diff --git a/Source/ADSRDisplay.cpp b/Source/ADSRDisplay.cpp
index 4c1b8d969..06f7c5d81 100644
--- a/Source/ADSRDisplay.cpp
+++ b/Source/ADSRDisplay.cpp
@@ -269,7 +269,7 @@ void ADSRDisplay::Render()
             ofRect(0, 0, mWidth, 10);
             ofRect(ofMap(mMaxTime, 10, 10000, 0, mWidth - 3, K(clamp)), 0, 3, 10);
             ofSetColor(255, 255, 255, .8f * gModuleDrawAlpha);
-            DrawTextNormal(ofToString(mMaxTime, 0) + " ms", 3, 8, 10);
+            DrawTextNormal(ofToString(mMaxTime, 0) + " ms", 3, 8, 8);
             break;
          case kAdjustAttackAR:
             ofRect(0, 0, mWidth * .5f, mHeight);
diff --git a/Source/Arpeggiator.cpp b/Source/Arpeggiator.cpp
index 11232d987..bcfe7126f 100644
--- a/Source/Arpeggiator.cpp
+++ b/Source/Arpeggiator.cpp
@@ -96,8 +96,8 @@ void Arpeggiator::DrawModule()
       }
       else
       {
-         float w = gFont.GetStringWidth(pad, 15);
-         DrawTextNormal(GetArpNoteDisplay(mChord[i].pitch), 5 + w + pad.length() / 5.0f, 16);
+         float w = gFont.GetStringWidth(pad, 13);
+         DrawTextNormal(GetArpNoteDisplay(mChord[i].pitch), 6 + w, 16);
          break;
       }
    }
diff --git a/Source/CanvasElement.cpp b/Source/CanvasElement.cpp
index ea36567eb..e3d3ab24e 100644
--- a/Source/CanvasElement.cpp
+++ b/Source/CanvasElement.cpp
@@ -552,7 +552,7 @@ void SampleCanvasElement::DrawContents(bool clamp, bool wrapped, ofVec2f offset)
       float lengthOriginalSpeed = lengthMs / TheTransport->GetDuration(sampleCanvas->GetInterval());
       float speed = lengthOriginalSpeed / mLength;
       ofSetColor(255, 255, 255);
-      DrawTextBold(ofToString(speed, 2), rect.x + 3, rect.y + 10, 12);
+      DrawTextBold(ofToString(speed, 2), rect.x + 3, rect.y + 10, 10);
    }
 
    if (mMute)
diff --git a/Source/ChaosEngine.cpp b/Source/ChaosEngine.cpp
index 18c011c9f..558c94751 100644
--- a/Source/ChaosEngine.cpp
+++ b/Source/ChaosEngine.cpp
@@ -252,10 +252,10 @@ void ChaosEngine::DrawModule()
 
    if (!mHideBeat)
       gFont.DrawString(ofToString(beat), 42, 10, 50);
-   gFont.DrawString(ofToString(TheTransport->GetTimeSigTop()) + "/" + ofToString(TheTransport->GetTimeSigBottom()), 42, 90, 50);
-   gFont.DrawString(ofToString(TheTransport->GetTempo(), 0) + "bpm", 42, 230, 50);
+   gFont.DrawString(ofToString(TheTransport->GetTimeSigTop()) + "/" + ofToString(TheTransport->GetTimeSigBottom()), 40, 90, 50);
+   gFont.DrawString(ofToString(TheTransport->GetTempo(), 0) + "bpm", 420, 230, 50);
 
-   gFont.DrawString(NoteName(TheScale->ScaleRoot()) + " " + TheScale->GetType(), 42, 10, 100);
+   gFont.DrawString(NoteName(TheScale->ScaleRoot()) + " " + TheScale->GetType(), 40, 10, 100);
 
    DrawKeyboard(10, 150);
    DrawGuitar(10, 285);
@@ -280,7 +280,7 @@ void ChaosEngine::DrawModule()
       Chord displayChord;
       displayChord.SetFromDegreeAndScale(mChordProgression[i].mDegree, scale);
       displayChord.mInversion = mChordProgression[i].mInversion;
-      gFont.DrawString(displayChord.Name(true, false, &scale), 48, x, y);
+      gFont.DrawString(displayChord.Name(true, false, &scale), 46, x, y);
 
       std::string accidentalList;
       for (int j = 0; j < mChordProgression[i].mAccidentals.size(); ++j)
diff --git a/Source/CodeEntry.cpp b/Source/CodeEntry.cpp
index 10ffde788..af9fc0dcb 100644
--- a/Source/CodeEntry.cpp
+++ b/Source/CodeEntry.cpp
@@ -1337,8 +1337,8 @@ void CodeEntry::SetStyleFromJSON(const ofxJSONElement& vdict)
 {
    auto fs = vdict.get("font-size", 14);
    auto fsi = fs.asInt();
-   if (fsi > 2 && fsi < 200)
-      mFontSize = fsi;
+   if (fsi > 4 && fsi < 200)
+      mFontSize = fsi - 2;
 
    auto fromRGB = [vdict](const std::string& key, ofColor& onto)
    {
diff --git a/Source/CodeEntry.h b/Source/CodeEntry.h
index d173c484c..afb67e7bb 100644
--- a/Source/CodeEntry.h
+++ b/Source/CodeEntry.h
@@ -207,5 +207,5 @@ class CodeEntry : public IUIControl, public IKeyboardFocusListener
    ofColor jediParamsHighlight{ 230, 230, 255 };
    ofColor unknownColor = ofColor::white;
 
-   float mFontSize{ 14 };
+   float mFontSize{ 12 };
 };
diff --git a/Source/DotGrid.cpp b/Source/DotGrid.cpp
index f8fa9fd35..3ec88eda1 100644
--- a/Source/DotGrid.cpp
+++ b/Source/DotGrid.cpp
@@ -147,7 +147,7 @@ void DotGrid::Render()
                {
                   DotData& data = mData[GetDataIndex(mCurrentHover.mCol, mCurrentHover.mRow)];
                   ofSetColor(0, 255, 0);
-                  DrawTextNormal(ofToString(data.mVelocity, 2), GetX(i), GetY(j), 10.0f);
+                  DrawTextNormal(ofToString(data.mVelocity, 2), GetX(i), GetY(j), 8.0f);
                }
             }
             else
diff --git a/Source/DotSequencer.cpp b/Source/DotSequencer.cpp
index 64a2352a5..cc2d08d02 100644
--- a/Source/DotSequencer.cpp
+++ b/Source/DotSequencer.cpp
@@ -135,7 +135,7 @@ void DotSequencer::DrawModule()
          ofSetColor(128, 128, 128, gModuleDrawAlpha * .8f);
       }
 
-      float scale = std::min(mDotGrid->IClickable::GetDimensions().y / mDotGrid->GetRows(), 12.0f);
+      float scale = std::min(mDotGrid->IClickable::GetDimensions().y / mDotGrid->GetRows() - 2, 10.0f);
       DrawTextRightJustify(NoteName(RowToPitch(i), false, true) + "(" + ofToString(RowToPitch(i)) + ")", pos.x - 3, pos.y - ysize * .5f + (scale / 2), scale);
    }
    ofPopStyle();
diff --git a/Source/DrumPlayer.cpp b/Source/DrumPlayer.cpp
index 8b32718a0..a03fd4564 100644
--- a/Source/DrumPlayer.cpp
+++ b/Source/DrumPlayer.cpp
@@ -839,7 +839,7 @@ void DrumPlayer::DrawModule()
             }
 
             ofSetColor(255, 255, 255, gModuleDrawAlpha);
-            gFont.DrawStringWrap(mDrumHits[sampleIdx].mSample.Name(), 12, i * 70 + 5, j * 70 + 10, 60);
+            gFont.DrawStringWrap(mDrumHits[sampleIdx].mSample.Name(), 10, i * 70 + 5, j * 70 + 10, 60);
          }
       }
       ofPopStyle();
diff --git a/Source/FubbleModule.cpp b/Source/FubbleModule.cpp
index 7506d7a86..6bf0d6b22 100644
--- a/Source/FubbleModule.cpp
+++ b/Source/FubbleModule.cpp
@@ -309,7 +309,7 @@ void FubbleModule::DrawModuleUnclipped()
    if (Minimized() || IsVisible() == false)
       return;
 
