFont(ctx->fs, state->fontId);
+
+	cverts = nvg__maxi(2, (int)(end - string)) * 6; // conservative estimate.
+	verts = nvg__allocTempVerts(ctx, cverts);
+	if (verts == NULL) return x;
+
+	fonsTextIterInit(ctx->fs, &iter, x*scale, y*scale, string, end);
+	prevIter = iter;
+	while (fonsTextIterNext(ctx->fs, &iter, &q)) {
+		float c[4*2];
+		if (iter.prevGlyphIndex == -1) { // can not retrieve glyph?
+			if (!nvg__allocTextAtlas(ctx))
+				break; // no memory :(
+			if (nverts != 0) {
+				nvg__renderText(ctx, verts, nverts);
+				nverts = 0;
+			}
+			iter = prevIter;
+			fonsTextIterNext(ctx->fs, &iter, &q); // try again
+			if (iter.prevGlyphIndex == -1) // still can not find glyph?
+				break;
+		}
+		prevIter = iter;
+		// Transform corners.
+		nvgTransformPoint(&c[0],&c[1], state->xform, q.x0*invscale, q.y0*invscale);
+		nvgTransformPoint(&c[2],&c[3], state->xform, q.x1*invscale, q.y0*invscale);
+		nvgTransformPoint(&c[4],&c[5], state->xform, q.x1*invscale, q.y1*invscale);
+		nvgTransformPoint(&c[6],&c[7], state->xform, q.x0*invscale, q.y1*invscale);
+		// Create triangles
+		if (nverts+6 <= cverts) {
+			nvg__vset(&verts[nverts], c[0], c[1], q.s0, q.t0); nverts++;
+			nvg__vset(&verts[nverts], c[4], c[5], q.s1, q.t1); nverts++;
+			nvg__vset(&verts[nverts], c[2], c[3], q.s1, q.t0); nverts++;
+			nvg__vset(&verts[nverts], c[0], c[1], q.s0, q.t0); nverts++;
+			nvg__vset(&verts[nverts], c[6], c[7], q.s0, q.t1); nverts++;
+			nvg__vset(&verts[nverts], c[4], c[5], q.s1, q.t1); nverts++;
+		}
+	}
+
+	// TODO: add back-end bit to do this just once per frame. 
+	nvg__flushTextTexture(ctx);
+
+	nvg__renderText(ctx, verts, nverts);
+
+	return iter.x;
+}
+
+void nvgTextBox(NVGcontext* ctx, float x, float y, float breakRowWidth, const char* string, const char* end)
+{
+	NVGstate* state = nvg__getState(ctx);
+	NVGtextRow rows[2];
+	int nrows = 0, i;
+	int oldAlign = state->textAlign;
+	int haling = state->textAlign & (NVG_ALIGN_LEFT | NVG_ALIGN_CENTER | NVG_ALIGN_RIGHT);
+	int valign = state->textAlign & (NVG_ALIGN_TOP | NVG_ALIGN_MIDDLE | NVG_ALIGN_BOTTOM | NVG_ALIGN_BASELINE);
+	float lineh = 0;
+
+	if (state->fontId == FONS_INVALID) return;
+
+	nvgTextMetrics(ctx, NULL, NULL, &lineh);
+
+	state->textAlign = NVG_ALIGN_LEFT | valign;
+
+	while ((nrows = nvgTextBreakLines(ctx, string, end, breakRowWidth, rows, 2))) {
+		for (i = 0; i < nrows; i++) {
+			NVGtextRow* row = &rows[i];
+			if (haling & NVG_ALIGN_LEFT)
+				nvgText(ctx, x, y, row->start, row->end);
+			else if (haling & NVG_ALIGN_CENTER)
+				nvgText(ctx, x + breakRowWidth*0.5f - row->width*0.5f, y, row->start, row->end);
+			else if (haling & NVG_ALIGN_RIGHT)
+				nvgText(ctx, x + breakRowWidth - row->width, y, row->start, row->end);
+			y += lineh * state->lineHeight;
+		}
+		string = rows[nrows-1].next;
+	}
+
+	state->textAlign = oldAlign;
+}
+
+int nvgTextGlyphPositions(NVGcontext* ctx, float x, float y, const char* string, const char* end, NVGglyphPosition* positions, int maxPositions)
+{
+	NVGstate* state = nvg__getState(ctx);
+	float scale = nvg__getFontScale(state) * ctx->devicePxRatio;
+	float invscale = 1.0f / scale;
+	FONStextIter iter, prevIter;
+	FONSquad q;
+	int npos = 0;
+
+	if (state->fontId == FONS_INVALID) return 0;
+
+	if (end == NULL)
+		end = string + strlen(string);
+
+	if (string == end)
+		return 0;
+
+	fonsSetSize(ctx->fs, state->fontSize*scale);
+	fonsSetSpacing(ctx->fs, state->letterSpacing*scale);
+	fonsSetBlur(ctx->fs, state->fontBlur*scale);
+	fonsSetAlign(ctx->fs, state->textAlign);
+	fonsSetFont(ctx->fs, state->fontId);
+
+	fonsTextIterInit(ctx->fs, &iter, x*scale, y*scale, string, end);
+	prevIter = iter;
+	while (fonsTextIterNext(ctx->fs, &iter, &q)) {
+		if (iter.prevGlyphIndex < 0 && nvg__allocTextAtlas(ctx)) { // can not retrieve glyph?
+			iter = prevIter;
+			fonsTextIterNext(ctx->fs, &iter, &q); // try again
+		}
+		prevIter = iter;
+		positions[npos].str = iter.str;
+		positions[npos].x = iter.x * invscale;
+		positions[npos].minx = nvg__minf(iter.x, q.x0) * invscale;
+		positions[npos].maxx = nvg__maxf(iter.nextx, q.x1) * invscale;
+		npos++;
+		if (npos >= maxPositions)
+			break;
+	}
+
+	return npos;
+}
+
+enum NVGcodepointType {
+	NVG_SPACE,
+	NVG_NEWLINE,
+	NVG_CHAR,
+};
+
+int nvgTextBreakLines(NVGcontext* ctx, const char* string, const char* end, float breakRowWidth, NVGtextRow* rows, int maxRows)
+{
+	NVGstate* state = nvg__getState(ctx);
+	float scale = nvg__getFontScale(state) * ctx->devicePxRatio;
+	float invscale = 1.0f / scale;
+	FONStextIter iter, prevIter;
+	FONSquad q;
+	int nrows = 0;
+	float rowStartX = 0;
+	float rowWidth = 0;
+	float rowMinX = 0;
+	float rowMaxX = 0;
+	const char* rowStart = NULL;
+	const char* rowEnd = NULL;
+	const char* wordStart = NULL;
+	float wordStartX = 0;
+	float wordMinX = 0;
+	const char* breakEnd = NULL;
+	float breakWidth = 0;
+	float breakMaxX = 0;
+	int type = NVG_SPACE, ptype = NVG_SPACE;
+	unsigned int pcodepoint = 0;
+
+	if (maxRows == 0) return 0;
+	if (state->fontId == FONS_INVALID) return 0;
+
+	if (end == NULL)
+		end = string + strlen(string);
+
+	if (string == end) return 0;
+
+	fonsSetSize(ctx->fs, state->fontSize*scale);
+	fonsSetSpacing(ctx->fs, state->letterSpacing*scale);
+	fonsSetBlur(ctx->fs, state->fontBlur*scale);
+	fonsSetAlign(ctx->fs, state->textAlign);
+	fonsSetFont(ctx->fs, state->fontId);
+
+	breakRowWidth *= scale;
+
+	fonsTextIterInit(ctx->fs, &iter, 0, 0, string, end);
+	prevIter = iter;
+	while (fonsTextIterNext(ctx->fs, &iter, &q)) {
+		if (iter.prevGlyphIndex < 0 && nvg__allocTextAtlas(ctx)) { // can not retrieve glyph?
+			iter = prevIter;
+			fonsTextIterNext(ctx->fs, &iter, &q); // try again
+		}
+		prevIter = iter;
+		switch (iter.codepoint) {
+			case 9:			// \t
+			case 11:		// \v
+			case 12:		// \f
+			case 32:		// space
+			case 0x00a0:	// NBSP
+				type = NVG_SPACE;
+				break;
+			case 10:		// \n
+				type = pcodepoint == 13 ? NVG_SPACE : NVG_NEWLINE;
+				break;
+			case 13:		// \r
+				type = pcodepoint == 10 ? NVG_SPACE : NVG_NEWLINE;
+				break;
+			case 0x0085:	// NEL
+				type = NVG_NEWLINE;
+				break;
+			default:
+				type = NVG_CHAR;
+				break;
+		}
+
+		if (type == NVG_NEWLINE) {
+			// Always handle new lines.
+			rows[nrows].start = rowStart != NULL ? rowStart : iter.str;
+			rows[nrows].end = rowEnd != NULL ? rowEnd : iter.str;
+			rows[nrows].width = rowWidth * invscale;
+			rows[nrows].minx = rowMinX * invscale;
+			rows[nrows].maxx = rowMaxX * invscale;
+			rows[nrows].next = iter.next;
+			nrows++;
+			if (nrows >= maxRows)
+				return nrows;
+			// Set null break point
+			breakEnd = rowStart;
+			breakWidth = 0.0;
+			breakMaxX = 0.0;
+			// Indicate to skip the white space at the beginning of the row.
+			rowStart = NULL;
+			rowEnd = NULL;
+			rowWidth = 0;
+			rowMinX = rowMaxX = 0;
+		} else {
+			if (rowStart == NULL) {
+				// Skip white space until the beginning of the line
+				if (type == NVG_CHAR) {
+					// The current char is the row so far
+					rowStartX = iter.x;
+					rowStart = iter.str;
+					rowEnd = iter.next;
+					rowWidth = iter.nextx - rowStartX; // q.x1 - rowStartX;
+					rowMinX = q.x0 - rowStartX;
+					rowMaxX = q.x1 - rowStartX;
+					wordStart = iter.str;
+					wordStartX = iter.x;
+					wordMinX = q.x0 - rowStartX;
+					// Set null break point
+					breakEnd = rowStart;
+					breakWidth = 0.0;
+					breakMaxX = 0.0;
+				}
+			} else {
+				float nextWidth = iter.nextx - rowStartX;
+
+				// track last non-white space character
+				if (type == NVG_CHAR) {
+					rowEnd = iter.next;
+					rowWidth = iter.nextx - rowStartX;
+					rowMaxX = q.x1 - rowStartX;
+				}
+				// track last end of a word
+				if (ptype == NVG_CHAR && type == NVG_SPACE) {
+					breakEnd = iter.str;
+					breakWidth = rowWidth;
+					breakMaxX = rowMaxX;
+				}
+				// track last beginning of a word
+				if (ptype == NVG_SPACE && type == NVG_CHAR) {
+					wordStart = iter.str;
+					wordStartX = iter.x;
+					wordMinX = q.x0 - rowStartX;
+				}
+
+				// Break to new line when a character is beyond break width.
+				if (type == NVG_CHAR && nextWidth > breakRowWidth) {
+					// The run length is too long, need to break to new line.
+					if (breakEnd == rowStart) {
+						// The current word is longer than the row length, just break it from here.
+						rows[nrows].start = rowStart;
+						rows[nrows].end = iter.str;
+						rows[nrows].width = rowWidth * invscale;
+						rows[nrows].minx = rowMinX * invscale;
+						rows[nrows].maxx = rowMaxX * invscale;
+						rows[nrows].next = iter.str;
+						nrows++;
+						if (nrows >= maxRows)
+							return nrows;
+						rowStartX = iter.x;
+						rowStart = iter.str;
+						rowEnd = iter.next;
+						rowWidth = iter.nextx - rowStartX;
+						rowMinX = q.x0 - rowStartX;
+						rowMaxX = q.x1 - rowStartX;
+						wordStart = iter.str;
+						wordStartX = iter.x;
+						wordMinX = q.x0 - rowStartX;
+					} else {
+						// Break the line from the end of the last word, and start new line from the beginning of the new.
+						rows[nrows].start = rowStart;
+						rows[nrows].end = breakEnd;
+						rows[nrows].width = breakWidth * invscale;
+						rows[nrows].minx = rowMinX * invscale;
+						rows[nrows].maxx = breakMaxX * invscale;
+						rows[nrows].next = wordStart;
+						nrows++;
+						if (nrows >= maxRows)
+							return nrows;
+						rowStartX = wordStartX;
+						rowStart = wordStart;
+						rowEnd = iter.next;
+						rowWidth = iter.nextx - rowStartX;
+						rowMinX = wordMinX;
+						rowMaxX = q.x1 - rowStartX;
+						// No change to the word start
+					}
+					// Set null break point
+					breakEnd = rowStart;
+					breakWidth = 0.0;
+					breakMaxX = 0.0;
+				}
+			}
+		}
+
+		pcodepoint = iter.codepoint;
+		ptype = type;
+	}
+
+	// Break the line from the end of the last word, and start new line from the beginning of the new.
+	if (rowStart != NULL) {
+		rows[nrows].start = rowStart;
+		rows[nrows].end = rowEnd;
+		rows[nrows].width = rowWidth * invscale;
+		rows[nrows].minx = rowMinX * invscale;
+		rows[nrows].maxx = rowMaxX * invscale;
+		rows[nrows].next = end;
+		nrows++;
+	}
+
+	return nrows;
+}
+
+float nvgTextBounds(NVGcontext* ctx, float x, float y, const char* string, const char* end, float* bounds)
+{
+	NVGstate* state = nvg__getState(ctx);
+	float scale = nvg__getFontScale(state) * ctx->devicePxRatio;
+	float invscale = 1.0f / scale;
+	float width;
+
+	if (state->fontId == FONS_INVALID) return 0;
+
+	fonsSetSize(ctx->fs, state->fontSize*scale);
+	fonsSetSpacing(ctx->fs, state->letterSpacing*scale);
+	fonsSetBlur(ctx->fs, state->fontBlur*scale);
+	fonsSetAlign(ctx->fs, state->textAlign);
+	fonsSetFont(ctx->fs, state->fontId);
+
+	width = fonsTextBounds(ctx->fs, x*scale, y*scale, string, end, bounds);
+	if (bounds != NULL) {
+		// Use line bounds for height.
+		fonsLineBounds(ctx->fs, y*scale, &bounds[1], &bounds[3]);
+		bounds[0] *= invscale;
+		bounds[1] *= invscale;
+		bounds[2] *= invscale;
+		bounds[3] *= invscale;
+	}
+	return width * invscale;
+}
+
+void nvgTextBoxBounds(NVGcontext* ctx, float x, float y, float breakRowWidth, const char* string, const char* end, float* bounds)
+{
+	NVGstate* state = nvg__getState(ctx);
+	NVGtextRow rows[2];
+	float scale = nvg__getFontScale(state) * ctx->devicePxRatio;
+	float invscale = 1.0f / scale;
+	int nrows = 0, i;
+	int oldAlign = state->textAlign;
+	int haling = state->textAlign & (NVG_ALIGN_LEFT | NVG_ALIGN_CENTER | NVG_ALIGN_RIGHT);
+	int valign = state->textAlign & (NVG_ALIGN_TOP | NVG_ALIGN_MIDDLE | NVG_ALIGN_BOTTOM | NVG_ALIGN_BASELINE);
+	float lineh = 0, rminy = 0, rmaxy = 0;
+	float minx, miny, maxx, maxy;
+
+	if (state->fontId == FONS_INVALID) {
+		if (bounds != NULL)
+			bounds[0] = bounds[1] = bounds[2] = bounds[3] = 0.0f;
+		return;
+	}
+
+	nvgTextMetrics(ctx, NULL, NULL, &lineh);
+
+	state->textAlign = NVG_ALIGN_LEFT | valign;
+
+	minx = maxx = x;
+	miny = maxy = y;
+
+	fonsSetSize(ctx->fs, state->fontSize*scale);
+	fonsSetSpacing(ctx->fs, state->letterSpacing*scale);
+	fonsSetBlur(ctx->fs, state->fontBlur*scale);
+	fonsSetAlign(ctx->fs, state->textAlign);
+	fonsSetFont(ctx->fs, state->fontId);
+	fonsLineBounds(ctx->fs, 0, &rminy, &rmaxy);
+	rminy *= invscale;
+	rmaxy *= invscale;
+
+	while ((nrows = nvgTextBreakLines(ctx, string, end, breakRowWidth, rows, 2))) {
+		for (i = 0; i < nrows; i++) {
+			NVGtextRow* row = &rows[i];
+			float rminx, rmaxx, dx = 0;
+			// Horizontal bounds
+			if (haling & NVG_ALIGN_LEFT)
+				dx = 0;
+			else if (haling & NVG_ALIGN_CENTER)
+				dx = breakRowWidth*0.5f - row->width*0.5f;
+			else if (haling & NVG_ALIGN_RIGHT)
+				dx = breakRowWidth - row->width;
+			rminx = x + row->minx + dx;
+			rmaxx = x + row->maxx + dx;
+			minx = nvg__minf(minx, rminx);
+			maxx = nvg__maxf(maxx, rmaxx);
+			// Vertical bounds.
+			miny = nvg__minf(miny, y + rminy);
+			maxy = nvg__maxf(maxy, y + rmaxy);
+
+			y += lineh * state->lineHeight;
+		}
+		string = rows[nrows-1].next;
+	}
+
+	state->textAlign = oldAlign;
+
+	if (bounds != NULL) {
+		bounds[0] = minx;
+		bounds[1] = miny;
+		bounds[2] = maxx;
+		bounds[3] = maxy;
+	}
+}
+
+void nvgTextMetrics(NVGcontext* ctx, float* ascender, float* descender, float* lineh)
+{
+	NVGstate* state = nvg__getState(ctx);
+	float scale = nvg__getFontScale(state) * ctx->devicePxRatio;
+	float invscale = 1.0f / scale;
+
+	if (state->fontId == FONS_INVALID) return;
+
+	fonsSetSize(ctx->fs, state->fontSize*scale);
+	fonsSetSpacing(ctx->fs, state->letterSpacing*scale);
+	fonsSetBlur(ctx->fs, state->fontBlur*scale);
+	fonsSetAlign(ctx->fs, state->textAlign);
+	fonsSetFont(ctx->fs, state->fontId);
+
+	fonsVertMetrics(ctx->fs, ascender, descender, lineh);
+	if (ascender != NULL)
+		*ascender *= invscale;
+	if (descender != NULL)
+		*descender *= invscale;
+	if (lineh != NULL)
+		*lineh *= invscale;
+}
+// vim: ft=c nu noet ts=4
diff --git a/dgl/src/nanovg2/nanovg.h b/dgl/src/nanovg2/nanovg.h
new file mode 100644
index 00000000..410ce660
--- /dev/null
+++ b/dgl/src/nanovg2/nanovg.h
@@ -0,0 +1,620 @@
+//
+// Copyright (c) 2013 Mikko Mononen memon@inside.org
+//
+// This software is provided 'as-is', without any express or implied
+// warranty.  In no event will the authors be held liable for any damages
+// arising from the use of this software.
+// Permission is granted to anyone to use this software for any purpose,
+// including commercial applications, and to alter it and redistribute it
+// freely, subject to the following restrictions:
+// 1. The origin of this software must not be misrepresented; you must not
+//    claim that you wrote the original software. If you use this software
+//    in a product, an acknowledgment in the product documentation would be
+//    appreciated but is not required.
+// 2. Altered source versions must be plainly marked as such, and must not be
+//    misrepresented as being the original software.
+// 3. This notice may not be removed or altered from any source distribution.
+//
+
+#ifndef NANOVG_H
+#define NANOVG_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#define NVG_PI 3.14159265358979323846264338327f
+
+#ifdef _MSC_VER
+#pragma warning(push)
+#pragma warning(disable: 4201)  // nonstandard extension used : nameless struct/union
+#endif
+
+typedef struct NVGcontext NVGcontext;
+
+struct NVGcolor {
+	union {
+		float rgba[4];
+		struct {
+			float r,g,b,a;
+		};
+	};
+};
+typedef struct NVGcolor NVGcolor;
+
+struct NVGpaint {
+	float xform[6];
+	float extent[2];
+	float radius;
+	float feather;
+	NVGcolor innerColor;
+	NVGcolor outerColor;
+	int image;
+};
+typedef struct NVGpaint NVGpaint;
+
+enum NVGwinding {
+	NVG_CCW = 1,			// Winding for solid shapes
+	NVG_CW = 2,				// Winding for holes
+};
+
+enum NVGsolidity {
+	NVG_SOLID = 1,			// CCW
+	NVG_HOLE = 2,			// CW
+};
+
+enum NVGlineCap {
+	NVG_BUTT,
+	NVG_ROUND,
+	NVG_SQUARE,
+	NVG_BEVEL,
+	NVG_MITER,
+};
+
+enum NVGalign {
+	// Horizontal align
+	NVG_ALIGN_LEFT 		= 1<<0,	// Default, align text horizontally to left.
+	NVG_ALIGN_CENTER 	= 1<<1,	// Align text horizontally to center.
+	NVG_ALIGN_RIGHT 	= 1<<2,	// Align text horizontally to right.
+	// Vertical align
+	NVG_ALIGN_TOP 		= 1<<3,	// Align text vertically to top.
+	NVG_ALIGN_MIDDLE	= 1<<4,	// Align text vertically to middle.
+	NVG_ALIGN_BOTTOM	= 1<<5,	// Align text vertically to bottom. 
+	NVG_ALIGN_BASELINE	= 1<<6, // Default, align text vertically to baseline. 
+};
+
+struct NVGglyphPosition {
+	const char* str;	// Position of the glyph in the input string.
+	float x;			// The x-coordinate of the logical glyph position.
+	float minx, maxx;	// The bounds of the glyph shape.
+};
+typedef struct NVGglyphPosition NVGglyphPosition;
+
+struct NVGtextRow {
+	const char* start;	// Pointer to the input text where the row starts.
+	const char* end;	// Pointer to the input text where the row ends (one past the last character).
+	const char* next;	// Pointer to the beginning of the next row.
+	float width;		// Logical width of the row.
+	float minx, maxx;	// Actual bounds of the row. Logical with and bounds can differ because of kerning and some parts over extending.
+};
+typedef struct NVGtextRow NVGtextRow;
+
+enum NVGimageFlags {
+	NVG_IMAGE_GENERATE_MIPMAPS		= 1<<0,		// Generate mipmaps during creation of the image.
+	NVG_IMAGE_REPEATX			= 1<<1,		// Repeat image in X direction.
+	NVG_IMAGE_REPEATY			= 1<<2,		// Repeat image in Y direction.
+	NVG_IMAGE_FLIPY				= 1<<3,		// Flips (inverses) image in Y direction when rendered.
+	NVG_IMAGE_PREMULTIPLIED			= 1<<4,		// Image data has premultiplied alpha.
+};
+
+// Begin drawing a new frame
+// Calls to nanovg drawing API should be wrapped in nvgBeginFrame() & nvgEndFrame()
+// nvgBeginFrame() defines the size of the window to render to in relation currently
+// set viewport (i.e. glViewport on GL backends). Device pixel ration allows to
+// control the rendering on Hi-DPI devices.
+// For example, GLFW returns two dimension for an opened window: window size and
+// frame buffer size. In that case you would set windowWidth/Height to the window size
+// devicePixelRatio to: frameBufferWidth / windowWidth.
+void nvgBeginFrame(NVGcontext* ctx, int windowWidth, int windowHeight, float devicePixelRatio);
+
+// Cancels drawing the current frame.
+void nvgCancelFrame(NVGcontext* ctx);
+
+// Ends drawing flushing remaining render state.
+void nvgEndFrame(NVGcontext* ctx);
+
+//
+// Color utils
+//
+// Colors in NanoVG are stored as unsigned ints in ABGR format.
+
+// Returns a color value from red, green, blue values. Alpha will be set to 255 (1.0f).
+NVGcolor nvgRGB(unsigned char r, unsigned char g, unsigned char b);
+
+// Returns a color value from red, green, blue values. Alpha will be set to 1.0f.
+NVGcolor nvgRGBf(float r, float g, float b);
+
+
+// Returns a color value from red, green, blue and alpha values.
+NVGcolor nvgRGBA(unsigned char r, unsigned char g, unsigned char b, unsigned char a);
+
+// Returns a color value from red, green, blue and alpha values.
+NVGcolor nvgRGBAf(float r, float g, float b, float a);
+
+
+// Linearly interpolates from color c0 to c1, and returns resulting color value.
+NVGcolor nvgLerpRGBA(NVGcolor c0, NVGcolor c1, float u);
+
+// Sets transparency of a color value.
+NVGcolor nvgTransRGBA(NVGcolor c0, unsigned char a);
+
+// Sets transparency of a color value.
+NVGcolor nvgTransRGBAf(NVGcolor c0, float a);
+
+// Returns color value specified by hue, saturation and lightness.
+// HSL values are all in range [0..1], alpha will be set to 255.
+NVGcolor nvgHSL(float h, float s, float l);
+
+// Returns color value specified by hue, saturation and lightness and alpha.
+// HSL values are all in range [0..1], alpha in range [0..255]
+NVGcolor nvgHSLA(float h, float s, float l, unsigned char a);
+
+//
+// State Handling
+//
+// NanoVG contains state which represents how paths will be rendered.
+// The state contains transform, fill and stroke styles, text and font styles,
+// and scissor clipping.
+
+// Pushes and saves the current render state into a state stack.
+// A matching nvgRestore() must be used to restore the state.
+void nvgSave(NVGcontext* ctx);
+
+// Pops and restores current render state.
+void nvgRestore(NVGcontext* ctx);
+
+// Resets current render state to default values. Does not affect the render state stack.
+void nvgReset(NVGcontext* ctx);
+
+//
+// Render styles
+//
+// Fill and stroke render style can be either a solid color or a paint which is a gradient or a pattern.
+// Solid color is simply defined as a color value, different kinds of paints can be created
+// using nvgLinearGradient(), nvgBoxGradient(), nvgRadialGradient() and nvgImagePattern().
+//
+// Current render style can be saved and restored using nvgSave() and nvgRestore(). 
+
+// Sets current stroke style to a solid color.
+void nvgStrokeColor(NVGcontext* ctx, NVGcolor color);
+
+// Sets current stroke style to a paint, which can be a one of the gradients or a pattern.
+void nvgStrokePaint(NVGcontext* ctx, NVGpaint paint);
+
+// Sets current fill style to a solid color.
+void nvgFillColor(NVGcontext* ctx, NVGcolor color);
+
+// Sets current fill style to a paint, which can be a one of the gradients or a pattern.
+void nvgFillPaint(NVGcontext* ctx, NVGpaint paint);
+
+// Sets the miter limit of the stroke style.
+// Miter limit controls when a sharp corner is beveled.
+void nvgMiterLimit(NVGcontext* ctx, float limit);
+
+// Sets the stroke width of the stroke style.
+void nvgStrokeWidth(NVGcontext* ctx, float size);
+
+// Sets how the end of the line (cap) is drawn,
+// Can be one of: NVG_BUTT (default), NVG_ROUND, NVG_SQUARE.
+void nvgLineCap(NVGcontext* ctx, int cap);
+
+// Sets how sharp path corners are drawn.
+// Can be one of NVG_MITER (default), NVG_ROUND, NVG_BEVEL.
+void nvgLineJoin(NVGcontext* ctx, int join);
+
+// Sets the transparency applied to all rendered shapes.
+// Already transparent paths will get proportionally more transparent as well.
+void nvgGlobalAlpha(NVGcontext* ctx, float alpha);
+
+//
+// Transforms
+//
+// The paths, gradients, patterns and scissor region are transformed by an transformation
+// matrix at the time when they are passed to the API.
+// The current transformation matrix is a affine matrix:
+//   [sx kx tx]
+//   [ky sy ty]
+//   [ 0  0  1]
+// Where: sx,sy define scaling, kx,ky skewing, and tx,ty translation.
+// The last row is assumed to be 0,0,1 and is not stored.
+//
+// Apart from nvgResetTransform(), each transformation function first creates
+// specific transformation matrix and pre-multiplies the current transformation by it.
+//
+// Current coordinate system (transformation) can be saved and restored using nvgSave() and nvgRestore(). 
+
+// Resets current transform to a identity matrix.
+void nvgResetTransform(NVGcontext* ctx);
+
+// Premultiplies current coordinate system by specified matrix.
+// The parameters are interpreted as matrix as follows:
+//   [a c e]
+//   [b d f]
+//   [0 0 1]
+void nvgTransform(NVGcontext* ctx, float a, float b, float c, float d, float e, float f);
+
+// Translates current coordinate system.
+void nvgTranslate(NVGcontext* ctx, float x, float y);
+
+// Rotates current coordinate system. Angle is specified in radians.
+void nvgRotate(NVGcontext* ctx, float angle);
+
+// Skews the current coordinate system along X axis. Angle is specified in radians.
+void nvgSkewX(NVGcontext* ctx, float angle);
+
+// Skews the current coordinate system along Y axis. Angle is specified in radians.
+void nvgSkewY(NVGcontext* ctx, float angle);
+
+// Scales the current coordinate system.
+void nvgScale(NVGcontext* ctx, float x, float y);
+
+// Stores the top part (a-f) of the current transformation matrix in to the specified buffer.
+//   [a c e]
+//   [b d f]
+//   [0 0 1]
+// There should be space for 6 floats in the return buffer for the values a-f.
+void nvgCurrentTransform(NVGcontext* ctx, float* xform);
+
+
+// The following functions can be used to make calculations on 2x3 transformation matrices.
+// A 2x3 matrix is represented as float[6].
+
+// Sets the transform to identity matrix.
+void nvgTransformIdentity(float* dst);
+
+// Sets the transform to translation matrix matrix.
+void nvgTransformTranslate(float* dst, float tx, float ty);
+
+// Sets the transform to scale matrix.
+void nvgTransformScale(float* dst, float sx, float sy);
+
+// Sets the transform to rotate matrix. Angle is specified in radians.
+void nvgTransformRotate(float* dst, float a);
+
+// Sets the transform to skew-x matrix. Angle is specified in radians.
+void nvgTransformSkewX(float* dst, float a);
+
+// Sets the transform to skew-y matrix. Angle is specified in radians.
+void nvgTransformSkewY(float* dst, float a);
+
+// Sets the transform to the result of multiplication of two transforms, of A = A*B.
+void nvgTransformMultiply(float* dst, const float* src);
+
+// Sets the transform to the result of multiplication of two transforms, of A = B*A.
+void nvgTransformPremultiply(float* dst, const float* src);
+
+// Sets the destination to inverse of specified transform.
+// Returns 1 if the inverse could be calculated, else 0.
+int nvgTransformInverse(float* dst, const float* src);
+
+// Transform a point by given transform.
+void nvgTransformPoint(float* dstx, float* dsty, const float* xform, float srcx, float srcy);
+
+// Converts degrees to radians and vice versa.
+float nvgDegToRad(float deg);
+float nvgRadToDeg(float rad);
+
+//
+// Images
+//
+// NanoVG allows you to load jpg, png, psd, tga, pic and gif files to be used for rendering.
+// In addition you can upload your own image. The image loading is provided by stb_image.
+// The parameter imageFlags is combination of flags defined in NVGimageFlags.
+
+// Creates image by loading it from the disk from specified file name.
+// Returns handle to the image.
+int nvgCreateImage(NVGcontext* ctx, const char* filename, int imageFlags);
+
+// Creates image by loading it from the specified chunk of memory.
+// Returns handle to the image.
+int nvgCreateImageMem(NVGcontext* ctx, int imageFlags, unsigned char* data, int ndata);
+
+// Creates image from specified image data.
+// Returns handle to the image.
+int nvgCreateImageRGBA(NVGcontext* ctx, int w, int h, int imageFlags, const unsigned char* data);
+
+// Updates image data specified by image handle.
+void nvgUpdateImage(NVGcontext* ctx, int image, const unsigned char* data);
+
+// Returns the dimensions of a created image.
+void nvgImageSize(NVGcontext* ctx, int image, int* w, int* h);
+
+// Deletes created image.
+void nvgDeleteImage(NVGcontext* ctx, int image);
+
+//
+// Paints
+//
+// NanoVG supports four types of paints: linear gradient, box gradient, radial gradient and image pattern.
+// These can be used as paints for strokes and fills.
+
+// Creates and returns a linear gradient. Parameters (sx,sy)-(ex,ey) specify the start and end coordinates
+// of the linear gradient, icol specifies the start color and ocol the end color.
+// The gradient is transformed by the current transform when it is passed to nvgFillPaint() or nvgStrokePaint().
+NVGpaint nvgLinearGradient(NVGcontext* ctx, float sx, float sy, float ex, float ey,
+						   NVGcolor icol, NVGcolor ocol);
+
+// Creates and returns a box gradient. Box gradient is a feathered rounded rectangle, it is useful for rendering
+// drop shadows or highlights for boxes. Parameters (x,y) define the top-left corner of the rectangle,
+// (w,h) define the size of the rectangle, r defines the corner radius, and f feather. Feather defines how blurry
+// the border of the rectangle is. Parameter icol specifies the inner color and ocol the outer color of the gradient.
+// The gradient is transformed by the current transform when it is passed to nvgFillPaint() or nvgStrokePaint().
+NVGpaint nvgBoxGradient(NVGcontext* ctx, float x, float y, float w, float h,
+						float r, float f, NVGcolor icol, NVGcolor ocol);
+
+// Creates and returns a radial gradient. Parameters (cx,cy) specify the center, inr and outr specify
+// the inner and outer radius of the gradient, icol specifies the start color and ocol the end color.
+// The gradient is transformed by the current transform when it is passed to nvgFillPaint() or nvgStrokePaint().
+NVGpaint nvgRadialGradient(NVGcontext* ctx, float cx, float cy, float inr, float outr,
+						   NVGcolor icol, NVGcolor ocol);
+
+// Creates and returns an image patter. Parameters (ox,oy) specify the left-top location of the image pattern,
+// (ex,ey) the size of one image, angle rotation around the top-left corner, image is handle to the image to render.
+// The gradient is transformed by the current transform when it is passed to nvgFillPaint() or nvgStrokePaint().
+NVGpaint nvgImagePattern(NVGcontext* ctx, float ox, float oy, float ex, float ey,
+						 float angle, int image, float alpha);
+
+//
+// Scissoring
+//
+// Scissoring allows you to clip the rendering into a rectangle. This is useful for various
+// user interface cases like rendering a text edit or a timeline. 
+
+// Sets the current scissor rectangle.
+// The scissor rectangle is transformed by the current transform.
+void nvgScissor(NVGcontext* ctx, float x, float y, float w, float h);
+
+// Intersects current scissor rectangle with the specified rectangle.
+// The scissor rectangle is transformed by the current transform.
+// Note: in case the rotation of previous scissor rect differs from
+// the current one, the intersection will be done between the specified
+// rectangle and the previous scissor rectangle transformed in the current
+// transform space. The resulting shape is always rectangle.
+void nvgIntersectScissor(NVGcontext* ctx, float x, float y, float w, float h);
+
+// Reset and disables scissoring.
+void nvgResetScissor(NVGcontext* ctx);
+
+//
+// Paths
+//
+// Drawing a new shape starts with nvgBeginPath(), it clears all the currently defined paths.
+// Then you define one or more paths and sub-paths which describe the shape. The are functions
+// to draw common shapes like rectangles and circles, and lower level step-by-step functions,
+// which allow to define a path curve by curve.
+//
+// NanoVG uses even-odd fill rule to draw the shapes. Solid shapes should have counter clockwise
+// winding and holes should have counter clockwise order. To specify winding of a path you can
+// call nvgPathWinding(). This is useful especially for the common shapes, which are drawn CCW.
+//
+// Finally you can fill the path using current fill style by calling nvgFill(), and stroke it
+// with current stroke style by calling nvgStroke().
+//
+// The curve segments and sub-paths are transformed by the current transform.
+
+// Clears the current path and sub-paths.
+void nvgBeginPath(NVGcontext* ctx);
+
+// Starts new sub-path with specified point as first point.
+void nvgMoveTo(NVGcontext* ctx, float x, float y);
+
+// Adds line segment from the last point in the path to the specified point.
+void nvgLineTo(NVGcontext* ctx, float x, float y);
+
+// Adds cubic bezier segment from last point in the path via two control points to the specified point.
+void nvgBezierTo(NVGcontext* ctx, float c1x, float c1y, float c2x, float c2y, float x, float y);
+
+// Adds quadratic bezier segment from last point in the path via a control point to the specified point.
+void nvgQuadTo(NVGcontext* ctx, float cx, float cy, float x, float y);
+
+// Adds an arc segment at the corner defined by the last path point, and two specified points.
+void nvgArcTo(NVGcontext* ctx, float x1, float y1, float x2, float y2, float radius);
+
+// Closes current sub-path with a line segment.
+void nvgClosePath(NVGcontext* ctx);
+
+// Sets the current sub-path winding, see NVGwinding and NVGsolidity. 
+void nvgPathWinding(NVGcontext* ctx, int dir);
+
+// Creates new circle arc shaped sub-path. The arc center is at cx,cy, the arc radius is r,
+// and the arc is drawn from angle a0 to a1, and swept in direction dir (NVG_CCW, or NVG_CW).
+// Angles are specified in radians.
+void nvgArc(NVGcontext* ctx, float cx, float cy, float r, float a0, float a1, int dir);
+
+// Creates new rectangle shaped sub-path.
+void nvgRect(NVGcontext* ctx, float x, float y, float w, float h);
+
+// Creates new rounded rectangle shaped sub-path.
+void nvgRoundedRect(NVGcontext* ctx, float x, float y, float w, float h, float r);
+
+// Creates new ellipse shaped sub-path.
+void nvgEllipse(NVGcontext* ctx, float cx, float cy, float rx, float ry);
+
+// Creates new circle shaped sub-path. 
+void nvgCircle(NVGcontext* ctx, float cx, float cy, float r);
+
+// Fills the current path with current fill style.
+void nvgFill(NVGcontext* ctx);
+
+// Fills the current path with current stroke style.
+void nvgStroke(NVGcontext* ctx);
+
+
+//
+// Text
+//
+// NanoVG allows you to load .ttf files and use the font to render text.
+//
+// The appearance of the text can be defined by setting the current text style
+// and by specifying the fill color. Common text and font settings such as
+// font size, letter spacing and text align are supported. Font blur allows you
+// to create simple text effects such as drop shadows.
+//
+// At render time the font face can be set based on the font handles or name.
+//
+// Font measure functions return values in local space, the calculations are
+// carried in the same resolution as the final rendering. This is done because
+// the text glyph positions are snapped to the nearest pixels sharp rendering.
+//
+// The local space means that values are not rotated or scale as per the current
+// transformation. For example if you set font size to 12, which would mean that
+// line height is 16, then regardless of the current scaling and rotation, the
+// returned line height is always 16. Some measures may vary because of the scaling
+// since aforementioned pixel snapping.
+//
+// While this may sound a little odd, the setup allows you to always render the
+// same way regardless of scaling. I.e. following works regardless of scaling:
+//
+//		const char* txt = "Text me up.";
+//		nvgTextBounds(vg, x,y, txt, NULL, bounds);
+//		nvgBeginPath(vg);
+//		nvgRoundedRect(vg, bounds[0],bounds[1], bounds[2]-bounds[0], bounds[3]-bounds[1]);
+//		nvgFill(vg);
+//
+// Note: currently only solid color fill is supported for text.
+
+// Creates font by loading it from the disk from specified file name.
+// Returns handle to the font.
+int nvgCreateFont(NVGcontext* ctx, const char* name, const char* filename);
+
+// Creates image by loading it from the specified memory chunk.
+// Returns handle to the font.
+int nvgCreateFontMem(NVGcontext* ctx, const char* name, unsigned char* data, int ndata, int freeData);
+
+// Finds a loaded font of specified name, and returns handle to it, or -1 if the font is not found.
+int nvgFindFont(NVGcontext* ctx, const char* name);
+
+// Sets the font size of current text style.
+void nvgFontSize(NVGcontext* ctx, float size);
+
+// Sets the blur of current text style.
+void nvgFontBlur(NVGcontext* ctx, float blur);
+
+// Sets the letter spacing of current text style.
+void nvgTextLetterSpacing(NVGcontext* ctx, float spacing);
+
+// Sets the proportional line height of current text style. The line height is specified as multiple of font size. 
+void nvgTextLineHeight(NVGcontext* ctx, float lineHeight);
+
+// Sets the text align of current text style, see NVGalign for options.
+void nvgTextAlign(NVGcontext* ctx, int align);
+
+// Sets the font face based on specified id of current text style.
+void nvgFontFaceId(NVGcontext* ctx, int font);
+
+// Sets the font face based on specified name of current text style.
+void nvgFontFace(NVGcontext* ctx, const char* font);
+
+// Draws text string at specified location. If end is specified only the sub-string up to the end is drawn.
+float nvgText(NVGcontext* ctx, float x, float y, const char* string, const char* end);
+
+// Draws multi-line text string at specified location wrapped at the specified width. If end is specified only the sub-string up to the end is drawn.
+// White space is stripped at the beginning of the rows, the text is split at word boundaries or when new-line characters are encountered.
+// Words longer than the max width are slit at nearest character (i.e. no hyphenation).
+void nvgTextBox(NVGcontext* ctx, float x, float y, float breakRowWidth, const char* string, const char* end);
+
+// Measures the specified text string. Parameter bounds should be a pointer to float[4],
+// if the bounding box of the text should be returned. The bounds value are [xmin,ymin, xmax,ymax]
+// Returns the horizontal advance of the measured text (i.e. where the next character should drawn).
+// Measured values are returned in local coordinate space.
+float nvgTextBounds(NVGcontext* ctx, float x, float y, const char* string, const char* end, float* bounds);
+
+// Measures the specified multi-text string. Parameter bounds should be a pointer to float[4],
+// if the bounding box of the text should be returned. The bounds value are [xmin,ymin, xmax,ymax]
+// Measured values are returned in local coordinate space.
+void nvgTextBoxBounds(NVGcontext* ctx, float x, float y, float breakRowWidth, const char* string, const char* end, float* bounds);
+
+// Calculates the glyph x positions of the specified text. If end is specified only the sub-string will be used.
+// Measured values are returned in local coordinate space.
+int nvgTextGlyphPositions(NVGcontext* ctx, float x, float y, const char* string, const char* end, NVGglyphPosition* positions, int maxPositions);
+
+// Returns the vertical metrics based on the current text style.
+// Measured values are returned in local coordinate space.
+void nvgTextMetrics(NVGcontext* ctx, float* ascender, float* descender, float* lineh);
+
+// Breaks the specified text into lines. If end is specified only the sub-string will be used.