-   DrawTextNormal("(concept by @_ojack_)", 60, -3, 11);
+   DrawTextNormal("(concept by @_ojack_)", 60, -3, 9);
 }
 
 ofRectangle FubbleModule::GetFubbleRect()
diff --git a/Source/GridModule.cpp b/Source/GridModule.cpp
index 30aab38d9..08ccfbe2d 100644
--- a/Source/GridModule.cpp
+++ b/Source/GridModule.cpp
@@ -146,7 +146,7 @@ void GridModule::DrawModule()
    for (int i = 0; i < mGrid->GetRows() && i < (int)mLabels.size(); ++i)
    {
       ofVec2f pos = mGrid->GetCellPosition(0, i) + mGrid->GetPosition(true);
-      float scale = MIN(mGrid->IClickable::GetDimensions().y / mGrid->GetRows(), 12);
+      float scale = MIN(mGrid->IClickable::GetDimensions().y / mGrid->GetRows() - 2, 10);
       DrawTextNormal(mLabels[i], 2, pos.y - (scale / 8), scale);
    }
    ofPopStyle();
diff --git a/Source/IDrawableModule.cpp b/Source/IDrawableModule.cpp
index 0580480ab..0941c0440 100644
--- a/Source/IDrawableModule.cpp
+++ b/Source/IDrawableModule.cpp
@@ -353,7 +353,7 @@ void IDrawableModule::DrawFrame(float w, float h, bool drawModule, float& titleB
    if (HasTitleBar())
    {
       ofSetColor(color * (1 - GetBeaconAmount()) + ofColor::yellow * GetBeaconAmount(), gModuleDrawAlpha);
-      DrawTextBold(GetTitleLabel(), 5 + enableToggleOffset, 10 - titleBarHeight, 16);
+      DrawTextBold(GetTitleLabel(), 5 + enableToggleOffset, 10 - titleBarHeight, 14);
    }
 
    bool groupSelected = !TheSynth->GetGroupSelectedModules().empty() && VectorContains(this, TheSynth->GetGroupSelectedModules());
@@ -853,7 +853,7 @@ float IDrawableModule::GetMinimizedWidth()
    if (titleLabel != mLastTitleLabel)
    {
       mLastTitleLabel = titleLabel;
-      mTitleLabelWidth = gFont.GetStringWidth(GetTitleLabel(), 16);
+      mTitleLabelWidth = gFont.GetStringWidth(GetTitleLabel(), 14);
    }
    float width = mTitleLabelWidth;
    width += 10; //padding
diff --git a/Source/KeyboardDisplay.cpp b/Source/KeyboardDisplay.cpp
index 8d00c171d..1d4953c1f 100644
--- a/Source/KeyboardDisplay.cpp
+++ b/Source/KeyboardDisplay.cpp
@@ -250,7 +250,7 @@ void KeyboardDisplay::DrawKeyboard(int x, int y, int w, int h)
       for (int i = 0; i < NumKeys(); i += 7)
       {
          ofSetColor(108, 37, 62, 255);
-         DrawTextNormal("C" + std::to_string(oct), keySpace * 0.5f - 6.5f + i * keySpace, h - 8, 14);
+         DrawTextNormal("C" + std::to_string(oct), keySpace * 0.5f - 6.5f + i * keySpace, h - 8, 12);
          oct++;
       }
    }
diff --git a/Source/LabelDisplay.cpp b/Source/LabelDisplay.cpp
index 07209b107..8bfda45ad 100644
--- a/Source/LabelDisplay.cpp
+++ b/Source/LabelDisplay.cpp
@@ -66,9 +66,9 @@ void LabelDisplay::CreateUIControls()
    UIBLOCK_SHIFTRIGHT();
    INTSLIDER(mLabelSizeSlider, "size", &mLabelSize, 25, 500);
    UIBLOCK_SHIFTRIGHT();
-   DROPDOWN(mLabelFontDropdown, "font", &mLabelFontIndex, 60);
+   DROPDOWN(mLabelFontDropdown, "font", &mLabelFontIndex, 90);
    UIBLOCK_SHIFTRIGHT();
-   DROPDOWN(mLabelColorDropdown, "color", &mLabelColorIndex, 60);
+   DROPDOWN(mLabelColorDropdown, "color", &mLabelColorIndex, 72);
    ENDUIBLOCK(mWidth, mHeight);
 
    for (int i = 0; i < mFonts.size(); i++)
diff --git a/Source/LabelDisplay.h b/Source/LabelDisplay.h
index 1ff5d78a0..5ef7b2db8 100644
--- a/Source/LabelDisplay.h
+++ b/Source/LabelDisplay.h
@@ -68,7 +68,7 @@ class LabelDisplay : public IDrawableModule, public ITextEntryListener, public I
 
    char mLabel[MAX_TEXTENTRY_LENGTH]{ "Label" };
    TextEntry* mLabelEntry{ nullptr };
-   int mLabelSize{ 42 };
+   int mLabelSize{ 40 };
    IntSlider* mLabelSizeSlider{ nullptr };
 
    RetinaTrueTypeFont mLabelFont{ gFont };
diff --git a/Source/LaunchpadKeyboard.cpp b/Source/LaunchpadKeyboard.cpp
index f4b6dbbc0..abcf09a90 100644
--- a/Source/LaunchpadKeyboard.cpp
+++ b/Source/LaunchpadKeyboard.cpp
@@ -464,7 +464,7 @@ void LaunchpadKeyboard::DrawModuleUnclipped()
       DrawTextNormal(mDebugDisplayText, 0, 90);
 
       for (int i = 0; i < 128; ++i)
-         DrawTextNormal(ofToString(i) + " " + ofToString(mCurrentNotes[i]), 180 + (i / 24) * 20, (i % 24) * 9, 8);
+         DrawTextNormal(ofToString(i) + " " + ofToString(mCurrentNotes[i]), 180 + (i / 24) * 20, (i % 24) * 9, 6);
    }
 }
 
diff --git a/Source/MidiController.cpp b/Source/MidiController.cpp
index a07642543..38000a656 100644
--- a/Source/MidiController.cpp
+++ b/Source/MidiController.cpp
@@ -936,13 +936,13 @@ void MidiController::DrawModule()
       else
          ofSetColor(255, 0, 0, 255 * (1 - (gTime - mLastActivityTime) / 200));
       ofFill();
-      ofRect(30 + gFont.GetStringWidth(Name(), 15), -11, 10, 10);
+      ofRect(30 + gFont.GetStringWidth(Name(), 13), -11, 10, 10);
       ofPopStyle();
    }
 
    if (!mIsConnected)
    {
-      float xStart = 30 + gFont.GetStringWidth(Name(), 15);
+      float xStart = 30 + gFont.GetStringWidth(Name(), 13);
       float yStart = -11;
 
       ofPushStyle();
@@ -1179,7 +1179,7 @@ void MidiController::DrawModule()
       ofPopStyle();
 
       if (mLayoutLoadError != "")
-         gFont.DrawStringWrap(mLayoutLoadError, 15, 3, kLayoutControlsY + 160, 235);
+         gFont.DrawStringWrap(mLayoutLoadError, 13, 3, kLayoutControlsY + 160, 235);
 
       if (mHighlightedLayoutElement != -1)
       {
diff --git a/Source/ModularSynth.cpp b/Source/ModularSynth.cpp
index c7f2780a5..4f6ce0833 100644
--- a/Source/ModularSynth.cpp
+++ b/Source/ModularSynth.cpp
@@ -513,7 +513,7 @@ void ModularSynth::Draw(void* vg)
       DrawFallbackText(("bespoke " + GetBuildInfoString()).c_str(), 100, 50);
 
       if (gFont.IsLoaded())
-         DrawTextNormal(mFatalError, 100, 100, 20);
+         DrawTextNormal(mFatalError, 100, 100, 18);
       else
          DrawFallbackText(mFatalError.c_str(), 100, 100);
    }
@@ -532,14 +532,14 @@ void ModularSynth::Draw(void* vg)
    if (gTime == 1 && mFatalError == "")
    {
       std::string loading("Bespoke is initializing audio...");
-      DrawTextNormal(loading, ofGetWidth() / 2 - GetStringWidth(loading, 30) / 2, ofGetHeight() / 2 - 6, 30);
+      DrawTextNormal(loading, ofGetWidth() / 2 - GetStringWidth(loading, 28) / 2, ofGetHeight() / 2 - 6, 28);
       return;
    }
 
    if (!mInitialized && mFatalError == "")
    {
       std::string loading("Bespoke is loading...");
-      DrawTextNormal(loading, ofGetWidth() / 2 - GetStringWidth(loading, 30) / 2, ofGetHeight() / 2 - 6, 30);
+      DrawTextNormal(loading, ofGetWidth() / 2 - GetStringWidth(loading, 28) / 2, ofGetHeight() / 2 - 6, 28);
       return;
    }
 
@@ -786,7 +786,7 @@ void ModularSynth::Draw(void* vg)
 
       float maxWidth = 300;
 
-      float fontSize = 15;
+      float fontSize = 13;
       nvgFontFaceId(gNanoVG, gFont.GetFontHandle());
       nvgFontSize(gNanoVG, fontSize);
       float bounds[4];
@@ -934,7 +934,7 @@ void ModularSynth::DrawConsole()
             ofSetColor(255, 255, 0);
          else
             ofSetColor(255, 255, 255);
-         gFontFixedWidth.DrawString(it->text, 15, 10, consoleY);
+         gFontFixedWidth.DrawString(it->text, 13, 10, consoleY);
          std::vector<std::string> lines = ofSplitString(it->text, "\n");
          ofPopStyle();
          consoleY += 15 * lines.size();
@@ -947,7 +947,7 @@ void ModularSynth::DrawConsole()
          ofSetColor(255, 0, 0);
          for (auto it = mErrors.begin(); it != mErrors.end(); ++it)
          {
-            gFontFixedWidth.DrawString(*it, 15, 600, consoleY);
+            gFontFixedWidth.DrawString(*it, 13, 600, consoleY);
             std::vector<std::string> lines = ofSplitString(*it, "\n");
             consoleY += 15 * lines.size();
          }
diff --git a/Source/NoteCanvas.cpp b/Source/NoteCanvas.cpp
index 2eddc3666..2070f2972 100644
--- a/Source/NoteCanvas.cpp
+++ b/Source/NoteCanvas.cpp
@@ -357,7 +357,7 @@ void NoteCanvas::DrawModule()
       int pitch = 127 - mCanvas->GetRowOffset() - i;
       float boxHeight = (float(mCanvas->GetHeight()) / mCanvas->GetNumVisibleRows());
       float y = mCanvas->GetPosition(true).y + i * boxHeight;
-      float scale = MIN(boxHeight, 20);
+      float scale = MIN(boxHeight - 2, 18);
       DrawTextNormal(NoteName(pitch, false, true) + "(" + ofToString(pitch) + ")", mCanvas->GetPosition(true).x + 2, y - (scale / 8) + boxHeight, scale);
    }
    ofPopStyle();
diff --git a/Source/NoteStepSequencer.cpp b/Source/NoteStepSequencer.cpp
index 291bc62ce..649804073 100644
--- a/Source/NoteStepSequencer.cpp
+++ b/Source/NoteStepSequencer.cpp
@@ -240,7 +240,7 @@ void NoteStepSequencer::DrawModule()
    for (int i = 0; i < mGrid->GetRows(); ++i)
    {
       ofVec2f pos = mGrid->GetCellPosition(0, i - 1) + mGrid->GetPosition(true);
-      float scale = MIN(mGrid->IClickable::GetDimensions().y / mGrid->GetRows(), 20);
+      float scale = MIN(mGrid->IClickable::GetDimensions().y / mGrid->GetRows() - 2, 18);
       DrawTextNormal(NoteName(RowToPitch(i), false, true) + "(" + ofToString(RowToPitch(i)) + ")", pos.x + 1, pos.y - (scale / 8), scale);
    }
    ofPopStyle();
diff --git a/Source/NoteTable.cpp b/Source/NoteTable.cpp
index 431ad1dd1..6d1e6396c 100644
--- a/Source/NoteTable.cpp
+++ b/Source/NoteTable.cpp
@@ -137,7 +137,7 @@ void NoteTable::DrawModule()
    for (int i = 0; i < mGrid->GetRows(); ++i)
    {
       ofVec2f pos = mGrid->GetCellPosition(0, i - 1) + mGrid->GetPosition(true);
-      float scale = MIN(mGrid->IClickable::GetDimensions().y / mGrid->GetRows(), 20);
+      float scale = MIN(mGrid->IClickable::GetDimensions().y / mGrid->GetRows() - 2, 18);
       DrawTextNormal(NoteName(RowToPitch(i), false, true) + "(" + ofToString(RowToPitch(i)) + ")", pos.x + 4, pos.y - (scale / 8), scale);
    }
    ofPopStyle();
diff --git a/Source/OutputChannel.cpp b/Source/OutputChannel.cpp
index e62b66111..c091be0c1 100644
--- a/Source/OutputChannel.cpp
+++ b/Source/OutputChannel.cpp
@@ -175,7 +175,7 @@ void OutputChannel::DrawModule()
       {
          ofPushStyle();
          ofSetColor(ofColor::red);
-         DrawTextBold("clipped", kPaddingOutside + 10, 20 + i * (kBarHeight + 2) + 8, 12.0f);
+         DrawTextBold("clipped", kPaddingOutside + 10, 20 + i * (kBarHeight + 2) + 8, 10.0f);
          ofPopStyle();
       }
    }
diff --git a/Source/PanicButton.cpp b/Source/PanicButton.cpp
index d83b33d7a..b5c8f612a 100644
--- a/Source/PanicButton.cpp
+++ b/Source/PanicButton.cpp
@@ -42,13 +42,13 @@ void PanicButton::OnClicked(float x, float y, bool right)
 
 void PanicButton::DrawModule()
 {
-   gFont.DrawString("If anything goes horribly awry, click this\nand the party will be restored", 15, 5, 12);
+   gFont.DrawString("If anything goes horribly awry, click this\nand the party will be restored", 13, 5, 12);
 
    ofPushStyle();
    ofFill();
    ofSetColor(70, 0, 0);
    ofRect(10, 40, 270, 80);
    ofSetColor(255, 0, 0);
-   gFont.DrawString("PANIC", 108, 15, 115);
+   gFont.DrawString("PANIC", 106, 15, 115);
    ofPopStyle();
 }
\ No newline at end of file
diff --git a/Source/PatchCable.cpp b/Source/PatchCable.cpp
index c3cf81223..90e30a4d8 100644
--- a/Source/PatchCable.cpp
+++ b/Source/PatchCable.cpp
@@ -400,7 +400,7 @@ void PatchCable::Render()
             ofSetColor(255, 255, 0);
             ofCircle(cable.plug.x, cable.plug.y, 6);
             ofSetColor(0, 0, 0);
-            DrawTextBold("!", cable.plug.x - 2, cable.plug.y + 5, 17);
+            DrawTextBold("!", cable.plug.x - 2, cable.plug.y + 5, 15);
          }
       }
       else
diff --git a/Source/Profiler.cpp b/Source/Profiler.cpp
index b55424d16..e76c5b6d6 100644
--- a/Source/Profiler.cpp
+++ b/Source/Profiler.cpp
@@ -139,7 +139,7 @@ void Profiler::Draw()
       long maxCost = cost.MaxCost();
 
       ofSetColor(255, 255, 255);
-      gFont.DrawString(std::string(sCosts[i].mName) + ": " + ofToString(maxCost / 1000), 15, 0, 0);
+      gFont.DrawString(std::string(sCosts[i].mName) + ": " + ofToString(maxCost / 1000), 13, 0, 0);
 
       if (maxCost > entireFrameUs)
          ofSetColor(255, 0, 0);
diff --git a/Source/Push2Control.cpp b/Source/Push2Control.cpp
index fa48e1504..1c039bc39 100644
--- a/Source/Push2Control.cpp
+++ b/Source/Push2Control.cpp
@@ -510,7 +510,7 @@ void Push2Control::DrawToFramebuffer(NVGcontext* vg, NVGLUframebuffer* fb, float
          ofRect(1, 1, ableton::Push2DisplayBitmap::kWidth - 2, ableton::Push2DisplayBitmap::kHeight - 2);
 
          ofSetColor(0, 0, 0);
-         DrawTextBold(stateInfo, 10, 147, 20);
+         DrawTextBold(stateInfo, 10, 147, 18);
 
          ofPopStyle();
       }
@@ -530,14 +530,14 @@ void Push2Control::DrawToFramebuffer(NVGcontext* vg, NVGLUframebuffer* fb, float
             ofSetColor(IDrawableModule::GetColor(TheSynth->GetModuleFactory()->GetModuleCategory(moduleTypeToSpawn)));
             text = "\ntap grid to spawn " + moduleTypeToSpawn;
          }
-         DrawTextBold(text, 5, 80, 20);
+         DrawTextBold(text, 5, 80, 18);
 
          ofSetColor(IDrawableModule::GetColor(kModuleCategory_Other));
          ofNoFill();
 
          ofTranslate(-kColumnSpacing * mModuleViewOffsetSmoothed, 0);
 