+// White space is stripped at the beginning of the rows, the text is split at word boundaries or when new-line characters are encountered.
+// Words longer than the max width are slit at nearest character (i.e. no hyphenation).
+int nvgTextBreakLines(NVGcontext* ctx, const char* string, const char* end, float breakRowWidth, NVGtextRow* rows, int maxRows);
+
+//
+// Internal Render API
+//
+enum NVGtexture {
+	NVG_TEXTURE_ALPHA = 0x01,
+	NVG_TEXTURE_RGBA = 0x02,
+};
+
+struct NVGscissor {
+	float xform[6];
+	float extent[2];
+};
+typedef struct NVGscissor NVGscissor;
+
+struct NVGvertex {
+	float x,y,u,v;
+};
+typedef struct NVGvertex NVGvertex;
+
+struct NVGpath {
+	int first;
+	int count;
+	unsigned char closed;
+	int nbevel;
+	NVGvertex* fill;
+	int nfill;
+	NVGvertex* stroke;
+	int nstroke;
+	int winding;
+	int convex;
+};
+typedef struct NVGpath NVGpath;
+
+struct NVGparams {
+	void* userPtr;
+	int edgeAntiAlias;
+	int (*renderCreate)(void* uptr);
+	int (*renderCreateTexture)(void* uptr, int type, int w, int h, int imageFlags, const unsigned char* data);
+	int (*renderDeleteTexture)(void* uptr, int image);
+	int (*renderUpdateTexture)(void* uptr, int image, int x, int y, int w, int h, const unsigned char* data);
+	int (*renderGetTextureSize)(void* uptr, int image, int* w, int* h);
+	void (*renderViewport)(void* uptr, int width, int height);
+	void (*renderCancel)(void* uptr);
+	void (*renderFlush)(void* uptr);
+	void (*renderFill)(void* uptr, NVGpaint* paint, NVGscissor* scissor, float fringe, const float* bounds, const NVGpath* paths, int npaths);
+	void (*renderStroke)(void* uptr, NVGpaint* paint, NVGscissor* scissor, float fringe, float strokeWidth, const NVGpath* paths, int npaths);
+	void (*renderTriangles)(void* uptr, NVGpaint* paint, NVGscissor* scissor, const NVGvertex* verts, int nverts);
+	void (*renderDelete)(void* uptr);
+};
+typedef struct NVGparams NVGparams;
+
+// Constructor and destructor, called by the render back-end.
+NVGcontext* nvgCreateInternal(NVGparams* params);
+void nvgDeleteInternal(NVGcontext* ctx);
+
+NVGparams* nvgInternalParams(NVGcontext* ctx);
+
+// Debug function to dump cached path data.
+void nvgDebugDumpPathCache(NVGcontext* ctx);
+
+#ifdef _MSC_VER
+#pragma warning(pop)
+#endif
+
+#define NVG_NOTUSED(v) for (;;) { (void)(1 ? (void)0 : ( (void)(v) ) ); break; }
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif // NANOVG_H
diff --git a/dgl/src/nanovg2/nanovg_gl.h b/dgl/src/nanovg2/nanovg_gl.h
new file mode 100644
index 00000000..52357951
--- /dev/null
+++ b/dgl/src/nanovg2/nanovg_gl.h
@@ -0,0 +1,1525 @@
+//
+// Copyright (c) 2009-2013 Mikko Mononen memon@inside.org
+//
+// This software is provided 'as-is', without any express or implied
+// warranty.  In no event will the authors be held liable for any damages
+// arising from the use of this software.
+// Permission is granted to anyone to use this software for any purpose,
+// including commercial applications, and to alter it and redistribute it
+// freely, subject to the following restrictions:
+// 1. The origin of this software must not be misrepresented; you must not
+//    claim that you wrote the original software. If you use this software
+//    in a product, an acknowledgment in the product documentation would be
+//    appreciated but is not required.
+// 2. Altered source versions must be plainly marked as such, and must not be
+//    misrepresented as being the original software.
+// 3. This notice may not be removed or altered from any source distribution.
+//
+#ifndef NANOVG_GL_H
+#define NANOVG_GL_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+// Create flags
+
+enum NVGcreateFlags {
+	// Flag indicating if geometry based anti-aliasing is used (may not be needed when using MSAA).
+	NVG_ANTIALIAS 		= 1<<0,
+	// Flag indicating if strokes should be drawn using stencil buffer. The rendering will be a little
+	// slower, but path overlaps (i.e. self-intersecting or sharp turns) will be drawn just once.
+	NVG_STENCIL_STROKES	= 1<<1,
+	// Flag indicating that additional debug checks are done.
+	NVG_DEBUG 			= 1<<2,
+};
+
+#if defined NANOVG_GL2_IMPLEMENTATION
+#  define NANOVG_GL2 1
+#  define NANOVG_GL_IMPLEMENTATION 1
+#elif defined NANOVG_GL3_IMPLEMENTATION
+#  define NANOVG_GL3 1
+#  define NANOVG_GL_IMPLEMENTATION 1
+#  define NANOVG_GL_USE_UNIFORMBUFFER 1
+#elif defined NANOVG_GLES2_IMPLEMENTATION
+#  define NANOVG_GLES2 1
+#  define NANOVG_GL_IMPLEMENTATION 1
+#elif defined NANOVG_GLES3_IMPLEMENTATION
+#  define NANOVG_GLES3 1
+#  define NANOVG_GL_IMPLEMENTATION 1
+#endif
+
+#define NANOVG_GL_USE_STATE_FILTER (1)
+
+// Creates NanoVG contexts for different OpenGL (ES) versions.
+// Flags should be combination of the create flags above.
+
+#if defined NANOVG_GL2
+
+NVGcontext* nvgCreateGL2(int flags);
+void nvgDeleteGL2(NVGcontext* ctx);
+
+#endif
+
+#if defined NANOVG_GL3
+
+NVGcontext* nvgCreateGL3(int flags);
+void nvgDeleteGL3(NVGcontext* ctx);
+
+#endif
+
+#if defined NANOVG_GLES2
+
+NVGcontext* nvgCreateGLES2(int flags);
+void nvgDeleteGLES2(NVGcontext* ctx);
+
+#endif
+
+#if defined NANOVG_GLES3
+
+NVGcontext* nvgCreateGLES3(int flags);
+void nvgDeleteGLES3(NVGcontext* ctx);
+
+#endif
+
+// These are additional flags on top of NVGimageFlags.
+enum NVGimageFlagsGL {
+	NVG_IMAGE_NODELETE			= 1<<16,	// Do not delete GL texture handle.
+};
+
+int nvglCreateImageFromHandle(NVGcontext* ctx, GLuint textureId, int w, int h, int flags);
+GLuint nvglImageHandle(NVGcontext* ctx, int image);
+
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* NANOVG_GL_H */
+
+#ifdef NANOVG_GL_IMPLEMENTATION
+
+#include <stdlib.h>
+#include <stdio.h>
+#include <string.h>
+#include <math.h>
+#include "nanovg.h"
+
+enum GLNVGuniformLoc {
+	GLNVG_LOC_VIEWSIZE,
+	GLNVG_LOC_TEX,
+	GLNVG_LOC_FRAG,
+	GLNVG_MAX_LOCS
+};
+
+enum GLNVGshaderType {
+	NSVG_SHADER_FILLGRAD,
+	NSVG_SHADER_FILLIMG,
+	NSVG_SHADER_SIMPLE,
+	NSVG_SHADER_IMG
+};
+
+#if NANOVG_GL_USE_UNIFORMBUFFER
+enum GLNVGuniformBindings {
+	GLNVG_FRAG_BINDING = 0,
+};
+#endif
+
+struct GLNVGshader {
+	GLuint prog;
+	GLuint frag;
+	GLuint vert;
+	GLint loc[GLNVG_MAX_LOCS];
+};
+typedef struct GLNVGshader GLNVGshader;
+
+struct GLNVGtexture {
+	int id;
+	GLuint tex;
+	int width, height;
+	int type;
+	int flags;
+};
+typedef struct GLNVGtexture GLNVGtexture;
+
+enum GLNVGcallType {
+	GLNVG_NONE = 0,
+	GLNVG_FILL,
+	GLNVG_CONVEXFILL,
+	GLNVG_STROKE,
+	GLNVG_TRIANGLES,
+};
+
+struct GLNVGcall {
+	int type;
+	int image;
+	int pathOffset;
+	int pathCount;
+	int triangleOffset;
+	int triangleCount;
+	int uniformOffset;
+};
+typedef struct GLNVGcall GLNVGcall;
+
+struct GLNVGpath {
+	int fillOffset;
+	int fillCount;
+	int strokeOffset;
+	int strokeCount;
+};
+typedef struct GLNVGpath GLNVGpath;
+
+struct GLNVGfragUniforms {
+	#if NANOVG_GL_USE_UNIFORMBUFFER
+		float scissorMat[12]; // matrices are actually 3 vec4s
+		float paintMat[12];
+		struct NVGcolor innerCol;
+		struct NVGcolor outerCol;
+		float scissorExt[2];
+		float scissorScale[2];
+		float extent[2];
+		float radius;
+		float feather;
+		float strokeMult;
+		float strokeThr;
+		int texType;
+		int type;
+	#else
+		// note: after modifying layout or size of uniform array,
+		// don't forget to also update the fragment shader source!
+		#define NANOVG_GL_UNIFORMARRAY_SIZE 11
+		union {
+			struct {
+				float scissorMat[12]; // matrices are actually 3 vec4s
+				float paintMat[12];
+				struct NVGcolor innerCol;
+				struct NVGcolor outerCol;
+				float scissorExt[2];
+				float scissorScale[2];
+				float extent[2];
+				float radius;
+				float feather;
+				float strokeMult;
+				float strokeThr;
+				float texType;
+				float type;
+			};
+			float uniformArray[NANOVG_GL_UNIFORMARRAY_SIZE][4];
+		};
+	#endif
+};
+typedef struct GLNVGfragUniforms GLNVGfragUniforms;
+
+struct GLNVGcontext {
+	GLNVGshader shader;
+	GLNVGtexture* textures;
+	float view[2];
+	int ntextures;
+	int ctextures;
+	int textureId;
+	GLuint vertBuf;
+#if defined NANOVG_GL3
+	GLuint vertArr;
+#endif
+#if NANOVG_GL_USE_UNIFORMBUFFER
+	GLuint fragBuf;
+#endif
+	int fragSize;
+	int flags;
+
+	// Per frame buffers
+	GLNVGcall* calls;
+	int ccalls;
+	int ncalls;
+	GLNVGpath* paths;
+	int cpaths;
+	int npaths;
+	struct NVGvertex* verts;
+	int cverts;
+	int nverts;
+	unsigned char* uniforms;
+	int cuniforms;
+	int nuniforms;
+
+	// cached state
+	#if NANOVG_GL_USE_STATE_FILTER
+	GLuint boundTexture;
+	GLuint stencilMask;
+	GLenum stencilFunc;
+	GLint stencilFuncRef;
+	GLuint stencilFuncMask;
+	#endif
+};
+typedef struct GLNVGcontext GLNVGcontext;
+
+static int glnvg__maxi(int a, int b) { return a > b ? a : b; }
+
+#ifdef NANOVG_GLES2
+static unsigned int glnvg__nearestPow2(unsigned int num)
+{
+	unsigned n = num > 0 ? num - 1 : 0;
+	n |= n >> 1;
+	n |= n >> 2;
+	n |= n >> 4;
+	n |= n >> 8;
+	n |= n >> 16;
+	n++;
+	return n;
+}
+#endif
+
+static void glnvg__bindTexture(GLNVGcontext* gl, GLuint tex)
+{
+#if NANOVG_GL_USE_STATE_FILTER
+	if (gl->boundTexture != tex) {
+		gl->boundTexture = tex;
+		glBindTexture(GL_TEXTURE_2D, tex);
+	}
+#else
+	glBindTexture(GL_TEXTURE_2D, tex);
+#endif
+}
+
+static void glnvg__stencilMask(GLNVGcontext* gl, GLuint mask)
+{
+#if NANOVG_GL_USE_STATE_FILTER
+	if (gl->stencilMask != mask) {
+		gl->stencilMask = mask;
+		glStencilMask(mask);
+	}
+#else
+	glStencilMask(mask);
+#endif
+}
+
+static void glnvg__stencilFunc(GLNVGcontext* gl, GLenum func, GLint ref, GLuint mask)
+{
+#if NANOVG_GL_USE_STATE_FILTER
+	if ((gl->stencilFunc != func) ||
+		(gl->stencilFuncRef != ref) ||
+		(gl->stencilFuncMask != mask)) {
+		
+		gl->stencilFunc = func;
+		gl->stencilFuncRef = ref;
+		gl->stencilFuncMask = mask;
+		glStencilFunc(func, ref, mask);
+	}
+#else
+	glStencilFunc(func, ref, mask);
+#endif
+}
+
+static GLNVGtexture* glnvg__allocTexture(GLNVGcontext* gl)
+{
+	GLNVGtexture* tex = NULL;
+	int i;
+
+	for (i = 0; i < gl->ntextures; i++) {
+		if (gl->textures[i].id == 0) {
+			tex = &gl->textures[i];
+			break;
+		}
+	}
+	if (tex == NULL) {
+		if (gl->ntextures+1 > gl->ctextures) {
+			GLNVGtexture* textures;
+			int ctextures = glnvg__maxi(gl->ntextures+1, 4) +  gl->ctextures/2; // 1.5x Overallocate
+			textures = (GLNVGtexture*)realloc(gl->textures, sizeof(GLNVGtexture)*ctextures);
+			if (textures == NULL) return NULL;
+			gl->textures = textures;
+			gl->ctextures = ctextures;
+		}
+		tex = &gl->textures[gl->ntextures++];
+	}
+	
+	memset(tex, 0, sizeof(*tex));
+	tex->id = ++gl->textureId;
+	
+	return tex;
+}
+
+static GLNVGtexture* glnvg__findTexture(GLNVGcontext* gl, int id)
+{
+	int i;
+	for (i = 0; i < gl->ntextures; i++)
+		if (gl->textures[i].id == id)
+			return &gl->textures[i];
+	return NULL;
+}
+
+static int glnvg__deleteTexture(GLNVGcontext* gl, int id)
+{
+	int i;
+	for (i = 0; i < gl->ntextures; i++) {
+		if (gl->textures[i].id == id) {
+			if (gl->textures[i].tex != 0 && (gl->textures[i].flags & NVG_IMAGE_NODELETE) == 0)
+				glDeleteTextures(1, &gl->textures[i].tex);
+			memset(&gl->textures[i], 0, sizeof(gl->textures[i]));
+			return 1;
+		}
+	}
+	return 0;
+}
+
+static void glnvg__dumpShaderError(GLuint shader, const char* name, const char* type)
+{
+	char str[512+1];
+	int len = 0;
+	glGetShaderInfoLog(shader, 512, &len, str);
+	if (len > 512) len = 512;
+	str[len] = '\0';
+	printf("Shader %s/%s error:\n%s\n", name, type, str);
+}
+
+static void glnvg__dumpProgramError(GLuint prog, const char* name)
+{
+	char str[512+1];
+	int len = 0;
+	glGetProgramInfoLog(prog, 512, &len, str);
+	if (len > 512) len = 512;
+	str[len] = '\0';
+	printf("Program %s error:\n%s\n", name, str);
+}
+
+static void glnvg__checkError(GLNVGcontext* gl, const char* str)
+{
+	GLenum err;
+	if ((gl->flags & NVG_DEBUG) == 0) return;
+	err = glGetError();
+	if (err != GL_NO_ERROR) {
+		printf("Error %08x after %s\n", err, str);
+		return;
+	}
+}
+
+static int glnvg__createShader(GLNVGshader* shader, const char* name, const char* header, const char* opts, const char* vshader, const char* fshader)
+{
+	GLint status;
+	GLuint prog, vert, frag;
+	const char* str[3];
+	str[0] = header;
+	str[1] = opts != NULL ? opts : "";
+
+	memset(shader, 0, sizeof(*shader));
+
+	prog = glCreateProgram();
+	vert = glCreateShader(GL_VERTEX_SHADER);
+	frag = glCreateShader(GL_FRAGMENT_SHADER);
+	str[2] = vshader;
+	glShaderSource(vert, 3, str, 0);
+	str[2] = fshader;
+	glShaderSource(frag, 3, str, 0);
+
+	glCompileShader(vert);
+	glGetShaderiv(vert, GL_COMPILE_STATUS, &status);
+	if (status != GL_TRUE) {
+		glnvg__dumpShaderError(vert, name, "vert");
+		return 0;
+	}
+
+	glCompileShader(frag);
+	glGetShaderiv(frag, GL_COMPILE_STATUS, &status);
+	if (status != GL_TRUE) {
+		glnvg__dumpShaderError(frag, name, "frag");
+		return 0;
+	}
+
+	glAttachShader(prog, vert);
+	glAttachShader(prog, frag);
+
+	glBindAttribLocation(prog, 0, "vertex");
+	glBindAttribLocation(prog, 1, "tcoord");
+
+	glLinkProgram(prog);
+	glGetProgramiv(prog, GL_LINK_STATUS, &status);
+	if (status != GL_TRUE) {
+		glnvg__dumpProgramError(prog, name);
+		return 0;
+	}
+
+	shader->prog = prog;
+	shader->vert = vert;
+	shader->frag = frag;
+
+	return 1;
+}
+
+static void glnvg__deleteShader(GLNVGshader* shader)
+{
+	if (shader->prog != 0)
+		glDeleteProgram(shader->prog);
+	if (shader->vert != 0)
+		glDeleteShader(shader->vert);
+	if (shader->frag != 0)
+		glDeleteShader(shader->frag);
+}
+
+static void glnvg__getUniforms(GLNVGshader* shader)
+{
+	shader->loc[GLNVG_LOC_VIEWSIZE] = glGetUniformLocation(shader->prog, "viewSize");
+	shader->loc[GLNVG_LOC_TEX] = glGetUniformLocation(shader->prog, "tex");
+
+#if NANOVG_GL_USE_UNIFORMBUFFER
+	shader->loc[GLNVG_LOC_FRAG] = glGetUniformBlockIndex(shader->prog, "frag");
+#else
+	shader->loc[GLNVG_LOC_FRAG] = glGetUniformLocation(shader->prog, "frag");
+#endif
+}
+
+static int glnvg__renderCreate(void* uptr)
+{
+	GLNVGcontext* gl = (GLNVGcontext*)uptr;
+	int align = 4;
+
+	// TODO: mediump float may not be enough for GLES2 in iOS.
+	// see the following discussion: https://github.com/memononen/nanovg/issues/46
+	static const char* shaderHeader =
+#if defined NANOVG_GL2
+		"#define NANOVG_GL2 1\n"
+#elif defined NANOVG_GL3
+		"#version 150 core\n"
+		"#define NANOVG_GL3 1\n"
+#elif defined NANOVG_GLES2
+		"#version 100\n"
+		"#define NANOVG_GL2 1\n"
+#elif defined NANOVG_GLES3
+		"#version 300 es\n"
+		"#define NANOVG_GL3 1\n"
+#endif
+
+#if NANOVG_GL_USE_UNIFORMBUFFER
+	"#define USE_UNIFORMBUFFER 1\n"
+#else
+	"#define UNIFORMARRAY_SIZE 11\n"
+#endif
+	"\n";
+
+	static const char* fillVertShader =
+		"#ifdef NANOVG_GL3\n"
+		"	uniform vec2 viewSize;\n"
+		"	in vec2 vertex;\n"
+		"	in vec2 tcoord;\n"
+		"	out vec2 ftcoord;\n"
+		"	out vec2 fpos;\n"
+		"#else\n"
+		"	uniform vec2 viewSize;\n"
+		"	attribute vec2 vertex;\n"
+		"	attribute vec2 tcoord;\n"
+		"	varying vec2 ftcoord;\n"
+		"	varying vec2 fpos;\n"
+		"#endif\n"
+		"void main(void) {\n"
+		"	ftcoord = tcoord;\n"
+		"	fpos = vertex;\n"
+		"	gl_Position = vec4(2.0*vertex.x/viewSize.x - 1.0, 1.0 - 2.0*vertex.y/viewSize.y, 0, 1);\n"
+		"}\n";
+
+	static const char* fillFragShader = 
+		"#ifdef GL_ES\n"
+		"#if defined(GL_FRAGMENT_PRECISION_HIGH) || defined(NANOVG_GL3)\n"
+		" precision highp float;\n"
+		"#else\n"
+		" precision mediump float;\n"
+		"#endif\n"
+		"#endif\n"
+		"#ifdef NANOVG_GL3\n"
+		"#ifdef USE_UNIFORMBUFFER\n"
+		"	layout(std140) uniform frag {\n"
+		"		mat3 scissorMat;\n"
+		"		mat3 paintMat;\n"
+		"		vec4 innerCol;\n"
+		"		vec4 outerCol;\n"
+		"		vec2 scissorExt;\n"
+		"		vec2 scissorScale;\n"
+		"		vec2 extent;\n"
+		"		float radius;\n"
+		"		float feather;\n"
+		"		float strokeMult;\n"
+		"		float strokeThr;\n"
+		"		int texType;\n"
+		"		int type;\n"
+		"	};\n"
+		"#else\n" // NANOVG_GL3 && !USE_UNIFORMBUFFER
+		"	uniform vec4 frag[UNIFORMARRAY_SIZE];\n"
+		"#endif\n"
+		"	uniform sampler2D tex;\n"
+		"	in vec2 ftcoord;\n"
+		"	in vec2 fpos;\n"
+		"	out vec4 outColor;\n"
+		"#else\n" // !NANOVG_GL3
+		"	uniform vec4 frag[UNIFORMARRAY_SIZE];\n"
+		"	uniform sampler2D tex;\n"
+		"	varying vec2 ftcoord;\n"
+		"	varying vec2 fpos;\n"
+		"#endif\n"
+		"#ifndef USE_UNIFORMBUFFER\n"
+		"	#define scissorMat mat3(frag[0].xyz, frag[1].xyz, frag[2].xyz)\n"
+		"	#define paintMat mat3(frag[3].xyz, frag[4].xyz, frag[5].xyz)\n"
+		"	#define innerCol frag[6]\n"
+		"	#define outerCol frag[7]\n"
+		"	#define scissorExt frag[8].xy\n"
+		"	#define scissorScale frag[8].zw\n"
+		"	#define extent frag[9].xy\n"
+		"	#define radius frag[9].z\n"
+		"	#define feather frag[9].w\n"
+		"	#define strokeMult frag[10].x\n"
+		"	#define strokeThr frag[10].y\n"
+		"	#define texType int(frag[10].z)\n"
+		"	#define type int(frag[10].w)\n"
+		"#endif\n"
+		"\n"
+		"float sdroundrect(vec2 pt, vec2 ext, float rad) {\n"
+		"	vec2 ext2 = ext - vec2(rad,rad);\n"
+		"	vec2 d = abs(pt) - ext2;\n"
+		"	return min(max(d.x,d.y),0.0) + length(max(d,0.0)) - rad;\n"
+		"}\n"
+		"\n"
+		"// Scissoring\n"
+		"float scissorMask(vec2 p) {\n"
+		"	vec2 sc = (abs((scissorMat * vec3(p,1.0)).xy) - scissorExt);\n"
+		"	sc = vec2(0.5,0.5) - sc * scissorScale;\n"
+		"	return clamp(sc.x,0.0,1.0) * clamp(sc.y,0.0,1.0);\n"
+		"}\n"
+		"#ifdef EDGE_AA\n"
+		"// Stroke - from [0..1] to clipped pyramid, where the slope is 1px.\n"
+		"float strokeMask() {\n"
+		"	return min(1.0, (1.0-abs(ftcoord.x*2.0-1.0))*strokeMult) * min(1.0, ftcoord.y);\n"
+		"}\n"
+		"#endif\n"
+		"\n"
+		"void main(void) {\n"
+		"   vec4 result;\n"
+		"	float scissor = scissorMask(fpos);\n"
+		"#ifdef EDGE_AA\n"
+		"	float strokeAlpha = strokeMask();\n"
+		"#else\n"
+		"	float strokeAlpha = 1.0;\n"
+		"#endif\n"
+		"	if (type == 0) {			// Gradient\n"
+		"		// Calculate gradient color using box gradient\n"
+		"		vec2 pt = (paintMat * vec3(fpos,1.0)).xy;\n"
+		"		float d = clamp((sdroundrect(pt, extent, radius) + feather*0.5) / feather, 0.0, 1.0);\n"
+		"		vec4 color = mix(innerCol,outerCol,d);\n"
+		"		// Combine alpha\n"
+		"		color *= strokeAlpha * scissor;\n"
+		"		result = color;\n"
+		"	} else if (type == 1) {		// Image\n"
+		"		// Calculate color fron texture\n"
+		"		vec2 pt = (paintMat * vec3(fpos,1.0)).xy / extent;\n"
+		"#ifdef NANOVG_GL3\n"
+		"		vec4 color = texture(tex, pt);\n"
+		"#else\n"
+		"		vec4 color = texture2D(tex, pt);\n"
+		"#endif\n"
+		"		if (texType == 1) color = vec4(color.xyz*color.w,color.w);"
+		"		if (texType == 2) color = vec4(color.x);"
+		"		// Apply color tint and alpha.\n"
+		"		color *= innerCol;\n"
+		"		// Combine alpha\n"
+		"		color *= strokeAlpha * scissor;\n"
+		"		result = color;\n"
+		"	} else if (type == 2) {		// Stencil fill\n"
+		"		result = vec4(1,1,1,1);\n"
+		"	} else if (type == 3) {		// Textured tris\n"
+		"#ifdef NANOVG_GL3\n"
+		"		vec4 color = texture(tex, ftcoord);\n"
+		"#else\n"
+		"		vec4 color = texture2D(tex, ftcoord);\n"
+		"#endif\n"
+		"		if (texType == 1) color = vec4(color.xyz*color.w,color.w);"
+		"		if (texType == 2) color = vec4(color.x);"
+		"		color *= scissor;\n"
+		"		result = color * innerCol;\n"
+		"	}\n"
+		"#ifdef EDGE_AA\n"
+		"	if (strokeAlpha < strokeThr) discard;\n"
+		"#endif\n"
+		"#ifdef NANOVG_GL3\n"
+		"	outColor = result;\n"
+		"#else\n"
+		"	gl_FragColor = result;\n"
+		"#endif\n"
+		"}\n";
+
+	glnvg__checkError(gl, "init");
+
+	if (gl->flags & NVG_ANTIALIAS) {
+		if (glnvg__createShader(&gl->shader, "shader", shaderHeader, "#define EDGE_AA 1\n", fillVertShader, fillFragShader) == 0)
+			return 0;
+	} else {
+		if (glnvg__createShader(&gl->shader, "shader", shaderHeader, NULL, fillVertShader, fillFragShader) == 0)
+			return 0;
+	}
+
+	glnvg__checkError(gl, "uniform locations");
+	glnvg__getUniforms(&gl->shader);
+
+	// Create dynamic vertex array
+#if defined NANOVG_GL3
+	glGenVertexArrays(1, &gl->vertArr);
+#endif
+	glGenBuffers(1, &gl->vertBuf);
+
+#if NANOVG_GL_USE_UNIFORMBUFFER
+	// Create UBOs
+	glUniformBlockBinding(gl->shader.prog, gl->shader.loc[GLNVG_LOC_FRAG], GLNVG_FRAG_BINDING);
+	glGenBuffers(1, &gl->fragBuf); 
+	glGetIntegerv(GL_UNIFORM_BUFFER_OFFSET_ALIGNMENT, &align);
+#endif
+	gl->fragSize = sizeof(GLNVGfragUniforms) + align - sizeof(GLNVGfragUniforms) % align;
+
+	glnvg__checkError(gl, "create done");
+
+	glFinish();
+
+	return 1;
+}
+
+static int glnvg__renderCreateTexture(void* uptr, int type, int w, int h, int imageFlags, const unsigned char* data)
+{
+	GLNVGcontext* gl = (GLNVGcontext*)uptr;
+	GLNVGtexture* tex = glnvg__allocTexture(gl);
+
+	if (tex == NULL) return 0;
+
+#ifdef NANOVG_GLES2
+	// Check for non-power of 2.
+	if (glnvg__nearestPow2(w) != (unsigned int)w || glnvg__nearestPow2(h) != (unsigned int)h) {
+		// No repeat
+		if ((imageFlags & NVG_IMAGE_REPEATX) != 0 || (imageFlags & NVG_IMAGE_REPEATY) != 0) {
+			printf("Repeat X/Y is not supported for non power-of-two textures (%d x %d)\n", w, h);
+			imageFlags &= ~(NVG_IMAGE_REPEATX | NVG_IMAGE_REPEATY);
+		}
+		// No mips. 
+		if (imageFlags & NVG_IMAGE_GENERATE_MIPMAPS) {
+			printf("Mip-maps is not support for non power-of-two textures (%d x %d)\n", w, h);
+			imageFlags &= ~NVG_IMAGE_GENERATE_MIPMAPS;
+		}
+	}
+#endif
+
+	glGenTextures(1, &tex->tex);
+	tex->width = w;
+	tex->height = h;
+	tex->type = type;
+	tex->flags = imageFlags;
+	glnvg__bindTexture(gl, tex->tex);
+
+	glPixelStorei(GL_UNPACK_ALIGNMENT,1);
+#ifndef NANOVG_GLES2
+	glPixelStorei(GL_UNPACK_ROW_LENGTH, tex->width);
+	glPixelStorei(GL_UNPACK_SKIP_PIXELS, 0);
+	glPixelStorei(GL_UNPACK_SKIP_ROWS, 0);
+#endif
+
+#if defined (NANOVG_GL2)
+	// GL 1.4 and later has support for generating mipmaps using a tex parameter.
+	if (imageFlags & NVG_IMAGE_GENERATE_MIPMAPS) {
+		glTexParameteri(GL_TEXTURE_2D, GL_GENERATE_MIPMAP, GL_TRUE);
+	}
+#endif
+
+	if (type == NVG_TEXTURE_RGBA)
+		glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA, w, h, 0, GL_RGBA, GL_UNSIGNED_BYTE, data);
+	else
+#if defined(NANOVG_GLES2)
+		glTexImage2D(GL_TEXTURE_2D, 0, GL_LUMINANCE, w, h, 0, GL_LUMINANCE, GL_UNSIGNED_BYTE, data);
+#elif defined(NANOVG_GLES3)
+		glTexImage2D(GL_TEXTURE_2D, 0, GL_R8, w, h, 0, GL_RED, GL_UNSIGNED_BYTE, data);
+#else
+		glTexImage2D(GL_TEXTURE_2D, 0, GL_RED, w, h, 0, GL_RED, GL_UNSIGNED_BYTE, data);
+#endif
+
+	if (imageFlags & NVG_IMAGE_GENERATE_MIPMAPS) {
+		glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR_MIPMAP_LINEAR);
+	} else {
+		glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
+	}
+	glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
+
+	if (imageFlags & NVG_IMAGE_REPEATX)
+		glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_REPEAT);
+	else
+		glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
+
+	if (imageFlags & NVG_IMAGE_REPEATY)
+		glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_REPEAT);
+	else
+		glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
+
+	glPixelStorei(GL_UNPACK_ALIGNMENT, 4);
+#ifndef NANOVG_GLES2
+	glPixelStorei(GL_UNPACK_ROW_LENGTH, 0);
+	glPixelStorei(GL_UNPACK_SKIP_PIXELS, 0);
+	glPixelStorei(GL_UNPACK_SKIP_ROWS, 0);
+#endif
+
+	// The new way to build mipmaps on GLES and GL3
+#if !defined(NANOVG_GL2)
+	if (imageFlags & NVG_IMAGE_GENERATE_MIPMAPS) {
+		glGenerateMipmap(GL_TEXTURE_2D);
+	}
+#endif
+
+	glnvg__checkError(gl, "create tex");
+	glnvg__bindTexture(gl, 0);
+
+	return tex->id;
+}
+
+
+static int glnvg__renderDeleteTexture(void* uptr, int image)
+{
+	GLNVGcontext* gl = (GLNVGcontext*)uptr;
+	return glnvg__deleteTexture(gl, image);
+}
+
+static int glnvg__renderUpdateTexture(void* uptr, int image, int x, int y, int w, int h, const unsigned char* data)
+{
+	GLNVGcontext* gl = (GLNVGcontext*)uptr;
+	GLNVGtexture* tex = glnvg__findTexture(gl, image);
+
+	if (tex == NULL) return 0;
+	glnvg__bindTexture(gl, tex->tex);
+
+	glPixelStorei(GL_UNPACK_ALIGNMENT,1);
+
+#ifndef NANOVG_GLES2
+	glPixelStorei(GL_UNPACK_ROW_LENGTH, tex->width);
+	glPixelStorei(GL_UNPACK_SKIP_PIXELS, x);
+	glPixelStorei(GL_UNPACK_SKIP_ROWS, y);
+#else
+	// No support for all of skip, need to update a whole row at a time.
+	if (tex->type == NVG_TEXTURE_RGBA)
+		data += y*tex->width*4;
+	else
+		data += y*tex->width;
+	x = 0;
+	w = tex->width;
+#endif
+
+	if (tex->type == NVG_TEXTURE_RGBA)
+		glTexSubImage2D(GL_TEXTURE_2D, 0, x,y, w,h, GL_RGBA, GL_UNSIGNED_BYTE, data);
+	else
+#ifdef NANOVG_GLES2
+		glTexSubImage2D(GL_TEXTURE_2D, 0, x,y, w,h, GL_LUMINANCE, GL_UNSIGNED_BYTE, data);
+#else
+		glTexSubImage2D(GL_TEXTURE_2D, 0, x,y, w,h, GL_RED, GL_UNSIGNED_BYTE, data);
+#endif
+
+	glPixelStorei(GL_UNPACK_ALIGNMENT, 4);
+#ifndef NANOVG_GLES2
+	glPixelStorei(GL_UNPACK_ROW_LENGTH, 0);
+	glPixelStorei(GL_UNPACK_SKIP_PIXELS, 0);
+	glPixelStorei(GL_UNPACK_SKIP_ROWS, 0);
+#endif
+
+	glnvg__bindTexture(gl, 0);
+
+	return 1;
+}
+
+static int glnvg__renderGetTextureSize(void* uptr, int image, int* w, int* h)
+{
+	GLNVGcontext* gl = (GLNVGcontext*)uptr;
+	GLNVGtexture* tex = glnvg__findTexture(gl, image);
+	if (tex == NULL) return 0;
+	*w = tex->width;
+	*h = tex->height;
+	return 1;
+}
+
+static void glnvg__xformToMat3x4(float* m3, float* t)
+{
+	m3[0] = t[0];
+	m3[1] = t[1];
+	m3[2] = 0.0f;
+	m3[3] = 0.0f;
+	m3[4] = t[2];
+	m3[5] = t[3];
+	m3[6] = 0.0f;
+	m3[7] = 0.0f;
+	m3[8] = t[4];
+	m3[9] = t[5];
+	m3[10] = 1.0f;
+	m3[11] = 0.0f;
+}
+
+static NVGcolor glnvg__premulColor(NVGcolor c)
+{
+	c.r *= c.a;
+	c.g *= c.a;
+	c.b *= c.a;
+	return c;
+}
+
+static int glnvg__convertPaint(GLNVGcontext* gl, GLNVGfragUniforms* frag, NVGpaint* paint,
+							   NVGscissor* scissor, float width, float fringe, float strokeThr)
+{
+	GLNVGtexture* tex = NULL;
+	float invxform[6];
+
+	memset(frag, 0, sizeof(*frag));
+
+	frag->innerCol = glnvg__premulColor(paint->innerColor);
+	frag->outerCol = glnvg__premulColor(paint->outerColor);
+
+	if (scissor->extent[0] < -0.5f || scissor->extent[1] < -0.5f) {
+		memset(frag->scissorMat, 0, sizeof(frag->scissorMat));
+		frag->scissorExt[0] = 1.0f;
+		frag->scissorExt[1] = 1.0f;
+		frag->scissorScale[0] = 1.0f;
+		frag->scissorScale[1] = 1.0f;
+	} else {
+		nvgTransformInverse(invxform, scissor->xform);
+		glnvg__xformToMat3x4(frag->scissorMat, invxform);
+		frag->scissorExt[0] = scissor->extent[0];
+		frag->scissorExt[1] = scissor->extent[1];
+		frag->scissorScale[0] = sqrtf(scissor->xform[0]*scissor->xform[0] + scissor->xform[2]*scissor->xform[2]) / fringe;
+		frag->scissorScale[1] = sqrtf(scissor->xform[1]*scissor->xform[1] + scissor->xform[3]*scissor->xform[3]) / fringe;
+	}
+
+	memcpy(frag->extent, paint->extent, sizeof(frag->extent));
+	frag->strokeMult = (width*0.5f + fringe*0.5f) / fringe;
+	frag->strokeThr = strokeThr;
+
+	if (paint->image != 0) {
+		tex = glnvg__findTexture(gl, paint->image);
+		if (tex == NULL) return 0;
+		if ((tex->flags & NVG_IMAGE_FLIPY) != 0) {
+			float flipped[6];
+			nvgTransformScale(flipped, 1.0f, -1.0f);
+			nvgTransformMultiply(flipped, paint->xform);
+			nvgTransformInverse(invxform, flipped);
+		} else {
+			nvgTransformInverse(invxform, paint->xform);
+		}
+		frag->type = NSVG_SHADER_FILLIMG;
+
+		if (tex->type == NVG_TEXTURE_RGBA)
+			frag->texType = (tex->flags & NVG_IMAGE_PREMULTIPLIED) ? 0 : 1;
+		else
+			frag->texType = 2;
+//		printf("frag->texType = %d\n", frag->texType);
+	} else {
+		frag->type = NSVG_SHADER_FILLGRAD;
+		frag->radius = paint->radius;
+		frag->feather = paint->feather;
+		nvgTransformInverse(invxform, paint->xform);
+	}
+
+	glnvg__xformToMat3x4(frag->paintMat, invxform);
+
+	return 1;
+}
+
+static GLNVGfragUniforms* nvg__fragUniformPtr(GLNVGcontext* gl, int i);
+
+static void glnvg__setUniforms(GLNVGcontext* gl, int uniformOffset, int image)
+{
+#if NANOVG_GL_USE_UNIFORMBUFFER
+	glBindBufferRange(GL_UNIFORM_BUFFER, GLNVG_FRAG_BINDING, gl->fragBuf, uniformOffset, sizeof(GLNVGfragUniforms));
+#else
+	GLNVGfragUniforms* frag = nvg__fragUniformPtr(gl, uniformOffset);
+	glUniform4fv(gl->shader.loc[GLNVG_LOC_FRAG], NANOVG_GL_UNIFORMARRAY_SIZE, &(frag->uniformArray[0][0]));
+#endif
+
+	if (image != 0) {
+		GLNVGtexture* tex = glnvg__findTexture(gl, image);
+		glnvg__bindTexture(gl, tex != NULL ? tex->tex : 0);
+		glnvg__checkError(gl, "tex paint tex");
+	} else {
+		glnvg__bindTexture(gl, 0);
+	}
+}
+
+static void glnvg__renderViewport(void* uptr, int width, int height)
+{
+	GLNVGcontext* gl = (GLNVGcontext*)uptr;
+	gl->view[0] = (float)width;
+	gl->view[1] = (float)height;
+}
+
+static void glnvg__fill(GLNVGcontext* gl, GLNVGcall* call)
+{
+	GLNVGpath* paths = &gl->paths[call->pathOffset];
+	int i, npaths = call->pathCount;
+
+	// Draw shapes
+	glEnable(GL_STENCIL_TEST);
+	glnvg__stencilMask(gl, 0xff);
+	glnvg__stencilFunc(gl, GL_ALWAYS, 0, 0xff);
+	glColorMask(GL_FALSE, GL_FALSE, GL_FALSE, GL_FALSE);
+
+	// set bindpoint for solid loc
+	glnvg__setUniforms(gl, call->uniformOffset, 0);
+	glnvg__checkError(gl, "fill simple");
+
+	glStencilOpSeparate(GL_FRONT, GL_KEEP, GL_KEEP, GL_INCR_WRAP);
+	glStencilOpSeparate(GL_BACK, GL_KEEP, GL_KEEP, GL_DECR_WRAP);
+	glDisable(GL_CULL_FACE);
+	for (i = 0; i < npaths; i++)
+		glDrawArrays(GL_TRIANGLE_FAN, paths[i].fillOffset, paths[i].fillCount);
+	glEnable(GL_CULL_FACE);
+
+	// Draw anti-aliased pixels
+	glColorMask(GL_TRUE, GL_TRUE, GL_TRUE, GL_TRUE);
+
+	glnvg__setUniforms(gl, call->uniformOffset + gl->fragSize, call->image);
+	glnvg__checkError(gl, "fill fill");
+
+	if (gl->flags & NVG_ANTIALIAS) {
+		glnvg__stencilFunc(gl, GL_EQUAL, 0x00, 0xff);
+		glStencilOp(GL_KEEP, GL_KEEP, GL_KEEP);
+		// Draw fringes
+		for (i = 0; i < npaths; i++)
+			glDrawArrays(GL_TRIANGLE_STRIP, paths[i].strokeOffset, paths[i].strokeCount);
+	}
+
+	// Draw fill
+	glnvg__stencilFunc(gl, GL_NOTEQUAL, 0x0, 0xff);
+	glStencilOp(GL_ZERO, GL_ZERO, GL_ZERO);
+	glDrawArrays(GL_TRIANGLES, call->triangleOffset, call->triangleCount);
+
+	glDisable(GL_STENCIL_TEST);
+}
+
+static void glnvg__convexFill(GLNVGcontext* gl, GLNVGcall* call)
+{
+	GLNVGpath* paths = &gl->paths[call->pathOffset];
+	int i, npaths = call->pathCount;
+
+	glnvg__setUniforms(gl, call->uniformOffset, call->image);
+	glnvg__checkError(gl, "convex fill");
+
+	for (i = 0; i < npaths; i++)
+		glDrawArrays(GL_TRIANGLE_FAN, paths[i].fillOffset, paths[i].fillCount);
+	if (gl->flags & NVG_ANTIALIAS) {
+		// Draw fringes
+		for (i = 0; i < npaths; i++)
+			glDrawArrays(GL_TRIANGLE_STRIP, paths[i].strokeOffset, paths[i].strokeCount);
+	}
+}
+
+static void glnvg__stroke(GLNVGcontext* gl, GLNVGcall* call)
+{
+	GLNVGpath* paths = &gl->paths[call->pathOffset];
+	int npaths = call->pathCount, i;
+
+	if (gl->flags & NVG_STENCIL_STROKES) {
+
+		glEnable(GL_STENCIL_TEST);
+		glnvg__stencilMask(gl, 0xff);
+
+		// Fill the stroke base without overlap
+		glnvg__stencilFunc(gl, GL_EQUAL, 0x0, 0xff);
+		glStencilOp(GL_KEEP, GL_KEEP, GL_INCR);
+		glnvg__setUniforms(gl, call->uniformOffset + gl->fragSize, call->image);
+		glnvg__checkError(gl, "stroke fill 0");
+		for (i = 0; i < npaths; i++)
+			glDrawArrays(GL_TRIANGLE_STRIP, paths[i].strokeOffset, paths[i].strokeCount);
+
+		// Draw anti-aliased pixels.