-         nvgFontSize(sVG, 16);
+         nvgFontSize(sVG, 12);
          DrawControls(mButtonControls, false, 60);
          DrawControls(mSliderControls, true, 20);
       }
@@ -1103,9 +1103,9 @@ void Push2Control::DrawControls(std::vector<IUIControl*> controls, bool sliders,
       if (adsr == nullptr)
       {
          if (mDisplayModule == this)
-            DrawTextBold(juce::String(controls[i]->Path()).replace("~", "\n").toRawUTF8(), kColumnSpacing * i + 3, yPos - 12, 10);
+            DrawTextBold(juce::String(controls[i]->Path()).replace("~", "\n").toRawUTF8(), kColumnSpacing * i + 3, yPos - 12, 8);
          else
-            DrawTextBold(controls[i]->Name(), kColumnSpacing * i + 3, yPos - 5, 16);
+            DrawTextBold(controls[i]->Name(), kColumnSpacing * i + 3, yPos - 5, 14);
       }
       controls[i]->Render();
       ofPopStyle();
diff --git a/Source/QuickSpawnMenu.cpp b/Source/QuickSpawnMenu.cpp
index 901a20d7e..8595da993 100644
--- a/Source/QuickSpawnMenu.cpp
+++ b/Source/QuickSpawnMenu.cpp
@@ -471,9 +471,9 @@ void QuickSpawnMenu::DrawModule()
 void QuickSpawnMenu::DrawModuleUnclipped()
 {
    if (mMenuMode == MenuMode::SingleLetter)
-      DrawTextBold(mHeldKeys.toStdString(), 3, -2, 17);
+      DrawTextBold(mHeldKeys.toStdString(), 3, -2, 15);
    if (mMenuMode == MenuMode::Search)
-      DrawTextBold(mSearchString.toStdString(), 3, -2, 17);
+      DrawTextBold(mSearchString.toStdString(), 3, -2, 15);
 }
 
 bool QuickSpawnMenu::MouseMoved(float x, float y)
diff --git a/Source/SampleBrowser.cpp b/Source/SampleBrowser.cpp
index 2e6182bbd..01b95549a 100644
--- a/Source/SampleBrowser.cpp
+++ b/Source/SampleBrowser.cpp
@@ -72,7 +72,7 @@ void SampleBrowser::DrawModule()
    if (Minimized() || IsVisible() == false)
       return;
 
-   float fontSize = 15;
+   float fontSize = 13;
    float stringWidth = gFont.GetStringWidth(mCurrentDirectory.toStdString(), fontSize);
    float moduleWidth, moduleHeight;
    GetModuleDimensions(moduleWidth, moduleHeight);
diff --git a/Source/SamplePlayer.cpp b/Source/SamplePlayer.cpp
index 41028567d..4bccbb3e0 100644
--- a/Source/SamplePlayer.cpp
+++ b/Source/SamplePlayer.cpp
@@ -1039,7 +1039,7 @@ void SamplePlayer::DrawModule()
          ofSetColor(255, 255, 255, 50);
          ofRect(0, 0, (mWidth - 10) * mSample->GetSampleLoadProgress(), mHeight - 65);
          ofSetColor(40, 40, 40);
-         DrawTextNormal("loading sample...", 10, 10, 10);
+         DrawTextNormal("loading sample...", 10, 10, 8);
          ofPopStyle();
       }
    }
@@ -1054,7 +1054,7 @@ void SamplePlayer::DrawModule()
       ofSetColor(255, 255, 255, 50);
       ofRect(0, 0, mWidth - 10, mHeight - 65);
       ofSetColor(220, 0, 0);
-      DrawTextNormal(mErrorString, 10, 10, 10);
+      DrawTextNormal(mErrorString, 10, 10, 8);
       ofPopStyle();
    }
    else if (mSample && mSample->LengthInSamples() > 0)
@@ -1080,7 +1080,7 @@ void SamplePlayer::DrawModule()
       if (playPosition >= 0)
       {
          float x = ofMap(playPosition, GetZoomStartSample(), GetZoomEndSample(), 0, sampleWidth);
-         DrawTextNormal(ofToString(playPosition / (gSampleRate * mSample->GetSampleRateRatio()), 1), x + 2, mHeight - 65, 11);
+         DrawTextNormal(ofToString(playPosition / (gSampleRate * mSample->GetSampleRateRatio()), 1), x + 2, mHeight - 65, 9);
       }
 
       if (mShowGrid)
@@ -1113,7 +1113,7 @@ void SamplePlayer::DrawModule()
       ofSetColor(255, 255, 255, 50);
       ofRect(0, 0, mWidth - 10, mHeight - 65);
       ofSetColor(40, 40, 40);
-      DrawTextNormal("drag and drop a sample here...", 10, 10, 10);
+      DrawTextNormal("drag and drop a sample here...", 10, 10, 8);
       ofPopStyle();
    }
 
@@ -1138,7 +1138,7 @@ void SamplePlayer::DrawModule()
                ofRect(x, 0, 15, 10);
                ofFill();
             }
-            DrawTextNormal(ofToString((int)i), x + 2, 8, 11);
+            DrawTextNormal(ofToString((int)i), x + 2, 8, 9);
 
             if (i == mHoveredCuePointIndex)
             {
diff --git a/Source/ScriptModule.cpp b/Source/ScriptModule.cpp
index 08f1c2e64..8f055b67f 100644
--- a/Source/ScriptModule.cpp
+++ b/Source/ScriptModule.cpp
@@ -358,7 +358,7 @@ void ScriptModule::DrawModule()
       float y = buttonRect.getCenter().y;
       ofCircle(x, y, 6);
       ofSetColor(0, 0, 0);
-      DrawTextBold("!", x - 2, y + 5, 17);
+      DrawTextBold("!", x - 2, y + 5, 15);
       ofPopStyle();
 
       if (mShowJediWarning)
diff --git a/Source/Slider.cpp b/Source/Slider.cpp
index f89f70dd3..695199ef3 100644
--- a/Source/Slider.cpp
+++ b/Source/Slider.cpp
@@ -188,11 +188,11 @@ void FloatSlider::Render()
    DrawHover(mX, mY, mWidth, mHeight);
 
    std::string display;
-   float textSize = 15;
+   float textSize = 13;
    if (showSmoothAdjustmentUI)
    {
       display = "smooth: " + ofToString(mSmooth, 3);
-      textSize = 11;
+      textSize = 9;
    }
    else
    {
@@ -258,12 +258,12 @@ void FloatSlider::Render()
 
       ofPushMatrix();
       ofClipWindow(mX, mY, mWidth * .4f, mHeight, true);
-      DrawTextNormal(ofToString(mMin), mX + 2, mY + 4 + mHeight / 2, 12);
+      DrawTextNormal(ofToString(mMin), mX + 2, mY + 4 + mHeight / 2, 10);
       ofPopMatrix();
 
       ofPushMatrix();
       ofClipWindow(mX + mWidth * .6f, mY, mWidth * .4f, mHeight, true);
-      DrawTextRightJustify(ofToString(mMax), mX + mWidth - 2, mY + 4 + mHeight / 2, 12);
+      DrawTextRightJustify(ofToString(mMax), mX + mWidth - 2, mY + 4 + mHeight / 2, 10);
       ofPopMatrix();
 
       ofPopStyle();
@@ -1009,8 +1009,8 @@ void IntSlider::Render()
       ofSetColor(255, 255, 255);
       ofRect(mX, mY, mWidth * .4f, mHeight);
       ofRect(mX + mWidth * .6f, mY, mWidth * .4f, mHeight);
-      DrawTextNormal(ofToString(mMin), mX + 2, mY + 4 + mHeight / 2, 12);
-      DrawTextRightJustify(ofToString(mMax), mX + mWidth - 2, mY + 4 + mHeight / 2, 12);
+      DrawTextNormal(ofToString(mMin), mX + 2, mY + 4 + mHeight / 2, 10);
+      DrawTextRightJustify(ofToString(mMax), mX + mWidth - 2, mY + 4 + mHeight / 2, 10);
       ofPopStyle();
    }
 
diff --git a/Source/SongBuilder.cpp b/Source/SongBuilder.cpp
index 9e3006523..a2003afea 100644
--- a/Source/SongBuilder.cpp
+++ b/Source/SongBuilder.cpp
@@ -1143,7 +1143,7 @@ void SongBuilder::ControlTarget::Draw(float x, float y, int numRows)
       for (; cursor + kSliceSize < (int)text.size(); cursor += kSliceSize)
          displayString += text.substr(cursor, kSliceSize) + "\n";
       displayString += text.substr(cursor, (int)text.size() - cursor);
-      DrawTextNormal(displayString, x + 2, y + 9, 9);
+      DrawTextNormal(displayString, x + 2, y + 9, 7);
       ofPopMatrix();
    }
    float bottomY = y + kTargetTabHeightTop + kSpacingY + numRows * (kRowHeight + kSpacingY);
diff --git a/Source/SynthGlobals.h b/Source/SynthGlobals.h
index 16175662b..f01f8fc28 100644
--- a/Source/SynthGlobals.h
+++ b/Source/SynthGlobals.h
@@ -188,10 +188,10 @@ float Interp(float a, float start, float end);
 double GetPhaseInc(float freq);
 float FloatWrap(float num, float space);
 double DoubleWrap(double num, float space);
-void DrawTextNormal(std::string text, int x, int y, float size = 15);
-void DrawTextRightJustify(std::string text, int x, int y, float size = 15);
-void DrawTextBold(std::string text, int x, int y, float size = 15);
-float GetStringWidth(std::string text, float size = 15);
+void DrawTextNormal(std::string text, int x, int y, float size = 13);
+void DrawTextRightJustify(std::string text, int x, int y, float size = 13);
+void DrawTextBold(std::string text, int x, int y, float size = 13);
+float GetStringWidth(std::string text, float size = 13);
 void AssertIfDenormal(float input);
 float GetInterpolatedSample(double offset, const float* buffer, int bufferSize);
 float GetInterpolatedSample(double offset, ChannelBuffer* buffer, int bufferSize, float channelBlend);
diff --git a/Source/TextEntry.cpp b/Source/TextEntry.cpp
index 64135cbaf..0e7ba3989 100644
--- a/Source/TextEntry.cpp
+++ b/Source/TextEntry.cpp
@@ -96,7 +96,7 @@ void TextEntry::Construct(ITextEntryListener* owner, const char* name, int x, in
    UpdateDisplayString();
 
    SetName(name);
-   mLabelSize = gFont.GetStringWidth(name, 15) + 3;
+   mLabelSize = gFont.GetStringWidth(name, 13) + 3;
    SetPosition(x, y);
    assert(owner);
    IDrawableModule* module = dynamic_cast<IDrawableModule*>(owner);
@@ -150,7 +150,7 @@ void TextEntry::Render()
    ofNoFill();
    ofRect(mX + xOffset, mY, w - xOffset, h);
 
-   gFontFixedWidth.DrawString(mString, 14, mX + 2 + xOffset, mY + 12);
+   gFontFixedWidth.DrawString(mString, 12, mX + 2 + xOffset, mY + 12);
 
    if (IKeyboardFocusListener::GetActiveKeyboardFocus() == this)
    {
@@ -163,7 +163,7 @@ void TextEntry::Render()
             char beforeCaret[MAX_TEXTENTRY_LENGTH];
             strncpy(beforeCaret, mString, mCaretPosition);
             beforeCaret[mCaretPosition] = 0;
-            caretX += gFontFixedWidth.GetStringWidth(beforeCaret, 14);
+            caretX += gFontFixedWidth.GetStringWidth(beforeCaret, 12);
          }
          ofFill();
          ofRect(caretX, caretY, 1, 12, L(corner, 1));
@@ -191,13 +191,13 @@ void TextEntry::Render()
       char selectionTmp[MAX_TEXTENTRY_LENGTH];
       strncpy(selectionTmp, mString, start);
       selectionTmp[start] = 0;
-      selStartX += gFontFixedWidth.GetStringWidth(selectionTmp, 14);
+      selStartX += gFontFixedWidth.GetStringWidth(selectionTmp, 12);
 
       //
       int end = MAX(mCaretPosition, mCaretPosition2);
       strncpy(selectionTmp, mString, end);
       selectionTmp[end] = 0;
-      selEndX += gFontFixedWidth.GetStringWidth(selectionTmp, 14);
+      selEndX += gFontFixedWidth.GetStringWidth(selectionTmp, 12);
 
       ofRect(selStartX, selY, selEndX - selStartX, 12, 0);
 
@@ -208,7 +208,7 @@ void TextEntry::Render()
    {
       ofSetColor(100, 100, 100, .8f*gModuleDrawAlpha);
       ofFill();
-      ofRect(mX+xOffset,mY-12,GetStringWidth(Name()),12);
+      ofRect(mX+xOffset,mY-12,GetStringWidth(Name()),10);
       ofSetColor(255, 255, 255, gModuleDrawAlpha);
       DrawTextNormal(Name(), mX+xOffset, mY);
    }*/
@@ -221,7 +221,7 @@ void TextEntry::Render()
 void TextEntry::GetDimensions(float& width, float& height)
 {
    if (mFlexibleWidth)
-      width = MAX(30.0f, gFontFixedWidth.GetStringWidth(mString, 14) + 4);
+      width = MAX(30.0f, gFontFixedWidth.GetStringWidth(mString, 12) + 4);
    else
       width = mCharWidth * 9;
 
@@ -251,12 +251,12 @@ void TextEntry::OnClicked(float x, float y, bool right)
       char caretCheck[MAX_TEXTENTRY_LENGTH];
       size_t checkLength = strnlen(mString, MAX_TEXTENTRY_LENGTH);
       strncpy(caretCheck, mString, checkLength);
-      int lastSubstrWidth = gFontFixedWidth.GetStringWidth(caretCheck, 14);
+      int lastSubstrWidth = gFontFixedWidth.GetStringWidth(caretCheck, 12);
       for (int i = (int)checkLength - 1; i >= 0; --i)
       {
          caretCheck[i] = 0; //shorten string by one
 
-         int substrWidth = gFontFixedWidth.GetStringWidth(caretCheck, 14);
+         int substrWidth = gFontFixedWidth.GetStringWidth(caretCheck, 12);
          //ofLog() << x << " " << i << " " << (xOffset + substrWidth);
          if (x > xOffset + ((substrWidth + lastSubstrWidth) * .5f))
          {
diff --git a/Source/TimerDisplay.cpp b/Source/TimerDisplay.cpp
index a1f613245..02f85665b 100644
--- a/Source/TimerDisplay.cpp
+++ b/Source/TimerDisplay.cpp
@@ -69,6 +69,6 @@ void TimerDisplay::DrawModule()
 
    ofPushStyle();
    ofSetColor(255, 255, 255);
-   gFont.DrawString(zeroPadMins + ofToString(mins) + ":" + zeroPadSecs + ofToString(secs), 54, 15, 36);
+   gFont.DrawString(zeroPadMins + ofToString(mins) + ":" + zeroPadSecs + ofToString(secs), 52, 15, 36);
    ofPopStyle();
 }
diff --git a/Source/TitleBar.cpp b/Source/TitleBar.cpp
index 23e0c4d8c..e1a6535c8 100644
--- a/Source/TitleBar.cpp
+++ b/Source/TitleBar.cpp
@@ -342,16 +342,16 @@ void TitleBar::DrawModule()
    ofPushStyle();
    if (gHoveredModule == this && mLeftCornerHovered)
       ofSetColor(ofColor::lerp(ofColor::black, ofColor::white, ofMap(sin(gTime / 1000 * PI * 2), -1, 1, .7f, .9f)));
-   DrawTextBold("bespoke", 2, 28, 36);
+   DrawTextBold("bespoke", 2, 28, 34);
 #if BESPOKE_NIGHTLY && !BESPOKE_SUPPRESS_NIGHTLY_LABEL
-   DrawTextNormal("nightly", 90, 35, 10);
+   DrawTextNormal("nightly", 90, 35, 8);
 #endif
 #if DEBUG
    ofFill();
    ofSetColor(0, 0, 0, 180);
    ofRect(13, 12, 90, 17);
    ofSetColor(255, 0, 0);
-   DrawTextBold("debug build", 17, 25, 19);
+   DrawTextBold("debug build", 17, 25, 17);
 #endif
    ofPopStyle();
 
@@ -439,7 +439,7 @@ void TitleBar::DrawModuleUnclipped()
       TheTitleBar->GetDimensions(titleBarWidth, titleBarHeight);
       float x = 100;
       float y = 50 + titleBarHeight;
-      gFontBold.DrawString("please close plugin manager to continue", 50, x, y);
+      gFontBold.DrawString("please close plugin manager to continue", 48, x, y);
       ofPopStyle();
       return;
    }
@@ -453,7 +453,7 @@ void TitleBar::DrawModuleUnclipped()
       TheTitleBar->GetDimensions(titleBarWidth, titleBarHeight);
       float x = 100;
       float y = 40 + titleBarHeight;
-      gFontBold.DrawString(mDisplayMessage, 50, x, y);
+      gFontBold.DrawString(mDisplayMessage, 48, x, y);
       ofPopStyle();
    }
 
@@ -465,7 +465,7 @@ void TitleBar::DrawModuleUnclipped()
       ofPushStyle();
       ofSetColor(255, 255, 255);
       std::string text = "click ? to view help and toggle tooltips";
-      float size = 28;
+      float size = 26;
       float titleBarWidth, titleBarHeight;
       TheTitleBar->GetDimensions(titleBarWidth, titleBarHeight);
       ofRectangle helpButtonRect = mDisplayHelpButton->GetRect(true);
@@ -512,7 +512,7 @@ void TitleBar::DrawModuleUnclipped()
          ofSetColor(120, 120, 120, 150);
          ofRect(pos.x, pos.y, kWidth * drawControl->GetMidiValue(), kHeight);
          ofSetColor(255, 255, 255);
-         DrawTextBold(displayString, pos.x + 20, pos.y + 50, 40);
+         DrawTextBold(displayString, pos.x + 20, pos.y + 50, 38);
          ofPopStyle();
       }
    }
diff --git a/Source/UserPrefsEditor.cpp b/Source/UserPrefsEditor.cpp
index 575f11c27..7a482b163 100644
--- a/Source/UserPrefsEditor.cpp
+++ b/Source/UserPrefsEditor.cpp
@@ -307,7 +307,7 @@ void UserPrefsEditor::DrawModule()
       ofRectangle rect = UserPrefs.audio_output_device.GetControl()->GetRect(true);
       ofPushStyle();
       ofSetColor(ofColor::white);
-      DrawTextNormal("note: " + UserPrefs.devicetype.GetDropdown()->GetLabel(UserPrefs.devicetype.GetIndex()) + " uses the same audio device for output and input", rect.x, rect.getMaxY() + 14, 13);
+      DrawTextNormal("note: " + UserPrefs.devicetype.GetDropdown()->GetLabel(UserPrefs.devicetype.GetIndex()) + " uses the same audio device for output and input", rect.x, rect.getMaxY() + 14, 11);
       ofPopStyle();
    }
 