+		glnvg__setUniforms(gl, call->uniformOffset, call->image);
+		glnvg__stencilFunc(gl, GL_EQUAL, 0x00, 0xff);
+		glStencilOp(GL_KEEP, GL_KEEP, GL_KEEP);
+		for (i = 0; i < npaths; i++)
+			glDrawArrays(GL_TRIANGLE_STRIP, paths[i].strokeOffset, paths[i].strokeCount);
+
+		// Clear stencil buffer.		
+		glColorMask(GL_FALSE, GL_FALSE, GL_FALSE, GL_FALSE);
+		glnvg__stencilFunc(gl, GL_ALWAYS, 0x0, 0xff);
+		glStencilOp(GL_ZERO, GL_ZERO, GL_ZERO);
+		glnvg__checkError(gl, "stroke fill 1");
+		for (i = 0; i < npaths; i++)
+			glDrawArrays(GL_TRIANGLE_STRIP, paths[i].strokeOffset, paths[i].strokeCount);
+		glColorMask(GL_TRUE, GL_TRUE, GL_TRUE, GL_TRUE);
+
+		glDisable(GL_STENCIL_TEST);
+
+//		glnvg__convertPaint(gl, nvg__fragUniformPtr(gl, call->uniformOffset + gl->fragSize), paint, scissor, strokeWidth, fringe, 1.0f - 0.5f/255.0f);
+
+	} else {
+		glnvg__setUniforms(gl, call->uniformOffset, call->image);
+		glnvg__checkError(gl, "stroke fill");
+		// Draw Strokes
+		for (i = 0; i < npaths; i++)
+			glDrawArrays(GL_TRIANGLE_STRIP, paths[i].strokeOffset, paths[i].strokeCount);
+	}
+}
+
+static void glnvg__triangles(GLNVGcontext* gl, GLNVGcall* call)
+{
+	glnvg__setUniforms(gl, call->uniformOffset, call->image);
+	glnvg__checkError(gl, "triangles fill");
+
+	glDrawArrays(GL_TRIANGLES, call->triangleOffset, call->triangleCount);
+}
+
+static void glnvg__renderCancel(void* uptr) {
+	GLNVGcontext* gl = (GLNVGcontext*)uptr;
+	gl->nverts = 0;
+	gl->npaths = 0;
+	gl->ncalls = 0;
+	gl->nuniforms = 0;
+}
+
+static void glnvg__renderFlush(void* uptr)
+{
+	GLNVGcontext* gl = (GLNVGcontext*)uptr;
+	int i;
+
+	if (gl->ncalls > 0) {
+
+		// Setup require GL state.
+		glUseProgram(gl->shader.prog);
+
+		glBlendFunc(GL_ONE, GL_ONE_MINUS_SRC_ALPHA);
+		glEnable(GL_CULL_FACE);
+		glCullFace(GL_BACK);
+		glFrontFace(GL_CCW);
+		glEnable(GL_BLEND);
+		glDisable(GL_DEPTH_TEST);
+		glDisable(GL_SCISSOR_TEST);
+		glColorMask(GL_TRUE, GL_TRUE, GL_TRUE, GL_TRUE);
+		glStencilMask(0xffffffff);
+		glStencilOp(GL_KEEP, GL_KEEP, GL_KEEP);
+		glStencilFunc(GL_ALWAYS, 0, 0xffffffff);
+		glActiveTexture(GL_TEXTURE0);
+		glBindTexture(GL_TEXTURE_2D, 0);
+		#if NANOVG_GL_USE_STATE_FILTER
+		gl->boundTexture = 0;
+		gl->stencilMask = 0xffffffff;
+		gl->stencilFunc = GL_ALWAYS;
+		gl->stencilFuncRef = 0;
+		gl->stencilFuncMask = 0xffffffff;
+		#endif
+
+#if NANOVG_GL_USE_UNIFORMBUFFER
+		// Upload ubo for frag shaders
+		glBindBuffer(GL_UNIFORM_BUFFER, gl->fragBuf);
+		glBufferData(GL_UNIFORM_BUFFER, gl->nuniforms * gl->fragSize, gl->uniforms, GL_STREAM_DRAW);
+#endif
+
+		// Upload vertex data
+#if defined NANOVG_GL3
+		glBindVertexArray(gl->vertArr);
+#endif
+		glBindBuffer(GL_ARRAY_BUFFER, gl->vertBuf);
+		glBufferData(GL_ARRAY_BUFFER, gl->nverts * sizeof(NVGvertex), gl->verts, GL_STREAM_DRAW);
+		glEnableVertexAttribArray(0);
+		glEnableVertexAttribArray(1);
+		glVertexAttribPointer(0, 2, GL_FLOAT, GL_FALSE, sizeof(NVGvertex), (const GLvoid*)(size_t)0);
+		glVertexAttribPointer(1, 2, GL_FLOAT, GL_FALSE, sizeof(NVGvertex), (const GLvoid*)(0 + 2*sizeof(float)));
+
+		// Set view and texture just once per frame.
+		glUniform1i(gl->shader.loc[GLNVG_LOC_TEX], 0);
+		glUniform2fv(gl->shader.loc[GLNVG_LOC_VIEWSIZE], 1, gl->view);
+
+#if NANOVG_GL_USE_UNIFORMBUFFER
+		glBindBuffer(GL_UNIFORM_BUFFER, gl->fragBuf);
+#endif
+
+		for (i = 0; i < gl->ncalls; i++) {
+			GLNVGcall* call = &gl->calls[i];
+			if (call->type == GLNVG_FILL)
+				glnvg__fill(gl, call);
+			else if (call->type == GLNVG_CONVEXFILL)
+				glnvg__convexFill(gl, call);
+			else if (call->type == GLNVG_STROKE)
+				glnvg__stroke(gl, call);
+			else if (call->type == GLNVG_TRIANGLES)
+				glnvg__triangles(gl, call);
+		}
+
+		glDisableVertexAttribArray(0);
+		glDisableVertexAttribArray(1);
+#if defined NANOVG_GL3
+		glBindVertexArray(0);
+#endif	
+		glDisable(GL_CULL_FACE);
+			glBindBuffer(GL_ARRAY_BUFFER, 0);
+		glUseProgram(0);
+		glnvg__bindTexture(gl, 0);
+	}
+
+	// Reset calls
+	gl->nverts = 0;
+	gl->npaths = 0;
+	gl->ncalls = 0;
+	gl->nuniforms = 0;
+}
+
+static int glnvg__maxVertCount(const NVGpath* paths, int npaths)
+{
+	int i, count = 0;
+	for (i = 0; i < npaths; i++) {
+		count += paths[i].nfill;
+		count += paths[i].nstroke;
+	}
+	return count;
+}
+
+static GLNVGcall* glnvg__allocCall(GLNVGcontext* gl)
+{
+	GLNVGcall* ret = NULL;
+	if (gl->ncalls+1 > gl->ccalls) {
+		GLNVGcall* calls;
+		int ccalls = glnvg__maxi(gl->ncalls+1, 128) + gl->ccalls/2; // 1.5x Overallocate
+		calls = (GLNVGcall*)realloc(gl->calls, sizeof(GLNVGcall) * ccalls);
+		if (calls == NULL) return NULL;
+		gl->calls = calls;
+		gl->ccalls = ccalls;
+	}
+	ret = &gl->calls[gl->ncalls++];
+	memset(ret, 0, sizeof(GLNVGcall));
+	return ret;
+}
+
+static int glnvg__allocPaths(GLNVGcontext* gl, int n)
+{
+	int ret = 0;
+	if (gl->npaths+n > gl->cpaths) {
+		GLNVGpath* paths;
+		int cpaths = glnvg__maxi(gl->npaths + n, 128) + gl->cpaths/2; // 1.5x Overallocate
+		paths = (GLNVGpath*)realloc(gl->paths, sizeof(GLNVGpath) * cpaths);
+		if (paths == NULL) return -1;
+		gl->paths = paths;
+		gl->cpaths = cpaths;
+	}
+	ret = gl->npaths;
+	gl->npaths += n;
+	return ret;
+}
+
+static int glnvg__allocVerts(GLNVGcontext* gl, int n)
+{
+	int ret = 0;
+	if (gl->nverts+n > gl->cverts) {
+		NVGvertex* verts;
+		int cverts = glnvg__maxi(gl->nverts + n, 4096) + gl->cverts/2; // 1.5x Overallocate
+		verts = (NVGvertex*)realloc(gl->verts, sizeof(NVGvertex) * cverts);
+		if (verts == NULL) return -1;
+		gl->verts = verts;
+		gl->cverts = cverts;
+	}
+	ret = gl->nverts;
+	gl->nverts += n;
+	return ret;
+}
+
+static int glnvg__allocFragUniforms(GLNVGcontext* gl, int n)
+{
+	int ret = 0, structSize = gl->fragSize;
+	if (gl->nuniforms+n > gl->cuniforms) {
+		unsigned char* uniforms;
+		int cuniforms = glnvg__maxi(gl->nuniforms+n, 128) + gl->cuniforms/2; // 1.5x Overallocate
+		uniforms = (unsigned char*)realloc(gl->uniforms, structSize * cuniforms);
+		if (uniforms == NULL) return -1;
+		gl->uniforms = uniforms;
+		gl->cuniforms = cuniforms;
+	}
+	ret = gl->nuniforms * structSize;
+	gl->nuniforms += n;
+	return ret;
+}
+
+static GLNVGfragUniforms* nvg__fragUniformPtr(GLNVGcontext* gl, int i)
+{
+	return (GLNVGfragUniforms*)&gl->uniforms[i];
+}
+
+static void glnvg__vset(NVGvertex* vtx, float x, float y, float u, float v)
+{
+	vtx->x = x;
+	vtx->y = y;
+	vtx->u = u;
+	vtx->v = v;
+}
+
+static void glnvg__renderFill(void* uptr, NVGpaint* paint, NVGscissor* scissor, float fringe,
+							  const float* bounds, const NVGpath* paths, int npaths)
+{
+	GLNVGcontext* gl = (GLNVGcontext*)uptr;
+	GLNVGcall* call = glnvg__allocCall(gl);
+	NVGvertex* quad;
+	GLNVGfragUniforms* frag;
+	int i, maxverts, offset;
+
+	if (call == NULL) return;
+
+	call->type = GLNVG_FILL;
+	call->pathOffset = glnvg__allocPaths(gl, npaths);
+	if (call->pathOffset == -1) goto error;
+	call->pathCount = npaths;
+	call->image = paint->image;
+
+	if (npaths == 1 && paths[0].convex)
+		call->type = GLNVG_CONVEXFILL;
+
+	// Allocate vertices for all the paths.
+	maxverts = glnvg__maxVertCount(paths, npaths) + 6;
+	offset = glnvg__allocVerts(gl, maxverts);
+	if (offset == -1) goto error;
+
+	for (i = 0; i < npaths; i++) {
+		GLNVGpath* copy = &gl->paths[call->pathOffset + i];
+		const NVGpath* path = &paths[i];
+		memset(copy, 0, sizeof(GLNVGpath));
+		if (path->nfill > 0) {
+			copy->fillOffset = offset;
+			copy->fillCount = path->nfill;
+			memcpy(&gl->verts[offset], path->fill, sizeof(NVGvertex) * path->nfill);
+			offset += path->nfill;
+		}
+		if (path->nstroke > 0) {
+			copy->strokeOffset = offset;
+			copy->strokeCount = path->nstroke;
+			memcpy(&gl->verts[offset], path->stroke, sizeof(NVGvertex) * path->nstroke);
+			offset += path->nstroke;
+		}
+	}
+
+	// Quad
+	call->triangleOffset = offset;
+	call->triangleCount = 6;
+	quad = &gl->verts[call->triangleOffset];
+	glnvg__vset(&quad[0], bounds[0], bounds[3], 0.5f, 1.0f);
+	glnvg__vset(&quad[1], bounds[2], bounds[3], 0.5f, 1.0f);
+	glnvg__vset(&quad[2], bounds[2], bounds[1], 0.5f, 1.0f);
+
+	glnvg__vset(&quad[3], bounds[0], bounds[3], 0.5f, 1.0f);
+	glnvg__vset(&quad[4], bounds[2], bounds[1], 0.5f, 1.0f);
+	glnvg__vset(&quad[5], bounds[0], bounds[1], 0.5f, 1.0f);
+
+	// Setup uniforms for draw calls
+	if (call->type == GLNVG_FILL) {
+		call->uniformOffset = glnvg__allocFragUniforms(gl, 2);
+		if (call->uniformOffset == -1) goto error;
+		// Simple shader for stencil
+		frag = nvg__fragUniformPtr(gl, call->uniformOffset);
+		memset(frag, 0, sizeof(*frag));
+		frag->strokeThr = -1.0f;
+		frag->type = NSVG_SHADER_SIMPLE;
+		// Fill shader
+		glnvg__convertPaint(gl, nvg__fragUniformPtr(gl, call->uniformOffset + gl->fragSize), paint, scissor, fringe, fringe, -1.0f);
+	} else {
+		call->uniformOffset = glnvg__allocFragUniforms(gl, 1);
+		if (call->uniformOffset == -1) goto error;
+		// Fill shader
+		glnvg__convertPaint(gl, nvg__fragUniformPtr(gl, call->uniformOffset), paint, scissor, fringe, fringe, -1.0f);
+	}
+
+	return;
+
+error:
+	// We get here if call alloc was ok, but something else is not.
+	// Roll back the last call to prevent drawing it.
+	if (gl->ncalls > 0) gl->ncalls--;
+}
+
+static void glnvg__renderStroke(void* uptr, NVGpaint* paint, NVGscissor* scissor, float fringe,
+								float strokeWidth, const NVGpath* paths, int npaths)
+{
+	GLNVGcontext* gl = (GLNVGcontext*)uptr;
+	GLNVGcall* call = glnvg__allocCall(gl);
+	int i, maxverts, offset;
+
+	if (call == NULL) return;
+
+	call->type = GLNVG_STROKE;
+	call->pathOffset = glnvg__allocPaths(gl, npaths);
+	if (call->pathOffset == -1) goto error;
+	call->pathCount = npaths;
+	call->image = paint->image;
+
+	// Allocate vertices for all the paths.
+	maxverts = glnvg__maxVertCount(paths, npaths);
+	offset = glnvg__allocVerts(gl, maxverts);
+	if (offset == -1) goto error;
+
+	for (i = 0; i < npaths; i++) {
+		GLNVGpath* copy = &gl->paths[call->pathOffset + i];
+		const NVGpath* path = &paths[i];
+		memset(copy, 0, sizeof(GLNVGpath));
+		if (path->nstroke) {
+			copy->strokeOffset = offset;
+			copy->strokeCount = path->nstroke;
+			memcpy(&gl->verts[offset], path->stroke, sizeof(NVGvertex) * path->nstroke);
+			offset += path->nstroke;
+		}
+	}
+
+	if (gl->flags & NVG_STENCIL_STROKES) {
+		// Fill shader
+		call->uniformOffset = glnvg__allocFragUniforms(gl, 2);
+		if (call->uniformOffset == -1) goto error;
+
+		glnvg__convertPaint(gl, nvg__fragUniformPtr(gl, call->uniformOffset), paint, scissor, strokeWidth, fringe, -1.0f);
+		glnvg__convertPaint(gl, nvg__fragUniformPtr(gl, call->uniformOffset + gl->fragSize), paint, scissor, strokeWidth, fringe, 1.0f - 0.5f/255.0f);
+
+	} else {
+		// Fill shader
+		call->uniformOffset = glnvg__allocFragUniforms(gl, 1);
+		if (call->uniformOffset == -1) goto error;
+		glnvg__convertPaint(gl, nvg__fragUniformPtr(gl, call->uniformOffset), paint, scissor, strokeWidth, fringe, -1.0f);
+	}
+
+	return;
+
+error:
+	// We get here if call alloc was ok, but something else is not.
+	// Roll back the last call to prevent drawing it.
+	if (gl->ncalls > 0) gl->ncalls--;
+}
+
+static void glnvg__renderTriangles(void* uptr, NVGpaint* paint, NVGscissor* scissor,
+								   const NVGvertex* verts, int nverts)
+{
+	GLNVGcontext* gl = (GLNVGcontext*)uptr;
+	GLNVGcall* call = glnvg__allocCall(gl);
+	GLNVGfragUniforms* frag;
+
+	if (call == NULL) return;
+
+	call->type = GLNVG_TRIANGLES;
+	call->image = paint->image;
+
+	// Allocate vertices for all the paths.
+	call->triangleOffset = glnvg__allocVerts(gl, nverts);
+	if (call->triangleOffset == -1) goto error;
+	call->triangleCount = nverts;
+
+	memcpy(&gl->verts[call->triangleOffset], verts, sizeof(NVGvertex) * nverts);
+
+	// Fill shader
+	call->uniformOffset = glnvg__allocFragUniforms(gl, 1);
+	if (call->uniformOffset == -1) goto error;
+	frag = nvg__fragUniformPtr(gl, call->uniformOffset);
+	glnvg__convertPaint(gl, frag, paint, scissor, 1.0f, 1.0f, -1.0f);
+	frag->type = NSVG_SHADER_IMG;
+
+	return;
+
+error:
+	// We get here if call alloc was ok, but something else is not.
+	// Roll back the last call to prevent drawing it.
+	if (gl->ncalls > 0) gl->ncalls--;
+}
+
+static void glnvg__renderDelete(void* uptr)
+{
+	GLNVGcontext* gl = (GLNVGcontext*)uptr;
+	int i;
+	if (gl == NULL) return;
+
+	glnvg__deleteShader(&gl->shader);
+
+#if NANOVG_GL3
+#if NANOVG_GL_USE_UNIFORMBUFFER
+	if (gl->fragBuf != 0)
+		glDeleteBuffers(1, &gl->fragBuf);
+#endif
+	if (gl->vertArr != 0)
+		glDeleteVertexArrays(1, &gl->vertArr);
+#endif
+	if (gl->vertBuf != 0)
+		glDeleteBuffers(1, &gl->vertBuf);
+
+	for (i = 0; i < gl->ntextures; i++) {
+		if (gl->textures[i].tex != 0 && (gl->textures[i].flags & NVG_IMAGE_NODELETE) == 0)
+			glDeleteTextures(1, &gl->textures[i].tex);
+	}
+	free(gl->textures);
+
+	free(gl->paths);
+	free(gl->verts);
+	free(gl->uniforms);
+	free(gl->calls);
+
+	free(gl);
+}
+
+
+#if defined NANOVG_GL2
+NVGcontext* nvgCreateGL2(int flags)
+#elif defined NANOVG_GL3
+NVGcontext* nvgCreateGL3(int flags)
+#elif defined NANOVG_GLES2
+NVGcontext* nvgCreateGLES2(int flags)
+#elif defined NANOVG_GLES3
+NVGcontext* nvgCreateGLES3(int flags)
+#endif
+{
+	NVGparams params;
+	NVGcontext* ctx = NULL;
+	GLNVGcontext* gl = (GLNVGcontext*)malloc(sizeof(GLNVGcontext));
+	if (gl == NULL) goto error;
+	memset(gl, 0, sizeof(GLNVGcontext));
+
+	memset(&params, 0, sizeof(params));
+	params.renderCreate = glnvg__renderCreate;
+	params.renderCreateTexture = glnvg__renderCreateTexture;
+	params.renderDeleteTexture = glnvg__renderDeleteTexture;
+	params.renderUpdateTexture = glnvg__renderUpdateTexture;
+	params.renderGetTextureSize = glnvg__renderGetTextureSize;
+	params.renderViewport = glnvg__renderViewport;
+	params.renderCancel = glnvg__renderCancel;
+	params.renderFlush = glnvg__renderFlush;
+	params.renderFill = glnvg__renderFill;
+	params.renderStroke = glnvg__renderStroke;
+	params.renderTriangles = glnvg__renderTriangles;
+	params.renderDelete = glnvg__renderDelete;
+	params.userPtr = gl;
+	params.edgeAntiAlias = flags & NVG_ANTIALIAS ? 1 : 0;
+
+	gl->flags = flags;
+
+	ctx = nvgCreateInternal(&params);
+	if (ctx == NULL) goto error;
+
+	return ctx;
+
+error:
+	// 'gl' is freed by nvgDeleteInternal.
+	if (ctx != NULL) nvgDeleteInternal(ctx);
+	return NULL;
+}
+
+#if defined NANOVG_GL2
+void nvgDeleteGL2(NVGcontext* ctx)
+#elif defined NANOVG_GL3
+void nvgDeleteGL3(NVGcontext* ctx)
+#elif defined NANOVG_GLES2
+void nvgDeleteGLES2(NVGcontext* ctx)
+#elif defined NANOVG_GLES3
+void nvgDeleteGLES3(NVGcontext* ctx)
+#endif
+{
+	nvgDeleteInternal(ctx);
+}
+
+int nvglCreateImageFromHandle(NVGcontext* ctx, GLuint textureId, int w, int h, int imageFlags)
+{
+	GLNVGcontext* gl = (GLNVGcontext*)nvgInternalParams(ctx)->userPtr;
+	GLNVGtexture* tex = glnvg__allocTexture(gl);
+
+	if (tex == NULL) return 0;
+
+	tex->type = NVG_TEXTURE_RGBA;
+	tex->tex = textureId;
+	tex->flags = imageFlags;
+	tex->width = w;
+	tex->height = h;
+
+	return tex->id;
+}
+
+GLuint nvglImageHandle(NVGcontext* ctx, int image)
+{
+	GLNVGcontext* gl = (GLNVGcontext*)nvgInternalParams(ctx)->userPtr;
+	GLNVGtexture* tex = glnvg__findTexture(gl, image);
+	return tex->tex;
+}
+
+#endif /* NANOVG_GL_IMPLEMENTATION */
diff --git a/dgl/src/nanovg2/nanovg_gl_utils.h b/dgl/src/nanovg2/nanovg_gl_utils.h
new file mode 100644
index 00000000..89ae4811
--- /dev/null
+++ b/dgl/src/nanovg2/nanovg_gl_utils.h
@@ -0,0 +1,132 @@
+//
+// Copyright (c) 2009-2013 Mikko Mononen memon@inside.org
+//
+// This software is provided 'as-is', without any express or implied
+// warranty.  In no event will the authors be held liable for any damages
+// arising from the use of this software.
+// Permission is granted to anyone to use this software for any purpose,
+// including commercial applications, and to alter it and redistribute it
+// freely, subject to the following restrictions:
+// 1. The origin of this software must not be misrepresented; you must not
+//    claim that you wrote the original software. If you use this software
+//    in a product, an acknowledgment in the product documentation would be
+//    appreciated but is not required.
+// 2. Altered source versions must be plainly marked as such, and must not be
+//    misrepresented as being the original software.
+// 3. This notice may not be removed or altered from any source distribution.
+//
+#ifndef NANOVG_GL_UTILS_H
+#define NANOVG_GL_UTILS_H
+
+struct NVGLUframebuffer {
+	NVGcontext* ctx;
+	GLuint fbo;
+	GLuint rbo;
+	GLuint texture;
+	int image;
+};
+typedef struct NVGLUframebuffer NVGLUframebuffer;
+
+// Helper function to create GL frame buffer to render to.
+void nvgluBindFramebuffer(NVGLUframebuffer* fb);
+NVGLUframebuffer* nvgluCreateFramebuffer(NVGcontext* ctx, int w, int h, int imageFlags);
+void nvgluDeleteFramebuffer(NVGcontext* ctx, NVGLUframebuffer* fb);
+
+#endif // NANOVG_GL_UTILS_H
+
+#ifdef NANOVG_GL_IMPLEMENTATION
+
+#if defined(NANOVG_GL3) || defined(NANOVG_GLES2) || defined(NANOVG_GLES3)
+// FBO is core in OpenGL 3>.
+#	define NANOVG_FBO_VALID 1
+#elif defined(NANOVG_GL2)
+// On OS X including glext defines FBO on GL2 too.
+#	ifdef __APPLE__
+#		include <OpenGL/glext.h>
+#		define NANOVG_FBO_VALID 1
+#	endif
+#endif
+
+static GLint defaultFBO = -1;
+
+NVGLUframebuffer* nvgluCreateFramebuffer(NVGcontext* ctx, int w, int h, int imageFlags)
+{
+#ifdef NANOVG_FBO_VALID
+	GLint defaultFBO;
+	GLint defaultRBO;
+	NVGLUframebuffer* fb = NULL;
+
+	glGetIntegerv(GL_FRAMEBUFFER_BINDING, &defaultFBO);
+	glGetIntegerv(GL_RENDERBUFFER_BINDING, &defaultRBO);
+
+	fb = (NVGLUframebuffer*)malloc(sizeof(NVGLUframebuffer));
+	if (fb == NULL) goto error;
+	memset(fb, 0, sizeof(NVGLUframebuffer));
+
+	fb->image = nvgCreateImageRGBA(ctx, w, h, imageFlags | NVG_IMAGE_FLIPY | NVG_IMAGE_PREMULTIPLIED, NULL);
+	fb->texture = nvglImageHandle(ctx, fb->image);
+
+	// frame buffer object
+	glGenFramebuffers(1, &fb->fbo);
+	glBindFramebuffer(GL_FRAMEBUFFER, fb->fbo);
+
+	// render buffer object
+	glGenRenderbuffers(1, &fb->rbo);
+	glBindRenderbuffer(GL_RENDERBUFFER, fb->rbo);
+	glRenderbufferStorage(GL_RENDERBUFFER, GL_STENCIL_INDEX8, w, h);
+
+	// combine all
+	glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, fb->texture, 0);
+	glFramebufferRenderbuffer(GL_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, GL_RENDERBUFFER, fb->rbo);
+
+	if (glCheckFramebufferStatus(GL_FRAMEBUFFER) != GL_FRAMEBUFFER_COMPLETE) goto error;
+
+	glBindFramebuffer(GL_FRAMEBUFFER, defaultFBO);
+	glBindRenderbuffer(GL_RENDERBUFFER, defaultRBO);
+	return fb;
+error:
+	glBindFramebuffer(GL_FRAMEBUFFER, defaultFBO);
+	glBindRenderbuffer(GL_RENDERBUFFER, defaultRBO);
+	nvgluDeleteFramebuffer(ctx, fb);
+	return NULL;
+#else
+	NVG_NOTUSED(ctx);
+	NVG_NOTUSED(w);
+	NVG_NOTUSED(h);
+	NVG_NOTUSED(imageFlags);
+	return NULL;
+#endif
+}
+
+void nvgluBindFramebuffer(NVGLUframebuffer* fb)
+{
+#ifdef NANOVG_FBO_VALID
+	if (defaultFBO == -1) glGetIntegerv(GL_FRAMEBUFFER_BINDING, &defaultFBO);
+	glBindFramebuffer(GL_FRAMEBUFFER, fb != NULL ? fb->fbo : defaultFBO);
+#else
+	NVG_NOTUSED(fb);
+#endif
+}
+
+void nvgluDeleteFramebuffer(NVGcontext* ctx, NVGLUframebuffer* fb)
+{
+#ifdef NANOVG_FBO_VALID
+	if (fb == NULL) return;
+	if (fb->fbo != 0)
+		glDeleteFramebuffers(1, &fb->fbo);
+	if (fb->rbo != 0)
+		glDeleteRenderbuffers(1, &fb->rbo);
+	if (fb->image >= 0)
+		nvgDeleteImage(ctx, fb->image);
+	fb->fbo = 0;
+	fb->rbo = 0;
+	fb->texture = 0;
+	fb->image = -1;
+	free(fb);
+#else
+	NVG_NOTUSED(ctx);
+	NVG_NOTUSED(fb);
+#endif
+}
+
+#endif // NANOVG_GL_IMPLEMENTATION
diff --git a/dgl/src/nanovg2/stb_image.h b/dgl/src/nanovg2/stb_image.h
new file mode 100644
index 00000000..4da7ec65
--- /dev/null
+++ b/dgl/src/nanovg2/stb_image.h
@@ -0,0 +1,4647 @@
+/* stb_image - v1.39 - public domain JPEG/PNG reader - http://nothings.org/stb_image.c
+   when you control the images you're loading
+                                     no warranty implied; use at your own risk
+
+   Do this:
+      #define STB_IMAGE_IMPLEMENTATION
+   before you include this file in *one* C or C++ file to create the implementation.
+
+   QUICK NOTES:
+      Primarily of interest to game developers and other people who can
+          avoid problematic images and only need the trivial interface
+
+      JPEG baseline (no JPEG progressive)
+      PNG 8-bit-per-channel only
+
+      TGA (not sure what subset, if a subset)
+      BMP non-1bpp, non-RLE
+      PSD (composited view only, no extra channels)
+
+      GIF (*comp always reports as 4-channel)
+      HDR (radiance rgbE format)
+      PIC (Softimage PIC)
+
+      - stbi__jpeg_huff_decode from memory or through FILE (define STBI_NO_STDIO to remove code)
+      - stbi__jpeg_huff_decode from arbitrary I/O callbacks
+      - overridable dequantizing-IDCT, YCbCr-to-RGB conversion (define STBI_SIMD)
+
+   Latest revisions:
+      1.39 (2014-06-15) TGA optimization fix, multiple BMP fixes
+      1.38 (2014-06-06) suppress MSVC run-time warnings, fix accidental rename of 'skip'
+      1.37 (2014-06-04) remove duplicate typedef
+      1.36 (2014-06-03) converted to header file, allow reading incorrect iphoned-images without iphone flag
+      1.35 (2014-05-27) warnings, bugfixes, TGA optimization, etc
+      1.34 (unknown   ) warning fix
+      1.33 (2011-07-14) minor fixes suggested by Dave Moore
+
+   See end of file for full revision history.
+
+   TODO:
+      stbi_info support for BMP,PSD,HDR,PIC
+
+
+ ============================    Contributors    =========================
+              
+ Image formats                                Bug fixes & warning fixes
+    Sean Barrett (jpeg, png, bmp)                Marc LeBlanc
+    Nicolas Schulz (hdr, psd)                    Christpher Lloyd
+    Jonathan Dummer (tga)                        Dave Moore
+    Jean-Marc Lienher (gif)                      Won Chun
+    Tom Seddon (pic)                             the Horde3D community
+    Thatcher Ulrich (psd)                        Janez Zemva
+                                                 Jonathan Blow
+                                                 Laurent Gomila
+ Extensions, features                            Aruelien Pocheville
+    Jetro Lauha (stbi_info)                      Ryamond Barbiero
+    James "moose2000" Brown (iPhone PNG)         David Woo
+    Ben "Disch" Wenger (io callbacks)            Roy Eltham
+    Martin "SpartanJ" Golini                     Luke Graham
+                                                 Thomas Ruf
+                                                 John Bartholomew
+ Optimizations & bugfixes                        Ken Hamada
+    Fabian "ryg" Giesen                          Cort Stratton
+    Arseny Kapoulkine                            Blazej Dariusz Roszkowski
+                                                 Thibault Reuille
+ If your name should be here but                 Paul Du Bois
+ isn't, let Sean know.                           Guillaume George
+                                                 Jerry Jansson
+                                                 Hayaki Saito
+*/
+
+#ifndef STBI_INCLUDE_STB_IMAGE_H
+#define STBI_INCLUDE_STB_IMAGE_H
+
+// Limitations:
+//    - no jpeg progressive support
+//    - non-HDR formats support 8-bit samples only (jpeg, png)
+//    - no delayed line count (jpeg) -- IJG doesn't support either
+//    - no 1-bit BMP
+//    - GIF always returns *comp=4
+//
+// Basic usage (see HDR discussion below):
+//    int x,y,n;
+//    unsigned char *data = stbi_load(filename, &x, &y, &n, 0);
+//    // ... process data if not NULL ... 
+//    // ... x = width, y = height, n = # 8-bit components per pixel ...
+//    // ... replace '0' with '1'..'4' to force that many components per pixel
+//    // ... but 'n' will always be the number that it would have been if you said 0
+//    stbi_image_free(data)
+//
+// Standard parameters:
+//    int *x       -- outputs image width in pixels
+//    int *y       -- outputs image height in pixels
+//    int *comp    -- outputs # of image components in image file
+//    int req_comp -- if non-zero, # of image components requested in result
+//
+// The return value from an image loader is an 'unsigned char *' which points
+// to the pixel data. The pixel data consists of *y scanlines of *x pixels,
+// with each pixel consisting of N interleaved 8-bit components; the first
+// pixel pointed to is top-left-most in the image. There is no padding between
+// image scanlines or between pixels, regardless of format. The number of
+// components N is 'req_comp' if req_comp is non-zero, or *comp otherwise.
+// If req_comp is non-zero, *comp has the number of components that _would_
+// have been output otherwise. E.g. if you set req_comp to 4, you will always
+// get RGBA output, but you can check *comp to easily see if it's opaque.
+//
+// An output image with N components has the following components interleaved
+// in this order in each pixel:
+//
+//     N=#comp     components
+//       1           grey
+//       2           grey, alpha
+//       3           red, green, blue
+//       4           red, green, blue, alpha
+//
+// If image loading fails for any reason, the return value will be NULL,
+// and *x, *y, *comp will be unchanged. The function stbi_failure_reason()
+// can be queried for an extremely brief, end-user unfriendly explanation
+// of why the load failed. Define STBI_NO_FAILURE_STRINGS to avoid
+// compiling these strings at all, and STBI_FAILURE_USERMSG to get slightly
+// more user-friendly ones.
+//
+// Paletted PNG, BMP, GIF, and PIC images are automatically depalettized.
+//
+// ===========================================================================
+//
+// iPhone PNG support:
+//
+// By default we convert iphone-formatted PNGs back to RGB; nominally they
+// would silently load as BGR, except the existing code should have just
+// failed on such iPhone PNGs. But you can disable this conversion by
+// by calling stbi_convert_iphone_png_to_rgb(0), in which case
+// you will always just get the native iphone "format" through.
+//
+// Call stbi_set_unpremultiply_on_load(1) as well to force a divide per
+// pixel to remove any premultiplied alpha *only* if the image file explicitly
+// says there's premultiplied data (currently only happens in iPhone images,
+// and only if iPhone convert-to-rgb processing is on).
+//
+// ===========================================================================
+//
+// HDR image support   (disable by defining STBI_NO_HDR)
+//
+// stb_image now supports loading HDR images in general, and currently
+// the Radiance .HDR file format, although the support is provided
+// generically. You can still load any file through the existing interface;
+// if you attempt to load an HDR file, it will be automatically remapped to
+// LDR, assuming gamma 2.2 and an arbitrary scale factor defaulting to 1;
+// both of these constants can be reconfigured through this interface:
+//
+//     stbi_hdr_to_ldr_gamma(2.2f);
+//     stbi_hdr_to_ldr_scale(1.0f);
+//
+// (note, do not use _inverse_ constants; stbi_image will invert them
+// appropriately).
+//
+// Additionally, there is a new, parallel interface for loading files as
+// (linear) floats to preserve the full dynamic range:
+//
+//    float *data = stbi_loadf(filename, &x, &y, &n, 0);
+// 
+// If you load LDR images through this interface, those images will
+// be promoted to floating point values, run through the inverse of
+// constants corresponding to the above:
+//
+//     stbi_ldr_to_hdr_scale(1.0f);
+//     stbi_ldr_to_hdr_gamma(2.2f);
+//
+// Finally, given a filename (or an open file or memory block--see header
+// file for details) containing image data, you can query for the "most
+// appropriate" interface to use (that is, whether the image is HDR or
+// not), using:
+//
+//     stbi_is_hdr(char *filename);
+//
+// ===========================================================================
+//
+// 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).
+
+
+#ifndef STBI_NO_STDIO
+
+#if defined(_MSC_VER) && _MSC_VER >= 1400
+#define _CRT_SECURE_NO_WARNINGS // suppress warnings about fopen()
+#pragma warning(push)
+#pragma warning(disable:4996)   // suppress even more warnings about fopen()
+#endif
+#include <stdio.h>
+#endif // STBI_NO_STDIO
+
+#define STBI_VERSION 1
+
+enum
+{
+   STBI_default = 0, // only used for req_comp
+
+   STBI_grey       = 1,
+   STBI_grey_alpha = 2,
+   STBI_rgb        = 3,
+   STBI_rgb_alpha  = 4
+};
+
+typedef unsigned char stbi_uc;
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#ifdef STB_IMAGE_STATIC
+#define STBIDEF static
+#else
+#define STBIDEF extern
+#endif
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// PRIMARY API - works on images of any type
+//
+
+//
+// load image by filename, open file, or memory buffer
+//
+
+STBIDEF stbi_uc *stbi_load_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp);
+
+#ifndef STBI_NO_STDIO
+STBIDEF stbi_uc *stbi_load            (char const *filename,     int *x, int *y, int *comp, int req_comp);
+STBIDEF stbi_uc *stbi_load_from_file  (FILE *f,                  int *x, int *y, int *comp, int req_comp);
+// for stbi_load_from_file, file pointer is left pointing immediately after image
+#endif
+
+typedef struct
+{
+   int      (*read)  (void *user,char *data,int size);   // fill 'data' with 'size' bytes.  return number of bytes actually read 
+   void     (*skip)  (void *user,int n);                 // skip the next 'n' bytes, or 'unget' the last -n bytes if negative
+   int      (*eof)   (void *user);                       // returns nonzero if we are at end of file/data
+} stbi_io_callbacks;
+
+STBIDEF stbi_uc *stbi_load_from_callbacks  (stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp, int req_comp);
+
+#ifndef STBI_NO_HDR
+   STBIDEF float *stbi_loadf_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp);
+
+   #ifndef STBI_NO_STDIO
+   STBIDEF float *stbi_loadf            (char const *filename,   int *x, int *y, int *comp, int req_comp);
+   STBIDEF float *stbi_loadf_from_file  (FILE *f,                int *x, int *y, int *comp, int req_comp);
+   #endif
+   
+   STBIDEF float *stbi_loadf_from_callbacks  (stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp, int req_comp);
+
+   STBIDEF void   stbi_hdr_to_ldr_gamma(float gamma);
+   STBIDEF void   stbi_hdr_to_ldr_scale(float scale);
+
+   STBIDEF void   stbi_ldr_to_hdr_gamma(float gamma);
+   STBIDEF void   stbi_ldr_to_hdr_scale(float scale);
+#endif // STBI_NO_HDR
+
+// stbi_is_hdr is always defined
+STBIDEF int    stbi_is_hdr_from_callbacks(stbi_io_callbacks const *clbk, void *user);
+STBIDEF int    stbi_is_hdr_from_memory(stbi_uc const *buffer, int len);
+#ifndef STBI_NO_STDIO
+STBIDEF int      stbi_is_hdr          (char const *filename);
+STBIDEF int      stbi_is_hdr_from_file(FILE *f);
+#endif // STBI_NO_STDIO
+
+
+// get a VERY brief reason for failure
+// NOT THREADSAFE
+STBIDEF const char *stbi_failure_reason  (void); 
+
+// free the loaded image -- this is just free()
+STBIDEF void     stbi_image_free      (void *retval_from_stbi_load);
+
+// get image dimensions & components without fully decoding
+STBIDEF int      stbi_info_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp);
+STBIDEF int      stbi_info_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp);
+
+#ifndef STBI_NO_STDIO
+STBIDEF int      stbi_info            (char const *filename,     int *x, int *y, int *comp);
+STBIDEF int      stbi_info_from_file  (FILE *f,                  int *x, int *y, int *comp);
+
+#endif
+
+
+
+// for image formats that explicitly notate that they have premultiplied alpha,
+// we just return the colors as stored in the file. set this flag to force
+// unpremultiplication. results are undefined if the unpremultiply overflow.
+STBIDEF void stbi_set_unpremultiply_on_load(int flag_true_if_should_unpremultiply);
+
+// indicate whether we should process iphone images back to canonical format,
+// or just pass them through "as-is"
+STBIDEF void stbi_convert_iphone_png_to_rgb(int flag_true_if_should_convert);
+
+
+// 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);
+
+
+// define faster low-level operations (typically SIMD support)
+#ifdef STBI_SIMD
+typedef void (*stbi_idct_8x8)(stbi_uc *out, int out_stride, short data[64], unsigned short *dequantize);
+// compute an integer IDCT on "input"
+//     input[x] = data[x] * dequantize[x]
+//     write results to 'out': 64 samples, each run of 8 spaced by 'out_stride'
+//                             CLAMP results to 0..255
+typedef void (*stbi_YCbCr_to_RGB_run)(stbi_uc *output, stbi_uc const  *y, stbi_uc const *cb, stbi_uc const *cr, int count, int step);
+// compute a conversion from YCbCr to RGB
+//     'count' pixels
+//     write pixels to 'output'; each pixel is 'step' bytes (either 3 or 4; if 4, write '255' as 4th), order R,G,B
+//     y: Y input channel
+//     cb: Cb input channel; scale/biased to be 0..255
+//     cr: Cr input channel; scale/biased to be 0..255
+
+STBIDEF void stbi_install_idct(stbi_idct_8x8 func);
+STBIDEF void stbi_install_YCbCr_to_RGB(stbi_YCbCr_to_RGB_run func);
+#endif // STBI_SIMD
+
+
+#ifdef __cplusplus
+}
+#endif
+
+//
+//
+////   end header file   /////////////////////////////////////////////////////
+#endif // STBI_INCLUDE_STB_IMAGE_H
+
+#ifdef STB_IMAGE_IMPLEMENTATION
+
+#ifndef STBI_NO_HDR
+#include <math.h>  // ldexp
+#include <string.h> // strcmp, strtok
+#endif
+
+#ifndef STBI_NO_STDIO
+#include <stdio.h>
+#endif
+#include <stdlib.h>
+#include <memory.h>
+#include <assert.h>
+#include <stdarg.h>
+#include <stddef.h> // ptrdiff_t on osx
+
+#ifndef _MSC_VER
+   #ifdef __cplusplus
+   #define stbi_inline inline
+   #else
+   #define stbi_inline
+   #endif
+#else
+   #define stbi_inline __forceinline
+#endif
+
+
+#ifdef _MSC_VER
+typedef unsigned short stbi__uint16;
+typedef   signed short stbi__int16;
+typedef unsigned int   stbi__uint32;
+typedef   signed int   stbi__int32;
+#else
+#include <stdint.h>
+typedef uint16_t stbi__uint16;
+typedef int16_t  stbi__int16;
+typedef uint32_t stbi__uint32;
+typedef int32_t  stbi__int32;
+#endif
+
+// should produce compiler error if size is wrong
+typedef unsigned char validate_uint32[sizeof(stbi__uint32)==4 ? 1 : -1];
+
+#ifdef _MSC_VER
+#define STBI_NOTUSED(v)  (void)(v)
+#else
+#define STBI_NOTUSED(v)  (void)sizeof(v)
+#endif
+
+#ifdef _MSC_VER
+#define STBI_HAS_LROTL
+#endif
+
+#ifdef STBI_HAS_LROTL
+   #define stbi_lrot(x,y)  _lrotl(x,y)
+#else
+   #define stbi_lrot(x,y)  (((x) << (y)) | ((x) >> (32 - (y))))
+#endif
+
+///////////////////////////////////////////////
+//
+//  stbi__context struct and start_xxx functions
+
+// stbi__context structure is our basic context used by all images, so it
+// contains all the IO context, plus some basic image information
+typedef struct
+{
+   stbi__uint32 img_x, img_y;
+   int img_n, img_out_n;
+   
+   stbi_io_callbacks io;
+   void *io_user_data;
+
+   int read_from_callbacks;
+   int buflen;
+   stbi_uc buffer_start[128];
+
+   stbi_uc *img_buffer, *img_buffer_end;
+   stbi_uc *img_buffer_original;
+} stbi__context;
+
+
+static void stbi__refill_buffer(stbi__context *s);
+
+// initialize a memory-stbi__jpeg_huff_decode context
+static void stbi__start_mem(stbi__context *s, stbi_uc const *buffer, int len)
+{
+   s->io.read = NULL;
+   s->read_from_callbacks = 0;
+   s->img_buffer = s->img_buffer_original = (stbi_uc *) buffer;
+   s->img_buffer_end = (stbi_uc *) buffer+len;
+}
+
+// initialize a callback-based context
+static void stbi__start_callbacks(stbi__context *s, stbi_io_callbacks *c, void *user)
+{
+   s->io = *c;
+   s->io_user_data = user;
+   s->buflen = sizeof(s->buffer_start);
+   s->read_from_callbacks = 1;
+   s->img_buffer_original = s->buffer_start;
+   stbi__refill_buffer(s);
+}
+
+#ifndef STBI_NO_STDIO
+
+static int stbi__stdio_read(void *user, char *data, int size)
+{
+   return (int) fread(data,1,size,(FILE*) user);
+}
+
+static void stbi__stdio_skip(void *user, int n)
+{
+   fseek((FILE*) user, n, SEEK_CUR);
+}
+
+static int stbi__stdio_eof(void *user)
+{
+   return feof((FILE*) user);
+}
+
+static stbi_io_callbacks stbi__stdio_callbacks =
+{
+   stbi__stdio_read,
+   stbi__stdio_skip,
+   stbi__stdio_eof,
+};
+
+static void stbi__start_file(stbi__context *s, FILE *f)
+{
+   stbi__start_callbacks(s, &stbi__stdio_callbacks, (void *) f);
+}
+
+//static void stop_file(stbi__context *s) { }
+
+#endif // !STBI_NO_STDIO
+
+static void stbi__rewind(stbi__context *s)
+{
+   // conceptually rewind SHOULD rewind to the beginning of the stream,
+   // but we just rewind to the beginning of the initial buffer, because
+   // we only use it after doing 'test', which only ever looks at at most 92 bytes
+   s->img_buffer = s->img_buffer_original;
+}
+
+static int      stbi__jpeg_test(stbi__context *s);
+static stbi_uc *stbi__jpeg_load(stbi__context *s, int *x, int *y, int *comp, int req_comp);
+static int      stbi__jpeg_info(stbi__context *s, int *x, int *y, int *comp);
+static int      stbi__png_test(stbi__context *s);
+static stbi_uc *stbi__png_load(stbi__context *s, int *x, int *y, int *comp, int req_comp);
+static int      stbi__png_info(stbi__context *s, int *x, int *y, int *comp);
+static int      stbi__bmp_test(stbi__context *s);
+static stbi_uc *stbi__bmp_load(stbi__context *s, int *x, int *y, int *comp, int req_comp);
+static int      stbi__tga_test(stbi__context *s);
+static stbi_uc *stbi__tga_load(stbi__context *s, int *x, int *y, int *comp, int req_comp);
+static int      stbi__tga_info(stbi__context *s, int *x, int *y, int *comp);
+static int      stbi__psd_test(stbi__context *s);
+static stbi_uc *stbi__psd_load(stbi__context *s, int *x, int *y, int *comp, int req_comp);
+#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);
+#endif
+static int      stbi__pic_test(stbi__context *s);
+static stbi_uc *stbi__pic_load(stbi__context *s, int *x, int *y, int *comp, int req_comp);
+static int      stbi__gif_test(stbi__context *s);
+static stbi_uc *stbi__gif_load(stbi__context *s, int *x, int *y, int *comp, int req_comp);
+static int      stbi__gif_info(stbi__context *s, int *x, int *y, int *comp);
+
+
+// this is not threadsafe
+static const char *stbi__g_failure_reason;
+
+STBIDEF const char *stbi_failure_reason(void)
+{
+   return stbi__g_failure_reason;
+}
+
+static int stbi__err(const char *str)
+{
+   stbi__g_failure_reason = str;
+   return 0;
+}
+
+// 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 *) (stbi__err(x,y)?NULL:NULL))
+#define stbi__errpuc(x,y)  ((unsigned char *) (stbi__err(x,y)?NULL:NULL))
+
+STBIDEF void stbi_image_free(void *retval_from_stbi_load)
+{
+   free(retval_from_stbi_load);
+}
+
+#ifndef STBI_NO_HDR
+static float   *stbi__ldr_to_hdr(stbi_uc *data, int x, int y, int comp);
+static stbi_uc *stbi__hdr_to_ldr(float   *data, int x, int y, int comp);
+#endif
+
+static unsigned char *stbi_load_main(stbi__context *s, int *x, int *y, int *comp, int req_comp)
+{
+   if (stbi__jpeg_test(s)) return stbi__jpeg_load(s,x,y,comp,req_comp);
+   if (stbi__png_test(s))  return stbi__png_load(s,x,y,comp,req_comp);
+   if (stbi__bmp_test(s))  return stbi__bmp_load(s,x,y,comp,req_comp);
+   if (stbi__gif_test(s))  return stbi__gif_load(s,x,y,comp,req_comp);
+   if (stbi__psd_test(s))  return stbi__psd_load(s,x,y,comp,req_comp);
+   if (stbi__pic_test(s))  return stbi__pic_load(s,x,y,comp,req_comp);
+
+   #ifndef STBI_NO_HDR
+   if (stbi__hdr_test(s)) {
+      float *hdr = stbi__hdr_load(s, x,y,comp,req_comp);
+      return stbi__hdr_to_ldr(hdr, *x, *y, req_comp ? req_comp : *comp);
+   }
+   #endif
+
+   // test tga last because it's a crappy test!
+   if (stbi__tga_test(s))
+      return stbi__tga_load(s,x,y,comp,req_comp);
+   return stbi__errpuc("unknown image type", "Image not of any known type, or corrupt");
+}
+
+#ifndef STBI_NO_STDIO
+STBIDEF unsigned char *stbi_load(char const *filename, int *x, int *y, int *comp, int req_comp)
+{
+   FILE *f = 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 unsigned char *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_main(&s,x,y,comp,req_comp);
+   if (result) {
+      // need to 'unget' all the characters in the IO buffer
+      fseek(f, - (int) (s.img_buffer_end - s.img_buffer), SEEK_CUR);
+   }
+   return result;
+}
+#endif //!STBI_NO_STDIO
+
+STBIDEF unsigned char *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_main(&s,x,y,comp,req_comp);
+}
+
+unsigned char *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_main(&s,x,y,comp,req_comp);
+}
+
+#ifndef STBI_NO_HDR
+
+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))
+      return stbi__hdr_load(s,x,y,comp,req_comp);
+   #endif
+   data = stbi_load_main(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");
+}
+
+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);
+}
+
+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
+float *stbi_loadf(char const *filename, int *x, int *y, int *comp, int req_comp)
+{
+   FILE *f = fopen(filename, "rb");
+   float *result;
+   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;
+}
+
+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_HDR
+
+// these is-hdr-or-not is defined independent of whether STBI_NO_HDR is
+// defined, for API simplicity; if STBI_NO_HDR is defined, it always
+// reports false!
+
+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 = fopen(filename, "rb");
+   int result=0;
+   if (f) {
+      result = stbi_is_hdr_from_file(f);
+      fclose(f);
+   }
+   return result;
+}
+
+STBIDEF int      stbi_is_hdr_from_file(FILE *f)
+{
+   #ifndef STBI_NO_HDR
+   stbi__context s;
+   stbi__start_file(&s,f);
+   return stbi__hdr_test(&s);
+   #else
+   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
+   return 0;
+   #endif
+}
+
+#ifndef STBI_NO_HDR
+static float stbi__h2l_gamma_i=1.0f/2.2f, stbi__h2l_scale_i=1.0f;
+static float stbi__l2h_gamma=2.2f, stbi__l2h_scale=1.0f;
+
+void   stbi_hdr_to_ldr_gamma(float gamma) { stbi__h2l_gamma_i = 1/gamma; }
+void   stbi_hdr_to_ldr_scale(float scale) { stbi__h2l_scale_i = 1/scale; }
+
+void   stbi_ldr_to_hdr_gamma(float gamma) { stbi__l2h_gamma = gamma; }
+void   stbi_ldr_to_hdr_scale(float scale) { stbi__l2h_scale = scale; }
+#endif
+
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// Common code used by all image loaders
+//
+
+enum
+{
+   SCAN_load=0,
+   SCAN_type,
+   SCAN_header
+};
+
+static void stbi__refill_buffer(stbi__context *s)
+{
+   int n = (s->io.read)(s->io_user_data,(char*)s->buffer_start,s->buflen);
+   if (n == 0) {
+      // at end of file, treat same as if from memory, but need to handle case
+      // where s->img_buffer isn't pointing to safe memory, stbi__err.g. 0-byte file
+      s->read_from_callbacks = 0;
+      s->img_buffer = s->buffer_start;
+      s->img_buffer_end = s->buffer_start+1;
+      *s->img_buffer = 0;
+   } else {
+      s->img_buffer = s->buffer_start;
+      s->img_buffer_end = s->buffer_start + n;
+   }
+}
+
+stbi_inline static stbi_uc stbi__get8(stbi__context *s)
+{
+   if (s->img_buffer < s->img_buffer_end)
+      return *s->img_buffer++;
+   if (s->read_from_callbacks) {
+      stbi__refill_buffer(s);
+      return *s->img_buffer++;
+   }
+   return 0;
+}
+
+stbi_inline static int stbi__at_eof(stbi__context *s)
+{
+   if (s->io.read) {
+      if (!(s->io.eof)(s->io_user_data)) return 0;
+      // if feof() is true, check if buffer = end
+      // special case: we've only got the special 0 character at the end
+      if (s->read_from_callbacks == 0) return 1;
+   }
+
+   return s->img_buffer >= s->img_buffer_end;   
+}
+
+static void stbi__skip(stbi__context *s, int n)
+{
+   if (s->io.read) {
+      int blen = (int) (s->img_buffer_end - s->img_buffer);
+      if (blen < n) {
+         s->img_buffer = s->img_buffer_end;
+         (s->io.skip)(s->io_user_data, n - blen);
+         return;
+      }
+   }
+   s->img_buffer += n;
+}
+
+static int stbi__getn(stbi__context *s, stbi_uc *buffer, int n)
+{
+   if (s->io.read) {
+      int blen = (int) (s->img_buffer_end - s->img_buffer);
+      if (blen < n) {
+         int res, count;
+
+         memcpy(buffer, s->img_buffer, blen);
+         
+         count = (s->io.read)(s->io_user_data, (char*) buffer + blen, n - blen);
+         res = (count == (n-blen));
+         s->img_buffer = s->img_buffer_end;
+         return res;
+      }
+   }
+
+   if (s->img_buffer+n <= s->img_buffer_end) {
+      memcpy(buffer, s->img_buffer, n);
+      s->img_buffer += n;
+      return 1;
+   } else
+      return 0;
+}
+
+static int stbi__get16be(stbi__context *s)
+{
+   int z = stbi__get8(s);
+   return (z << 8) + stbi__get8(s);
+}
+
+static stbi__uint32 stbi__get32be(stbi__context *s)
+{
+   stbi__uint32 z = stbi__get16be(s);
+   return (z << 16) + stbi__get16be(s);
+}
+
+static int stbi__get16le(stbi__context *s)
+{
+   int z = stbi__get8(s);
+   return z + (stbi__get8(s) << 8);
+}
+
+static stbi__uint32 stbi__get32le(stbi__context *s)
+{
+   stbi__uint32 z = stbi__get16le(s);
+   return z + (stbi__get16le(s) << 16);
+}
+
+//////////////////////////////////////////////////////////////////////////////
+//
+//  generic converter from built-in img_n to req_comp
+//    individual types do this automatically as much as possible (stbi__err.g. jpeg
+//    does all cases internally since it needs to colorspace convert anyway,
+//    and it never has alpha, so very few cases ). png can automatically
+//    interleave an alpha=255 channel, but falls back to this for other cases
+//
+//  assume data buffer is malloced, so malloc a new one and free that one
+//  only failure mode is malloc failing
+
+static stbi_uc stbi__compute_y(int r, int g, int b)
+{
+   return (stbi_uc) (((r*77) + (g*150) +  (29*b)) >> 8);
+}
+
+static unsigned char *stbi__convert_format(unsigned char *data, int img_n, int req_comp, unsigned int x, unsigned int y)
+{
+   int i,j;
+   unsigned char *good;
+
+   if (req_comp == img_n) return data;
+   assert(req_comp >= 1 && req_comp <= 4);
+
+   good = (unsigned char *) malloc(req_comp * x * y);
+   if (good == NULL) {
+      free(data);
+      return stbi__errpuc("outofmem", "Out of memory");
+   }
+
+   for (j=0; j < (int) y; ++j) {
+      unsigned char *src  = data + j * x * img_n   ;
+      unsigned char *dest = good + j * x * req_comp;
+
+      #define COMBO(a,b)  ((a)*8+(b))
+      #define CASE(a,b)   case COMBO(a,b): for(i=x-1; i >= 0; --i, src += a, dest += b)
+      // convert source image with img_n components to one with req_comp components;
+      // avoid switch per pixel, so use switch per scanline and massive macros
+      switch (COMBO(img_n, req_comp)) {
+         CASE(1,2) dest[0]=src[0], dest[1]=255; break;
+         CASE(1,3) dest[0]=dest[1]=dest[2]=src[0]; break;
+         CASE(1,4) dest[0]=dest[1]=dest[2]=src[0], dest[3]=255; break;
+         CASE(2,1) dest[0]=src[0]; break;
+         CASE(2,3) dest[0]=dest[1]=dest[2]=src[0]; break;
+         CASE(2,4) dest[0]=dest[1]=dest[2]=src[0], dest[3]=src[1]; break;
+         CASE(3,4) dest[0]=src[0],dest[1]=src[1],dest[2]=src[2],dest[3]=255; break;
+         CASE(3,1) dest[0]=stbi__compute_y(src[0],src[1],src[2]); break;
+         CASE(3,2) dest[0]=stbi__compute_y(src[0],src[1],src[2]), dest[1] = 255; break;
+         CASE(4,1) dest[0]=stbi__compute_y(src[0],src[1],src[2]); break;
+         CASE(4,2) dest[0]=stbi__compute_y(src[0],src[1],src[2]), dest[1] = src[3]; break;
+         CASE(4,3) dest[0]=src[0],dest[1]=src[1],dest[2]=src[2]; break;
+         default: assert(0);
+      }
+      #undef CASE
+   }
+
+   free(data);
+   return good;
+}
+
+#ifndef STBI_NO_HDR
+static float   *stbi__ldr_to_hdr(stbi_uc *data, int x, int y, int comp)
+{
+   int i,k,n;
+   float *output = (float *) malloc(x * y * comp * sizeof(float));
+   if (output == NULL) { free(data); return stbi__errpf("outofmem", "Out of memory"); }
+   // compute number of non-alpha components
+   if (comp & 1) n = comp; else n = comp-1;
+   for (i=0; i < x*y; ++i) {
+      for (k=0; k < n; ++k) {
+         output[i*comp + k] = (float) (pow(data[i*comp+k]/255.0f, stbi__l2h_gamma) * stbi__l2h_scale);
+      }
+      if (k < comp) output[i*comp + k] = data[i*comp+k]/255.0f;
+   }
+   free(data);
+   return output;
+}
+
+#define stbi__float2int(x)   ((int) (x))
+static stbi_uc *stbi__hdr_to_ldr(float   *data, int x, int y, int comp)
+{
+   int i,k,n;
+   stbi_uc *output = (stbi_uc *) malloc(x * y * comp);
+   if (output == NULL) { 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);
+      }
+   }
+   free(data);
+   return output;
+}
+#endif
+
+//////////////////////////////////////////////////////////////////////////////
+//
+//  "baseline" JPEG/JFIF decoder (not actually fully baseline implementation)
+//
+//    simple implementation
+//      - channel subsampling of at most 2 in each dimension
+//      - 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
+//      - uses a lot of intermediate memory, could cache poorly
+//      - load http://nothings.org/remote/anemones.jpg 3 times on 2.8Ghz P4
+//          stb_jpeg:   1.34 seconds (MSVC6, default release build)
+//          stb_jpeg:   1.06 seconds (MSVC6, processor = Pentium Pro)
+//          IJL11.dll:  1.08 seconds (compiled by intel)
+//          IJG 1998:   0.98 seconds (MSVC6, makefile provided by IJG)
+//          IJG 1998:   0.95 seconds (MSVC6, makefile + proc=PPro)
+
+// 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
+{
+   #ifdef STBI_SIMD
+   unsigned short dequant2[4][64];
+   #endif
+   stbi__context *s;
+   stbi__huffman huff_dc[4];
+   stbi__huffman huff_ac[4];
+   stbi_uc dequant[4][64];
+
+// 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;
+      stbi_uc *linebuf;
+   } 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 scan_n, order[4];
+   int restart_interval, todo;
+} stbi__jpeg;
+
+static int stbi__build_huffman(stbi__huffman *h, int *count)
+{
+   int i,j,k=0,code;
+   // build size list for each symbol (from JPEG spec)
+   for (i=0; i < 16; ++i)
+      for (j=0; j < count[i]; ++j)
+         h->size[k++] = (stbi_uc) (i+1);
+   h->size[k] = 0;
+
+   // compute actual symbols (from jpeg spec)
+   code = 0;
+   k = 0;
+   for(j=1; j <= 16; ++j) {
+      // compute delta to add to code to compute symbol id
+      h->delta[j] = k - code;
+      if (h->size[k] == j) {
+         while (h->size[k] == j)
+            h->code[k++] = (stbi__uint16) (code++);
+         if (code-1 >= (1 << j)) return stbi__err("bad code lengths","Corrupt JPEG");
+      }
+      // compute largest code + 1 for this size, preshifted as needed later
+      h->maxcode[j] = code << (16-j);
+      code <<= 1;
+   }
+   h->maxcode[j] = 0xffffffff;
+
+   // build non-spec acceleration table; 255 is flag for not-accelerated
+   memset(h->fast, 255, 1 << FAST_BITS);
+   for (i=0; i < k; ++i) {
+      int s = h->size[i];
+      if (s <= FAST_BITS) {
+         int c = h->code[i] << (FAST_BITS-s);
+         int m = 1 << (FAST_BITS-s);
+         for (j=0; j < m; ++j) {
+            h->fast[c+j] = (stbi_uc) i;
+         }
+      }
+   }
+   return 1;
+}
+
+static void stbi__grow_buffer_unsafe(stbi__jpeg *j)
+{
+   do {
+      int b = j->nomore ? 0 : stbi__get8(j->s);
+      if (b == 0xff) {
+         int c = stbi__get8(j->s);
+         if (c != 0) {
+            j->marker = (unsigned char) c;
+            j->nomore = 1;
+            return;
+         }
+      }
+      j->code_buffer |= b << (24 - j->code_bits);
+      j->code_bits += 8;
+   } while (j->code_bits <= 24);
+}
+
+// (1 << n) - 1
+static stbi__uint32 stbi__bmask[17]={0,1,3,7,15,31,63,127,255,511,1023,2047,4095,8191,16383,32767,65535};
+
+// stbi__jpeg_huff_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];
+   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];
+}
+
+// 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 m = 1 << (n-1);
+   unsigned int k;
+   if (j->code_bits < n) stbi__grow_buffer_unsafe(j);
+
+   #if 1
+   k = stbi_lrot(j->code_buffer, n);
+   j->code_buffer = k & ~stbi__bmask[n];
+   k &= stbi__bmask[n];
+   j->code_bits -= n;
+   #else
+   k = (j->code_buffer >> (32 - n)) & stbi__bmask[n];
+   j->code_bits -= n;
+   j->code_buffer <<= n;
+   #endif
+   // the following test is probably a random branch that won't
+   // predict well. I tried to table accelerate it but failed.
+   // maybe it's compiling as a conditional move?
+   if (k < m)
+      return (-1 << n) + k + 1;
+   else
+      return k;
+}
+
+// given a value that's at position X in the zigzag stream,
+// where does it appear in the 8x8 matrix coded as row-major?
+static stbi_uc stbi__jpeg_dezigzag[64+15] =
+{
+    0,  1,  8, 16,  9,  2,  3, 10,
+   17, 24, 32, 25, 18, 11,  4,  5,
+   12, 19, 26, 33, 40, 48, 41, 34,
+   27, 20, 13,  6,  7, 14, 21, 28,
+   35, 42, 49, 56, 57, 50, 43, 36,
+   29, 22, 15, 23, 30, 37, 44, 51,
+   58, 59, 52, 45, 38, 31, 39, 46,
+   53, 60, 61, 54, 47, 55, 62, 63,
+   // let corrupt input sample past end
+   63, 63, 63, 63, 63, 63, 63, 63,
+   63, 63, 63, 63, 63, 63, 63
+};
+
+// stbi__jpeg_huff_decode one 64-entry block--
+static int stbi__jpeg_decode_block(stbi__jpeg *j, short data[64], stbi__huffman *hdc, stbi__huffman *hac, int b)
+{
+   int diff,dc,k;
+   int t = stbi__jpeg_huff_decode(j, hdc);
+   if (t < 0) return stbi__err("bad huffman code","Corrupt JPEG");
+
+   // 0 all the ac values now so we can do it 32-bits at a time
+   memset(data,0,64*sizeof(data[0]));
+
+   diff = t ? stbi__extend_receive(j, t) : 0;
+   dc = j->img_comp[b].dc_pred + diff;
+   j->img_comp[b].dc_pred = dc;
+   data[0] = (short) dc;
+
+   // stbi__jpeg_huff_decode AC components, see JPEG spec
+   k = 1;
+   do {
+      int r,s;
+      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;
+         // stbi__jpeg_huff_decode into unzigzag'd location
+         data[stbi__jpeg_dezigzag[k++]] = (short) stbi__extend_receive(j,s);
+      }
+   } while (k < 64);
+   return 1;
+}
+
+// take a -128..127 value and stbi__clamp it and convert to 0..255
+stbi_inline static stbi_uc stbi__clamp(int x)
+{
+   // trick to use a single test to catch both cases
+   if ((unsigned int) x > 255) {
+      if (x < 0) return 0;
+      if (x > 255) return 255;
+   }
+   return (stbi_uc) x;
+}
+
+#define stbi__f2f(x)  (int) (((x) * 4096 + 0.5))
+#define stbi__fsh(x)  ((x) << 12)
+
+// derived from jidctint -- DCT_ISLOW
+#define STBI__IDCT_1D(s0,s1,s2,s3,s4,s5,s6,s7)       \
+   int t0,t1,t2,t3,p1,p2,p3,p4,p5,x0,x1,x2,x3; \
+   p2 = s2;                                    \
+   p3 = s6;                                    \
+   p1 = (p2+p3) * stbi__f2f(0.5411961f);             \
+   t2 = p1 + p3*stbi__f2f(-1.847759065f);            \
+   t3 = p1 + p2*stbi__f2f( 0.765366865f);            \
+   p2 = s0;                                    \
+   p3 = s4;                                    \
+   t0 = stbi__fsh(p2+p3);                            \
+   t1 = stbi__fsh(p2-p3);                            \
+   x0 = t0+t3;                                 \
+   x3 = t0-t3;                                 \
+   x1 = t1+t2;                                 \
+   x2 = t1-t2;                                 \
+   t0 = s7;                                    \
+   t1 = s5;                                    \
+   t2 = s3;                                    \
+   t3 = s1;                                    \
+   p3 = t0+t2;                                 \
+   p4 = t1+t3;                                 \
+   p1 = t0+t3;                                 \
+   p2 = t1+t2;                                 \
+   p5 = (p3+p4)*stbi__f2f( 1.175875602f);            \
+   t0 = t0*stbi__f2f( 0.298631336f);                 \
+   t1 = t1*stbi__f2f( 2.053119869f);                 \
+   t2 = t2*stbi__f2f( 3.072711026f);                 \
+   t3 = t3*stbi__f2f( 1.501321110f);                 \
+   p1 = p5 + p1*stbi__f2f(-0.899976223f);            \
+   p2 = p5 + p2*stbi__f2f(-2.562915447f);            \
+   p3 = p3*stbi__f2f(-1.961570560f);                 \
+   p4 = p4*stbi__f2f(-0.390180644f);                 \
+   t3 += p1+p4;                                \
+   t2 += p2+p3;                                \
+   t1 += p2+p4;                                \
+   t0 += p1+p3;
+
+#ifdef STBI_SIMD
+typedef unsigned short stbi_dequantize_t;
+#else
+typedef stbi_uc stbi_dequantize_t;
+#endif
+
+// .344 seconds on 3*anemones.jpg
+static void stbi__idct_block(stbi_uc *out, int out_stride, short data[64], stbi_dequantize_t *dequantize)
+{
+   int i,val[64],*v=val;
+   stbi_dequantize_t *dq = dequantize;
+   stbi_uc *o;
+   short *d = data;
+
+   // columns
+   for (i=0; i < 8; ++i,++d,++dq, ++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] * dq[0] << 2;
+         v[0] = v[8] = v[16] = v[24] = v[32] = v[40] = v[48] = v[56] = dcterm;
+      } else {
+         STBI__IDCT_1D(d[ 0]*dq[ 0],d[ 8]*dq[ 8],d[16]*dq[16],d[24]*dq[24],
+                 d[32]*dq[32],d[40]*dq[40],d[48]*dq[48],d[56]*dq[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_SIMD
+static stbi_idct_8x8 stbi__idct_installed = stbi__idct_block;
+
+STBIDEF void stbi_install_idct(stbi_idct_8x8 func)
+{
+   stbi__idct_installed = func;
+}
+#endif
+
+#define STBI__MARKER_none  0xff
+// if there's a pending marker from the entropy stream, return that
+// otherwise, fetch from the stream and get a marker. if there's no
+// marker, return 0xff, which is never a valid marker value
+static stbi_uc stbi__get_marker(stbi__jpeg *j)
+{
+   stbi_uc x;
+   if (j->marker != STBI__MARKER_none) { x = j->marker; j->marker = STBI__MARKER_none; return x; }
+   x = stbi__get8(j->s);
+   if (x != 0xff) return STBI__MARKER_none;
+   while (x == 0xff)
+      x = stbi__get8(j->s);
+   return x;
+}
+
+// in each scan, we'll have scan_n components, and the order
+// of the components is specified by order[]
+#define STBI__RESTART(x)     ((x) >= 0xd0 && (x) <= 0xd7)
+
+// after a restart interval, stbi__jpeg_reset the entropy decoder and
+// the dc prediction
+static void stbi__jpeg_reset(stbi__jpeg *j)
+{
+   j->code_bits = 0;
+   j->code_buffer = 0;
+   j->nomore = 0;
+   j->img_comp[0].dc_pred = j->img_comp[1].dc_pred = j->img_comp[2].dc_pred = 0;
+   j->marker = STBI__MARKER_none;
+   j->todo = j->restart_interval ? j->restart_interval : 0x7fffffff;
+   // 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->scan_n == 1) {
+      int i,j;
+      #ifdef STBI_SIMD
+      __declspec(align(16))
+      #endif
+      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) {
+            if (!stbi__jpeg_decode_block(z, data, z->huff_dc+z->img_comp[n].hd, z->huff_ac+z->img_comp[n].ha, n)) return 0;
+            #ifdef STBI_SIMD
+            stbi__idct_installed(z->img_comp[n].data+z->img_comp[n].w2*j*8+i*8, z->img_comp[n].w2, data, z->dequant2[z->img_comp[n].tq]);
+            #else
+            stbi__idct_block(z->img_comp[n].data+z->img_comp[n].w2*j*8+i*8, z->img_comp[n].w2, data, z->dequant[z->img_comp[n].tq]);
+            #endif
+            // 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);
+            }
+         }
+      }
+   } else { // interleaved!
+      int i,j,k,x,y;
+      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;
+                     if (!stbi__jpeg_decode_block(z, data, z->huff_dc+z->img_comp[n].hd, z->huff_ac+z->img_comp[n].ha, n)) return 0;
+                     #ifdef STBI_SIMD
+                     stbi__idct_installed(z->img_comp[n].data+z->img_comp[n].w2*y2+x2, z->img_comp[n].w2, data, z->dequant2[z->img_comp[n].tq]);
+                     #else
+                     stbi__idct_block(z->img_comp[n].data+z->img_comp[n].w2*y2+x2, z->img_comp[n].w2, data, z->dequant[z->img_comp[n].tq]);
+                     #endif
+                  }
+               }
+            }
+            // 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 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;
+}
+
+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 0xC2: // stbi__SOF - progressive
+         return stbi__err("progressive jpeg","JPEG format not supported (progressive)");
+
+      case 0xDD: // DRI - specify restart interval
+         if (stbi__get16be(z->s) != 4) return stbi__err("bad DRI len","Corrupt JPEG");
+         z->restart_interval = stbi__get16be(z->s);
+         return 1;
+
+      case 0xDB: // DQT - define quantization table
+         L = stbi__get16be(z->s)-2;
+         while (L > 0) {
+            int q = stbi__get8(z->s);
+            int p = q >> 4;
+            int t = q & 15,i;
+            if (p != 0) return stbi__err("bad DQT type","Corrupt JPEG");
+            if (t > 3) return stbi__err("bad DQT table","Corrupt JPEG");
+            for (i=0; i < 64; ++i)
+               z->dequant[t][stbi__jpeg_dezigzag[i]] = stbi__get8(z->s);
+            #ifdef STBI_SIMD
+            for (i=0; i < 64; ++i)
+               z->dequant2[t][i] = z->dequant[t][i];
+            #endif
+            L -= 65;
+         }
+         return L==0;
+
+      case 0xC4: // DHT - define huffman table
+         L = stbi__get16be(z->s)-2;
+         while (L > 0) {
+            stbi_uc *v;
+            int sizes[16],i,n=0;
+            int q = stbi__get8(z->s);
+            int tc = q >> 4;
+            int th = q & 15;
+            if (tc > 1 || th > 3) return stbi__err("bad DHT header","Corrupt JPEG");
+            for (i=0; i < 16; ++i) {
+               sizes[i] = stbi__get8(z->s);
+               n += sizes[i];
+            }
+            L -= 17;
+            if (tc == 0) {
+               if (!stbi__build_huffman(z->huff_dc+th, sizes)) return 0;
+               v = z->huff_dc[th].values;
+            } else {
+               if (!stbi__build_huffman(z->huff_ac+th, sizes)) return 0;
+               v = z->huff_ac[th].values;
+            }
+            for (i=0; i < n; ++i)
+               v[i] = stbi__get8(z->s);
+            L -= n;
+         }
+         return L==0;
+   }
+   // check for comment block or APP blocks
+   if ((m >= 0xE0 && m <= 0xEF) || m == 0xFE) {
+      stbi__skip(z->s, stbi__get16be(z->s)-2);
+      return 1;
+   }
+   return 0;
+}
+
+// after we see stbi__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 stbi__SOS component count","Corrupt JPEG");
+   if (Ls != 6+2*z->scan_n) return stbi__err("bad stbi__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;
+      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;
+   }
+   if (stbi__get8(z->s) != 0) return stbi__err("bad stbi__SOS","Corrupt JPEG");
+   stbi__get8(z->s); // should be 63, but might be 0
+   if (stbi__get8(z->s) != 0) return stbi__err("bad stbi__SOS","Corrupt JPEG");
+
+   return 1;
+}
+
+static int stbi__process_frame_header(stbi__jpeg *z, int scan)
+{
+   stbi__context *s = z->s;
+   int Lf,p,i,q, h_max=1,v_max=1,c;
+   Lf = stbi__get16be(s);         if (Lf < 11) return stbi__err("bad stbi__SOF len","Corrupt JPEG"); // JPEG
+   p  = stbi__get8(s);          if (p != 8) return stbi__err("only 8-bit","JPEG format not supported: 8-bit only"); // JPEG baseline
+   s->img_y = stbi__get16be(s);   if (s->img_y == 0) return stbi__err("no header height", "JPEG format not supported: delayed height"); // Legal, but we don't handle it--but neither does IJG
+   s->img_x = stbi__get16be(s);   if (s->img_x == 0) return stbi__err("0 width","Corrupt JPEG"); // JPEG requires
+   c = stbi__get8(s);
+   if (c != 3 && c != 1) return stbi__err("bad component count","Corrupt JPEG");    // JFIF requires
+   s->img_n = c;
+   for (i=0; i < c; ++i) {
+      z->img_comp[i].data = NULL;
+      z->img_comp[i].linebuf = NULL;
+   }
+
+   if (Lf != 8+3*s->img_n) return stbi__err("bad stbi__SOF len","Corrupt JPEG");
+
+   for (i=0; i < s->img_n; ++i) {
+      z->img_comp[i].id = stbi__get8(s);
+      if (z->img_comp[i].id != i+1)   // JFIF requires
+         if (z->img_comp[i].id != i)  // some version of jpegtran outputs non-JFIF-compliant files!
+            return stbi__err("bad component ID","Corrupt JPEG");
+      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 != SCAN_load) return 1;
+
+   if ((1 << 30) / s->img_x / s->img_n < s->img_y) return stbi__err("too large", "Image too large to stbi__jpeg_huff_decode");
+
+   for (i=0; i < s->img_n; ++i) {
+      if (z->img_comp[i].h > h_max) h_max = z->img_comp[i].h;
+      if (z->img_comp[i].v > v_max) v_max = z->img_comp[i].v;
+   }
+
+   // compute interleaved mcu info
+   z->img_h_max = h_max;
+   z->img_v_max = v_max;
+   z->img_mcu_w = h_max * 8;
+   z->img_mcu_h = v_max * 8;
+   z->img_mcu_x = (s->img_x + z->img_mcu_w-1) / z->img_mcu_w;
+   z->img_mcu_y = (s->img_y + z->img_mcu_h-1) / z->img_mcu_h;
+
+   for (i=0; i < s->img_n; ++i) {
+      // number of effective pixels (stbi__err.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 stbi__jpeg_huff_decode
+      // the bogus oversized data from using interleaved MCUs and their
+      // big blocks (stbi__err.g. a 16x16 iMCU on an image of width 33); we won't
+      // discard the extra data until colorspace conversion
+      z->img_comp[i].w2 = z->img_mcu_x * z->img_comp[i].h * 8;
+      z->img_comp[i].h2 = z->img_mcu_y * z->img_comp[i].v * 8;
+      z->img_comp[i].raw_data = malloc(z->img_comp[i].w2 * z->img_comp[i].h2+15);
+      if (z->img_comp[i].raw_data == NULL) {
+         for(--i; i >= 0; --i) {
+            free(z->img_comp[i].raw_data);
+            z->img_comp[i].data = NULL;
+         }
+         return stbi__err("outofmem", "Out of memory");
+      }
+      // align blocks for installable-idct using mmx/sse
+      z->img_comp[i].data = (stbi_uc*) (((size_t) z->img_comp[i].raw_data + 15) & ~15);
+      z->img_comp[i].linebuf = NULL;
+   }
+
+   return 1;
+}
+
+// use comparisons since in some cases we handle more than one case (stbi__err.g. stbi__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)
+#define stbi__SOS(x)         ((x) == 0xda)
+
+static int decode_jpeg_header(stbi__jpeg *z, int scan)
+{
+   int m;
+   z->marker = STBI__MARKER_none; // initialize cached marker to empty
+   m = stbi__get_marker(z);
+   if (!stbi__SOI(m)) return stbi__err("no stbi__SOI","Corrupt JPEG");
+   if (scan == 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 stbi__SOF", "Corrupt JPEG");
+         m = stbi__get_marker(z);
+      }
+   }
+   if (!stbi__process_frame_header(z, scan)) return 0;
+   return 1;
+}
+
+static int decode_jpeg_image(stbi__jpeg *j)
+{
+   int m;
+   j->restart_interval = 0;
+   if (!decode_jpeg_header(j, SCAN_load)) return 0;
+   m = stbi__get_marker(j);
+   while (!stbi__EOI(m)) {
+      if (stbi__SOS(m)) {
+         if (!stbi__process_scan_header(j)) return 0;
+         if (!stbi__parse_entropy_coded_data(j)) return 0;
+         if (j->marker == STBI__MARKER_none ) {
+            // handle 0s at the end of image data from IP Kamera 9060
+            while (!stbi__at_eof(j->s)) {
+               int x = stbi__get8(j->s);
+               if (x == 255) {
+                  j->marker = stbi__get8(j->s);
+                  break;
+               } else if (x != 0) {
+                  return 0;
+               }
+            }
+            // if we reach eof without hitting a marker, stbi__get_marker() below will fail and we'll eventually return 0
+         }
+      } else {
+         if (!stbi__process_marker(j, m)) return 0;
+      }
+      m = stbi__get_marker(j);
+   }
+   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;
+}
+
+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;
+}
+
+#define float2fixed(x)  ((int) ((x) * 65536 + 0.5))
+
+// 0.38 seconds on 3*anemones.jpg   (0.25 with processor = Pro)
+// VC6 without processor=Pro is generating multiple LEAs per multiply!