@@ -355,7 +355,7 @@ void UserPrefsEditor::DrawRightLabel(IUIControl* control, std::string text, ofCo
       ofRectangle rect = control->GetRect(true);
       ofPushStyle();
       ofSetColor(color);
-      DrawTextNormal(text, rect.getMaxX() + offsetX, rect.getMaxY() - 3, 13);
+      DrawTextNormal(text, rect.getMaxX() + offsetX, rect.getMaxY() - 3, 11);
       ofPopStyle();
    }
 }
diff --git a/libs/nanovg/nanovg/fontstash.h b/libs/nanovg/nanovg/fontstash.h
index 44e733121..b3feeaab0 100755
--- a/libs/nanovg/nanovg/fontstash.h
+++ b/libs/nanovg/nanovg/fontstash.h
@@ -38,6 +38,11 @@ enum FONSalign {
 	FONS_ALIGN_BASELINE	= 1<<6, // Default
 };
 
+enum FONSglyphBitmap {
+	FONS_GLYPH_BITMAP_OPTIONAL = 1,
+	FONS_GLYPH_BITMAP_REQUIRED = 2,
+};
+
 enum FONSerrorCode {
 	// Font atlas is full.
 	FONS_ATLAS_FULL = 1,
@@ -78,6 +83,7 @@ struct FONStextIter {
 	const char* next;
 	const char* end;
 	unsigned int utf8state;
+	int bitmapOption;
 };
 typedef struct FONStextIter FONStextIter;
 
@@ -96,8 +102,8 @@ int fonsExpandAtlas(FONScontext* s, int width, int height);
 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 fonsAddFont(FONScontext* s, const char* name, const char* path, int fontIndex);
+int fonsAddFontMem(FONScontext* s, const char* name, unsigned char* data, int ndata, int freeData, int fontIndex);
 int fonsGetFontByName(FONScontext* s, const char* name);
 
 // State handling
@@ -122,7 +128,7 @@ 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 fonsTextIterInit(FONScontext* stash, FONStextIter* iter, float x, float y, const char* str, const char* end, int bitmapOption);
 int fonsTextIterNext(FONScontext* stash, FONStextIter* iter, struct FONSquad* quad);
 
 // Pull texture changes
@@ -151,94 +157,10 @@ struct FONSttFontImpl {
 };
 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)
@@ -250,61 +172,10 @@ struct FONSttFontImpl {
 };
 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 64000
+#	define FONS_SCRATCH_BUF_SIZE 96000
 #endif
 #ifndef FONS_HASH_LUT_SIZE
 #	define FONS_HASH_LUT_SIZE 256
@@ -324,6 +195,9 @@ int fons__tt_getGlyphKernAdvance(FONSttFontImpl *font, int glyph1, int glyph2)
 #ifndef FONS_MAX_STATES
 #	define FONS_MAX_STATES 20
 #endif
+#ifndef FONS_MAX_FALLBACKS
+#	define FONS_MAX_FALLBACKS 20
+#endif
 
 static unsigned int fons__hashint(unsigned int a)
 {
@@ -371,6 +245,8 @@ struct FONSfont
 	int cglyphs;
 	int nglyphs;
 	int lut[FONS_HASH_LUT_SIZE];
+	int fallbacks[FONS_MAX_FALLBACKS];
+	int nfallbacks;
 };
 typedef struct FONSfont FONSfont;
 
@@ -419,8 +295,173 @@ struct FONScontext
 	int nstates;
 	void (*handleError)(void* uptr, int error, int val);
 	void* errorUptr;
+#ifdef FONS_USE_FREETYPE
+	FT_Library ftLibrary;
+#endif
 };
 
+#ifdef FONS_USE_FREETYPE
+
+int fons__tt_init(FONScontext *context)
+{
+	FT_Error ftError;
+	FONS_NOTUSED(context);
+	ftError = FT_Init_FreeType(&context->ftLibrary);
+	return ftError == 0;
+}
+
+int fons__tt_done(FONScontext *context)
+{
+	FT_Error ftError;
+	FONS_NOTUSED(context);
+	ftError = FT_Done_FreeType(context->ftLibrary);
+	return ftError == 0;
+}
+
+int fons__tt_loadFont(FONScontext *context, FONSttFontImpl *font, unsigned char *data, int dataSize, int fontIndex)
+{
+	FT_Error ftError;
+	FONS_NOTUSED(context);
+
+	ftError = FT_New_Memory_Face(context->ftLibrary, (const FT_Byte*)data, dataSize, fontIndex, &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->units_per_EM;
+}
+
+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;
+	FT_Fixed advFixed;
+	FONS_NOTUSED(scale);
+
+	ftError = FT_Set_Pixel_Sizes(font->font, 0, size);
+	if (ftError) return 0;
+	ftError = FT_Load_Glyph(font->font, glyph, FT_LOAD_RENDER | FT_LOAD_FORCE_AUTOHINT | FT_LOAD_TARGET_LIGHT);
+	if (ftError) return 0;
+	ftError = FT_Get_Advance(font->font, glyph, FT_LOAD_NO_SCALE, &advFixed);
+	if (ftError) return 0;
+	ftGlyph = font->font->glyph;
+	*advance = (int)advFixed;
+	*lsb = (int)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;
+	unsigned 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 (int)((ftKerning.x + 32) >> 6);  // Round up and convert to integer
+}
+
+#else
+
+int fons__tt_init(FONScontext *context)
+{
+	FONS_NOTUSED(context);
+	return 1;
+}
+
+int fons__tt_done(FONScontext *context)
+{
+	FONS_NOTUSED(context);
+	return 1;
+}
+
+int fons__tt_loadFont(FONScontext *context, FONSttFontImpl *font, unsigned char *data, int dataSize, int fontIndex)
+{
+	int offset, stbError;
+	FONS_NOTUSED(dataSize);
+
+	font->font.userdata = context;
+	offset = stbtt_GetFontOffsetForIndex(data, fontIndex);
+	if (offset == -1) {
+		stbError = 0;
+	} else {
+		stbError = stbtt_InitFont(&font->font, data, offset);
+	}
+	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_ScaleForMappingEmToPixels(&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
+
+#ifdef STB_TRUETYPE_IMPLEMENTATION
+
 static void* fons__tmpalloc(size_t size, void* up)
 {
 	unsigned char* ptr;
@@ -446,6 +487,8 @@ static void fons__tmpfree(void* ptr, void* up)
 	// empty
 }
 
+#endif // STB_TRUETYPE_IMPLEMENTATION
+
 // Copyright (c) 2008-2010 Bjoern Hoehrmann <bjoern@hoehrmann.de>
 // See http://bjoern.hoehrmann.de/utf-8/decoder/dfa/ for details.
 
@@ -751,6 +794,27 @@ static FONSstate* fons__getState(FONScontext* stash)
 	return &stash->states[stash->nstates-1];
 }
 
+int fonsAddFallbackFont(FONScontext* stash, int base, int fallback)
+{
+	FONSfont* baseFont = stash->fonts[base];
+	if (baseFont->nfallbacks < FONS_MAX_FALLBACKS) {
+		baseFont->fallbacks[baseFont->nfallbacks++] = fallback;
+		return 1;
+	}
+	return 0;
+}
+
+void fonsResetFallbackFont(FONScontext* stash, int base)
+{
+	int i;
+
+	FONSfont* baseFont = stash->fonts[base];
+	baseFont->nfallbacks = 0;
+	baseFont->nglyphs = 0;
+	for (i = 0; i < FONS_HASH_LUT_SIZE; i++)
+		baseFont->lut[i] = -1;
+}
+
 void fonsSetSize(FONScontext* stash, float size)
 {
 	fons__getState(stash)->size = size;
@@ -849,10 +913,11 @@ static int fons__allocFont(FONScontext* stash)
 	return FONS_INVALID;
 }
 
-int fonsAddFont(FONScontext* stash, const char* name, const char* path)
+int fonsAddFont(FONScontext* stash, const char* name, const char* path, int fontIndex)
 {
 	FILE* fp = 0;
 	int dataSize = 0;
+	size_t readed;
 	unsigned char* data = NULL;
 