+static void stbi__YCbCr_to_RGB_row(stbi_uc *out, const stbi_uc *y, const stbi_uc *pcb, const stbi_uc *pcr, int count, int step)
+{
+   int i;
+   for (i=0; i < count; ++i) {
+      int y_fixed = (y[i] << 16) + 32768; // rounding
+      int r,g,b;
+      int cr = pcr[i] - 128;
+      int cb = pcb[i] - 128;
+      r = y_fixed + cr*float2fixed(1.40200f);
+      g = y_fixed - cr*float2fixed(0.71414f) - cb*float2fixed(0.34414f);
+      b = y_fixed                            + cb*float2fixed(1.77200f);
+      r >>= 16;
+      g >>= 16;
+      b >>= 16;
+      if ((unsigned) r > 255) { if (r < 0) r = 0; else r = 255; }
+      if ((unsigned) g > 255) { if (g < 0) g = 0; else g = 255; }
+      if ((unsigned) b > 255) { if (b < 0) b = 0; else b = 255; }
+      out[0] = (stbi_uc)r;
+      out[1] = (stbi_uc)g;
+      out[2] = (stbi_uc)b;
+      out[3] = 255;
+      out += step;
+   }
+}
+
+#ifdef STBI_SIMD
+static stbi_YCbCr_to_RGB_run stbi__YCbCr_installed = stbi__YCbCr_to_RGB_row;
+
+STBIDEF void stbi_install_YCbCr_to_RGB(stbi_YCbCr_to_RGB_run func)
+{
+   stbi__YCbCr_installed = func;
+}
+#endif
+
+
+// clean up the temporary component buffers
+static void stbi__cleanup_jpeg(stbi__jpeg *j)
+{
+   int i;
+   for (i=0; i < j->s->img_n; ++i) {
+      if (j->img_comp[i].data) {
+         free(j->img_comp[i].raw_data);
+         j->img_comp[i].data = NULL;
+      }
+      if (j->img_comp[i].linebuf) {
+         free(j->img_comp[i].linebuf);
+         j->img_comp[i].linebuf = NULL;
+      }
+   }
+}
+
+typedef struct
+{
+   resample_row_func resample;
+   stbi_uc *line0,*line1;
+   int hs,vs;   // expansion factor in each axis
+   int w_lores; // horizontal pixels pre-expansion 
+   int ystep;   // how far through vertical expansion we are
+   int ypos;    // which pre-expansion row we're on
+} stbi__resample;
+
+static stbi_uc *load_jpeg_image(stbi__jpeg *z, int *out_x, int *out_y, int *comp, int req_comp)
+{
+   int n, decode_n;
+   // validate req_comp
+   if (req_comp < 0 || req_comp > 4) return stbi__errpuc("bad req_comp", "Internal error");
+   z->s->img_n = 0;
+
+   // load a jpeg image from whichever source
+   if (!decode_jpeg_image(z)) { stbi__cleanup_jpeg(z); return NULL; }
+
+   // determine actual number of components to generate
+   n = req_comp ? req_comp : z->s->img_n;
+
+   if (z->s->img_n == 3 && n < 3)
+      decode_n = 1;
+   else
+      decode_n = z->s->img_n;
+
+   // resample and color-convert
+   {
+      int k;
+      unsigned int i,j;
+      stbi_uc *output;
+      stbi_uc *coutput[4];
+
+      stbi__resample res_comp[4];
+
+      for (k=0; k < decode_n; ++k) {
+         stbi__resample *r = &res_comp[k];
+
+         // allocate line buffer big enough for upsampling off the edges
+         // with upsample factor of 4
+         z->img_comp[k].linebuf = (stbi_uc *) 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 = stbi__resample_row_hv_2;
+         else                               r->resample = stbi__resample_row_generic;
+      }
+
+      // can't error after this so, this is safe
+      output = (stbi_uc *) malloc(n * z->s->img_x * z->s->img_y + 1);
+      if (!output) { stbi__cleanup_jpeg(z); return stbi__errpuc("outofmem", "Out of memory"); }
+
+      // now go ahead and resample
+      for (j=0; j < z->s->img_y; ++j) {
+         stbi_uc *out = output + n * z->s->img_x * j;
+         for (k=0; k < decode_n; ++k) {
+            stbi__resample *r = &res_comp[k];
+            int y_bot = r->ystep >= (r->vs >> 1);
+            coutput[k] = r->resample(z->img_comp[k].linebuf,
+                                     y_bot ? r->line1 : r->line0,
+                                     y_bot ? r->line0 : r->line1,
+                                     r->w_lores, r->hs);
+            if (++r->ystep >= r->vs) {
+               r->ystep = 0;
+               r->line0 = r->line1;
+               if (++r->ypos < z->img_comp[k].y)
+                  r->line1 += z->img_comp[k].w2;
+            }
+         }
+         if (n >= 3) {
+            stbi_uc *y = coutput[0];
+            if (z->s->img_n == 3) {
+               #ifdef STBI_SIMD
+               stbi__YCbCr_installed(out, y, coutput[1], coutput[2], z->s->img_x, n);
+               #else
+               stbi__YCbCr_to_RGB_row(out, y, coutput[1], coutput[2], z->s->img_x, n);
+               #endif
+            } else
+               for (i=0; i < z->s->img_x; ++i) {
+                  out[0] = out[1] = out[2] = y[i];
+                  out[3] = 255; // not used if n==3
+                  out += n;
+               }
+         } else {
+            stbi_uc *y = coutput[0];
+            if (n == 1)
+               for (i=0; i < z->s->img_x; ++i) out[i] = y[i];
+            else
+               for (i=0; i < z->s->img_x; ++i) *out++ = y[i], *out++ = 255;
+         }
+      }
+      stbi__cleanup_jpeg(z);
+      *out_x = z->s->img_x;
+      *out_y = z->s->img_y;
+      if (comp) *comp  = z->s->img_n; // report original components, not output
+      return output;
+   }
+}
+
+static unsigned char *stbi__jpeg_load(stbi__context *s, int *x, int *y, int *comp, int req_comp)
+{
+   stbi__jpeg j;
+   j.s = s;
+   return load_jpeg_image(&j, x,y,comp,req_comp);
+}
+
+static int stbi__jpeg_test(stbi__context *s)
+{
+   int r;
+   stbi__jpeg j;
+   j.s = s;
+   r = decode_jpeg_header(&j, SCAN_type);
+   stbi__rewind(s);
+   return r;
+}
+
+static int stbi__jpeg_info_raw(stbi__jpeg *j, int *x, int *y, int *comp)
+{
+   if (!decode_jpeg_header(j, SCAN_header)) {
+      stbi__rewind( j->s );
+      return 0;
+   }
+   if (x) *x = j->s->img_x;
+   if (y) *y = j->s->img_y;
+   if (comp) *comp = j->s->img_n;
+   return 1;
+}
+
+static int stbi__jpeg_info(stbi__context *s, int *x, int *y, int *comp)
+{
+   stbi__jpeg j;
+   j.s = s;
+   return stbi__jpeg_info_raw(&j, x, y, comp);
+}
+
+// public domain zlib stbi__jpeg_huff_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
+
+// fast-way is faster to check than jpeg huffman, but slow way is slower
+#define STBI__ZFAST_BITS  9 // accelerate all cases in default tables
+#define STBI__ZFAST_MASK  ((1 << STBI__ZFAST_BITS) - 1)
+
+// zlib-style huffman encoding
+// (jpegs packs from left, zlib from right, so can't share code)
+typedef struct
+{
+   stbi__uint16 fast[1 << STBI__ZFAST_BITS];
+   stbi__uint16 firstcode[16];
+   int maxcode[17];
+   stbi__uint16 firstsymbol[16];
+   stbi_uc  size[288];
+   stbi__uint16 value[288]; 
+} stbi__zhuffman;
+
+stbi_inline static int stbi__bitreverse16(int n)
+{
+  n = ((n & 0xAAAA) >>  1) | ((n & 0x5555) << 1);
+  n = ((n & 0xCCCC) >>  2) | ((n & 0x3333) << 2);
+  n = ((n & 0xF0F0) >>  4) | ((n & 0x0F0F) << 4);
+  n = ((n & 0xFF00) >>  8) | ((n & 0x00FF) << 8);
+  return n;
+}
+
+stbi_inline static int stbi__bit_reverse(int v, int bits)
+{
+   assert(bits <= 16);
+   // to bit reverse n bits, reverse 16 and shift
+   // stbi__err.g. 11 bits, bit reverse and shift away 5
+   return stbi__bitreverse16(v) >> (16-bits);
+}
+
+static int stbi__zbuild_huffman(stbi__zhuffman *z, stbi_uc *sizelist, int num)
+{
+   int i,k=0;
+   int code, next_code[16], sizes[17];
+
+   // DEFLATE spec for generating codes
+   memset(sizes, 0, sizeof(sizes));
+   memset(z->fast, 255, sizeof(z->fast));
+   for (i=0; i < num; ++i) 
+      ++sizes[sizelist[i]];
+   sizes[0] = 0;
+   for (i=1; i < 16; ++i)
+      assert(sizes[i] <= (1 << i));
+   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 JPEG");
+      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];
+         z->size [c] = (stbi_uc     ) s;
+         z->value[c] = (stbi__uint16) i;
+         if (s <= STBI__ZFAST_BITS) {
+            int k = stbi__bit_reverse(next_code[s],s);
+            while (k < (1 << STBI__ZFAST_BITS)) {
+               z->fast[k] = (stbi__uint16) c;
+               k += (1 << s);
+            }
+         }
+         ++next_code[s];
+      }
+   }
+   return 1;
+}
+
+// zlib-from-memory implementation for PNG reading
+//    because PNG allows splitting the zlib stream arbitrarily,
+//    and it's annoying structurally to have PNG call ZLIB call PNG,
+//    we require PNG read all the IDATs and combine them into a single
+//    memory buffer
+
+typedef struct
+{
+   stbi_uc *zbuffer, *zbuffer_end;
+   int num_bits;
+   stbi__uint32 code_buffer;
+
+   char *zout;
+   char *zout_start;
+   char *zout_end;
+   int   z_expandable;
+
+   stbi__zhuffman z_length, z_distance;
+} stbi__zbuf;
+
+stbi_inline static stbi_uc stbi__zget8(stbi__zbuf *z)
+{
+   if (z->zbuffer >= z->zbuffer_end) return 0;
+   return *z->zbuffer++;
+}
+
+static void stbi__fill_bits(stbi__zbuf *z)
+{
+   do {
+      assert(z->code_buffer < (1U << z->num_bits));
+      z->code_buffer |= 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;   
+}
+
+stbi_inline static int stbi__zhuffman_decode(stbi__zbuf *a, stbi__zhuffman *z)
+{
+   int b,s,k;
+   if (a->num_bits < 16) stbi__fill_bits(a);
+   b = z->fast[a->code_buffer & STBI__ZFAST_MASK];
+   if (b < 0xffff) {
+      s = z->size[b];
+      a->code_buffer >>= s;
+      a->num_bits -= s;
+      return z->value[b];
+   }
+
+   // 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];
+   assert(z->size[b] == s);
+   a->code_buffer >>= s;
+   a->num_bits -= s;
+   return z->value[b];
+}
+
+static int stbi__zexpand(stbi__zbuf *z, int n)  // need to make room for n bytes
+{
+   char *q;
+   int cur, limit;
+   if (!z->z_expandable) return stbi__err("output buffer limit","Corrupt PNG");
+   cur   = (int) (z->zout     - z->zout_start);
+   limit = (int) (z->zout_end - z->zout_start);
+   while (cur + n > limit)
+      limit *= 2;
+   q = (char *) realloc(z->zout_start, limit);
+   if (q == NULL) return stbi__err("outofmem", "Out of memory");
+   z->zout_start = q;
+   z->zout       = q + cur;
+   z->zout_end   = q + limit;
+   return 1;
+}
+
+static int stbi__zlength_base[31] = {
+   3,4,5,6,7,8,9,10,11,13,
+   15,17,19,23,27,31,35,43,51,59,
+   67,83,99,115,131,163,195,227,258,0,0 };
+
+static int stbi__zlength_extra[31]= 
+{ 0,0,0,0,0,0,0,0,1,1,1,1,2,2,2,2,3,3,3,3,4,4,4,4,5,5,5,5,0,0,0 };
+
+static int stbi__zdist_base[32] = { 1,2,3,4,5,7,9,13,17,25,33,49,65,97,129,193,
+257,385,513,769,1025,1537,2049,3073,4097,6145,8193,12289,16385,24577,0,0};
+
+static int stbi__zdist_extra[32] =
+{ 0,0,0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,8,8,9,9,10,10,11,11,12,12,13,13};
+
+static int stbi__parse_huffman_block(stbi__zbuf *a)
+{
+   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 (a->zout >= a->zout_end) if (!stbi__zexpand(a, 1)) return 0;
+         *a->zout++ = (char) z;
+      } else {
+         stbi_uc *p;
+         int len,dist;
+         if (z == 256) return 1;
+         z -= 257;
+         len = stbi__zlength_base[z];
+         if (stbi__zlength_extra[z]) len += stbi__zreceive(a, stbi__zlength_extra[z]);
+         z = stbi__zhuffman_decode(a, &a->z_distance);
+         if (z < 0) return stbi__err("bad huffman code","Corrupt PNG");
+         dist = stbi__zdist_base[z];
+         if (stbi__zdist_extra[z]) dist += stbi__zreceive(a, stbi__zdist_extra[z]);
+         if (a->zout - a->zout_start < dist) return stbi__err("bad dist","Corrupt PNG");
+         if (a->zout + len > a->zout_end) if (!stbi__zexpand(a, len)) return 0;
+         p = (stbi_uc *) (a->zout - dist);
+         while (len--)
+            *a->zout++ = *p++;
+      }
+   }
+}
+
+static int stbi__compute_huffman_codes(stbi__zbuf *a)
+{
+   static stbi_uc length_dezigzag[19] = { 16,17,18,0,8,7,9,6,10,5,11,4,12,3,13,2,14,1,15 };
+   stbi__zhuffman z_codelength;
+   stbi_uc lencodes[286+32+137];//padding for maximum single op
+   stbi_uc codelength_sizes[19];
+   int i,n;
+
+   int hlit  = stbi__zreceive(a,5) + 257;
+   int hdist = stbi__zreceive(a,5) + 1;
+   int hclen = stbi__zreceive(a,4) + 4;
+
+   memset(codelength_sizes, 0, sizeof(codelength_sizes));
+   for (i=0; i < hclen; ++i) {
+      int s = stbi__zreceive(a,3);
+      codelength_sizes[length_dezigzag[i]] = (stbi_uc) s;
+   }
+   if (!stbi__zbuild_huffman(&z_codelength, codelength_sizes, 19)) return 0;
+
+   n = 0;
+   while (n < hlit + hdist) {
+      int c = stbi__zhuffman_decode(a, &z_codelength);
+      assert(c >= 0 && c < 19);
+      if (c < 16)
+         lencodes[n++] = (stbi_uc) c;
+      else if (c == 16) {
+         c = stbi__zreceive(a,2)+3;
+         memset(lencodes+n, lencodes[n-1], c);
+         n += c;
+      } else if (c == 17) {
+         c = stbi__zreceive(a,3)+3;
+         memset(lencodes+n, 0, c);
+         n += c;
+      } else {
+         assert(c == 18);
+         c = stbi__zreceive(a,7)+11;
+         memset(lencodes+n, 0, c);
+         n += c;
+      }
+   }
+   if (n != hlit+hdist) return stbi__err("bad codelengths","Corrupt PNG");
+   if (!stbi__zbuild_huffman(&a->z_length, lencodes, hlit)) return 0;
+   if (!stbi__zbuild_huffman(&a->z_distance, lencodes+hlit, hdist)) return 0;
+   return 1;
+}
+
+static int stbi__parse_uncomperssed_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;
+   }
+   assert(a->num_bits == 0);
+   // now fill header the normal way
+   while (k < 4)
+      header[k++] = stbi__zget8(a);
+   len  = header[1] * 256 + header[0];
+   nlen = header[3] * 256 + header[2];
+   if (nlen != (len ^ 0xffff)) return stbi__err("zlib corrupt","Corrupt PNG");
+   if (a->zbuffer + len > a->zbuffer_end) return stbi__err("read past buffer","Corrupt PNG");
+   if (a->zout + len > a->zout_end)
+      if (!stbi__zexpand(a, len)) return 0;
+   memcpy(a->zout, a->zbuffer, len);
+   a->zbuffer += len;
+   a->zout += len;
+   return 1;
+}
+
+static int stbi__parse_zlib_header(stbi__zbuf *a)
+{
+   int cmf   = stbi__zget8(a);
+   int cm    = cmf & 15;
+   /* int cinfo = cmf >> 4; */
+   int flg   = stbi__zget8(a);
+   if ((cmf*256+flg) % 31 != 0) return stbi__err("bad zlib header","Corrupt PNG"); // zlib spec
+   if (flg & 32) return stbi__err("no preset dict","Corrupt PNG"); // preset dictionary not allowed in png
+   if (cm != 8) return stbi__err("bad compression","Corrupt PNG"); // DEFLATE required for png
+   // window = 1 << (8 + cinfo)... but who cares, we fully buffer output
+   return 1;
+}
+
+// @TODO: should statically initialize these for optimal thread safety
+static stbi_uc stbi__zdefault_length[288], stbi__zdefault_distance[32];
+static void stbi__init_zdefaults(void)
+{
+   int i;   // use <= to match clearly with spec
+   for (i=0; i <= 143; ++i)     stbi__zdefault_length[i]   = 8;
+   for (   ; i <= 255; ++i)     stbi__zdefault_length[i]   = 9;
+   for (   ; i <= 279; ++i)     stbi__zdefault_length[i]   = 7;
+   for (   ; i <= 287; ++i)     stbi__zdefault_length[i]   = 8;
+
+   for (i=0; i <=  31; ++i)     stbi__zdefault_distance[i] = 5;
+}
+
+static int stbi__parse_zlib(stbi__zbuf *a, int parse_header)
+{
+   int final, type;
+   if (parse_header)
+      if (!stbi__parse_zlib_header(a)) return 0;
+   a->num_bits = 0;
+   a->code_buffer = 0;
+   do {
+      final = stbi__zreceive(a,1);
+      type = stbi__zreceive(a,2);
+      if (type == 0) {
+         if (!stbi__parse_uncomperssed_block(a)) return 0;
+      } else if (type == 3) {
+         return 0;
+      } else {
+         if (type == 1) {
+            // use fixed code lengths
+            if (!stbi__zdefault_distance[31]) stbi__init_zdefaults();
+            if (!stbi__zbuild_huffman(&a->z_length  , stbi__zdefault_length  , 288)) return 0;
+            if (!stbi__zbuild_huffman(&a->z_distance, stbi__zdefault_distance,  32)) return 0;
+         } else {
+            if (!stbi__compute_huffman_codes(a)) return 0;
+         }
+         if (!stbi__parse_huffman_block(a)) return 0;
+      }
+   } while (!final);
+   return 1;
+}
+
+static int stbi__do_zlib(stbi__zbuf *a, char *obuf, int olen, int exp, int parse_header)
+{
+   a->zout_start = obuf;
+   a->zout       = obuf;
+   a->zout_end   = obuf + olen;
+   a->z_expandable = exp;
+
+   return stbi__parse_zlib(a, parse_header);
+}
+
+STBIDEF char *stbi_zlib_decode_malloc_guesssize(const char *buffer, int len, int initial_size, int *outlen)
+{
+   stbi__zbuf a;
+   char *p = (char *) 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 {
+      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 *) 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 {
+      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 *) 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 {
+      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;
+}
+
+// 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
+
+
+typedef struct
+{
+   stbi__uint32 length;
+   stbi__uint32 type;
+} stbi__pngchunk;
+
+#define PNG_TYPE(a,b,c,d)  (((a) << 24) + ((b) << 16) + ((c) << 8) + (d))
+
+static stbi__pngchunk stbi__get_chunk_header(stbi__context *s)
+{
+   stbi__pngchunk c;
+   c.length = stbi__get32be(s);
+   c.type   = stbi__get32be(s);
+   return c;
+}
+
+static int stbi__check_png_header(stbi__context *s)
+{
+   static stbi_uc png_sig[8] = { 137,80,78,71,13,10,26,10 };
+   int i;
+   for (i=0; i < 8; ++i)
+      if (stbi__get8(s) != png_sig[i]) return stbi__err("bad png sig","Not a PNG");
+   return 1;
+}
+
+typedef struct
+{
+   stbi__context *s;
+   stbi_uc *idata, *expanded, *out;
+} stbi__png;
+
+
+enum {
+   STBI__F_none=0, STBI__F_sub=1, STBI__F_up=2, STBI__F_avg=3, STBI__F_paeth=4,
+   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;
+}
+
+#define STBI__BYTECAST(x)  ((stbi_uc) ((x) & 255))  // truncate int to byte without warnings
+
+// 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)
+{
+   stbi__context *s = a->s;
+   stbi__uint32 i,j,stride = x*out_n;
+   int k;
+   int img_n = s->img_n; // copy it into a local for later
+   assert(out_n == s->img_n || out_n == s->img_n+1);
+   a->out = (stbi_uc *) malloc(x * y * out_n);
+   if (!a->out) return stbi__err("outofmem", "Out of memory");
+   if (s->img_x == x && s->img_y == y) {
+      if (raw_len != (img_n * x + 1) * y) return stbi__err("not enough pixels","Corrupt PNG");
+   } else { // interlaced:
+      if (raw_len < (img_n * x + 1) * y) return stbi__err("not enough pixels","Corrupt PNG");
+   }
+   for (j=0; j < y; ++j) {
+      stbi_uc *cur = a->out + stride*j;
+      stbi_uc *prior = cur - stride;
+      int filter = *raw++;
+      if (filter > 4) return stbi__err("invalid filter","Corrupt PNG");
+      // if first row, use special filter that doesn't sample previous row
+      if (j == 0) filter = first_row_filter[filter];
+      // handle first pixel explicitly
+      for (k=0; k < img_n; ++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 (img_n != out_n) cur[img_n] = 255;
+      raw += img_n;
+      cur += out_n;
+      prior += out_n;
+      // this is a little gross, so that we don't switch per-pixel or per-component
+      if (img_n == out_n) {
+         #define CASE(f) \
+             case f:     \
+                for (i=x-1; i >= 1; --i, raw+=img_n,cur+=img_n,prior+=img_n) \
+                   for (k=0; k < img_n; ++k)
+         switch (filter) {
+            CASE(STBI__F_none)         cur[k] = raw[k]; break;
+            CASE(STBI__F_sub)          cur[k] = STBI__BYTECAST(raw[k] + cur[k-img_n]); break;
+            CASE(STBI__F_up)           cur[k] = STBI__BYTECAST(raw[k] + prior[k]); break;
+            CASE(STBI__F_avg)          cur[k] = STBI__BYTECAST(raw[k] + ((prior[k] + cur[k-img_n])>>1)); break;
+            CASE(STBI__F_paeth)        cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k-img_n],prior[k],prior[k-img_n])); break;
+            CASE(STBI__F_avg_first)    cur[k] = STBI__BYTECAST(raw[k] + (cur[k-img_n] >> 1)); break;
+            CASE(STBI__F_paeth_first)  cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k-img_n],0,0)); break;
+         }
+         #undef CASE
+      } else {
+         assert(img_n+1 == out_n);
+         #define CASE(f) \
+             case f:     \
+                for (i=x-1; i >= 1; --i, cur[img_n]=255,raw+=img_n,cur+=out_n,prior+=out_n) \
+                   for (k=0; k < img_n; ++k)
+         switch (filter) {
+            CASE(STBI__F_none)         cur[k] = raw[k]; break;
+            CASE(STBI__F_sub)          cur[k] = STBI__BYTECAST(raw[k] + cur[k-out_n]); break;
+            CASE(STBI__F_up)           cur[k] = STBI__BYTECAST(raw[k] + prior[k]); break;
+            CASE(STBI__F_avg)          cur[k] = STBI__BYTECAST(raw[k] + ((prior[k] + cur[k-out_n])>>1)); break;
+            CASE(STBI__F_paeth)        cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k-out_n],prior[k],prior[k-out_n])); break;
+            CASE(STBI__F_avg_first)    cur[k] = STBI__BYTECAST(raw[k] + (cur[k-out_n] >> 1)); break;
+            CASE(STBI__F_paeth_first)  cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k-out_n],0,0)); break;
+         }
+         #undef CASE
+      }
+   }
+   return 1;
+}
+
+static int stbi__create_png_image(stbi__png *a, stbi_uc *raw, stbi__uint32 raw_len, int out_n, int interlaced)
+{
+   stbi_uc *final;
+   int p;
+   if (!interlaced)
+      return stbi__create_png_image_raw(a, raw, raw_len, out_n, a->s->img_x, a->s->img_y);
+
+   // de-interlacing
+   final = (stbi_uc *) malloc(a->s->img_x * a->s->img_y * out_n);
+   for (p=0; p < 7; ++p) {
+      int xorig[] = { 0,4,0,2,0,1,0 };
+      int yorig[] = { 0,0,4,0,2,0,1 };
+      int xspc[]  = { 8,8,4,4,2,2,1 };
+      int yspc[]  = { 8,8,8,4,4,2,2 };
+      int i,j,x,y;
+      // pass1_x[4] = 0, pass1_x[5] = 1, pass1_x[12] = 1
+      x = (a->s->img_x - xorig[p] + xspc[p]-1) / xspc[p];
+      y = (a->s->img_y - yorig[p] + yspc[p]-1) / yspc[p];
+      if (x && y) {
+         if (!stbi__create_png_image_raw(a, raw, raw_len, out_n, x, y)) {
+            free(final);
+            return 0;
+         }
+         for (j=0; j < y; ++j)
+            for (i=0; i < x; ++i)
+               memcpy(final + (j*yspc[p]+yorig[p])*a->s->img_x*out_n + (i*xspc[p]+xorig[p])*out_n,
+                      a->out + (j*x+i)*out_n, out_n);
+         free(a->out);
+         raw += (x*out_n+1)*y;
+         raw_len -= (x*out_n+1)*y;
+      }
+   }
+   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
+   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__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 *) malloc(pixel_count * pal_img_n);
+   if (p == NULL) return stbi__err("outofmem", "Out of memory");
+
+   // between here and free(out) below, exitting would leak
+   temp_out = p;
+
+   if (pal_img_n == 3) {
+      for (i=0; i < pixel_count; ++i) {
+         int n = orig[i]*4;
+         p[0] = palette[n  ];
+         p[1] = palette[n+1];
+         p[2] = palette[n+2];
+         p += 3;
+      }
+   } else {
+      for (i=0; i < pixel_count; ++i) {
+         int n = orig[i]*4;
+         p[0] = palette[n  ];
+         p[1] = palette[n+1];
+         p[2] = palette[n+2];
+         p[3] = palette[n+3];
+         p += 4;
+      }
+   }
+   free(a->out);
+   a->out = temp_out;
+
+   STBI_NOTUSED(len);
+
+   return 1;
+}
+
+static int stbi__unpremultiply_on_load = 0;
+static int stbi__de_iphone_flag = 0;
+
+STBIDEF void stbi_set_unpremultiply_on_load(int flag_true_if_should_unpremultiply)
+{
+   stbi__unpremultiply_on_load = flag_true_if_should_unpremultiply;
+}
+
+STBIDEF void stbi_convert_iphone_png_to_rgb(int flag_true_if_should_convert)
+{
+   stbi__de_iphone_flag = flag_true_if_should_convert;
+}
+
+static void stbi__de_iphone(stbi__png *z)
+{
+   stbi__context *s = z->s;
+   stbi__uint32 i, pixel_count = s->img_x * s->img_y;
+   stbi_uc *p = z->out;
+
+   if (s->img_out_n == 3) {  // convert bgr to rgb
+      for (i=0; i < pixel_count; ++i) {
+         stbi_uc t = p[0];
+         p[0] = p[2];
+         p[2] = t;
+         p += 3;
+      }
+   } else {
+      assert(s->img_out_n == 4);
+      if (stbi__unpremultiply_on_load) {
+         // convert bgr to rgb and unpremultiply
+         for (i=0; i < pixel_count; ++i) {
+            stbi_uc a = p[3];
+            stbi_uc t = p[0];
+            if (a) {
+               p[0] = p[2] * 255 / a;
+               p[1] = p[1] * 255 / a;
+               p[2] =  t   * 255 / a;
+            } else {
+               p[0] = p[2];
+               p[2] = t;
+            } 
+            p += 4;
+         }
+      } else {
+         // convert bgr to rgb
+         for (i=0; i < pixel_count; ++i) {
+            stbi_uc t = p[0];
+            p[0] = p[2];
+            p[2] = t;
+            p += 4;
+         }
+      }
+   }
+}
+
+static int stbi__parse_png_file(stbi__png *z, int scan, int req_comp)
+{
+   stbi_uc palette[1024], pal_img_n=0;
+   stbi_uc has_trans=0, tc[3];
+   stbi__uint32 ioff=0, idata_limit=0, i, pal_len=0;
+   int first=1,k,interlace=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 == SCAN_type) return 1;
+
+   for (;;) {
+      stbi__pngchunk c = stbi__get_chunk_header(s);
+      switch (c.type) {
+         case PNG_TYPE('C','g','B','I'):
+            is_iphone = 1;
+            stbi__skip(s, c.length);
+            break;
+         case PNG_TYPE('I','H','D','R'): {
+            int depth,color,comp,filter;
+            if (!first) return stbi__err("multiple IHDR","Corrupt PNG");
+            first = 0;
+            if (c.length != 13) return stbi__err("bad IHDR len","Corrupt PNG");
+            s->img_x = stbi__get32be(s); if (s->img_x > (1 << 24)) return stbi__err("too large","Very large image (corrupt?)");
+            s->img_y = stbi__get32be(s); if (s->img_y > (1 << 24)) return stbi__err("too large","Very large image (corrupt?)");
+            depth = stbi__get8(s);  if (depth != 8)        return stbi__err("8bit only","PNG not supported: 8-bit only");
+            color = stbi__get8(s);  if (color > 6)         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 stbi__jpeg_huff_decode");
+               if (scan == SCAN_header) return 1;
+            } else {
+               // if paletted, then pal_n is our final components, and
+               // img_n is # components to decompress/filter.
+               s->img_n = 1;
+               if ((1 << 30) / s->img_x / 4 < s->img_y) return stbi__err("too large","Corrupt PNG");
+               // if SCAN_header, have to scan to see if we have a tRNS
+            }
+            break;
+         }
+
+         case 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 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 == 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;
+               for (k=0; k < s->img_n; ++k)
+                  tc[k] = (stbi_uc) (stbi__get16be(s) & 255); // non 8-bit images will be larger
+            }
+            break;
+         }
+
+         case 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 == SCAN_header) { s->img_n = pal_img_n; return 1; }
+            if (ioff + c.length > 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;
+               p = (stbi_uc *) realloc(z->idata, 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 PNG_TYPE('I','E','N','D'): {
+            stbi__uint32 raw_len;
+            if (first) return stbi__err("first not IHDR", "Corrupt PNG");
+            if (scan != SCAN_load) return 1;
+            if (z->idata == NULL) return stbi__err("no IDAT","Corrupt PNG");
+            z->expanded = (stbi_uc *) stbi_zlib_decode_malloc_guesssize_headerflag((char *) z->idata, ioff, 16384, (int *) &raw_len, !is_iphone);
+            if (z->expanded == NULL) return 0; // zlib should set error
+            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, interlace)) return 0;
+            if (has_trans)
+               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;
+            }
+            free(z->expanded); z->expanded = NULL;
+            return 1;
+         }
+
+         default:
+            // if critical, fail
+            if (first) return stbi__err("first not IHDR", "Corrupt PNG");
+            if ((c.type & (1 << 29)) == 0) {
+               #ifndef STBI_NO_FAILURE_STRINGS
+               // not threadsafe
+               static char invalid_chunk[] = "XXXX PNG chunk not known";
+               invalid_chunk[0] = STBI__BYTECAST(c.type >> 24);
+               invalid_chunk[1] = STBI__BYTECAST(c.type >> 16);
+               invalid_chunk[2] = STBI__BYTECAST(c.type >>  8);
+               invalid_chunk[3] = STBI__BYTECAST(c.type >>  0);
+               #endif
+               return stbi__err(invalid_chunk, "PNG not supported: unknown PNG chunk type");
+            }
+            stbi__skip(s, c.length);
+            break;
+      }
+      // end of PNG chunk, read and skip CRC
+      stbi__get32be(s);
+   }
+}
+
+static unsigned char *stbi__do_png(stbi__png *p, int *x, int *y, int *n, int req_comp)
+{
+   unsigned char *result=NULL;
+   if (req_comp < 0 || req_comp > 4) return stbi__errpuc("bad req_comp", "Internal error");
+   if (stbi__parse_png_file(p, SCAN_load, req_comp)) {
+      result = p->out;
+      p->out = NULL;
+      if (req_comp && req_comp != p->s->img_out_n) {
+         result = stbi__convert_format(result, p->s->img_out_n, req_comp, p->s->img_x, p->s->img_y);
+         p->s->img_out_n = req_comp;
+         if (result == NULL) return result;
+      }
+      *x = p->s->img_x;
+      *y = p->s->img_y;
+      if (n) *n = p->s->img_n;
+   }
+   free(p->out);      p->out      = NULL;
+   free(p->expanded); p->expanded = NULL;
+   free(p->idata);    p->idata    = NULL;
+
+   return result;
+}
+
+static unsigned char *stbi__png_load(stbi__context *s, int *x, int *y, int *comp, int req_comp)
+{
+   stbi__png p;
+   p.s = s;
+   return stbi__do_png(&p, x,y,comp,req_comp);
+}
+
+static int stbi__png_test(stbi__context *s)
+{
+   int r;
+   r = stbi__check_png_header(s);
+   stbi__rewind(s);
+   return r;
+}
+
+static int stbi__png_info_raw(stbi__png *p, int *x, int *y, int *comp)
+{
+   if (!stbi__parse_png_file(p, 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);
+}
+
+// Microsoft/Windows BMP image
+static int stbi__bmp_test_raw(stbi__context *s)
+{
+   int r;
+   int sz;
+   if (stbi__get8(s) != 'B') return 0;
+   if (stbi__get8(s) != 'M') return 0;
+   stbi__get32le(s); // discard filesize
+   stbi__get16le(s); // discard reserved
+   stbi__get16le(s); // discard reserved
+   stbi__get32le(s); // discard data offset
+   sz = stbi__get32le(s);
+   r = (sz == 12 || sz == 40 || sz == 56 || sz == 108 || sz == 124);
+   return r;
+}
+
+static int stbi__bmp_test(stbi__context *s)
+{
+   int r = stbi__bmp_test_raw(s);
+   stbi__rewind(s);
+   return r;
+}
+
+
+// returns 0..31 for the highest set bit
+static int stbi__high_bit(unsigned int z)
+{
+   int n=0;
+   if (z == 0) return -1;
+   if (z >= 0x10000) n += 16, z >>= 16;
+   if (z >= 0x00100) n +=  8, z >>=  8;
+   if (z >= 0x00010) n +=  4, z >>=  4;
+   if (z >= 0x00004) n +=  2, z >>=  2;
+   if (z >= 0x00002) n +=  1, z >>=  1;
+   return n;
+}
+
+static int stbi__bitcount(unsigned int a)
+{
+   a = (a & 0x55555555) + ((a >>  1) & 0x55555555); // max 2
+   a = (a & 0x33333333) + ((a >>  2) & 0x33333333); // max 4
+   a = (a + (a >> 4)) & 0x0f0f0f0f; // max 8 per 4, now 8 bits
+   a = (a + (a >> 8)); // max 16 per 8 bits
+   a = (a + (a >> 16)); // max 32 per 8 bits
+   return a & 0xff;
+}
+
+static int stbi__shiftsigned(int v, int shift, int bits)
+{
+   int result;
+   int z=0;
+
+   if (shift < 0) v <<= -shift;
+   else v >>= shift;
+   result = v;
+
+   z = bits;
+   while (z < 8) {
+      result += v >> z;
+      z += bits;
+   }
+   return result;
+}
+
+static stbi_uc *stbi__bmp_load(stbi__context *s, int *x, int *y, int *comp, int req_comp)
+{
+   stbi_uc *out;
+   unsigned int mr=0,mg=0,mb=0,ma=0, fake_a=0;
+   stbi_uc pal[256][4];
+   int psize=0,i,j,compress=0,width;
+   int bpp, flip_vertically, pad, target, offset, 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
+   offset = stbi__get32le(s);
+   hsz = stbi__get32le(s);
+   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");
+   bpp = stbi__get16le(s);
+   if (bpp == 1) return stbi__errpuc("monochrome", "BMP type not supported: 1-bit");
+   flip_vertically = ((int) s->img_y) > 0;
+   s->img_y = abs((int) s->img_y);
+   if (hsz == 12) {
+      if (bpp < 24)
+         psize = (offset - 14 - 24) / 3;
+   } else {
+      compress = stbi__get32le(s);
+      if (compress == 1 || compress == 2) return stbi__errpuc("BMP RLE", "BMP type not supported: RLE");
+      stbi__get32le(s); // discard sizeof
+      stbi__get32le(s); // discard hres
+      stbi__get32le(s); // discard vres
+      stbi__get32le(s); // discard colorsused
+      stbi__get32le(s); // discard max important
+      if (hsz == 40 || hsz == 56) {
+         if (hsz == 56) {
+            stbi__get32le(s);
+            stbi__get32le(s);
+            stbi__get32le(s);
+            stbi__get32le(s);
+         }
+         if (bpp == 16 || bpp == 32) {
+            mr = mg = mb = 0;
+            if (compress == 0) {
+               if (bpp == 32) {
+                  mr = 0xffu << 16;
+                  mg = 0xffu <<  8;
+                  mb = 0xffu <<  0;
+                  ma = 0xffu << 24;
+                  fake_a = 1; // @TODO: check for cases like alpha value is all 0 and switch it to 255
+                  STBI_NOTUSED(fake_a);
+               } else {
+                  mr = 31u << 10;
+                  mg = 31u <<  5;
+                  mb = 31u <<  0;
+               }
+            } else if (compress == 3) {
+               mr = stbi__get32le(s);
+               mg = stbi__get32le(s);
+               mb = stbi__get32le(s);
+               // not documented, but generated by photoshop and handled by mspaint
+               if (mr == mg && mg == mb) {
+                  // ?!?!?
+                  return stbi__errpuc("bad BMP", "bad BMP");
+               }
+            } else
+               return stbi__errpuc("bad BMP", "bad BMP");
+         }
+      } else {
+         assert(hsz == 108 || hsz == 124);
+         mr = stbi__get32le(s);
+         mg = stbi__get32le(s);
+         mb = stbi__get32le(s);
+         ma = stbi__get32le(s);
+         stbi__get32le(s); // discard color space
+         for (i=0; i < 12; ++i)
+            stbi__get32le(s); // discard color space parameters
+         if (hsz == 124) {
+            stbi__get32le(s); // discard rendering intent
+            stbi__get32le(s); // discard offset of profile data
+            stbi__get32le(s); // discard size of profile data
+            stbi__get32le(s); // discard reserved
+         }
+      }
+      if (bpp < 16)
+         psize = (offset - 14 - hsz) >> 2;
+   }
+   s->img_n = ma ? 4 : 3;
+   if (req_comp && req_comp >= 3) // we can directly stbi__jpeg_huff_decode 3 or 4
+      target = req_comp;
+   else
+      target = s->img_n; // if they want monochrome, we'll post-convert
+   out = (stbi_uc *) malloc(target * s->img_x * s->img_y);
+   if (!out) return stbi__errpuc("outofmem", "Out of memory");
+   if (bpp < 16) {
+      int z=0;
+      if (psize == 0 || psize > 256) { 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 (hsz != 12) stbi__get8(s);
+         pal[i][3] = 255;
+      }
+      stbi__skip(s, offset - 14 - hsz - psize * (hsz == 12 ? 3 : 4));
+      if (bpp == 4) width = (s->img_x + 1) >> 1;
+      else if (bpp == 8) width = s->img_x;
+      else { free(out); return stbi__errpuc("bad bpp", "Corrupt BMP"); }
+      pad = (-width)&3;
+      for (j=0; j < (int) s->img_y; ++j) {
+         for (i=0; i < (int) s->img_x; i += 2) {
+            int v=stbi__get8(s),v2=0;
+            if (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 = (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, offset - 14 - hsz);
+      if (bpp == 24) width = 3 * s->img_x;
+      else if (bpp == 16) width = 2*s->img_x;
+      else /* bpp = 32 and pad = 0 */ width=0;
+      pad = (-width) & 3;
+      if (bpp == 24) {
+         easy = 1;
+      } else if (bpp == 32) {
+         if (mb == 0xff && mg == 0xff00 && mr == 0x00ff0000 && ma == 0xff000000)
+            easy = 2;
+      }
+      if (!easy) {
+         if (!mr || !mg || !mb) { free(out); return stbi__errpuc("bad masks", "Corrupt BMP"); }
+         // right shift amt to put high bit in position #7
+         rshift = stbi__high_bit(mr)-7; rcount = stbi__bitcount(mr);
+         gshift = stbi__high_bit(mg)-7; gcount = stbi__bitcount(mg);
+         bshift = stbi__high_bit(mb)-7; bcount = stbi__bitcount(mb);
+         ashift = stbi__high_bit(ma)-7; acount = stbi__bitcount(ma);
+      }
+      for (j=0; j < (int) s->img_y; ++j) {
+         if (easy) {
+            for (i=0; i < (int) s->img_x; ++i) {
+               unsigned char a;
+               out[z+2] = stbi__get8(s);
+               out[z+1] = stbi__get8(s);
+               out[z+0] = stbi__get8(s);
+               z += 3;
+               a = (easy == 2 ? stbi__get8(s) : 255);
+               if (target == 4) out[z++] = a;
+            }
+         } else {
+            for (i=0; i < (int) s->img_x; ++i) {
+               stbi__uint32 v = (stbi__uint32) (bpp == 16 ? stbi__get16le(s) : stbi__get32le(s));
+               int a;
+               out[z++] = STBI__BYTECAST(stbi__shiftsigned(v & mr, rshift, rcount));
+               out[z++] = STBI__BYTECAST(stbi__shiftsigned(v & mg, gshift, gcount));
+               out[z++] = STBI__BYTECAST(stbi__shiftsigned(v & mb, bshift, bcount));
+               a = (ma ? stbi__shiftsigned(v & ma, ashift, acount) : 255);
+               if (target == 4) out[z++] = STBI__BYTECAST(a); 
+            }
+         }
+         stbi__skip(s, pad);
+      }
+   }
+   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;
+}
+
+// Targa Truevision - TGA
+// by Jonathan Dummer
+
+static int stbi__tga_info(stbi__context *s, int *x, int *y, int *comp)
+{
+    int tga_w, tga_h, tga_comp;
+    int sz;
+    stbi__get8(s);                   // discard Offset
+    sz = stbi__get8(s);              // color type
+    if( sz > 1 ) {
+        stbi__rewind(s);
+        return 0;      // only RGB or indexed allowed
+    }
+    sz = stbi__get8(s);              // image type
+    // only RGB or grey allowed, +/- RLE
+    if ((sz != 1) && (sz != 2) && (sz != 3) && (sz != 9) && (sz != 10) && (sz != 11)) return 0;
+    stbi__skip(s,9);
+    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
+    }
+    sz = stbi__get8(s);               // bits per pixel
+    // only RGB or RGBA or grey allowed
+    if ((sz != 8) && (sz != 16) && (sz != 24) && (sz != 32)) {
+        stbi__rewind(s);
+        return 0;
+    }
+    tga_comp = sz;
+    if (x) *x = tga_w;
+    if (y) *y = tga_h;
+    if (comp) *comp = tga_comp / 8;
+    return 1;                   // seems to have passed everything
+}
+
+static int stbi__tga_test(stbi__context *s)
+{
+   int res;
+   int sz;
+   stbi__get8(s);      //   discard Offset
+   sz = stbi__get8(s);   //   color type
+   if ( sz > 1 ) return 0;   //   only RGB or indexed allowed
+   sz = stbi__get8(s);   //   image type
+   if ( (sz != 1) && (sz != 2) && (sz != 3) && (sz != 9) && (sz != 10) && (sz != 11) ) return 0;   //   only RGB or grey allowed, +/- RLE
+   stbi__get16be(s);      //   discard palette start
+   stbi__get16be(s);      //   discard palette length
+   stbi__get8(s);         //   discard bits per palette color entry
+   stbi__get16be(s);      //   discard x origin
+   stbi__get16be(s);      //   discard y origin
+   if ( stbi__get16be(s) < 1 ) return 0;      //   test width
+   if ( stbi__get16be(s) < 1 ) return 0;      //   test height
+   sz = stbi__get8(s);   //   bits per pixel
+   if ( (sz != 8) && (sz != 16) && (sz != 24) && (sz != 32) )
+      res = 0;
+   else
+      res = 1;
+   stbi__rewind(s);
+   return res;
+}
+
+static stbi_uc *stbi__tga_load(stbi__context *s, int *x, int *y, int *comp, int req_comp)
+{
+   //   read in the TGA header stuff
+   int tga_offset = stbi__get8(s);
+   int tga_indexed = stbi__get8(s);
+   int tga_image_type = stbi__get8(s);
+   int tga_is_RLE = 0;
+   int tga_palette_start = stbi__get16le(s);
+   int tga_palette_len = stbi__get16le(s);
+   int tga_palette_bits = stbi__get8(s);
+   int tga_x_origin = stbi__get16le(s);
+   int tga_y_origin = stbi__get16le(s);
+   int tga_width = stbi__get16le(s);
+   int tga_height = stbi__get16le(s);
+   int tga_bits_per_pixel = stbi__get8(s);
+   int tga_comp = tga_bits_per_pixel / 8;
+   int tga_inverted = stbi__get8(s);
+   //   image data
+   unsigned char *tga_data;
+   unsigned char *tga_palette = NULL;
+   int i, j;
+   unsigned char raw_data[4];
+   int RLE_count = 0;
+   int RLE_repeating = 0;
+   int read_next_pixel = 1;
+
+   //   do a tiny bit of precessing
+   if ( tga_image_type >= 8 )
+   {
+      tga_image_type -= 8;
+      tga_is_RLE = 1;
+   }
+   /* int tga_alpha_bits = tga_inverted & 15; */
+   tga_inverted = 1 - ((tga_inverted >> 5) & 1);
+
+   //   error check
+   if ( //(tga_indexed) ||
+      (tga_width < 1) || (tga_height < 1) ||
+      (tga_image_type < 1) || (tga_image_type > 3) ||
+      ((tga_bits_per_pixel != 8) && (tga_bits_per_pixel != 16) &&
+      (tga_bits_per_pixel != 24) && (tga_bits_per_pixel != 32))
+      )
+   {
+      return NULL; // we don't report this as a bad TGA because we don't even know if it's TGA
+   }
+
+   //   If I'm paletted, then I'll use the number of bits from the palette
+   if ( tga_indexed )
+   {
+      tga_comp = tga_palette_bits / 8;
+   }
+
+   //   tga info
+   *x = tga_width;
+   *y = tga_height;
+   if (comp) *comp = tga_comp;
+
+   tga_data = (unsigned char*)malloc( tga_width * tga_height * tga_comp );
+   if (!tga_data) return stbi__errpuc("outofmem", "Out of memory");
+
+   // skip to the data's starting position (offset usually = 0)
+   stbi__skip(s, tga_offset );
+
+   if ( !tga_indexed && !tga_is_RLE) {
+      for (i=0; i < tga_height; ++i) {
+         int y = tga_inverted ? tga_height -i - 1 : i;
+         stbi_uc *tga_row = tga_data + y*tga_width*tga_comp;
+         stbi__getn(s, tga_row, tga_width * tga_comp);
+      }
+   } else  {
+      //   do I need to load a palette?
+      if ( tga_indexed)
+      {
+         //   any data to skip? (offset usually = 0)
+         stbi__skip(s, tga_palette_start );
+         //   load the palette
+         tga_palette = (unsigned char*)malloc( tga_palette_len * tga_palette_bits / 8 );
+         if (!tga_palette) {
+            free(tga_data);
+            return stbi__errpuc("outofmem", "Out of memory");
+         }
+         if (!stbi__getn(s, tga_palette, tga_palette_len * tga_palette_bits / 8 )) {
+            free(tga_data);
+            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 1 byte, then perform the lookup
+               int pal_idx = stbi__get8(s);
+               if ( pal_idx >= tga_palette_len )
+               {
+                  //   invalid index
+                  pal_idx = 0;
+               }
+               pal_idx *= tga_bits_per_pixel / 8;
+               for (j = 0; j*8 < tga_bits_per_pixel; ++j)
+               {
+                  raw_data[j] = tga_palette[pal_idx+j];
+               }
+            } else
+            {
+               //   read in the data raw
+               for (j = 0; j*8 < tga_bits_per_pixel; ++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 )
+      {
+         free( tga_palette );
+      }
+   }
+
+   // swap RGB
+   if (tga_comp >= 3)
+   {
+      unsigned char* tga_pixel = tga_data;
+      for (i=0; i < tga_width * tga_height; ++i)
+      {
+         unsigned char temp = tga_pixel[0];
+         tga_pixel[0] = tga_pixel[2];
+         tga_pixel[2] = temp;
+         tga_pixel += tga_comp;
+      }
+   }
+
+   // convert to target component count
+   if (req_comp && req_comp != tga_comp)
+      tga_data = stbi__convert_format(tga_data, tga_comp, req_comp, tga_width, tga_height);
+
+   //   the things I do to get rid of an error message, and yet keep
+   //   Microsoft's C compilers happy... [8^(
+   tga_palette_start = tga_palette_len = tga_palette_bits =
+         tga_x_origin = tga_y_origin = 0;
+   //   OK, done
+   return tga_data;
+}
+
+// *************************************************************************************************
+// Photoshop PSD loader -- PD by Thatcher Ulrich, integration by Nicolas Schulz, tweaked by STB
+
+static int stbi__psd_test(stbi__context *s)
+{
+   int r = (stbi__get32be(s) == 0x38425053);
+   stbi__rewind(s);
+   return r;
+}
+
+static stbi_uc *stbi__psd_load(stbi__context *s, int *x, int *y, int *comp, int req_comp)
+{
+   int   pixelCount;
+   int channelCount, compression;
+   int channel, i, count, len;
+   int w,h;
+   stbi_uc *out;
+
+   // Check identifier
+   if (stbi__get32be(s) != 0x38425053)   // "8BPS"
+      return stbi__errpuc("not PSD", "Corrupt PSD image");
+
+   // Check file type version.
+   if (stbi__get16be(s) != 1)
+      return stbi__errpuc("wrong version", "Unsupported version of PSD image");
+
+   // Skip 6 reserved bytes.
+   stbi__skip(s, 6 );
+
+   // Read the number of channels (R, G, B, A, etc).
+   channelCount = stbi__get16be(s);
+   if (channelCount < 0 || channelCount > 16)
+      return stbi__errpuc("wrong channel count", "Unsupported number of channels in PSD image");
+
+   // Read the rows and columns of the image.
+   h = stbi__get32be(s);
+   w = stbi__get32be(s);
+   
+   // Make sure the depth is 8 bits.
+   if (stbi__get16be(s) != 8)
+      return stbi__errpuc("unsupported bit depth", "PSD bit depth is not 8 bit");
+
+   // Make sure the color mode is RGB.
+   // Valid options are:
+   //   0: Bitmap
+   //   1: Grayscale
+   //   2: Indexed color
+   //   3: RGB color
+   //   4: CMYK color
+   //   7: Multichannel
+   //   8: Duotone
+   //   9: Lab color
+   if (stbi__get16be(s) != 3)
+      return stbi__errpuc("wrong color format", "PSD is not in RGB color format");
+
+   // Skip the Mode Data.  (It's the palette for indexed color; other info for other modes.)
+   stbi__skip(s,stbi__get32be(s) );
+
+   // Skip the image resources.  (resolution, pen tool paths, etc)
+   stbi__skip(s, stbi__get32be(s) );
+
+   // Skip the reserved data.
+   stbi__skip(s, stbi__get32be(s) );
+
+   // Find out if the data is compressed.
+   // Known values:
+   //   0: no compression
+   //   1: RLE compressed
+   compression = stbi__get16be(s);
+   if (compression > 1)
+      return stbi__errpuc("bad compression", "PSD has an unknown compression format");
+
+   // Create the destination image.
+   out = (stbi_uc *) malloc(4 * w*h);
+   if (!out) return stbi__errpuc("outofmem", "Out of memory");
+   pixelCount = w*h;
+
+   // Initialize the data to zero.
+   //memset( out, 0, pixelCount * 4 );
+   
+   // Finally, the image data.
+   if (compression) {
+      // RLE as used by .PSD and .TIFF
+      // Loop until you get the number of unpacked bytes you are expecting:
+      //     Read the next source byte into n.
+      //     If n is between 0 and 127 inclusive, copy the next n+1 bytes literally.
+      //     Else if n is between -127 and -1 inclusive, copy the next byte -n+1 times.
+      //     Else if n is 128, noop.
+      // Endloop
+
+      // The RLE-compressed data is preceeded by a 2-byte data count for each row in the data,
+      // which we're going to just skip.
+      stbi__skip(s, h * channelCount * 2 );
+
+      // Read the RLE data by channel.
+      for (channel = 0; channel < 4; channel++) {
+         stbi_uc *p;
+         
+         p = out+channel;
+         if (channel >= channelCount) {
+            // Fill this channel with default data.
+            for (i = 0; i < pixelCount; i++) *p = (channel == 3 ? 255 : 0), p += 4;
+         } else {
+            // Read the RLE data.
+            count = 0;
+            while (count < pixelCount) {
+               len = stbi__get8(s);
+               if (len == 128) {
+                  // No-op.
+               } else if (len < 128) {
+                  // Copy next len+1 bytes literally.
+                  len++;
+                  count += len;
+                  while (len) {
+                     *p = stbi__get8(s);
+                     p += 4;
+                     len--;
+                  }
+               } else if (len > 128) {
+                  stbi_uc   val;
+                  // Next -len+1 bytes in the dest are replicated from next source byte.
+                  // (Interpret len as a negative 8-bit int.)
+                  len ^= 0x0FF;
+                  len += 2;
+                  val = stbi__get8(s);
+                  count += len;
+                  while (len) {
+                     *p = val;
+                     p += 4;
+                     len--;
+                  }
+               }
+            }
+         }
+      }
+      
+   } else {
+      // We're at the raw image data.  It's each channel in order (Red, Green, Blue, Alpha, ...)
+      // where each channel consists of an 8-bit value for each pixel in the image.
+      
+      // Read the data by channel.
+      for (channel = 0; channel < 4; channel++) {
+         stbi_uc *p;
+         
+         p = out + channel;
+         if (channel > channelCount) {
+            // Fill this channel with default data.
+            for (i = 0; i < pixelCount; i++) *p = channel == 3 ? 255 : 0, p += 4;
+         } else {
+            // Read the data.
+            for (i = 0; i < pixelCount; i++)
+               *p = stbi__get8(s), p += 4;
+         }
+      }
+   }
+
+   if (req_comp && req_comp != 4) {
+      out = stbi__convert_format(out, 4, req_comp, w, h);
+      if (out == NULL) return out; // stbi__convert_format frees input on failure
+   }
+
+   if (comp) *comp = channelCount;
+   *y = h;
+   *x = w;
+   
+   return out;
+}
+
+// *************************************************************************************************
+// 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/
+
+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];
+                     int i;
+
+                     if (count==128)
+                        count = stbi__get16be(s);
+                     else
+                        count -= 127;
+                     if (count > left)
+                        return stbi__errpuc("bad file","scanline overrun");
+
+                     if (!stbi__readval(s,packet->channel,value))
+                        return 0;
+
+                     for(i=0;i<count;++i, dest += 4)
+                        stbi__copyval(packet->channel,dest,value);
+                  } else { // Raw
+                     ++count;
+                     if (count>left) return stbi__errpuc("bad file","scanline overrun");
+
+                     for(i=0;i<count;++i, dest+=4)
+                        if (!stbi__readval(s,packet->channel,dest))
+                           return 0;
+                  }
+                  left-=count;
+               }
+               break;
+            }
+         }
+      }
+   }
+
+   return result;
+}
+
+static stbi_uc *stbi__pic_load(stbi__context *s,int *px,int *py,int *comp,int req_comp)
+{
+   stbi_uc *result;
+   int i, x,y;
+
+   for (i=0; i<92; ++i)
+      stbi__get8(s);
+
+   x = stbi__get16be(s);
+   y = stbi__get16be(s);
+   if (stbi__at_eof(s))  return stbi__errpuc("bad file","file too short (pic header)");
+   if ((1 << 28) / x < y) return stbi__errpuc("too large", "Image too large to stbi__jpeg_huff_decode");
+
+   stbi__get32be(s); //skip `ratio'
+   stbi__get16be(s); //skip `fields'
+   stbi__get16be(s); //skip `pad'
+
+   // intermediate buffer is RGBA
+   result = (stbi_uc *) malloc(x*y*4);
+   memset(result, 0xff, x*y*4);
+
+   if (!stbi__pic_load_core(s,x,y,comp, result)) {
+      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;
+}
+
+// *************************************************************************************************
+// GIF loader -- public domain by Jean-Marc Lienher -- simplified/shrunk by stb
+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)
+   int flags, bgindex, ratio, transparent, eflags;
+   stbi_uc  pal[256][4];
+   stbi_uc lpal[256][4];
+   stbi__gif_lzw codes[4096];
+   stbi_uc *color_table;
+   int parse, step;
+   int lflags;
+   int start_x, start_y;
+   int max_x, max_y;
+   int cur_x, cur_y;
+   int line_size;
+} stbi__gif;
+
+static int stbi__gif_test_raw(stbi__context *s)
+{
+   int sz;
+   if (stbi__get8(s) != 'G' || stbi__get8(s) != 'I' || stbi__get8(s) != 'F' || stbi__get8(s) != '8') return 0;
+   sz = stbi__get8(s);
+   if (sz != '9' && sz != '7') return 0;
+   if (stbi__get8(s) != 'a') return 0;
+   return 1;
+}
+
+static int stbi__gif_test(stbi__context *s)
+{
+   int r = stbi__gif_test_raw(s);
+   stbi__rewind(s);
+   return r;
+}
+
+static void stbi__gif_parse_colortable(stbi__context *s, stbi_uc pal[256][4], int num_entries, int transp)
+{
+   int i;
+   for (i=0; i < num_entries; ++i) {
+      pal[i][2] = stbi__get8(s);
+      pal[i][1] = stbi__get8(s);
+      pal[i][0] = stbi__get8(s);
+      pal[i][3] = transp ? 0 : 255;
+   }   
+}
+
+static int stbi__gif_header(stbi__context *s, stbi__gif *g, int *comp, int is_info)
+{
+   stbi_uc version;
+   if (stbi__get8(s) != 'G' || stbi__get8(s) != 'I' || stbi__get8(s) != 'F' || stbi__get8(s) != '8')
+      return stbi__err("not GIF", "Corrupt GIF");
+
+   version = stbi__get8(s);
+   if (version != '7' && version != '9')    return stbi__err("not GIF", "Corrupt GIF");
+   if (stbi__get8(s) != 'a')                      return stbi__err("not GIF", "Corrupt GIF");
+ 
+   stbi__g_failure_reason = "";
+   g->w = stbi__get16le(s);
+   g->h = stbi__get16le(s);
+   g->flags = stbi__get8(s);
+   g->bgindex = stbi__get8(s);
+   g->ratio = stbi__get8(s);
+   g->transparent = -1;
+
+   if (comp != 0) *comp = 4;  // can't actually tell whether it's 3 or 4 until we parse the comments
+
+   if (is_info) return 1;
+
+   if (g->flags & 0x80)
+      stbi__gif_parse_colortable(s,g->pal, 2 << (g->flags & 7), -1);
+
+   return 1;
+}
+
+static int stbi__gif_info_raw(stbi__context *s, int *x, int *y, int *comp)
+{
+   stbi__gif g;   
+   if (!stbi__gif_header(s, &g, comp, 1)) {
+      stbi__rewind( s );
+      return 0;
+   }
+   if (x) *x = g.w;
+   if (y) *y = g.h;
+   return 1;
+}
+
+static void stbi__out_gif_code(stbi__gif *g, stbi__uint16 code)
+{
+   stbi_uc *p, *c;
+
+   // recurse to stbi__jpeg_huff_decode the prefixes, since the linked-list is backwards,
+   // and working backwards through an interleaved image would be nasty
+   if (g->codes[code].prefix >= 0)
+      stbi__out_gif_code(g, g->codes[code].prefix);
+
+   if (g->cur_y >= g->max_y) return;
+  
+   p = &g->out[g->cur_x + g->cur_y];
+   c = &g->color_table[g->codes[code].suffix * 4];
+
+   if (c[3] >= 128) {
+      p[0] = c[2];
+      p[1] = c[1];
+      p[2] = c[0];
+      p[3] = c[3];
+   }
+   g->cur_x += 4;
+
+   if (g->cur_x >= g->max_x) {
+      g->cur_x = g->start_x;
+      g->cur_y += g->step;
+
+      while (g->cur_y >= g->max_y && g->parse > 0) {
+         g->step = (1 << g->parse) * g->line_size;
+         g->cur_y = g->start_y + (g->step >> 1);
+         --g->parse;
+      }
+   }
+}
+
+static stbi_uc *stbi__process_gif_raster(stbi__context *s, stbi__gif *g)
+{
+   stbi_uc lzw_cs;
+   stbi__int32 len, code;
+   stbi__uint32 first;
+   stbi__int32 codesize, codemask, avail, oldcode, bits, valid_bits, clear;
+   stbi__gif_lzw *p;
+
+   lzw_cs = stbi__get8(s);
+   clear = 1 << lzw_cs;
+   first = 1;
+   codesize = lzw_cs + 1;
+   codemask = (1 << codesize) - 1;
+   bits = 0;
+   valid_bits = 0;
+   for (code = 0; code < clear; code++) {
+      g->codes[code].prefix = -1;
+      g->codes[code].first = (stbi_uc) code;
+      g->codes[code].suffix = (stbi_uc) code;
+   }
+
+   // support no starting clear code
+   avail = clear+2;
+   oldcode = -1;
+
+   len = 0;
+   for(;;) {
+      if (valid_bits < codesize) {
+         if (len == 0) {
+            len = stbi__get8(s); // start new block
+            if (len == 0) 
+               return g->out;
+         }
+         --len;
+         bits |= (stbi__int32) stbi__get8(s) << valid_bits;
+         valid_bits += 8;
+      } else {
+         stbi__int32 code = bits & codemask;
+         bits >>= codesize;
+         valid_bits -= codesize;
+         // @OPTIMIZE: is there some way we can accelerate the non-clear path?
+         if (code == clear) {  // clear code
+            codesize = lzw_cs + 1;
+            codemask = (1 << codesize) - 1;
+            avail = clear + 2;
+            oldcode = -1;
+            first = 0;
+         } else if (code == clear + 1) { // end of stream code
+            stbi__skip(s, len);
+            while ((len = stbi__get8(s)) > 0)
+               stbi__skip(s,len);
+            return g->out;
+         } else if (code <= avail) {
+            if (first) return stbi__errpuc("no clear code", "Corrupt GIF");
+
+            if (oldcode >= 0) {
+               p = &g->codes[avail++];
+               if (avail > 4096)        return stbi__errpuc("too many codes", "Corrupt GIF");
+               p->prefix = (stbi__int16) oldcode;
+               p->first = g->codes[oldcode].first;
+               p->suffix = (code == avail) ? p->first : g->codes[code].first;
+            } else if (code == avail)
+               return stbi__errpuc("illegal code in raster", "Corrupt GIF");
+
+            stbi__out_gif_code(g, (stbi__uint16) code);
+
+            if ((avail & codemask) == 0 && avail <= 0x0FFF) {
+               codesize++;
+               codemask = (1 << codesize) - 1;
+            }
+
+            oldcode = code;
+         } else {
+            return stbi__errpuc("illegal code in raster", "Corrupt GIF");
+         }
+      } 
+   }
+}
+
+static void stbi__fill_gif_background(stbi__gif *g)
+{
+   int i;
+   stbi_uc *c = g->pal[g->bgindex];
+   // @OPTIMIZE: write a dword at a time
+   for (i = 0; i < g->w * g->h * 4; i += 4) {
+      stbi_uc *p  = &g->out[i];
+      p[0] = c[2];
+      p[1] = c[1];
+      p[2] = c[0];
+      p[3] = c[3];
+   }
+}
+
+// this function is designed to support animated gifs, although stb_image doesn't support it
+static stbi_uc *stbi__gif_load_next(stbi__context *s, stbi__gif *g, int *comp, int req_comp)
+{
+   int i;
+   stbi_uc *old_out = 0;
+
+   if (g->out == 0) {
+      if (!stbi__gif_header(s, g, comp,0))     return 0; // stbi__g_failure_reason set by stbi__gif_header
+      g->out = (stbi_uc *) malloc(4 * g->w * g->h);
+      if (g->out == 0)                      return stbi__errpuc("outofmem", "Out of memory");
+      stbi__fill_gif_background(g);
+   } else {
+      // animated-gif-only path
+      if (((g->eflags & 0x1C) >> 2) == 3) {
+         old_out = g->out;
+         g->out = (stbi_uc *) malloc(4 * g->w * g->h);
+         if (g->out == 0)                   return stbi__errpuc("outofmem", "Out of memory");
+         memcpy(g->out, old_out, g->w*g->h*4);
+      }
+   }
+    
+   for (;;) {
+      switch (stbi__get8(s)) {
+         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;
+
+            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) {
+               for (i=0; i < 256; ++i)  // @OPTIMIZE: stbi__jpeg_reset only the previous transparent
+                  g->pal[i][3] = 255; 
+               if (g->transparent >= 0 && (g->eflags & 0x01))
+                  g->pal[g->transparent][3] = 0;
+               g->color_table = (stbi_uc *) g->pal;
+            } else
+               return stbi__errpuc("missing color table", "Corrupt GIF");
+   
+            o = stbi__process_gif_raster(s, g);
+            if (o == NULL) return NULL;
+
+            if (req_comp && req_comp != 4)
+               o = stbi__convert_format(o, 4, req_comp, g->w, g->h);
+            return o;
+         }
+
+         case 0x21: // Comment Extension.
+         {
+            int len;
+            if (stbi__get8(s) == 0xF9) { // Graphic Control Extension.
+               len = stbi__get8(s);
+               if (len == 4) {
+                  g->eflags = stbi__get8(s);
+                  stbi__get16le(s); // delay
+                  g->transparent = stbi__get8(s);
+               } else {
+                  stbi__skip(s, len);
+                  break;
+               }
+            }
+            while ((len = stbi__get8(s)) != 0)
+               stbi__skip(s, len);
+            break;
+         }
+
+         case 0x3B: // gif stream termination code
+            return (stbi_uc *) 1;
+
+         default:
+            return stbi__errpuc("unknown code", "Corrupt GIF");
+      }
+   }
+}
+
+static stbi_uc *stbi__gif_load(stbi__context *s, int *x, int *y, int *comp, int req_comp)
+{
+   stbi_uc *u = 0;
+   stbi__gif g={0};
+
+   u = stbi__gif_load_next(s, &g, comp, req_comp);
+   if (u == (void *) 1) u = 0;  // end of animated gif marker
+   if (u) {
+      *x = g.w;
+      *y = g.h;
+   }
+
+   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);
+}
+
+
+// *************************************************************************************************
+// Radiance RGBE HDR loader
+// originally by Nicolas Schulz
+#ifndef STBI_NO_HDR
+static int stbi__hdr_test_core(stbi__context *s)
+{
+   const char *signature = "#?RADIANCE\n";
+   int i;
+   for (i=0; signature[i]; ++i)
+      if (stbi__get8(s) != signature[i])
+         return 0;
+   return 1;
+}
+
+static int stbi__hdr_test(stbi__context* s)
+{
+   int r = stbi__hdr_test_core(s);
+   stbi__rewind(s);
+   return r;
+}
+
+#define STBI__HDR_BUFLEN  1024
+static char *stbi__hdr_gettoken(stbi__context *z, char *buffer)
+{
+   int len=0;
+   char c = '\0';
+
+   c = (char) stbi__get8(z);
+
+   while (!stbi__at_eof(z) && c != '\n') {
+      buffer[len++] = c;
+      if (len == STBI__HDR_BUFLEN-1) {
+         // flush to end of line
+         while (!stbi__at_eof(z) && stbi__get8(z) != '\n')
+            ;
+         break;
+      }
+      c = (char) stbi__get8(z);
+   }
+
+   buffer[len] = 0;
+   return buffer;
+}
+
+static void stbi__hdr_convert(float *output, stbi_uc *input, int req_comp)
+{
+   if ( input[3] != 0 ) {
+      float f1;
+      // Exponent
+      f1 = (float) ldexp(1.0f, input[3] - (int)(128 + 8));
+      if (req_comp <= 2)
+         output[0] = (input[0] + input[1] + input[2]) * f1 / 3;
+      else {
+         output[0] = input[0] * f1;
+         output[1] = input[1] * f1;
+         output[2] = input[2] * f1;
+      }
+      if (req_comp == 2) output[1] = 1;
+      if (req_comp == 4) output[3] = 1;
+   } else {
+      switch (req_comp) {
+         case 4: output[3] = 1; /* fallthrough */
+         case 3: output[0] = output[1] = output[2] = 0;
+                 break;
+         case 2: output[1] = 1; /* fallthrough */
+         case 1: output[0] = 0;
+                 break;
+      }
+   }
+}
+
+static float *stbi__hdr_load(stbi__context *s, int *x, int *y, int *comp, int req_comp)
+{
+   char buffer[STBI__HDR_BUFLEN];
+   char *token;
+   int valid = 0;
+   int width, height;
+   stbi_uc *scanline;
+   float *hdr_data;
+   int len;
+   unsigned char count, value;
+   int i, j, k, c1,c2, z;
+
+
+   // Check identifier
+   if (strcmp(stbi__hdr_gettoken(s,buffer), "#?RADIANCE") != 0)
+      return stbi__errpf("not HDR", "Corrupt HDR image");
+   
+   // Parse header
+   for(;;) {
+      token = stbi__hdr_gettoken(s,buffer);
+      if (token[0] == 0) break;
+      if (strcmp(token, "FORMAT=32-bit_rle_rgbe") == 0) valid = 1;
+   }
+
+   if (!valid)    return stbi__errpf("unsupported format", "Unsupported HDR format");
+
+   // Parse width and height
+   // can't use sscanf() if we're not using stdio!
+   token = stbi__hdr_gettoken(s,buffer);
+   if (strncmp(token, "-Y ", 3))  return stbi__errpf("unsupported data layout", "Unsupported HDR format");
+   token += 3;
+   height = (int) strtol(token, &token, 10);
+   while (*token == ' ') ++token;
+   if (strncmp(token, "+X ", 3))  return stbi__errpf("unsupported data layout", "Unsupported HDR format");
+   token += 3;
+   width = (int) strtol(token, NULL, 10);
+
+   *x = width;
+   *y = height;
+
+   if (comp) *comp = 3;
+   if (req_comp == 0) req_comp = 3;
+
+   // Read data
+   hdr_data = (float *) malloc(height * width * req_comp * sizeof(float));
+
+   // Load image data
+   // image data is stored as some number of sca
+   if ( width < 8 || width >= 32768) {
+      // Read flat data
+      for (j=0; j < height; ++j) {
+         for (i=0; i < width; ++i) {
+            stbi_uc rgbe[4];
+           main_decode_loop:
+            stbi__getn(s, rgbe, 4);
+            stbi__hdr_convert(hdr_data + j * width * req_comp + i * req_comp, rgbe, req_comp);
+         }
+      }
+   } else {
+      // Read RLE-encoded data
+      scanline = NULL;
+
+      for (j = 0; j < height; ++j) {
+         c1 = stbi__get8(s);
+         c2 = stbi__get8(s);
+         len = stbi__get8(s);
+         if (c1 != 2 || c2 != 2 || (len & 0x80)) {
+            // not run-length encoded, so we have to actually use THIS data as a decoded
+            // pixel (note this can't be a valid pixel--one of RGB must be >= 128)
+            stbi_uc rgbe[4];
+            rgbe[0] = (stbi_uc) c1;
+            rgbe[1] = (stbi_uc) c2;
+            rgbe[2] = (stbi_uc) len;
+            rgbe[3] = (stbi_uc) stbi__get8(s);
+            stbi__hdr_convert(hdr_data, rgbe, req_comp);
+            i = 1;
+            j = 0;
+            free(scanline);
+            goto main_decode_loop; // yes, this makes no sense
+         }
+         len <<= 8;
+         len |= stbi__get8(s);
+         if (len != width) { free(hdr_data); free(scanline); return stbi__errpf("invalid decoded scanline length", "corrupt HDR"); }
+         if (scanline == NULL) scanline = (stbi_uc *) malloc(width * 4);
+            
+         for (k = 0; k < 4; ++k) {
+            i = 0;
+            while (i < width) {
+               count = stbi__get8(s);
+               if (count > 128) {
+                  // Run
+                  value = stbi__get8(s);
+                  count -= 128;
+                  for (z = 0; z < count; ++z)
+                     scanline[i++ * 4 + k] = value;
+               } else {
+                  // Dump
+                  for (z = 0; z < count; ++z)
+                     scanline[i++ * 4 + k] = stbi__get8(s);
+               }
+            }
+         }
+         for (i=0; i < width; ++i)
+            stbi__hdr_convert(hdr_data+(j*width + i)*req_comp, scanline + i*4, req_comp);
+      }
+      free(scanline);
+   }
+
+   return hdr_data;
+}
+
+static int stbi__hdr_info(stbi__context *s, int *x, int *y, int *comp)
+{
+   char buffer[STBI__HDR_BUFLEN];
+   char *token;
+   int valid = 0;
+
+   if (strcmp(stbi__hdr_gettoken(s,buffer), "#?RADIANCE") != 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
+
+static int stbi__bmp_info(stbi__context *s, int *x, int *y, int *comp)
+{
+   int hsz;
+   if (stbi__get8(s) != 'B' || stbi__get8(s) != 'M') {
+       stbi__rewind( s );
+       return 0;
+   }
+   stbi__skip(s,12);
+   hsz = stbi__get32le(s);
+   if (hsz != 12 && hsz != 40 && hsz != 56 && hsz != 108 && hsz != 124) {
+       stbi__rewind( s );
+       return 0;
+   }
+   if (hsz == 12) {
+      *x = stbi__get16le(s);
+      *y = stbi__get16le(s);
+   } else {
+      *x = stbi__get32le(s);
+      *y = stbi__get32le(s);
+   }
+   if (stbi__get16le(s) != 1) {
+       stbi__rewind( s );
+       return 0;
+   }
+   *comp = stbi__get16le(s) / 8;
+   return 1;
+}
+
+static int stbi__psd_info(stbi__context *s, int *x, int *y, int *comp)
+{
+   int channelCount;
+   if (stbi__get32be(s) != 0x38425053) {
+       stbi__rewind( s );
+       return 0;
+   }
+   if (stbi__get16be(s) != 1) {
+       stbi__rewind( s );
+       return 0;
+   }
+   stbi__skip(s, 6);
+   channelCount = stbi__get16be(s);
+   if (channelCount < 0 || channelCount > 16) {
+       stbi__rewind( s );
+       return 0;
+   }
+   *y = stbi__get32be(s);
+   *x = stbi__get32be(s);
+   if (stbi__get16be(s) != 8) {
+       stbi__rewind( s );
+       return 0;
+   }
+   if (stbi__get16be(s) != 3) {
+       stbi__rewind( s );
+       return 0;
+   }
+   *comp = 4;
+   return 1;
+}
+
+static int stbi__pic_info(stbi__context *s, int *x, int *y, int *comp)
+{
+   int act_comp=0,num_packets=0,chained;
+   stbi__pic_packet packets[10];
+
+   stbi__skip(s, 92);
+
+   *x = stbi__get16be(s);
+   *y = stbi__get16be(s);
+   if (stbi__at_eof(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;
+}
+
+static int stbi__info_main(stbi__context *s, int *x, int *y, int *comp)
+{
+   if (stbi__jpeg_info(s, x, y, comp))
+       return 1;
+   if (stbi__png_info(s, x, y, comp))
+       return 1;
+   if (stbi__gif_info(s, x, y, comp))
+       return 1;
+   if (stbi__bmp_info(s, x, y, comp))
+       return 1;
+   if (stbi__psd_info(s, x, y, comp))
+       return 1;
+   if (stbi__pic_info(s, x, y, comp))
+       return 1;
+   #ifndef STBI_NO_HDR
+   if (stbi__hdr_info(s, x, y, comp))
+       return 1;
+   #endif
+   // test tga last because it's a crappy test!
+   if (stbi__tga_info(s, x, y, comp))
+       return 1;
+   return stbi__err("unknown image type", "Image not of any known type, or corrupt");
+}
+
+#ifndef STBI_NO_STDIO
+STBIDEF int stbi_info(char const *filename, int *x, int *y, int *comp)
+{
+    FILE *f = fopen(filename, "rb");
+    int result;
+    if (!f) return stbi__err("can't fopen", "Unable to open file");
+    result = stbi_info_from_file(f, x, y, comp);
+    fclose(f);
+    return result;
+}
+
+STBIDEF int stbi_info_from_file(FILE *f, int *x, int *y, int *comp)
+{
+   int r;
+   stbi__context s;
+   long pos = ftell(f);
+   stbi__start_file(&s, f);
+   r = stbi__info_main(&s,x,y,comp);
+   fseek(f,pos,SEEK_SET);
+   return r;
+}
+#endif // !STBI_NO_STDIO
+
+STBIDEF int stbi_info_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp)
+{
+   stbi__context s;
+   stbi__start_mem(&s,buffer,len);
+   return stbi__info_main(&s,x,y,comp);
+}
+
+STBIDEF int stbi_info_from_callbacks(stbi_io_callbacks const *c, void *user, int *x, int *y, int *comp)
+{
+   stbi__context s;
+   stbi__start_callbacks(&s, (stbi_io_callbacks *) c, user);
+   return stbi__info_main(&s,x,y,comp);
+}
+
+#endif // STB_IMAGE_IMPLEMENTATION
+
+#if !defined(STBI_NO_STDIO) && defined(_MSC_VER) && _MSC_VER >= 1400
+#pragma warning(pop)
+#endif
+
+
+/*
+   revision history:
+      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   first released version
+*/
diff --git a/dgl/src/nanovg2/stb_truetype.h b/dgl/src/nanovg2/stb_truetype.h
new file mode 100644
index 00000000..570f236c
--- /dev/null
+++ b/dgl/src/nanovg2/stb_truetype.h
@@ -0,0 +1,2081 @@
+// stb_truetype.h - v0.8 - public domain
+// authored from 2009-2013 by Sean Barrett / RAD Game Tools
+//
+//   This library processes TrueType files:
+//        parse files
+//        extract glyph metrics
+//        extract glyph shapes
+//        render glyphs to one-channel bitmaps with antialiasing (box filter)
+//
+//   Todo:
+//        non-MS cmaps
+//        crashproof on bad data
+//        hinting? (no longer patented)
+//        cleartype-style AA?
+//        optimize: use simple memory allocator for intermediates
+//        optimize: build edge-list directly from curves
+//        optimize: rasterize directly from curves?
+//
+// ADDITIONAL CONTRIBUTORS
+//
+//   Mikko Mononen: compound shape support, more cmap formats
+//   Tor Andersson: kerning, subpixel rendering
+//
+//   Bug/warning reports:
+//       "Zer" on mollyrocket (with fix)
+//       Cass Everitt
+//       stoiko (Haemimont Games)
+//       Brian Hook 
+//       Walter van Niftrik
+//       David Gow
+//       David Given
+//       Ivan-Assen Ivanov
+//       Anthony Pesch
+//
+// VERSION HISTORY
+//
+//   0.8  (2014-05-25) fix a few more warnings
+//   0.7  (2013-09-25) bugfix: subpixel glyph bug fixed in 0.5 had come back
+//   0.6c (2012-07-24) improve documentation
+//   0.6b (2012-07-20) fix a few more warnings
+//   0.6  (2012-07-17) fix warnings; added stbtt_ScaleForMappingEmToPixels,
+//                        stbtt_GetFontBoundingBox, stbtt_IsGlyphEmpty
+//   0.5  (2011-12-09) bugfixes:
+//                        subpixel glyph renderer computed wrong bounding box
+//                        first vertex of shape can be off-curve (FreeSans)
+//   0.4b (2011-12-03) fixed an error in the font baking example
+//   0.4  (2011-12-01) kerning, subpixel rendering (tor)
+//                    bugfixes for:
+//                        codepoint-to-glyph conversion using table fmt=12
+//                        codepoint-to-glyph conversion using table fmt=4
+//                        stbtt_GetBakedQuad with non-square texture (Zer)
+//                    updated Hello World! sample to use kerning and subpixel
+//                    fixed some warnings
+//   0.3  (2009-06-24) cmap fmt=12, compound shapes (MM)
+//                    userdata, malloc-from-userdata, non-zero fill (STB)
+//   0.2  (2009-03-11) Fix unsigned/signed char warnings
+//   0.1  (2009-03-09) First public release
+//
+// LICENSE
+//
+//   This software is in the public domain. Where that dedication is not
+//   recognized, you are granted a perpetual, irrevokable license to copy
+//   and modify this file as you see fit.
+//
+// USAGE
+//
+//   Include this file in whatever places neeed to refer to it. In ONE C/C++
+//   file, write:
+//      #define STB_TRUETYPE_IMPLEMENTATION
+//   before the #include of this file. This expands out the actual
+//   implementation into that C/C++ file.
+//
+//   Simple 3D API (don't ship this, but it's fine for tools and quick start,
+//                  and you can cut and paste from it to move to more advanced)
+//           stbtt_BakeFontBitmap()               -- bake a font to a bitmap for use as texture
+//           stbtt_GetBakedQuad()                 -- compute quad to draw for a given char
+//
+//   "Load" a font file from a memory buffer (you have to keep the buffer loaded)
+//           stbtt_InitFont()
+//           stbtt_GetFontOffsetForIndex()        -- use for TTC font collections
+//
+//   Render a unicode codepoint to a bitmap
+//           stbtt_GetCodepointBitmap()           -- allocates and returns a bitmap
+//           stbtt_MakeCodepointBitmap()          -- renders into bitmap you provide
+//           stbtt_GetCodepointBitmapBox()        -- how big the bitmap must be
+//
+//   Character advance/positioning
+//           stbtt_GetCodepointHMetrics()
+//           stbtt_GetFontVMetrics()
+//           stbtt_GetCodepointKernAdvance()
+//
+// ADDITIONAL DOCUMENTATION
+//
+//   Immediately after this block comment are a series of sample programs.
+//
+//   After the sample programs is the "header file" section. This section
+//   includes documentation for each API function.
+//
+//   Some important concepts to understand to use this library:
+//
+//      Codepoint
+//         Characters are defined by unicode codepoints, e.g. 65 is
+//         uppercase A, 231 is lowercase c with a cedilla, 0x7e30 is
+//         the hiragana for "ma".
+//
+//      Glyph
+//         A visual character shape (every codepoint is rendered as
+//         some glyph)
+//
+//      Glyph index
+//         A font-specific integer ID representing a glyph
+//
+//      Baseline
+//         Glyph shapes are defined relative to a baseline, which is the
+//         bottom of uppercase characters. Characters extend both above
+//         and below the baseline.
+//
+//      Current Point
+//         As you draw text to the screen, you keep track of a "current point"
+//         which is the origin of each character. The current point's vertical
+//         position is the baseline. Even "baked fonts" use this model.
+//
+//      Vertical Font Metrics
+//         The vertical qualities of the font, used to vertically position
+//         and space the characters. See docs for stbtt_GetFontVMetrics.
+//
+//      Font Size in Pixels or Points
+//         The preferred interface for specifying font sizes in stb_truetype
+//         is to specify how tall the font's vertical extent should be in pixels.
+//         If that sounds good enough, skip the next paragraph.
+//
+//         Most font APIs instead use "points", which are a common typographic
+//         measurement for describing font size, defined as 72 points per inch.
+//         stb_truetype provides a point API for compatibility. However, true
+//         "per inch" conventions don't make much sense on computer displays
+//         since they different monitors have different number of pixels per
+//         inch. For example, Windows traditionally uses a convention that
+//         there are 96 pixels per inch, thus making 'inch' measurements have
+//         nothing to do with inches, and thus effectively defining a point to
+//         be 1.333 pixels. Additionally, the TrueType font data provides
+//         an explicit scale factor to scale a given font's glyphs to points,
+//         but the author has observed that this scale factor is often wrong
+//         for non-commercial fonts, thus making fonts scaled in points
+//         according to the TrueType spec incoherently sized in practice.
+//
+// ADVANCED USAGE
+//
+//   Quality:
+//
+//    - Use the functions with Subpixel at the end to allow your characters
+//      to have subpixel positioning. Since the font is anti-aliased, not
+//      hinted, this is very import for quality. (This is not possible with
+//      baked fonts.)
+//
+//    - Kerning is now supported, and if you're supporting subpixel rendering
+//      then kerning is worth using to give your text a polished look.
+//
+//   Performance:
+//
+//    - Convert Unicode codepoints to glyph indexes and operate on the glyphs;
+//      if you don't do this, stb_truetype is forced to do the conversion on
+//      every call.
+//
+//    - There are a lot of memory allocations. We should modify it to take
+//      a temp buffer and allocate from the temp buffer (without freeing),
+//      should help performance a lot.
+//
+// NOTES
+//
+//   The system uses the raw data found in the .ttf file without changing it
+//   and without building auxiliary data structures. This is a bit inefficient
+//   on little-endian systems (the data is big-endian), but assuming you're
+//   caching the bitmaps or glyph shapes this shouldn't be a big deal.
+//
+//   It appears to be very hard to programmatically determine what font a
+//   given file is in a general way. I provide an API for this, but I don't
+//   recommend it.
+//
+//
+// SOURCE STATISTICS (based on v0.6c, 2050 LOC)
+//
+//   Documentation & header file        520 LOC  \___ 660 LOC documentation
+//   Sample code                        140 LOC  /
+//   Truetype parsing                   620 LOC  ---- 620 LOC TrueType
+//   Software rasterization             240 LOC  \                           .
+//   Curve tesselation                  120 LOC   \__ 550 LOC Bitmap creation
+//   Bitmap management                  100 LOC   /
+//   Baked bitmap interface              70 LOC  /
+//   Font name matching & access        150 LOC  ---- 150 
+//   C runtime library abstraction       60 LOC  ----  60
+
+
+//////////////////////////////////////////////////////////////////////////////
+//////////////////////////////////////////////////////////////////////////////
+////
+////  SAMPLE PROGRAMS
+////
+//
+//  Incomplete text-in-3d-api example, which draws quads properly aligned to be lossless
+//
+#if 0
+#define STB_TRUETYPE_IMPLEMENTATION  // force following include to generate implementation
+#include "stb_truetype.h"
+
+char ttf_buffer[1<<20];
+unsigned char temp_bitmap[512*512];
+
+stbtt_bakedchar cdata[96]; // ASCII 32..126 is 95 glyphs
+GLstbtt_uint ftex;
+
+void my_stbtt_initfont(void)
+{
+   fread(ttf_buffer, 1, 1<<20, fopen("c:/windows/fonts/times.ttf", "rb"));
+   stbtt_BakeFontBitmap(data,0, 32.0, temp_bitmap,512,512, 32,96, cdata); // no guarantee this fits!