 	// Read in the font data.
@@ -863,11 +928,12 @@ int fonsAddFont(FONScontext* stash, const char* name, const char* path)
 	fseek(fp,0,SEEK_SET);
 	data = (unsigned char*)malloc(dataSize);
 	if (data == NULL) goto error;
-	fread(data, 1, dataSize, fp);
+	readed = fread(data, 1, dataSize, fp);
 	fclose(fp);
 	fp = 0;
+	if (readed != (size_t)dataSize) goto error;
 
-	return fonsAddFontMem(stash, name, data, dataSize, 1);
+	return fonsAddFontMem(stash, name, data, dataSize, 1, fontIndex);
 
 error:
 	if (data) free(data);
@@ -875,7 +941,7 @@ int fonsAddFont(FONScontext* stash, const char* name, const char* path)
 	return FONS_INVALID;
 }
 
-int fonsAddFontMem(FONScontext* stash, const char* name, unsigned char* data, int dataSize, int freeData)
+int fonsAddFontMem(FONScontext* stash, const char* name, unsigned char* data, int dataSize, int freeData, int fontIndex)
 {
 	int i, ascent, descent, fh, lineGap;
 	FONSfont* font;
@@ -900,15 +966,16 @@ int fonsAddFontMem(FONScontext* stash, const char* name, unsigned char* data, in
 
 	// Init font
 	stash->nscratch = 0;
-	if (!fons__tt_loadFont(stash, &font->font, data, dataSize)) goto error;
+	if (!fons__tt_loadFont(stash, &font->font, data, dataSize, fontIndex)) 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);
+	ascent += lineGap;
 	fh = ascent - descent;
 	font->ascender = (float)ascent / (float)fh;
 	font->descender = (float)descent / (float)fh;
-	font->lineh = (float)(fh + lineGap) / (float)fh;
+	font->lineh = font->ascender - font->descender;
 
 	return idx;
 
@@ -1007,7 +1074,7 @@ static void fons__blur(FONScontext* stash, unsigned char* dst, int w, int h, int
 }
 
 static FONSglyph* fons__getGlyph(FONScontext* stash, FONSfont* font, unsigned int codepoint,
-								 short isize, short iblur)
+								 short isize, short iblur, int bitmapOption)
 {
 	int i, g, advance, lsb, x0, y0, x1, y1, gw, gh, gx, gy, x, y;
 	float scale;
@@ -1017,6 +1084,7 @@ static FONSglyph* fons__getGlyph(FONScontext* stash, FONSfont* font, unsigned in
 	int pad, added;
 	unsigned char* bdst;
 	unsigned char* dst;
+	FONSfont* renderFont = font;
 
 	if (isize < 2) return NULL;
 	if (iblur > 20) iblur = 20;
@@ -1029,32 +1097,66 @@ static FONSglyph* fons__getGlyph(FONScontext* stash, FONSfont* font, unsigned in
 	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];
+		if (font->glyphs[i].codepoint == codepoint && font->glyphs[i].size == isize && font->glyphs[i].blur == iblur) {
+			glyph = &font->glyphs[i];
+			if (bitmapOption == FONS_GLYPH_BITMAP_OPTIONAL || (glyph->x0 >= 0 && glyph->y0 >= 0)) {
+			  return glyph;
+			}
+			// At this point, glyph exists but the bitmap data is not yet created.
+			break;
+		}
 		i = font->glyphs[i].next;
 	}
 
-	// Could not find glyph, create it.
-	scale = fons__tt_getPixelHeightScale(&font->font, size);
+	// Create a new glyph or rasterize bitmap data for a cached glyph.
 	g = fons__tt_getGlyphIndex(&font->font, codepoint);
-	fons__tt_buildGlyphBitmap(&font->font, g, size, scale, &advance, &lsb, &x0, &y0, &x1, &y1);
+	// Try to find the glyph in fallback fonts.
+	if (g == 0) {
+		for (i = 0; i < font->nfallbacks; ++i) {
+			FONSfont* fallbackFont = stash->fonts[font->fallbacks[i]];
+			int fallbackIndex = fons__tt_getGlyphIndex(&fallbackFont->font, codepoint);
+			if (fallbackIndex != 0) {
+				g = fallbackIndex;
+				renderFont = fallbackFont;
+				break;
+			}
+		}
+		// It is possible that we did not find a fallback glyph.
+		// In that case the glyph index 'g' is 0, and we'll proceed below and cache empty glyph.
+	}
+	scale = fons__tt_getPixelHeightScale(&renderFont->font, size);
+	fons__tt_buildGlyphBitmap(&renderFont->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);
+	// Determines the spot to draw glyph in the atlas.
+	if (bitmapOption == FONS_GLYPH_BITMAP_REQUIRED) {
+		// 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;
+	} else {
+		// Negative coordinate indicates there is no bitmap data created.
+		gx = -1;
+		gy = -1;
 	}
-	if (added == 0) return NULL;
 
 	// Init glyph.
-	glyph = fons__allocGlyph(font);
-	glyph->codepoint = codepoint;
-	glyph->size = isize;
-	glyph->blur = iblur;
+	if (glyph == NULL) {
+		glyph = fons__allocGlyph(font);
+		glyph->codepoint = codepoint;
+		glyph->size = isize;
+		glyph->blur = iblur;
+		glyph->next = 0;
+
+		// Insert char to hash lookup.
+		glyph->next = font->lut[h];
+		font->lut[h] = font->nglyphs-1;
+	}
 	glyph->index = g;
 	glyph->x0 = (short)gx;
 	glyph->y0 = (short)gy;
@@ -1063,15 +1165,14 @@ static FONSglyph* fons__getGlyph(FONScontext* stash, FONSfont* font, unsigned in
 	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;
+	if (bitmapOption == FONS_GLYPH_BITMAP_OPTIONAL) {
+		return glyph;
+	}
 
 	// 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);
+	fons__tt_renderGlyphBitmap(&renderFont->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];
@@ -1098,7 +1199,7 @@ static FONSglyph* fons__getGlyph(FONScontext* stash, FONSfont* font, unsigned in
 	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);
+		fons__blur(stash, bdst, gw, gh, stash->params.width, iblur);
 	}
 
 	stash->dirtyRect[0] = fons__mini(stash->dirtyRect[0], glyph->x0);
@@ -1131,8 +1232,8 @@ static void fons__getQuad(FONScontext* stash, FONSfont* font,
 	y1 = (float)(glyph->y1-1);
 
 	if (stash->params.flags & FONS_ZERO_TOPLEFT) {
-		rx = (float)(int)(*x + xoff);
-		ry = (float)(int)(*y + yoff);
+		rx = floorf(*x + xoff);
+		ry = floorf(*y + yoff);
 
 		q->x0 = rx;
 		q->y0 = ry;
@@ -1144,8 +1245,8 @@ static void fons__getQuad(FONScontext* stash, FONSfont* font,
 		q->s1 = x1 * stash->itw;
 		q->t1 = y1 * stash->ith;
 	} else {
-		rx = (float)(int)(*x + xoff);
-		ry = (float)(int)(*y - yoff);
+		rx = floorf(*x + xoff);
+		ry = floorf(*y - yoff);
 
 		q->x0 = rx;
 		q->y0 = ry;
@@ -1260,7 +1361,7 @@ float fonsDrawText(FONScontext* stash,
 	for (; str != end; ++str) {
 		if (fons__decutf8(&utf8state, &codepoint, *(const unsigned char*)str))
 			continue;
-		glyph = fons__getGlyph(stash, font, codepoint, isize, iblur);
+		glyph = fons__getGlyph(stash, font, codepoint, isize, iblur, FONS_GLYPH_BITMAP_REQUIRED);
 		if (glyph != NULL) {
 			fons__getQuad(stash, font, prevGlyphIndex, glyph, scale, state->spacing, &x, &y, &q);
 