+   // can free ttf_buffer at this point
+   glGenTextures(1, &ftex);
+   glBindTexture(GL_TEXTURE_2D, ftex);
+   glTexImage2D(GL_TEXTURE_2D, 0, GL_ALPHA, 512,512, 0, GL_ALPHA, GL_UNSIGNED_BYTE, temp_bitmap);
+   // can free temp_bitmap at this point
+   glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
+}
+
+void my_stbtt_print(float x, float y, char *text)
+{
+   // assume orthographic projection with units = screen pixels, origin at top left
+   glBindTexture(GL_TEXTURE_2D, ftex);
+   glBegin(GL_QUADS);
+   while (*text) {
+      if (*text >= 32 && *text < 128) {
+         stbtt_aligned_quad q;
+         stbtt_GetBakedQuad(cdata, 512,512, *text-32, &x,&y,&q,1);//1=opengl & d3d10+,0=d3d9
+         glTexCoord2f(q.s0,q.t1); glVertex2f(q.x0,q.y0);
+         glTexCoord2f(q.s1,q.t1); glVertex2f(q.x1,q.y0);
+         glTexCoord2f(q.s1,q.t0); glVertex2f(q.x1,q.y1);
+         glTexCoord2f(q.s0,q.t0); glVertex2f(q.x0,q.y1);
+      }
+      ++text;
+   }
+   glEnd();
+}
+#endif
+//
+//
+//////////////////////////////////////////////////////////////////////////////
+//
+// Complete program (this compiles): get a single bitmap, print as ASCII art
+//
+#if 0
+#include <stdio.h>
+#define STB_TRUETYPE_IMPLEMENTATION  // force following include to generate implementation
+#include "stb_truetype.h"
+
+char ttf_buffer[1<<25];
+
+int main(int argc, char **argv)
+{
+   stbtt_fontinfo font;
+   unsigned char *bitmap;
+   int w,h,i,j,c = (argc > 1 ? atoi(argv[1]) : 'a'), s = (argc > 2 ? atoi(argv[2]) : 20);
+
+   fread(ttf_buffer, 1, 1<<25, fopen(argc > 3 ? argv[3] : "c:/windows/fonts/arialbd.ttf", "rb"));
+
+   stbtt_InitFont(&font, ttf_buffer, stbtt_GetFontOffsetForIndex(ttf_buffer,0));
+   bitmap = stbtt_GetCodepointBitmap(&font, 0,stbtt_ScaleForPixelHeight(&font, s), c, &w, &h, 0,0);
+
+   for (j=0; j < h; ++j) {
+      for (i=0; i < w; ++i)
+         putchar(" .:ioVM@"[bitmap[j*w+i]>>5]);
+      putchar('\n');
+   }
+   return 0;
+}
+#endif 
+//
+// Output:
+//
+//     .ii.
+//    @@@@@@.
+//   V@Mio@@o
+//   :i.  V@V
+//     :oM@@M
+//   :@@@MM@M
+//   @@o  o@M
+//  :@@.  M@M
+//   @@@o@@@@
+//   :M@@V:@@.
+//  
+//////////////////////////////////////////////////////////////////////////////
+// 
+// Complete program: print "Hello World!" banner, with bugs
+//
+#if 0
+char buffer[24<<20];
+unsigned char screen[20][79];
+
+int main(int arg, char **argv)
+{
+   stbtt_fontinfo font;
+   int i,j,ascent,baseline,ch=0;
+   float scale, xpos=2; // leave a little padding in case the character extends left
+   char *text = "Heljo World!";
+
+   fread(buffer, 1, 1000000, fopen("c:/windows/fonts/arialbd.ttf", "rb"));
+   stbtt_InitFont(&font, buffer, 0);
+
+   scale = stbtt_ScaleForPixelHeight(&font, 15);
+   stbtt_GetFontVMetrics(&font, &ascent,0,0);
+   baseline = (int) (ascent*scale);
+
+   while (text[ch]) {
+      int advance,lsb,x0,y0,x1,y1;
+      float x_shift = xpos - (float) floor(xpos);
+      stbtt_GetCodepointHMetrics(&font, text[ch], &advance, &lsb);
+      stbtt_GetCodepointBitmapBoxSubpixel(&font, text[ch], scale,scale,x_shift,0, &x0,&y0,&x1,&y1);
+      stbtt_MakeCodepointBitmapSubpixel(&font, &screen[baseline + y0][(int) xpos + x0], x1-x0,y1-y0, 79, scale,scale,x_shift,0, text[ch]);
+      // note that this stomps the old data, so where character boxes overlap (e.g. 'lj') it's wrong
+      // because this API is really for baking character bitmaps into textures. if you want to render
+      // a sequence of characters, you really need to render each bitmap to a temp buffer, then
+      // "alpha blend" that into the working buffer
+      xpos += (advance * scale);
+      if (text[ch+1])
+         xpos += scale*stbtt_GetCodepointKernAdvance(&font, text[ch],text[ch+1]);
+      ++ch;
+   }
+
+   for (j=0; j < 20; ++j) {
+      for (i=0; i < 78; ++i)
+         putchar(" .:ioVM@"[screen[j][i]>>5]);
+      putchar('\n');
+   }
+
+   return 0;
+}
+#endif
+
+
+//////////////////////////////////////////////////////////////////////////////
+//////////////////////////////////////////////////////////////////////////////
+////
+////   INTEGRATION WITH YOUR CODEBASE
+////
+////   The following sections allow you to supply alternate definitions
+////   of C library functions used by stb_truetype.
+
+#ifdef STB_TRUETYPE_IMPLEMENTATION
+   // #define your own (u)stbtt_int8/16/32 before including to override this
+   #ifndef stbtt_uint8
+   typedef unsigned char   stbtt_uint8;
+   typedef signed   char   stbtt_int8;
+   typedef unsigned short  stbtt_uint16;
+   typedef signed   short  stbtt_int16;
+   typedef unsigned int    stbtt_uint32;
+   typedef signed   int    stbtt_int32;
+   #endif
+
+   typedef char stbtt__check_size32[sizeof(stbtt_int32)==4 ? 1 : -1];
+   typedef char stbtt__check_size16[sizeof(stbtt_int16)==2 ? 1 : -1];
+
+   // #define your own STBTT_sort() to override this to avoid qsort
+   #ifndef STBTT_sort
+   #include <stdlib.h>
+   #define STBTT_sort(data,num_items,item_size,compare_func)   qsort(data,num_items,item_size,compare_func)
+   #endif
+
+   // #define your own STBTT_ifloor/STBTT_iceil() to avoid math.h
+   #ifndef STBTT_ifloor
+   #include <math.h>
+   #define STBTT_ifloor(x)   ((int) floor(x))
+   #define STBTT_iceil(x)    ((int) ceil(x))
+   #endif
+
+   #ifndef STBTT_sqrt
+   #include <math.h>
+   #define STBTT_sqrt(x)      sqrt(x)
+   #endif
+
+   // #define your own functions "STBTT_malloc" / "STBTT_free" to avoid malloc.h
+   #ifndef STBTT_malloc
+   #include <stdlib.h>
+   #define STBTT_malloc(x,u)  ((void)(u),malloc(x))
+   #define STBTT_free(x,u)    free(x)
+   #endif
+
+   #ifndef STBTT_assert
+   #include <assert.h>
+   #define STBTT_assert(x)    assert(x)
+   #endif
+
+   #ifndef STBTT_strlen
+   #include <string.h>
+   #define STBTT_strlen(x)    strlen(x)
+   #endif
+
+   #ifndef STBTT_memcpy
+   #include <memory.h>
+   #define STBTT_memcpy       memcpy
+   #define STBTT_memset       memset
+   #endif
+#endif
+
+///////////////////////////////////////////////////////////////////////////////
+///////////////////////////////////////////////////////////////////////////////
+////
+////   INTERFACE
+////
+////
+
+#ifndef __STB_INCLUDE_STB_TRUETYPE_H__
+#define __STB_INCLUDE_STB_TRUETYPE_H__
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// TEXTURE BAKING API
+//
+// If you use this API, you only have to call two functions ever.
+//
+
+typedef struct
+{
+   unsigned short x0,y0,x1,y1; // coordinates of bbox in bitmap
+   float xoff,yoff,xadvance;   
+} stbtt_bakedchar;
+
+extern int stbtt_BakeFontBitmap(const unsigned char *data, int offset,  // font location (use offset=0 for plain .ttf)
+                                float pixel_height,                     // height of font in pixels
+                                unsigned char *pixels, int pw, int ph,  // bitmap to be filled in
+                                int first_char, int num_chars,          // characters to bake
+                                stbtt_bakedchar *chardata);             // you allocate this, it's num_chars long
+// if return is positive, the first unused row of the bitmap
+// if return is negative, returns the negative of the number of characters that fit
+// if return is 0, no characters fit and no rows were used
+// This uses a very crappy packing.
+
+typedef struct
+{
+   float x0,y0,s0,t0; // top-left
+   float x1,y1,s1,t1; // bottom-right
+} stbtt_aligned_quad;
+
+extern void stbtt_GetBakedQuad(stbtt_bakedchar *chardata, int pw, int ph,  // same data as above
+                               int char_index,             // character to display
+                               float *xpos, float *ypos,   // pointers to current position in screen pixel space
+                               stbtt_aligned_quad *q,      // output: quad to draw
+                               int opengl_fillrule);       // true if opengl fill rule; false if DX9 or earlier
+// Call GetBakedQuad with char_index = 'character - first_char', and it
+// creates the quad you need to draw and advances the current position.
+//
+// The coordinate system used assumes y increases downwards.
+//
+// Characters will extend both above and below the current position;
+// see discussion of "BASELINE" above.
+//
+// It's inefficient; you might want to c&p it and optimize it.
+
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// FONT LOADING
+//
+//
+
+extern int stbtt_GetFontOffsetForIndex(const unsigned char *data, int index);
+// Each .ttf/.ttc file may have more than one font. Each font has a sequential
+// index number starting from 0. Call this function to get the font offset for
+// a given index; it returns -1 if the index is out of range. A regular .ttf
+// file will only define one font and it always be at offset 0, so it will
+// return '0' for index 0, and -1 for all other indices. You can just skip
+// this step if you know it's that kind of font.
+
+
+// The following structure is defined publically so you can declare one on
+// the stack or as a global or etc, but you should treat it as opaque.
+typedef struct stbtt_fontinfo
+{
+   void           * userdata;
+   unsigned char  * data;              // pointer to .ttf file
+   int              fontstart;         // offset of start of font
+
+   int numGlyphs;                     // number of glyphs, needed for range checking
+
+   int loca,head,glyf,hhea,hmtx,kern; // table locations as offset from start of .ttf
+   int index_map;                     // a cmap mapping for our chosen character encoding
+   int indexToLocFormat;              // format needed to map from glyph index to glyph
+} stbtt_fontinfo;
+
+extern int stbtt_InitFont(stbtt_fontinfo *info, const unsigned char *data, int offset);
+// Given an offset into the file that defines a font, this function builds
+// the necessary cached info for the rest of the system. You must allocate
+// the stbtt_fontinfo yourself, and stbtt_InitFont will fill it out. You don't
+// need to do anything special to free it, because the contents are pure
+// value data with no additional data structures. Returns 0 on failure.
+
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// CHARACTER TO GLYPH-INDEX CONVERSIOn
+
+int stbtt_FindGlyphIndex(const stbtt_fontinfo *info, int unicode_codepoint);
+// If you're going to perform multiple operations on the same character
+// and you want a speed-up, call this function with the character you're
+// going to process, then use glyph-based functions instead of the
+// codepoint-based functions.
+
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// CHARACTER PROPERTIES
+//
+
+extern float stbtt_ScaleForPixelHeight(const stbtt_fontinfo *info, float pixels);
+// computes a scale factor to produce a font whose "height" is 'pixels' tall.
+// Height is measured as the distance from the highest ascender to the lowest
+// descender; in other words, it's equivalent to calling stbtt_GetFontVMetrics
+// and computing:
+//       scale = pixels / (ascent - descent)
+// so if you prefer to measure height by the ascent only, use a similar calculation.
+
+extern float stbtt_ScaleForMappingEmToPixels(const stbtt_fontinfo *info, float pixels);
+// computes a scale factor to produce a font whose EM size is mapped to
+// 'pixels' tall. This is probably what traditional APIs compute, but
+// I'm not positive.
+
+extern void stbtt_GetFontVMetrics(const stbtt_fontinfo *info, int *ascent, int *descent, int *lineGap);
+// ascent is the coordinate above the baseline the font extends; descent
+// is the coordinate below the baseline the font extends (i.e. it is typically negative)
+// lineGap is the spacing between one row's descent and the next row's ascent...
+// so you should advance the vertical position by "*ascent - *descent + *lineGap"
+//   these are expressed in unscaled coordinates, so you must multiply by
+//   the scale factor for a given size
+
+extern void stbtt_GetFontBoundingBox(const stbtt_fontinfo *info, int *x0, int *y0, int *x1, int *y1);
+// the bounding box around all possible characters
+
+extern void stbtt_GetCodepointHMetrics(const stbtt_fontinfo *info, int codepoint, int *advanceWidth, int *leftSideBearing);
+// leftSideBearing is the offset from the current horizontal position to the left edge of the character
+// advanceWidth is the offset from the current horizontal position to the next horizontal position
+//   these are expressed in unscaled coordinates
+
+extern int  stbtt_GetCodepointKernAdvance(const stbtt_fontinfo *info, int ch1, int ch2);
+// an additional amount to add to the 'advance' value between ch1 and ch2
+
+extern int stbtt_GetCodepointBox(const stbtt_fontinfo *info, int codepoint, int *x0, int *y0, int *x1, int *y1);
+// Gets the bounding box of the visible part of the glyph, in unscaled coordinates
+
+extern void stbtt_GetGlyphHMetrics(const stbtt_fontinfo *info, int glyph_index, int *advanceWidth, int *leftSideBearing);
+extern int  stbtt_GetGlyphKernAdvance(const stbtt_fontinfo *info, int glyph1, int glyph2);
+extern int  stbtt_GetGlyphBox(const stbtt_fontinfo *info, int glyph_index, int *x0, int *y0, int *x1, int *y1);
+// as above, but takes one or more glyph indices for greater efficiency
+
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// GLYPH SHAPES (you probably don't need these, but they have to go before
+// the bitmaps for C declaration-order reasons)
+//
+
+#ifndef STBTT_vmove // you can predefine these to use different values (but why?)
+   enum {
+      STBTT_vmove=1,
+      STBTT_vline,
+      STBTT_vcurve
+   };
+#endif
+
+#ifndef stbtt_vertex // you can predefine this to use different values
+                   // (we share this with other code at RAD)
+   #define stbtt_vertex_type short // can't use stbtt_int16 because that's not visible in the header file
+   typedef struct
+   {
+      stbtt_vertex_type x,y,cx,cy;
+      unsigned char type,padding;
+   } stbtt_vertex;
+#endif
+
+extern int stbtt_IsGlyphEmpty(const stbtt_fontinfo *info, int glyph_index);
+// returns non-zero if nothing is drawn for this glyph
+
+extern int stbtt_GetCodepointShape(const stbtt_fontinfo *info, int unicode_codepoint, stbtt_vertex **vertices);
+extern int stbtt_GetGlyphShape(const stbtt_fontinfo *info, int glyph_index, stbtt_vertex **vertices);
+// returns # of vertices and fills *vertices with the pointer to them
+//   these are expressed in "unscaled" coordinates
+//
+// The shape is a series of countours. Each one starts with
+// a STBTT_moveto, then consists of a series of mixed
+// STBTT_lineto and STBTT_curveto segments. A lineto
+// draws a line from previous endpoint to its x,y; a curveto
+// draws a quadratic bezier from previous endpoint to
+// its x,y, using cx,cy as the bezier control point.
+
+extern void stbtt_FreeShape(const stbtt_fontinfo *info, stbtt_vertex *vertices);
+// frees the data allocated above
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// BITMAP RENDERING
+//
+
+extern void stbtt_FreeBitmap(unsigned char *bitmap, void *userdata);
+// frees the bitmap allocated below
+
+extern unsigned char *stbtt_GetCodepointBitmap(const stbtt_fontinfo *info, float scale_x, float scale_y, int codepoint, int *width, int *height, int *xoff, int *yoff);
+// allocates a large-enough single-channel 8bpp bitmap and renders the
+// specified character/glyph at the specified scale into it, with
+// antialiasing. 0 is no coverage (transparent), 255 is fully covered (opaque).
+// *width & *height are filled out with the width & height of the bitmap,
+// which is stored left-to-right, top-to-bottom.
+//
+// xoff/yoff are the offset it pixel space from the glyph origin to the top-left of the bitmap
+
+extern unsigned char *stbtt_GetCodepointBitmapSubpixel(const stbtt_fontinfo *info, float scale_x, float scale_y, float shift_x, float shift_y, int codepoint, int *width, int *height, int *xoff, int *yoff);
+// the same as stbtt_GetCodepoitnBitmap, but you can specify a subpixel
+// shift for the character
+
+extern void stbtt_MakeCodepointBitmap(const stbtt_fontinfo *info, unsigned char *output, int out_w, int out_h, int out_stride, float scale_x, float scale_y, int codepoint);
+// the same as stbtt_GetCodepointBitmap, but you pass in storage for the bitmap
+// in the form of 'output', with row spacing of 'out_stride' bytes. the bitmap
+// is clipped to out_w/out_h bytes. Call stbtt_GetCodepointBitmapBox to get the
+// width and height and positioning info for it first.
+
+extern void stbtt_MakeCodepointBitmapSubpixel(const stbtt_fontinfo *info, unsigned char *output, int out_w, int out_h, int out_stride, float scale_x, float scale_y, float shift_x, float shift_y, int codepoint);
+// same as stbtt_MakeCodepointBitmap, but you can specify a subpixel
+// shift for the character
+
+extern void stbtt_GetCodepointBitmapBox(const stbtt_fontinfo *font, int codepoint, float scale_x, float scale_y, int *ix0, int *iy0, int *ix1, int *iy1);
+// get the bbox of the bitmap centered around the glyph origin; so the
+// bitmap width is ix1-ix0, height is iy1-iy0, and location to place
+// the bitmap top left is (leftSideBearing*scale,iy0).
+// (Note that the bitmap uses y-increases-down, but the shape uses
+// y-increases-up, so CodepointBitmapBox and CodepointBox are inverted.)
+
+extern void stbtt_GetCodepointBitmapBoxSubpixel(const stbtt_fontinfo *font, int codepoint, float scale_x, float scale_y, float shift_x, float shift_y, int *ix0, int *iy0, int *ix1, int *iy1);
+// same as stbtt_GetCodepointBitmapBox, but you can specify a subpixel
+// shift for the character
+
+// the following functions are equivalent to the above functions, but operate
+// on glyph indices instead of Unicode codepoints (for efficiency)
+extern unsigned char *stbtt_GetGlyphBitmap(const stbtt_fontinfo *info, float scale_x, float scale_y, int glyph, int *width, int *height, int *xoff, int *yoff);
+extern unsigned char *stbtt_GetGlyphBitmapSubpixel(const stbtt_fontinfo *info, float scale_x, float scale_y, float shift_x, float shift_y, int glyph, int *width, int *height, int *xoff, int *yoff);
+extern void stbtt_MakeGlyphBitmap(const stbtt_fontinfo *info, unsigned char *output, int out_w, int out_h, int out_stride, float scale_x, float scale_y, int glyph);
+extern void stbtt_MakeGlyphBitmapSubpixel(const stbtt_fontinfo *info, unsigned char *output, int out_w, int out_h, int out_stride, float scale_x, float scale_y, float shift_x, float shift_y, int glyph);
+extern void stbtt_GetGlyphBitmapBox(const stbtt_fontinfo *font, int glyph, float scale_x, float scale_y, int *ix0, int *iy0, int *ix1, int *iy1);
+extern void stbtt_GetGlyphBitmapBoxSubpixel(const stbtt_fontinfo *font, int glyph, float scale_x, float scale_y,float shift_x, float shift_y, int *ix0, int *iy0, int *ix1, int *iy1);
+
+
+// @TODO: don't expose this structure
+typedef struct
+{
+   int w,h,stride;
+   unsigned char *pixels;
+} stbtt__bitmap;
+
+extern void stbtt_Rasterize(stbtt__bitmap *result, float flatness_in_pixels, stbtt_vertex *vertices, int num_verts, float scale_x, float scale_y, float shift_x, float shift_y, int x_off, int y_off, int invert, void *userdata);
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// Finding the right font...
+//
+// You should really just solve this offline, keep your own tables
+// of what font is what, and don't try to get it out of the .ttf file.
+// That's because getting it out of the .ttf file is really hard, because
+// the names in the file can appear in many possible encodings, in many
+// possible languages, and e.g. if you need a case-insensitive comparison,
+// the details of that depend on the encoding & language in a complex way
+// (actually underspecified in truetype, but also gigantic).
+//
+// But you can use the provided functions in two possible ways:
+//     stbtt_FindMatchingFont() will use *case-sensitive* comparisons on
+//             unicode-encoded names to try to find the font you want;
+//             you can run this before calling stbtt_InitFont()
+//
+//     stbtt_GetFontNameString() lets you get any of the various strings
+//             from the file yourself and do your own comparisons on them.
+//             You have to have called stbtt_InitFont() first.
+
+
+extern int stbtt_FindMatchingFont(const unsigned char *fontdata, const char *name, int flags);
+// returns the offset (not index) of the font that matches, or -1 if none
+//   if you use STBTT_MACSTYLE_DONTCARE, use a font name like "Arial Bold".
+//   if you use any other flag, use a font name like "Arial"; this checks
+//     the 'macStyle' header field; i don't know if fonts set this consistently
+#define STBTT_MACSTYLE_DONTCARE     0
+#define STBTT_MACSTYLE_BOLD         1
+#define STBTT_MACSTYLE_ITALIC       2
+#define STBTT_MACSTYLE_UNDERSCORE   4
+#define STBTT_MACSTYLE_NONE         8   // <= not same as 0, this makes us check the bitfield is 0
+
+extern int stbtt_CompareUTF8toUTF16_bigendian(const char *s1, int len1, const char *s2, int len2);
+// returns 1/0 whether the first string interpreted as utf8 is identical to
+// the second string interpreted as big-endian utf16... useful for strings from next func
+
+extern const char *stbtt_GetFontNameString(const stbtt_fontinfo *font, int *length, int platformID, int encodingID, int languageID, int nameID);
+// returns the string (which may be big-endian double byte, e.g. for unicode)
+// and puts the length in bytes in *length.
+//
+// some of the values for the IDs are below; for more see the truetype spec:
+//     http://developer.apple.com/textfonts/TTRefMan/RM06/Chap6name.html
+//     http://www.microsoft.com/typography/otspec/name.htm
+
+enum { // platformID
+   STBTT_PLATFORM_ID_UNICODE   =0,
+   STBTT_PLATFORM_ID_MAC       =1,
+   STBTT_PLATFORM_ID_ISO       =2,
+   STBTT_PLATFORM_ID_MICROSOFT =3
+};
+
+enum { // encodingID for STBTT_PLATFORM_ID_UNICODE
+   STBTT_UNICODE_EID_UNICODE_1_0    =0,
+   STBTT_UNICODE_EID_UNICODE_1_1    =1,
+   STBTT_UNICODE_EID_ISO_10646      =2,
+   STBTT_UNICODE_EID_UNICODE_2_0_BMP=3,
+   STBTT_UNICODE_EID_UNICODE_2_0_FULL=4
+};
+
+enum { // encodingID for STBTT_PLATFORM_ID_MICROSOFT
+   STBTT_MS_EID_SYMBOL        =0,
+   STBTT_MS_EID_UNICODE_BMP   =1,
+   STBTT_MS_EID_SHIFTJIS      =2,
+   STBTT_MS_EID_UNICODE_FULL  =10
+};
+
+enum { // encodingID for STBTT_PLATFORM_ID_MAC; same as Script Manager codes
+   STBTT_MAC_EID_ROMAN        =0,   STBTT_MAC_EID_ARABIC       =4,
+   STBTT_MAC_EID_JAPANESE     =1,   STBTT_MAC_EID_HEBREW       =5,
+   STBTT_MAC_EID_CHINESE_TRAD =2,   STBTT_MAC_EID_GREEK        =6,
+   STBTT_MAC_EID_KOREAN       =3,   STBTT_MAC_EID_RUSSIAN      =7
+};
+
+enum { // languageID for STBTT_PLATFORM_ID_MICROSOFT; same as LCID...
+       // problematic because there are e.g. 16 english LCIDs and 16 arabic LCIDs
+   STBTT_MS_LANG_ENGLISH     =0x0409,   STBTT_MS_LANG_ITALIAN     =0x0410,
+   STBTT_MS_LANG_CHINESE     =0x0804,   STBTT_MS_LANG_JAPANESE    =0x0411,
+   STBTT_MS_LANG_DUTCH       =0x0413,   STBTT_MS_LANG_KOREAN      =0x0412,
+   STBTT_MS_LANG_FRENCH      =0x040c,   STBTT_MS_LANG_RUSSIAN     =0x0419,
+   STBTT_MS_LANG_GERMAN      =0x0407,   STBTT_MS_LANG_SPANISH     =0x0409,
+   STBTT_MS_LANG_HEBREW      =0x040d,   STBTT_MS_LANG_SWEDISH     =0x041D
+};
+
+enum { // languageID for STBTT_PLATFORM_ID_MAC
+   STBTT_MAC_LANG_ENGLISH      =0 ,   STBTT_MAC_LANG_JAPANESE     =11,
+   STBTT_MAC_LANG_ARABIC       =12,   STBTT_MAC_LANG_KOREAN       =23,
+   STBTT_MAC_LANG_DUTCH        =4 ,   STBTT_MAC_LANG_RUSSIAN      =32,
+   STBTT_MAC_LANG_FRENCH       =1 ,   STBTT_MAC_LANG_SPANISH      =6 ,
+   STBTT_MAC_LANG_GERMAN       =2 ,   STBTT_MAC_LANG_SWEDISH      =5 ,
+   STBTT_MAC_LANG_HEBREW       =10,   STBTT_MAC_LANG_CHINESE_SIMPLIFIED =33,
+   STBTT_MAC_LANG_ITALIAN      =3 ,   STBTT_MAC_LANG_CHINESE_TRAD =19
+};
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif // __STB_INCLUDE_STB_TRUETYPE_H__
+
+///////////////////////////////////////////////////////////////////////////////
+///////////////////////////////////////////////////////////////////////////////
+////
+////   IMPLEMENTATION
+////
+////
+
+#ifdef STB_TRUETYPE_IMPLEMENTATION
+
+//////////////////////////////////////////////////////////////////////////
+//
+// accessors to parse data from file
+//
+
+// on platforms that don't allow misaligned reads, if we want to allow
+// truetype fonts that aren't padded to alignment, define ALLOW_UNALIGNED_TRUETYPE
+
+#define ttBYTE(p)     (* (stbtt_uint8 *) (p))
+#define ttCHAR(p)     (* (stbtt_int8 *) (p))
+#define ttFixed(p)    ttLONG(p)
+
+#if defined(STB_TRUETYPE_BIGENDIAN) && !defined(ALLOW_UNALIGNED_TRUETYPE)
+
+   #define ttUSHORT(p)   (* (stbtt_uint16 *) (p))
+   #define ttSHORT(p)    (* (stbtt_int16 *) (p))
+   #define ttULONG(p)    (* (stbtt_uint32 *) (p))
+   #define ttLONG(p)     (* (stbtt_int32 *) (p))
+
+#else
+
+   stbtt_uint16 ttUSHORT(const stbtt_uint8 *p) { return p[0]*256 + p[1]; }
+   stbtt_int16 ttSHORT(const stbtt_uint8 *p)   { return p[0]*256 + p[1]; }
+   stbtt_uint32 ttULONG(const stbtt_uint8 *p)  { return (p[0]<<24) + (p[1]<<16) + (p[2]<<8) + p[3]; }
+   stbtt_int32 ttLONG(const stbtt_uint8 *p)    { return (p[0]<<24) + (p[1]<<16) + (p[2]<<8) + p[3]; }
+
+#endif
+
+#define stbtt_tag4(p,c0,c1,c2,c3) ((p)[0] == (c0) && (p)[1] == (c1) && (p)[2] == (c2) && (p)[3] == (c3))
+#define stbtt_tag(p,str)           stbtt_tag4(p,str[0],str[1],str[2],str[3])
+
+static int stbtt__isfont(const stbtt_uint8 *font)
+{
+   // check the version number
+   if (stbtt_tag4(font, '1',0,0,0))  return 1; // TrueType 1
+   if (stbtt_tag(font, "typ1"))   return 1; // TrueType with type 1 font -- we don't support this!
+   if (stbtt_tag(font, "OTTO"))   return 1; // OpenType with CFF
+   if (stbtt_tag4(font, 0,1,0,0)) return 1; // OpenType 1.0
+   return 0;
+}
+
+// @OPTIMIZE: binary search
+static stbtt_uint32 stbtt__find_table(stbtt_uint8 *data, stbtt_uint32 fontstart, const char *tag)
+{
+   stbtt_int32 num_tables = ttUSHORT(data+fontstart+4);
+   stbtt_uint32 tabledir = fontstart + 12;
+   stbtt_int32 i;
+   for (i=0; i < num_tables; ++i) {
+      stbtt_uint32 loc = tabledir + 16*i;
+      if (stbtt_tag(data+loc+0, tag))
+         return ttULONG(data+loc+8);
+   }
+   return 0;
+}
+
+int stbtt_GetFontOffsetForIndex(const unsigned char *font_collection, int index)
+{
+   // if it's just a font, there's only one valid index
+   if (stbtt__isfont(font_collection))
+      return index == 0 ? 0 : -1;
+
+   // check if it's a TTC
+   if (stbtt_tag(font_collection, "ttcf")) {
+      // version 1?
+      if (ttULONG(font_collection+4) == 0x00010000 || ttULONG(font_collection+4) == 0x00020000) {
+         stbtt_int32 n = ttLONG(font_collection+8);
+         if (index >= n)
+            return -1;
+         return ttULONG(font_collection+12+index*14);
+      }
+   }
+   return -1;
+}
+
+int stbtt_InitFont(stbtt_fontinfo *info, const unsigned char *data2, int fontstart)
+{
+   stbtt_uint8 *data = (stbtt_uint8 *) data2;
+   stbtt_uint32 cmap, t;
+   stbtt_int32 i,numTables;
+
+   info->data = data;
+   info->fontstart = fontstart;
+
+   cmap = stbtt__find_table(data, fontstart, "cmap");       // required
+   info->loca = stbtt__find_table(data, fontstart, "loca"); // required
+   info->head = stbtt__find_table(data, fontstart, "head"); // required
+   info->glyf = stbtt__find_table(data, fontstart, "glyf"); // required
+   info->hhea = stbtt__find_table(data, fontstart, "hhea"); // required
+   info->hmtx = stbtt__find_table(data, fontstart, "hmtx"); // required
+   info->kern = stbtt__find_table(data, fontstart, "kern"); // not required
+   if (!cmap || !info->loca || !info->head || !info->glyf || !info->hhea || !info->hmtx)
+      return 0;
+
+   t = stbtt__find_table(data, fontstart, "maxp");
+   if (t)
+      info->numGlyphs = ttUSHORT(data+t+4);
+   else
+      info->numGlyphs = 0xffff;
+
+   // find a cmap encoding table we understand *now* to avoid searching
+   // later. (todo: could make this installable)
+   // the same regardless of glyph.
+   numTables = ttUSHORT(data + cmap + 2);
+   info->index_map = 0;
+   for (i=0; i < numTables; ++i) {
+      stbtt_uint32 encoding_record = cmap + 4 + 8 * i;
+      // find an encoding we understand:
+      switch(ttUSHORT(data+encoding_record)) {
+         case STBTT_PLATFORM_ID_MICROSOFT:
+            switch (ttUSHORT(data+encoding_record+2)) {
+               case STBTT_MS_EID_UNICODE_BMP:
+               case STBTT_MS_EID_UNICODE_FULL:
+                  // MS/Unicode
+                  info->index_map = cmap + ttULONG(data+encoding_record+4);
+                  break;
+            }
+            break;
+      }
+   }
+   if (info->index_map == 0)
+      return 0;
+
+   info->indexToLocFormat = ttUSHORT(data+info->head + 50);
+   return 1;
+}
+
+int stbtt_FindGlyphIndex(const stbtt_fontinfo *info, int unicode_codepoint)
+{
+   stbtt_uint8 *data = info->data;
+   stbtt_uint32 index_map = info->index_map;
+
+   stbtt_uint16 format = ttUSHORT(data + index_map + 0);
+   if (format == 0) { // apple byte encoding
+      stbtt_int32 bytes = ttUSHORT(data + index_map + 2);
+      if (unicode_codepoint < bytes-6)
+         return ttBYTE(data + index_map + 6 + unicode_codepoint);
+      return 0;
+   } else if (format == 6) {
+      stbtt_uint32 first = ttUSHORT(data + index_map + 6);
+      stbtt_uint32 count = ttUSHORT(data + index_map + 8);
+      if ((stbtt_uint32) unicode_codepoint >= first && (stbtt_uint32) unicode_codepoint < first+count)
+         return ttUSHORT(data + index_map + 10 + (unicode_codepoint - first)*2);
+      return 0;
+   } else if (format == 2) {
+      STBTT_assert(0); // @TODO: high-byte mapping for japanese/chinese/korean
+      return 0;
+   } else if (format == 4) { // standard mapping for windows fonts: binary search collection of ranges
+      stbtt_uint16 segcount = ttUSHORT(data+index_map+6) >> 1;
+      stbtt_uint16 searchRange = ttUSHORT(data+index_map+8) >> 1;
+      stbtt_uint16 entrySelector = ttUSHORT(data+index_map+10);
+      stbtt_uint16 rangeShift = ttUSHORT(data+index_map+12) >> 1;
+      stbtt_uint16 item, offset, start, end;
+
+      // do a binary search of the segments
+      stbtt_uint32 endCount = index_map + 14;
+      stbtt_uint32 search = endCount;
+
+      if (unicode_codepoint > 0xffff)
+         return 0;
+
+      // they lie from endCount .. endCount + segCount
+      // but searchRange is the nearest power of two, so...
+      if (unicode_codepoint >= ttUSHORT(data + search + rangeShift*2))
+         search += rangeShift*2;
+
+      // now decrement to bias correctly to find smallest
+      search -= 2;
+      while (entrySelector) {
+         searchRange >>= 1;
+         start = ttUSHORT(data + search + searchRange*2 + segcount*2 + 2);
+         end = ttUSHORT(data + search + searchRange*2);
+         if (unicode_codepoint > end)
+            search += searchRange*2;
+         --entrySelector;
+      }
+      search += 2;
+
+      item = (stbtt_uint16) ((search - endCount) >> 1);
+
+      STBTT_assert(unicode_codepoint <= ttUSHORT(data + endCount + 2*item));
+      start = ttUSHORT(data + index_map + 14 + segcount*2 + 2 + 2*item);
+      end = ttUSHORT(data + index_map + 14 + 2 + 2*item);
+      if (unicode_codepoint < start)
+         return 0;
+
+      offset = ttUSHORT(data + index_map + 14 + segcount*6 + 2 + 2*item);
+      if (offset == 0)
+         return (stbtt_uint16) (unicode_codepoint + ttSHORT(data + index_map + 14 + segcount*4 + 2 + 2*item));
+
+      return ttUSHORT(data + offset + (unicode_codepoint-start)*2 + index_map + 14 + segcount*6 + 2 + 2*item);
+   } else if (format == 12 || format == 13) {
+      stbtt_uint32 ngroups = ttULONG(data+index_map+12);
+      stbtt_int32 low,high;
+      low = 0; high = (stbtt_int32)ngroups;
+      // Binary search the right group.
+      while (low < high) {
+         stbtt_int32 mid = low + ((high-low) >> 1); // rounds down, so low <= mid < high
+         stbtt_uint32 start_char = ttULONG(data+index_map+16+mid*12);
+         stbtt_uint32 end_char = ttULONG(data+index_map+16+mid*12+4);
+         if ((stbtt_uint32) unicode_codepoint < start_char)
+            high = mid;
+         else if ((stbtt_uint32) unicode_codepoint > end_char)
+            low = mid+1;
+         else {
+            stbtt_uint32 start_glyph = ttULONG(data+index_map+16+mid*12+8);
+            if (format == 12)
+               return start_glyph + unicode_codepoint-start_char;
+            else // format == 13
+               return start_glyph;
+         }
+      }
+      return 0; // not found
+   }
+   // @TODO
+   STBTT_assert(0);
+   return 0;
+}
+
+int stbtt_GetCodepointShape(const stbtt_fontinfo *info, int unicode_codepoint, stbtt_vertex **vertices)
+{
+   return stbtt_GetGlyphShape(info, stbtt_FindGlyphIndex(info, unicode_codepoint), vertices);
+}
+
+static void stbtt_setvertex(stbtt_vertex *v, stbtt_uint8 type, stbtt_int32 x, stbtt_int32 y, stbtt_int32 cx, stbtt_int32 cy)
+{
+   v->type = type;
+   v->x = (stbtt_int16) x;
+   v->y = (stbtt_int16) y;
+   v->cx = (stbtt_int16) cx;
+   v->cy = (stbtt_int16) cy;
+}
+
+static int stbtt__GetGlyfOffset(const stbtt_fontinfo *info, int glyph_index)
+{
+   int g1,g2;
+
+   if (glyph_index >= info->numGlyphs) return -1; // glyph index out of range
+   if (info->indexToLocFormat >= 2)    return -1; // unknown index->glyph map format
+
+   if (info->indexToLocFormat == 0) {
+      g1 = info->glyf + ttUSHORT(info->data + info->loca + glyph_index * 2) * 2;
+      g2 = info->glyf + ttUSHORT(info->data + info->loca + glyph_index * 2 + 2) * 2;
+   } else {
+      g1 = info->glyf + ttULONG (info->data + info->loca + glyph_index * 4);
+      g2 = info->glyf + ttULONG (info->data + info->loca + glyph_index * 4 + 4);
+   }
+
+   return g1==g2 ? -1 : g1; // if length is 0, return -1
+}
+
+int stbtt_GetGlyphBox(const stbtt_fontinfo *info, int glyph_index, int *x0, int *y0, int *x1, int *y1)
+{
+   int g = stbtt__GetGlyfOffset(info, glyph_index);
+   if (g < 0) return 0;
+
+   if (x0) *x0 = ttSHORT(info->data + g + 2);
+   if (y0) *y0 = ttSHORT(info->data + g + 4);
+   if (x1) *x1 = ttSHORT(info->data + g + 6);
+   if (y1) *y1 = ttSHORT(info->data + g + 8);
+   return 1;
+}
+
+int stbtt_GetCodepointBox(const stbtt_fontinfo *info, int codepoint, int *x0, int *y0, int *x1, int *y1)
+{
+   return stbtt_GetGlyphBox(info, stbtt_FindGlyphIndex(info,codepoint), x0,y0,x1,y1);
+}
+
+int stbtt_IsGlyphEmpty(const stbtt_fontinfo *info, int glyph_index)
+{
+   stbtt_int16 numberOfContours;
+   int g = stbtt__GetGlyfOffset(info, glyph_index);
+   if (g < 0) return 1;
+   numberOfContours = ttSHORT(info->data + g);
+   return numberOfContours == 0;
+}
+
+static int stbtt__close_shape(stbtt_vertex *vertices, int num_vertices, int was_off, int start_off,
+    stbtt_int32 sx, stbtt_int32 sy, stbtt_int32 scx, stbtt_int32 scy, stbtt_int32 cx, stbtt_int32 cy)
+{
+   if (start_off) {
+      if (was_off)
+         stbtt_setvertex(&vertices[num_vertices++], STBTT_vcurve, (cx+scx)>>1, (cy+scy)>>1, cx,cy);
+      stbtt_setvertex(&vertices[num_vertices++], STBTT_vcurve, sx,sy,scx,scy);
+   } else {
+      if (was_off)
+         stbtt_setvertex(&vertices[num_vertices++], STBTT_vcurve,sx,sy,cx,cy);
+      else
+         stbtt_setvertex(&vertices[num_vertices++], STBTT_vline,sx,sy,0,0);
+   }
+   return num_vertices;
+}
+
+int stbtt_GetGlyphShape(const stbtt_fontinfo *info, int glyph_index, stbtt_vertex **pvertices)
+{
+   stbtt_int16 numberOfContours;
+   stbtt_uint8 *endPtsOfContours;
+   stbtt_uint8 *data = info->data;
+   stbtt_vertex *vertices=0;
+   int num_vertices=0;
+   int g = stbtt__GetGlyfOffset(info, glyph_index);
+
+   *pvertices = NULL;
+
+   if (g < 0) return 0;
+
+   numberOfContours = ttSHORT(data + g);
+
+   if (numberOfContours > 0) {
+      stbtt_uint8 flags=0,flagcount;
+      stbtt_int32 ins, i,j=0,m,n, next_move, was_off=0, off, start_off=0;
+      stbtt_int32 x,y,cx,cy,sx,sy, scx,scy;
+      stbtt_uint8 *points;
+      endPtsOfContours = (data + g + 10);
+      ins = ttUSHORT(data + g + 10 + numberOfContours * 2);
+      points = data + g + 10 + numberOfContours * 2 + 2 + ins;
+
+      n = 1+ttUSHORT(endPtsOfContours + numberOfContours*2-2);
+
+      m = n + 2*numberOfContours;  // a loose bound on how many vertices we might need
+      vertices = (stbtt_vertex *) STBTT_malloc(m * sizeof(vertices[0]), info->userdata);
+      if (vertices == 0)
+         return 0;
+
+      next_move = 0;
+      flagcount=0;
+
+      // in first pass, we load uninterpreted data into the allocated array
+      // above, shifted to the end of the array so we won't overwrite it when
+      // we create our final data starting from the front
+
+      off = m - n; // starting offset for uninterpreted data, regardless of how m ends up being calculated
+
+      // first load flags
+
+      for (i=0; i < n; ++i) {
+         if (flagcount == 0) {
+            flags = *points++;
+            if (flags & 8)
+               flagcount = *points++;
+         } else
+            --flagcount;
+         vertices[off+i].type = flags;
+      }
+
+      // now load x coordinates
+      x=0;
+      for (i=0; i < n; ++i) {
+         flags = vertices[off+i].type;
+         if (flags & 2) {
+            stbtt_int16 dx = *points++;
+            x += (flags & 16) ? dx : -dx; // ???
+         } else {
+            if (!(flags & 16)) {
+               x = x + (stbtt_int16) (points[0]*256 + points[1]);
+               points += 2;
+            }
+         }
+         vertices[off+i].x = (stbtt_int16) x;
+      }
+
+      // now load y coordinates
+      y=0;
+      for (i=0; i < n; ++i) {
+         flags = vertices[off+i].type;
+         if (flags & 4) {
+            stbtt_int16 dy = *points++;
+            y += (flags & 32) ? dy : -dy; // ???
+         } else {
+            if (!(flags & 32)) {
+               y = y + (stbtt_int16) (points[0]*256 + points[1]);
+               points += 2;
+            }
+         }
+         vertices[off+i].y = (stbtt_int16) y;
+      }
+
+      // now convert them to our format
+      num_vertices=0;
+      sx = sy = cx = cy = scx = scy = 0;
+      for (i=0; i < n; ++i) {
+         flags = vertices[off+i].type;
+         x     = (stbtt_int16) vertices[off+i].x;
+         y     = (stbtt_int16) vertices[off+i].y;
+
+         if (next_move == i) {
+            if (i != 0)
+               num_vertices = stbtt__close_shape(vertices, num_vertices, was_off, start_off, sx,sy,scx,scy,cx,cy);
+
+            // now start the new one               
+            start_off = !(flags & 1);
+            if (start_off) {
+               // if we start off with an off-curve point, then when we need to find a point on the curve
+               // where we can start, and we need to save some state for when we wraparound.
+               scx = x;
+               scy = y;
+               if (!(vertices[off+i+1].type & 1)) {
+                  // next point is also a curve point, so interpolate an on-point curve
+                  sx = (x + (stbtt_int32) vertices[off+i+1].x) >> 1;
+                  sy = (y + (stbtt_int32) vertices[off+i+1].y) >> 1;
+               } else {
+                  // otherwise just use the next point as our start point
+                  sx = (stbtt_int32) vertices[off+i+1].x;
+                  sy = (stbtt_int32) vertices[off+i+1].y;
+                  ++i; // we're using point i+1 as the starting point, so skip it
+               }
+            } else {
+               sx = x;
+               sy = y;
+            }
+            stbtt_setvertex(&vertices[num_vertices++], STBTT_vmove,sx,sy,0,0);
+            was_off = 0;
+            next_move = 1 + ttUSHORT(endPtsOfContours+j*2);
+            ++j;
+         } else {
+            if (!(flags & 1)) { // if it's a curve
+               if (was_off) // two off-curve control points in a row means interpolate an on-curve midpoint
+                  stbtt_setvertex(&vertices[num_vertices++], STBTT_vcurve, (cx+x)>>1, (cy+y)>>1, cx, cy);
+               cx = x;
+               cy = y;
+               was_off = 1;
+            } else {
+               if (was_off)
+                  stbtt_setvertex(&vertices[num_vertices++], STBTT_vcurve, x,y, cx, cy);
+               else
+                  stbtt_setvertex(&vertices[num_vertices++], STBTT_vline, x,y,0,0);
+               was_off = 0;
+            }
+         }
+      }
+      num_vertices = stbtt__close_shape(vertices, num_vertices, was_off, start_off, sx,sy,scx,scy,cx,cy);
+   } else if (numberOfContours == -1) {
+      // Compound shapes.
+      int more = 1;
+      stbtt_uint8 *comp = data + g + 10;
+      num_vertices = 0;
+      vertices = 0;
+      while (more) {
+         stbtt_uint16 flags, gidx;
+         int comp_num_verts = 0, i;
+         stbtt_vertex *comp_verts = 0, *tmp = 0;
+         float mtx[6] = {1,0,0,1,0,0}, m, n;
+         
+         flags = ttSHORT(comp); comp+=2;
+         gidx = ttSHORT(comp); comp+=2;
+
+         if (flags & 2) { // XY values
+            if (flags & 1) { // shorts
+               mtx[4] = ttSHORT(comp); comp+=2;
+               mtx[5] = ttSHORT(comp); comp+=2;
+            } else {
+               mtx[4] = ttCHAR(comp); comp+=1;
+               mtx[5] = ttCHAR(comp); comp+=1;
+            }
+         }
+         else {
+            // @TODO handle matching point
+            STBTT_assert(0);
+         }
+         if (flags & (1<<3)) { // WE_HAVE_A_SCALE
+            mtx[0] = mtx[3] = ttSHORT(comp)/16384.0f; comp+=2;
+            mtx[1] = mtx[2] = 0;
+         } else if (flags & (1<<6)) { // WE_HAVE_AN_X_AND_YSCALE
+            mtx[0] = ttSHORT(comp)/16384.0f; comp+=2;
+            mtx[1] = mtx[2] = 0;
+            mtx[3] = ttSHORT(comp)/16384.0f; comp+=2;
+         } else if (flags & (1<<7)) { // WE_HAVE_A_TWO_BY_TWO
+            mtx[0] = ttSHORT(comp)/16384.0f; comp+=2;
+            mtx[1] = ttSHORT(comp)/16384.0f; comp+=2;
+            mtx[2] = ttSHORT(comp)/16384.0f; comp+=2;
+            mtx[3] = ttSHORT(comp)/16384.0f; comp+=2;
+         }
+         
+         // Find transformation scales.
+         m = (float) STBTT_sqrt(mtx[0]*mtx[0] + mtx[1]*mtx[1]);
+         n = (float) STBTT_sqrt(mtx[2]*mtx[2] + mtx[3]*mtx[3]);
+
+         // Get indexed glyph.
+         comp_num_verts = stbtt_GetGlyphShape(info, gidx, &comp_verts);
+         if (comp_num_verts > 0) {
+            // Transform vertices.
+            for (i = 0; i < comp_num_verts; ++i) {
+               stbtt_vertex* v = &comp_verts[i];
+               stbtt_vertex_type x,y;
+               x=v->x; y=v->y;
+               v->x = (stbtt_vertex_type)(m * (mtx[0]*x + mtx[2]*y + mtx[4]));
+               v->y = (stbtt_vertex_type)(n * (mtx[1]*x + mtx[3]*y + mtx[5]));
+               x=v->cx; y=v->cy;
+               v->cx = (stbtt_vertex_type)(m * (mtx[0]*x + mtx[2]*y + mtx[4]));
+               v->cy = (stbtt_vertex_type)(n * (mtx[1]*x + mtx[3]*y + mtx[5]));
+            }
+            // Append vertices.
+            tmp = (stbtt_vertex*)STBTT_malloc((num_vertices+comp_num_verts)*sizeof(stbtt_vertex), info->userdata);
+            if (!tmp) {
+               if (vertices) STBTT_free(vertices, info->userdata);
+               if (comp_verts) STBTT_free(comp_verts, info->userdata);
+               return 0;
+            }
+            if (num_vertices > 0) STBTT_memcpy(tmp, vertices, num_vertices*sizeof(stbtt_vertex));
+            STBTT_memcpy(tmp+num_vertices, comp_verts, comp_num_verts*sizeof(stbtt_vertex));
+            if (vertices) STBTT_free(vertices, info->userdata);
+            vertices = tmp;
+            STBTT_free(comp_verts, info->userdata);
+            num_vertices += comp_num_verts;
+         }
+         // More components ?
+         more = flags & (1<<5);
+      }
+   } else if (numberOfContours < 0) {
+      // @TODO other compound variations?
+      STBTT_assert(0);
+   } else {
+      // numberOfCounters == 0, do nothing
+   }
+
+   *pvertices = vertices;
+   return num_vertices;
+}
+
+void stbtt_GetGlyphHMetrics(const stbtt_fontinfo *info, int glyph_index, int *advanceWidth, int *leftSideBearing)
+{
+   stbtt_uint16 numOfLongHorMetrics = ttUSHORT(info->data+info->hhea + 34);
+   if (glyph_index < numOfLongHorMetrics) {
+      if (advanceWidth)     *advanceWidth    = ttSHORT(info->data + info->hmtx + 4*glyph_index);
+      if (leftSideBearing)  *leftSideBearing = ttSHORT(info->data + info->hmtx + 4*glyph_index + 2);
+   } else {
+      if (advanceWidth)     *advanceWidth    = ttSHORT(info->data + info->hmtx + 4*(numOfLongHorMetrics-1));
+      if (leftSideBearing)  *leftSideBearing = ttSHORT(info->data + info->hmtx + 4*numOfLongHorMetrics + 2*(glyph_index - numOfLongHorMetrics));
+   }
+}
+
+int  stbtt_GetGlyphKernAdvance(const stbtt_fontinfo *info, int glyph1, int glyph2)
+{
+   stbtt_uint8 *data = info->data + info->kern;
+   stbtt_uint32 needle, straw;
+   int l, r, m;
+
+   // we only look at the first table. it must be 'horizontal' and format 0.
+   if (!info->kern)
+      return 0;
+   if (ttUSHORT(data+2) < 1) // number of tables, need at least 1
+      return 0;
+   if (ttUSHORT(data+8) != 1) // horizontal flag must be set in format
+      return 0;
+
+   l = 0;
+   r = ttUSHORT(data+10) - 1;
+   needle = glyph1 << 16 | glyph2;
+   while (l <= r) {
+      m = (l + r) >> 1;
+      straw = ttULONG(data+18+(m*6)); // note: unaligned read
+      if (needle < straw)
+         r = m - 1;
+      else if (needle > straw)
+         l = m + 1;
+      else
+         return ttSHORT(data+22+(m*6));
+   }
+   return 0;
+}
+
+int  stbtt_GetCodepointKernAdvance(const stbtt_fontinfo *info, int ch1, int ch2)
+{
+   if (!info->kern) // if no kerning table, don't waste time looking up both codepoint->glyphs
+      return 0;
+   return stbtt_GetGlyphKernAdvance(info, stbtt_FindGlyphIndex(info,ch1), stbtt_FindGlyphIndex(info,ch2));
+}
+
+void stbtt_GetCodepointHMetrics(const stbtt_fontinfo *info, int codepoint, int *advanceWidth, int *leftSideBearing)
+{
+   stbtt_GetGlyphHMetrics(info, stbtt_FindGlyphIndex(info,codepoint), advanceWidth, leftSideBearing);
+}
+
+void stbtt_GetFontVMetrics(const stbtt_fontinfo *info, int *ascent, int *descent, int *lineGap)
+{
+   if (ascent ) *ascent  = ttSHORT(info->data+info->hhea + 4);
+   if (descent) *descent = ttSHORT(info->data+info->hhea + 6);
+   if (lineGap) *lineGap = ttSHORT(info->data+info->hhea + 8);
+}
+
+void stbtt_GetFontBoundingBox(const stbtt_fontinfo *info, int *x0, int *y0, int *x1, int *y1)
+{
+   *x0 = ttSHORT(info->data + info->head + 36);
+   *y0 = ttSHORT(info->data + info->head + 38);
+   *x1 = ttSHORT(info->data + info->head + 40);
+   *y1 = ttSHORT(info->data + info->head + 42);
+}
+
+float stbtt_ScaleForPixelHeight(const stbtt_fontinfo *info, float height)
+{
+   int fheight = ttSHORT(info->data + info->hhea + 4) - ttSHORT(info->data + info->hhea + 6);
+   return (float) height / fheight;
+}
+
+float stbtt_ScaleForMappingEmToPixels(const stbtt_fontinfo *info, float pixels)
+{
+   int unitsPerEm = ttUSHORT(info->data + info->head + 18);
+   return pixels / unitsPerEm;
+}
+
+void stbtt_FreeShape(const stbtt_fontinfo *info, stbtt_vertex *v)
+{
+   STBTT_free(v, info->userdata);
+}
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// antialiasing software rasterizer
+//
+
+void stbtt_GetGlyphBitmapBoxSubpixel(const stbtt_fontinfo *font, int glyph, float scale_x, float scale_y,float shift_x, float shift_y, int *ix0, int *iy0, int *ix1, int *iy1)
+{
+   int x0,y0,x1,y1;
+   if (!stbtt_GetGlyphBox(font, glyph, &x0,&y0,&x1,&y1))
+      x0=y0=x1=y1=0; // e.g. space character
+   // now move to integral bboxes (treating pixels as little squares, what pixels get touched)?
+   if (ix0) *ix0 =  STBTT_ifloor(x0 * scale_x + shift_x);
+   if (iy0) *iy0 = -STBTT_iceil (y1 * scale_y + shift_y);
+   if (ix1) *ix1 =  STBTT_iceil (x1 * scale_x + shift_x);
+   if (iy1) *iy1 = -STBTT_ifloor(y0 * scale_y + shift_y);
+}
+void stbtt_GetGlyphBitmapBox(const stbtt_fontinfo *font, int glyph, float scale_x, float scale_y, int *ix0, int *iy0, int *ix1, int *iy1)
+{
+   stbtt_GetGlyphBitmapBoxSubpixel(font, glyph, scale_x, scale_y,0.0f,0.0f, ix0, iy0, ix1, iy1);
+}
+
+void stbtt_GetCodepointBitmapBoxSubpixel(const stbtt_fontinfo *font, int codepoint, float scale_x, float scale_y, float shift_x, float shift_y, int *ix0, int *iy0, int *ix1, int *iy1)
+{
+   stbtt_GetGlyphBitmapBoxSubpixel(font, stbtt_FindGlyphIndex(font,codepoint), scale_x, scale_y,shift_x,shift_y, ix0,iy0,ix1,iy1);
+}
+
+void stbtt_GetCodepointBitmapBox(const stbtt_fontinfo *font, int codepoint, float scale_x, float scale_y, int *ix0, int *iy0, int *ix1, int *iy1)
+{
+   stbtt_GetCodepointBitmapBoxSubpixel(font, codepoint, scale_x, scale_y,0.0f,0.0f, ix0,iy0,ix1,iy1);
+}
+
+typedef struct stbtt__edge {
+   float x0,y0, x1,y1;
+   int invert;
+} stbtt__edge;
+
+typedef struct stbtt__active_edge
+{
+   int x,dx;
+   float ey;
+   struct stbtt__active_edge *next;
+   int valid;
+} stbtt__active_edge;
+
+#define FIXSHIFT   10
+#define FIX        (1 << FIXSHIFT)
+#define FIXMASK    (FIX-1)
+
+static stbtt__active_edge *new_active(stbtt__edge *e, int off_x, float start_point, void *userdata)
+{
+   stbtt__active_edge *z = (stbtt__active_edge *) STBTT_malloc(sizeof(*z), userdata); // @TODO: make a pool of these!!!
+   float dxdy = (e->x1 - e->x0) / (e->y1 - e->y0);
+   STBTT_assert(e->y0 <= start_point);
+   if (!z) return z;
+   // round dx down to avoid going too far
+   if (dxdy < 0)
+      z->dx = -STBTT_ifloor(FIX * -dxdy);
+   else
+      z->dx = STBTT_ifloor(FIX * dxdy);
+   z->x = STBTT_ifloor(FIX * (e->x0 + dxdy * (start_point - e->y0)));
+   z->x -= off_x * FIX;
+   z->ey = e->y1;
+   z->next = 0;
+   z->valid = e->invert ? 1 : -1;
+   return z;
+}
+
+// note: this routine clips fills that extend off the edges... ideally this
+// wouldn't happen, but it could happen if the truetype glyph bounding boxes
+// are wrong, or if the user supplies a too-small bitmap
+static void stbtt__fill_active_edges(unsigned char *scanline, int len, stbtt__active_edge *e, int max_weight)
+{
+   // non-zero winding fill
+   int x0=0, w=0;
+
+   while (e) {
+      if (w == 0) {
+         // if we're currently at zero, we need to record the edge start point
+         x0 = e->x; w += e->valid;
+      } else {
+         int x1 = e->x; w += e->valid;
+         // if we went to zero, we need to draw
+         if (w == 0) {
+            int i = x0 >> FIXSHIFT;
+            int j = x1 >> FIXSHIFT;
+
+            if (i < len && j >= 0) {
+               if (i == j) {
+                  // x0,x1 are the same pixel, so compute combined coverage
+                  scanline[i] = scanline[i] + (stbtt_uint8) ((x1 - x0) * max_weight >> FIXSHIFT);
+               } else {
+                  if (i >= 0) // add antialiasing for x0
+                     scanline[i] = scanline[i] + (stbtt_uint8) (((FIX - (x0 & FIXMASK)) * max_weight) >> FIXSHIFT);
+                  else
+                     i = -1; // clip
+
+                  if (j < len) // add antialiasing for x1
+                     scanline[j] = scanline[j] + (stbtt_uint8) (((x1 & FIXMASK) * max_weight) >> FIXSHIFT);
+                  else
+                     j = len; // clip
+
+                  for (++i; i < j; ++i) // fill pixels between x0 and x1
+                     scanline[i] = scanline[i] + (stbtt_uint8) max_weight;
+               }
+            }
+         }
+      }
+      
+      e = e->next;
+   }
+}
+
+static void stbtt__rasterize_sorted_edges(stbtt__bitmap *result, stbtt__edge *e, int n, int vsubsample, int off_x, int off_y, void *userdata)
+{
+   stbtt__active_edge *active = NULL;
+   int y,j=0;
+   int max_weight = (255 / vsubsample);  // weight per vertical scanline
+   int s; // vertical subsample index
+   unsigned char scanline_data[512], *scanline;
+
+   if (result->w > 512)
+      scanline = (unsigned char *) STBTT_malloc(result->w, userdata);
+   else
+      scanline = scanline_data;
+
+   y = off_y * vsubsample;
+   e[n].y0 = (off_y + result->h) * (float) vsubsample + 1;
+
+   while (j < result->h) {
+      STBTT_memset(scanline, 0, result->w);
+      for (s=0; s < vsubsample; ++s) {
+         // find center of pixel for this scanline
+         float scan_y = y + 0.5f;
+         stbtt__active_edge **step = &active;
+
+         // update all active edges;
+         // remove all active edges that terminate before the center of this scanline
+         while (*step) {
+            stbtt__active_edge * z = *step;
+            if (z->ey <= scan_y) {
+               *step = z->next; // delete from list
+               STBTT_assert(z->valid);
+               z->valid = 0;
+               STBTT_free(z, userdata);
+            } else {
+               z->x += z->dx; // advance to position for current scanline
+               step = &((*step)->next); // advance through list
+            }
+         }
+
+         // resort the list if needed
+         for(;;) {
+            int changed=0;
+            step = &active;
+            while (*step && (*step)->next) {
+               if ((*step)->x > (*step)->next->x) {
+                  stbtt__active_edge *t = *step;
+                  stbtt__active_edge *q = t->next;
+
+                  t->next = q->next;
+                  q->next = t;
+                  *step = q;
+                  changed = 1;
+               }
+               step = &(*step)->next;
+            }
+            if (!changed) break;
+         }
+
+         // insert all edges that start before the center of this scanline -- omit ones that also end on this scanline
+         while (e->y0 <= scan_y) {
+            if (e->y1 > scan_y) {
+               stbtt__active_edge *z = new_active(e, off_x, scan_y, userdata);
+               // find insertion point
+               if (active == NULL)
+                  active = z;
+               else if (z->x < active->x) {
+                  // insert at front
+                  z->next = active;
+                  active = z;
+               } else {
+                  // find thing to insert AFTER
+                  stbtt__active_edge *p = active;
+                  while (p->next && p->next->x < z->x)
+                     p = p->next;
+                  // at this point, p->next->x is NOT < z->x
+                  z->next = p->next;
+                  p->next = z;
+               }
+            }
+            ++e;
+         }
+
+         // now process all active edges in XOR fashion
+         if (active)
+            stbtt__fill_active_edges(scanline, result->w, active, max_weight);
+
+         ++y;
+      }
+      STBTT_memcpy(result->pixels + j * result->stride, scanline, result->w);
+      ++j;
+   }
+
+   while (active) {
+      stbtt__active_edge *z = active;
+      active = active->next;
+      STBTT_free(z, userdata);
+   }
+
+   if (scanline != scanline_data)
+      STBTT_free(scanline, userdata);
+}
+
+static int stbtt__edge_compare(const void *p, const void *q)
+{
+   stbtt__edge *a = (stbtt__edge *) p;
+   stbtt__edge *b = (stbtt__edge *) q;
+
+   if (a->y0 < b->y0) return -1;
+   if (a->y0 > b->y0) return  1;
+   return 0;
+}
+
+typedef struct
+{
+   float x,y;
+} stbtt__point;
+
+static void stbtt__rasterize(stbtt__bitmap *result, stbtt__point *pts, int *wcount, int windings, float scale_x, float scale_y, float shift_x, float shift_y, int off_x, int off_y, int invert, void *userdata)
+{
+   float y_scale_inv = invert ? -scale_y : scale_y;
+   stbtt__edge *e;
+   int n,i,j,k,m;
+   int vsubsample = result->h < 8 ? 15 : 5;
+   // vsubsample should divide 255 evenly; otherwise we won't reach full opacity
+
+   // now we have to blow out the windings into explicit edge lists
+   n = 0;
+   for (i=0; i < windings; ++i)
+      n += wcount[i];
+
+   e = (stbtt__edge *) STBTT_malloc(sizeof(*e) * (n+1), userdata); // add an extra one as a sentinel
+   if (e == 0) return;
+   n = 0;
+
+   m=0;
+   for (i=0; i < windings; ++i) {
+      stbtt__point *p = pts + m;
+      m += wcount[i];
+      j = wcount[i]-1;
+      for (k=0; k < wcount[i]; j=k++) {
+         int a=k,b=j;
+         // skip the edge if horizontal
+         if (p[j].y == p[k].y)
+            continue;
+         // add edge from j to k to the list
+         e[n].invert = 0;
+         if (invert ? p[j].y > p[k].y : p[j].y < p[k].y) {
+            e[n].invert = 1;
+            a=j,b=k;
+         }
+         e[n].x0 = p[a].x * scale_x + shift_x;
+         e[n].y0 = p[a].y * y_scale_inv * vsubsample + shift_y;
+         e[n].x1 = p[b].x * scale_x + shift_x;
+         e[n].y1 = p[b].y * y_scale_inv * vsubsample + shift_y;
+         ++n;
+      }
+   }
+
+   // now sort the edges by their highest point (should snap to integer, and then by x)
+   STBTT_sort(e, n, sizeof(e[0]), stbtt__edge_compare);
+
+   // now, traverse the scanlines and find the intersections on each scanline, use xor winding rule
+   stbtt__rasterize_sorted_edges(result, e, n, vsubsample, off_x, off_y, userdata);
+
+   STBTT_free(e, userdata);
+}
+
+static void stbtt__add_point(stbtt__point *points, int n, float x, float y)
+{
+   if (!points) return; // during first pass, it's unallocated
+   points[n].x = x;
+   points[n].y = y;
+}
+
+// tesselate until threshhold p is happy... @TODO warped to compensate for non-linear stretching
+static int stbtt__tesselate_curve(stbtt__point *points, int *num_points, float x0, float y0, float x1, float y1, float x2, float y2, float objspace_flatness_squared, int n)
+{
+   // midpoint
+   float mx = (x0 + 2*x1 + x2)/4;
+   float my = (y0 + 2*y1 + y2)/4;
+   // versus directly drawn line
+   float dx = (x0+x2)/2 - mx;
+   float dy = (y0+y2)/2 - my;
+   if (n > 16) // 65536 segments on one curve better be enough!
+      return 1;
+   if (dx*dx+dy*dy > objspace_flatness_squared) { // half-pixel error allowed... need to be smaller if AA
+      stbtt__tesselate_curve(points, num_points, x0,y0, (x0+x1)/2.0f,(y0+y1)/2.0f, mx,my, objspace_flatness_squared,n+1);
+      stbtt__tesselate_curve(points, num_points, mx,my, (x1+x2)/2.0f,(y1+y2)/2.0f, x2,y2, objspace_flatness_squared,n+1);
+   } else {
+      stbtt__add_point(points, *num_points,x2,y2);
+      *num_points = *num_points+1;
+   }
+   return 1;
+}
+
+// returns number of contours
+stbtt__point *stbtt_FlattenCurves(stbtt_vertex *vertices, int num_verts, float objspace_flatness, int **contour_lengths, int *num_contours, void *userdata)
+{
+   stbtt__point *points=0;
+   int num_points=0;
+
+   float objspace_flatness_squared = objspace_flatness * objspace_flatness;
+   int i,n=0,start=0, pass;
+
+   // count how many "moves" there are to get the contour count
+   for (i=0; i < num_verts; ++i)
+      if (vertices[i].type == STBTT_vmove)
+         ++n;
+
+   *num_contours = n;
+   if (n == 0) return 0;
+
+   *contour_lengths = (int *) STBTT_malloc(sizeof(**contour_lengths) * n, userdata);
+
+   if (*contour_lengths == 0) {
+      *num_contours = 0;
+      return 0;
+   }
+
+   // make two passes through the points so we don't need to realloc
+   for (pass=0; pass < 2; ++pass) {
+      float x=0,y=0;
+      if (pass == 1) {
+         points = (stbtt__point *) STBTT_malloc(num_points * sizeof(points[0]), userdata);
+         if (points == NULL) goto error;
+      }
+      num_points = 0;
+      n= -1;
+      for (i=0; i < num_verts; ++i) {
+         switch (vertices[i].type) {
+            case STBTT_vmove:
+               // start the next contour
+               if (n >= 0)
+                  (*contour_lengths)[n] = num_points - start;
+               ++n;
+               start = num_points;
+
+               x = vertices[i].x, y = vertices[i].y;
+               stbtt__add_point(points, num_points++, x,y);
+               break;
+            case STBTT_vline:
+               x = vertices[i].x, y = vertices[i].y;
+               stbtt__add_point(points, num_points++, x, y);
+               break;
+            case STBTT_vcurve:
+               stbtt__tesselate_curve(points, &num_points, x,y,
+                                        vertices[i].cx, vertices[i].cy,
+                                        vertices[i].x,  vertices[i].y,
+                                        objspace_flatness_squared, 0);
+               x = vertices[i].x, y = vertices[i].y;
+               break;
+         }
+      }
+      (*contour_lengths)[n] = num_points - start;
+   }
+
+   return points;
+error:
+   STBTT_free(points, userdata);
+   STBTT_free(*contour_lengths, userdata);
+   *contour_lengths = 0;
+   *num_contours = 0;
+   return NULL;
+}
+
+void stbtt_Rasterize(stbtt__bitmap *result, float flatness_in_pixels, stbtt_vertex *vertices, int num_verts, float scale_x, float scale_y, float shift_x, float shift_y, int x_off, int y_off, int invert, void *userdata)
+{
+   float scale = scale_x > scale_y ? scale_y : scale_x;
+   int winding_count, *winding_lengths;
+   stbtt__point *windings = stbtt_FlattenCurves(vertices, num_verts, flatness_in_pixels / scale, &winding_lengths, &winding_count, userdata);
+   if (windings) {
+      stbtt__rasterize(result, windings, winding_lengths, winding_count, scale_x, scale_y, shift_x, shift_y, x_off, y_off, invert, userdata);
+      STBTT_free(winding_lengths, userdata);
+      STBTT_free(windings, userdata);
+   }
+}
+
+void stbtt_FreeBitmap(unsigned char *bitmap, void *userdata)
+{
+   STBTT_free(bitmap, userdata);
+}
+
+unsigned char *stbtt_GetGlyphBitmapSubpixel(const stbtt_fontinfo *info, float scale_x, float scale_y, float shift_x, float shift_y, int glyph, int *width, int *height, int *xoff, int *yoff)
+{
+   int ix0,iy0,ix1,iy1;
+   stbtt__bitmap gbm;
+   stbtt_vertex *vertices;   
+   int num_verts = stbtt_GetGlyphShape(info, glyph, &vertices);
+
+   if (scale_x == 0) scale_x = scale_y;
+   if (scale_y == 0) {
+      if (scale_x == 0) return NULL;
+      scale_y = scale_x;
+   }
+
+   stbtt_GetGlyphBitmapBoxSubpixel(info, glyph, scale_x, scale_y, shift_x, shift_y, &ix0,&iy0,&ix1,&iy1);
+
+   // now we get the size
+   gbm.w = (ix1 - ix0);
+   gbm.h = (iy1 - iy0);
+   gbm.pixels = NULL; // in case we error
+
+   if (width ) *width  = gbm.w;
+   if (height) *height = gbm.h;
+   if (xoff  ) *xoff   = ix0;
+   if (yoff  ) *yoff   = iy0;
+   
+   if (gbm.w && gbm.h) {
+      gbm.pixels = (unsigned char *) STBTT_malloc(gbm.w * gbm.h, info->userdata);
+      if (gbm.pixels) {
+         gbm.stride = gbm.w;
+
+         stbtt_Rasterize(&gbm, 0.35f, vertices, num_verts, scale_x, scale_y, shift_x, shift_y, ix0, iy0, 1, info->userdata);
+      }
+   }
+   STBTT_free(vertices, info->userdata);
+   return gbm.pixels;
+}   
+
+unsigned char *stbtt_GetGlyphBitmap(const stbtt_fontinfo *info, float scale_x, float scale_y, int glyph, int *width, int *height, int *xoff, int *yoff)
+{
+   return stbtt_GetGlyphBitmapSubpixel(info, scale_x, scale_y, 0.0f, 0.0f, glyph, width, height, xoff, yoff);
+}
+
+void stbtt_MakeGlyphBitmapSubpixel(const stbtt_fontinfo *info, unsigned char *output, int out_w, int out_h, int out_stride, float scale_x, float scale_y, float shift_x, float shift_y, int glyph)
+{
+   int ix0,iy0;
+   stbtt_vertex *vertices;
+   int num_verts = stbtt_GetGlyphShape(info, glyph, &vertices);
+   stbtt__bitmap gbm;   
+
+   stbtt_GetGlyphBitmapBoxSubpixel(info, glyph, scale_x, scale_y, shift_x, shift_y, &ix0,&iy0,0,0);
+   gbm.pixels = output;
+   gbm.w = out_w;
+   gbm.h = out_h;
+   gbm.stride = out_stride;
+
+   if (gbm.w && gbm.h)
+      stbtt_Rasterize(&gbm, 0.35f, vertices, num_verts, scale_x, scale_y, shift_x, shift_y, ix0,iy0, 1, info->userdata);
+
+   STBTT_free(vertices, info->userdata);
+}
+
+void stbtt_MakeGlyphBitmap(const stbtt_fontinfo *info, unsigned char *output, int out_w, int out_h, int out_stride, float scale_x, float scale_y, int glyph)
+{
+   stbtt_MakeGlyphBitmapSubpixel(info, output, out_w, out_h, out_stride, scale_x, scale_y, 0.0f,0.0f, glyph);
+}
+
+unsigned char *stbtt_GetCodepointBitmapSubpixel(const stbtt_fontinfo *info, float scale_x, float scale_y, float shift_x, float shift_y, int codepoint, int *width, int *height, int *xoff, int *yoff)
+{
+   return stbtt_GetGlyphBitmapSubpixel(info, scale_x, scale_y,shift_x,shift_y, stbtt_FindGlyphIndex(info,codepoint), width,height,xoff,yoff);
+}   
+
+void stbtt_MakeCodepointBitmapSubpixel(const stbtt_fontinfo *info, unsigned char *output, int out_w, int out_h, int out_stride, float scale_x, float scale_y, float shift_x, float shift_y, int codepoint)
+{
+   stbtt_MakeGlyphBitmapSubpixel(info, output, out_w, out_h, out_stride, scale_x, scale_y, shift_x, shift_y, stbtt_FindGlyphIndex(info,codepoint));
+}
+
+unsigned char *stbtt_GetCodepointBitmap(const stbtt_fontinfo *info, float scale_x, float scale_y, int codepoint, int *width, int *height, int *xoff, int *yoff)
+{
+   return stbtt_GetCodepointBitmapSubpixel(info, scale_x, scale_y, 0.0f,0.0f, codepoint, width,height,xoff,yoff);
+}   
+
+void stbtt_MakeCodepointBitmap(const stbtt_fontinfo *info, unsigned char *output, int out_w, int out_h, int out_stride, float scale_x, float scale_y, int codepoint)
+{
+   stbtt_MakeCodepointBitmapSubpixel(info, output, out_w, out_h, out_stride, scale_x, scale_y, 0.0f,0.0f, codepoint);
+}
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// bitmap baking
+//
+// This is SUPER-CRAPPY packing to keep source code small
+
+extern int stbtt_BakeFontBitmap(const unsigned char *data, int offset,  // font location (use offset=0 for plain .ttf)
+                                float pixel_height,                     // height of font in pixels
+                                unsigned char *pixels, int pw, int ph,  // bitmap to be filled in
+                                int first_char, int num_chars,          // characters to bake
+                                stbtt_bakedchar *chardata)
+{
+   float scale;
+   int x,y,bottom_y, i;
+   stbtt_fontinfo f;
+   stbtt_InitFont(&f, data, offset);
+   STBTT_memset(pixels, 0, pw*ph); // background of 0 around pixels
+   x=y=1;
+   bottom_y = 1;
+
+   scale = stbtt_ScaleForPixelHeight(&f, pixel_height);
+
+   for (i=0; i < num_chars; ++i) {
+      int advance, lsb, x0,y0,x1,y1,gw,gh;
+      int g = stbtt_FindGlyphIndex(&f, first_char + i);
+      stbtt_GetGlyphHMetrics(&f, g, &advance, &lsb);
+      stbtt_GetGlyphBitmapBox(&f, g, scale,scale, &x0,&y0,&x1,&y1);
+      gw = x1-x0;
+      gh = y1-y0;
+      if (x + gw + 1 >= pw)
+         y = bottom_y, x = 1; // advance to next row
+      if (y + gh + 1 >= ph) // check if it fits vertically AFTER potentially moving to next row
+         return -i;
+      STBTT_assert(x+gw < pw);
+      STBTT_assert(y+gh < ph);
+      stbtt_MakeGlyphBitmap(&f, pixels+x+y*pw, gw,gh,pw, scale,scale, g);
+      chardata[i].x0 = (stbtt_int16) x;
+      chardata[i].y0 = (stbtt_int16) y;
+      chardata[i].x1 = (stbtt_int16) (x + gw);
+      chardata[i].y1 = (stbtt_int16) (y + gh);
+      chardata[i].xadvance = scale * advance;
+      chardata[i].xoff     = (float) x0;
+      chardata[i].yoff     = (float) y0;
+      x = x + gw + 2;
+      if (y+gh+2 > bottom_y)
+         bottom_y = y+gh+2;
+   }
+   return bottom_y;
+}
+
+void stbtt_GetBakedQuad(stbtt_bakedchar *chardata, int pw, int ph, int char_index, float *xpos, float *ypos, stbtt_aligned_quad *q, int opengl_fillrule)
+{
+   float d3d_bias = opengl_fillrule ? 0 : -0.5f;
+   float ipw = 1.0f / pw, iph = 1.0f / ph;
+   stbtt_bakedchar *b = chardata + char_index;
+   int round_x = STBTT_ifloor((*xpos + b->xoff) + 0.5);
+   int round_y = STBTT_ifloor((*ypos + b->yoff) + 0.5);
+
+   q->x0 = round_x + d3d_bias;
+   q->y0 = round_y + d3d_bias;
+   q->x1 = round_x + b->x1 - b->x0 + d3d_bias;
+   q->y1 = round_y + b->y1 - b->y0 + d3d_bias;
+
+   q->s0 = b->x0 * ipw;
+   q->t0 = b->y0 * iph;
+   q->s1 = b->x1 * ipw;
+   q->t1 = b->y1 * iph;
+
+   *xpos += b->xadvance;
+}
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// font name matching -- recommended not to use this
+//
+
+// check if a utf8 string contains a prefix which is the utf16 string; if so return length of matching utf8 string
+static stbtt_int32 stbtt__CompareUTF8toUTF16_bigendian_prefix(const stbtt_uint8 *s1, stbtt_int32 len1, const stbtt_uint8 *s2, stbtt_int32 len2) 
+{
+   stbtt_int32 i=0;
+
+   // convert utf16 to utf8 and compare the results while converting
+   while (len2) {
+      stbtt_uint16 ch = s2[0]*256 + s2[1];
+      if (ch < 0x80) {
+         if (i >= len1) return -1;
+         if (s1[i++] != ch) return -1;
+      } else if (ch < 0x800) {
+         if (i+1 >= len1) return -1;
+         if (s1[i++] != 0xc0 + (ch >> 6)) return -1;
+         if (s1[i++] != 0x80 + (ch & 0x3f)) return -1;
+      } else if (ch >= 0xd800 && ch < 0xdc00) {
+         stbtt_uint32 c;
+         stbtt_uint16 ch2 = s2[2]*256 + s2[3];
+         if (i+3 >= len1) return -1;
+         c = ((ch - 0xd800) << 10) + (ch2 - 0xdc00) + 0x10000;
+         if (s1[i++] != 0xf0 + (c >> 18)) return -1;
+         if (s1[i++] != 0x80 + ((c >> 12) & 0x3f)) return -1;
+         if (s1[i++] != 0x80 + ((c >>  6) & 0x3f)) return -1;
+         if (s1[i++] != 0x80 + ((c      ) & 0x3f)) return -1;
+         s2 += 2; // plus another 2 below
+         len2 -= 2;
+      } else if (ch >= 0xdc00 && ch < 0xe000) {
+         return -1;
+      } else {
+         if (i+2 >= len1) return -1;
+         if (s1[i++] != 0xe0 + (ch >> 12)) return -1;
+         if (s1[i++] != 0x80 + ((ch >> 6) & 0x3f)) return -1;
+         if (s1[i++] != 0x80 + ((ch     ) & 0x3f)) return -1;
+      }
+      s2 += 2;
+      len2 -= 2;
+   }
+   return i;
+}
+
+int stbtt_CompareUTF8toUTF16_bigendian(const char *s1, int len1, const char *s2, int len2) 
+{
+   return len1 == stbtt__CompareUTF8toUTF16_bigendian_prefix((const stbtt_uint8*) s1, len1, (const stbtt_uint8*) s2, len2);
+}
+
+// returns results in whatever encoding you request... but note that 2-byte encodings
+// will be BIG-ENDIAN... use stbtt_CompareUTF8toUTF16_bigendian() to compare
+const char *stbtt_GetFontNameString(const stbtt_fontinfo *font, int *length, int platformID, int encodingID, int languageID, int nameID)
+{
+   stbtt_int32 i,count,stringOffset;
+   stbtt_uint8 *fc = font->data;
+   stbtt_uint32 offset = font->fontstart;
+   stbtt_uint32 nm = stbtt__find_table(fc, offset, "name");
+   if (!nm) return NULL;
+
+   count = ttUSHORT(fc+nm+2);
+   stringOffset = nm + ttUSHORT(fc+nm+4);
+   for (i=0; i < count; ++i) {
+      stbtt_uint32 loc = nm + 6 + 12 * i;
+      if (platformID == ttUSHORT(fc+loc+0) && encodingID == ttUSHORT(fc+loc+2)
+          && languageID == ttUSHORT(fc+loc+4) && nameID == ttUSHORT(fc+loc+6)) {
+         *length = ttUSHORT(fc+loc+8);
+         return (const char *) (fc+stringOffset+ttUSHORT(fc+loc+10));
+      }
+   }
+   return NULL;
+}
+
+static int stbtt__matchpair(stbtt_uint8 *fc, stbtt_uint32 nm, stbtt_uint8 *name, stbtt_int32 nlen, stbtt_int32 target_id, stbtt_int32 next_id)
+{
+   stbtt_int32 i;
+   stbtt_int32 count = ttUSHORT(fc+nm+2);
+   stbtt_int32 stringOffset = nm + ttUSHORT(fc+nm+4);
+
+   for (i=0; i < count; ++i) {
+      stbtt_uint32 loc = nm + 6 + 12 * i;
+      stbtt_int32 id = ttUSHORT(fc+loc+6);
+      if (id == target_id) {
+         // find the encoding
+         stbtt_int32 platform = ttUSHORT(fc+loc+0), encoding = ttUSHORT(fc+loc+2), language = ttUSHORT(fc+loc+4);
+
+         // is this a Unicode encoding?
+         if (platform == 0 || (platform == 3 && encoding == 1) || (platform == 3 && encoding == 10)) {
+            stbtt_int32 slen = ttUSHORT(fc+loc+8);
+            stbtt_int32 off = ttUSHORT(fc+loc+10);
+
+            // check if there's a prefix match
+            stbtt_int32 matchlen = stbtt__CompareUTF8toUTF16_bigendian_prefix(name, nlen, fc+stringOffset+off,slen);
+            if (matchlen >= 0) {
+               // check for target_id+1 immediately following, with same encoding & language
+               if (i+1 < count && ttUSHORT(fc+loc+12+6) == next_id && ttUSHORT(fc+loc+12) == platform && ttUSHORT(fc+loc+12+2) == encoding && ttUSHORT(fc+loc+12+4) == language) {
+                  slen = ttUSHORT(fc+loc+12+8);
+                  off = ttUSHORT(fc+loc+12+10);
+                  if (slen == 0) {
+                     if (matchlen == nlen)
+                        return 1;
+                  } else if (matchlen < nlen && name[matchlen] == ' ') {
+                     ++matchlen;
+                     if (stbtt_CompareUTF8toUTF16_bigendian((char*) (name+matchlen), nlen-matchlen, (char*)(fc+stringOffset+off),slen))
+                        return 1;
+                  }
+               } else {
+                  // if nothing immediately following
+                  if (matchlen == nlen)
+                     return 1;
+               }
+            }
+         }
+
+         // @TODO handle other encodings
+      }
+   }
+   return 0;
+}
+
+static int stbtt__matches(stbtt_uint8 *fc, stbtt_uint32 offset, stbtt_uint8 *name, stbtt_int32 flags)
+{
+   stbtt_int32 nlen = (stbtt_int32) STBTT_strlen((char *) name);
+   stbtt_uint32 nm,hd;
+   if (!stbtt__isfont(fc+offset)) return 0;
+
+   // check italics/bold/underline flags in macStyle...
+   if (flags) {
+      hd = stbtt__find_table(fc, offset, "head");
+      if ((ttUSHORT(fc+hd+44) & 7) != (flags & 7)) return 0;
+   }
+
+   nm = stbtt__find_table(fc, offset, "name");
+   if (!nm) return 0;
+
+   if (flags) {
+      // if we checked the macStyle flags, then just check the family and ignore the subfamily
+      if (stbtt__matchpair(fc, nm, name, nlen, 16, -1))  return 1;
+      if (stbtt__matchpair(fc, nm, name, nlen,  1, -1))  return 1;
+      if (stbtt__matchpair(fc, nm, name, nlen,  3, -1))  return 1;
+   } else {
+      if (stbtt__matchpair(fc, nm, name, nlen, 16, 17))  return 1;
+      if (stbtt__matchpair(fc, nm, name, nlen,  1,  2))  return 1;
+      if (stbtt__matchpair(fc, nm, name, nlen,  3, -1))  return 1;
+   }
+
+   return 0;
+}
+
+int stbtt_FindMatchingFont(const unsigned char *font_collection, const char *name_utf8, stbtt_int32 flags)
+{
+   stbtt_int32 i;
+   for (i=0;;++i) {
+      stbtt_int32 off = stbtt_GetFontOffsetForIndex(font_collection, i);
+      if (off < 0) return off;
+      if (stbtt__matches((stbtt_uint8 *) font_collection, off, (stbtt_uint8*) name_utf8, flags))
+         return off;
+   }
+}
+
+#endif // STB_TRUETYPE_IMPLEMENTATION