@@ -1283,7 +1384,7 @@ float fonsDrawText(FONScontext* stash,
 }
 
 int fonsTextIterInit(FONScontext* stash, FONStextIter* iter,
-					 float x, float y, const char* str, const char* end)
+					 float x, float y, const char* str, const char* end, int bitmapOption)
 {
 	FONSstate* state = fons__getState(stash);
 	float width;
@@ -1323,6 +1424,7 @@ int fonsTextIterInit(FONScontext* stash, FONStextIter* iter,
 	iter->end = end;
 	iter->codepoint = 0;
 	iter->prevGlyphIndex = -1;
+	iter->bitmapOption = bitmapOption;
 
 	return 1;
 }
@@ -1343,7 +1445,8 @@ int fonsTextIterNext(FONScontext* stash, FONStextIter* iter, FONSquad* quad)
 		// Get glyph and quad
 		iter->x = iter->nextx;
 		iter->y = iter->nexty;
-		glyph = fons__getGlyph(stash, iter->font, iter->codepoint, iter->isize, iter->iblur);
+		glyph = fons__getGlyph(stash, iter->font, iter->codepoint, iter->isize, iter->iblur, iter->bitmapOption);
+		// If the iterator was initialized with FONS_GLYPH_BITMAP_OPTIONAL, then the UV coordinates of the quad will be invalid.
 		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;
@@ -1440,7 +1543,7 @@ float fonsTextBounds(FONScontext* stash,
 	for (; str != end; ++str) {
 		if (fons__decutf8(&utf8state, &codepoint, *(const unsigned char*)str))
 			continue;
-		glyph = fons__getGlyph(stash, font, codepoint, isize, iblur);
+		glyph = fons__getGlyph(stash, font, codepoint, isize, iblur, FONS_GLYPH_BITMAP_OPTIONAL);
 		if (glyph != NULL) {
 			fons__getQuad(stash, font, prevGlyphIndex, glyph, scale, state->spacing, &x, &y, &q);
 			if (q.x0 < minx) minx = q.x0;
@@ -1564,6 +1667,7 @@ void fonsDeleteInternal(FONScontext* stash)
 	if (stash->fonts) free(stash->fonts);
 	if (stash->texData) free(stash->texData);
 	if (stash->scratch) free(stash->scratch);
+	fons__tt_done(stash);
 	free(stash);
 }
 
diff --git a/libs/nanovg/nanovg/nanovg.h b/libs/nanovg/nanovg/nanovg.h
index 870a04d0d..0d90570ff 100755
--- a/libs/nanovg/nanovg/nanovg.h
+++ b/libs/nanovg/nanovg/nanovg.h
@@ -79,10 +79,46 @@ enum NVGalign {
 	// 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. 
+	NVG_ALIGN_BOTTOM	= 1<<5,	// Align text vertically to bottom.
+	NVG_ALIGN_BASELINE	= 1<<6, // Default, align text vertically to baseline.
 };
 
+enum NVGblendFactor {
+	NVG_ZERO = 1<<0,
+	NVG_ONE = 1<<1,
+	NVG_SRC_COLOR = 1<<2,
+	NVG_ONE_MINUS_SRC_COLOR = 1<<3,
+	NVG_DST_COLOR = 1<<4,
+	NVG_ONE_MINUS_DST_COLOR = 1<<5,
+	NVG_SRC_ALPHA = 1<<6,
+	NVG_ONE_MINUS_SRC_ALPHA = 1<<7,
+	NVG_DST_ALPHA = 1<<8,
+	NVG_ONE_MINUS_DST_ALPHA = 1<<9,
+	NVG_SRC_ALPHA_SATURATE = 1<<10,
+};
+
+enum NVGcompositeOperation {
+	NVG_SOURCE_OVER,
+	NVG_SOURCE_IN,
+	NVG_SOURCE_OUT,
+	NVG_ATOP,
+	NVG_DESTINATION_OVER,
+	NVG_DESTINATION_IN,
+	NVG_DESTINATION_OUT,
+	NVG_DESTINATION_ATOP,
+	NVG_LIGHTER,
+	NVG_COPY,
+	NVG_XOR,
+};
+
+struct NVGcompositeOperationState {
+	int srcRGB;
+	int dstRGB;
+	int srcAlpha;
+	int dstAlpha;
+};
+typedef struct NVGcompositeOperationState NVGcompositeOperationState;
+
 struct NVGglyphPosition {
 	const char* str;	// Position of the glyph in the input string.
 	float x;			// The x-coordinate of the logical glyph position.
@@ -105,6 +141,7 @@ enum NVGimageFlags {
 	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.
+	NVG_IMAGE_NEAREST			= 1<<5,		// Image interpolation is Nearest instead Linear
 };
 
 // Begin drawing a new frame
@@ -115,7 +152,7 @@ enum NVGimageFlags {
 // 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);
+void nvgBeginFrame(NVGcontext* ctx, float windowWidth, float windowHeight, float devicePixelRatio);
 
 // Cancels drawing the current frame.
 void nvgCancelFrame(NVGcontext* ctx);
@@ -123,6 +160,22 @@ void nvgCancelFrame(NVGcontext* ctx);
 // Ends drawing flushing remaining render state.
 void nvgEndFrame(NVGcontext* ctx);
 
+//
+// Composite operation
+//
+// The composite operations in NanoVG are modeled after HTML Canvas API, and
+// the blend func is based on OpenGL (see corresponding manuals for more info).
+// The colors in the blending state have premultiplied alpha.
+
+// Sets the composite operation. The op parameter should be one of NVGcompositeOperation.
+void nvgGlobalCompositeOperation(NVGcontext* ctx, int op);
+
+// Sets the composite operation with custom pixel arithmetic. The parameters should be one of NVGblendFactor.
+void nvgGlobalCompositeBlendFunc(NVGcontext* ctx, int sfactor, int dfactor);
+
+// Sets the composite operation with custom pixel arithmetic for RGB and alpha components separately. The parameters should be one of NVGblendFactor.
+void nvgGlobalCompositeBlendFuncSeparate(NVGcontext* ctx, int srcRGB, int dstRGB, int srcAlpha, int dstAlpha);
+
 //
 // Color utils
 //
@@ -183,7 +236,10 @@ void nvgReset(NVGcontext* ctx);
 // 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(). 
+// Current render style can be saved and restored using nvgSave() and nvgRestore().
+
+// Sets whether to draw antialias for nvgStroke() and nvgFill(). It's enabled by default.
+void nvgShapeAntiAlias(NVGcontext* ctx, int enabled);
 
 // Sets current stroke style to a solid color.
 void nvgStrokeColor(NVGcontext* ctx, NVGcolor color);
@@ -231,7 +287,7 @@ void nvgGlobalAlpha(NVGcontext* ctx, float alpha);
 // 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(). 
+// Current coordinate system (transformation) can be saved and restored using nvgSave() and nvgRestore().
 
 // Resets current transform to a identity matrix.
 void nvgResetTransform(NVGcontext* ctx);
@@ -358,7 +414,7 @@ NVGpaint nvgBoxGradient(NVGcontext* ctx, float x, float y, float w, float h,
 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,
+// Creates and returns an image pattern. 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,
@@ -368,7 +424,7 @@ NVGpaint nvgImagePattern(NVGcontext* ctx, float ox, float oy, float ex, float ey
 // 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. 
+// 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.
@@ -423,7 +479,7 @@ void nvgArcTo(NVGcontext* ctx, float x1, float y1, float x2, float y2, float rad
 // Closes current sub-path with a line segment.
 void nvgClosePath(NVGcontext* ctx);
 
-// Sets the current sub-path winding, see NVGwinding and NVGsolidity. 
+// 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,
@@ -437,10 +493,13 @@ 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 rounded rectangle shaped sub-path with varying radii for each corner.
+void nvgRoundedRectVarying(NVGcontext* ctx, float x, float y, float w, float h, float radTopLeft, float radTopRight, float radBottomRight, float radBottomLeft);
+
 // Creates new ellipse shaped sub-path.
 void nvgEllipse(NVGcontext* ctx, float cx, float cy, float rx, float ry);
 
-// Creates new circle shaped sub-path. 
+// Creates new circle shaped sub-path.
 void nvgCircle(NVGcontext* ctx, float cx, float cy, float r);
 
 // Fills the current path with current fill style.
@@ -478,7 +537,7 @@ void nvgStroke(NVGcontext* ctx);
 //		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]);
+//		nvgRect(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.
@@ -487,13 +546,31 @@ void nvgStroke(NVGcontext* ctx);
 // Returns handle to the font.
 int nvgCreateFont(NVGcontext* ctx, const char* name, const char* filename);
 
+// fontIndex specifies which font face to load from a .ttf/.ttc file.
+int nvgCreateFontAtIndex(NVGcontext* ctx, const char* name, const char* filename, const int fontIndex);
+
 // Creates font 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);
 
+// fontIndex specifies which font face to load from a .ttf/.ttc file.
+int nvgCreateFontMemAtIndex(NVGcontext* ctx, const char* name, unsigned char* data, int ndata, int freeData, const int fontIndex);
+
 // 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);
 
+// Adds a fallback font by handle.
+int nvgAddFallbackFontId(NVGcontext* ctx, int baseFont, int fallbackFont);
+
+// Adds a fallback font by name.
+int nvgAddFallbackFont(NVGcontext* ctx, const char* baseFont, const char* fallbackFont);
+
+// Resets fallback fonts by handle.
+void nvgResetFallbackFontsId(NVGcontext* ctx, int baseFont);
+
+// Resets fallback fonts by name.
+void nvgResetFallbackFonts(NVGcontext* ctx, const char* baseFont);
+
 // Sets the font size of current text style.
 void nvgFontSize(NVGcontext* ctx, float size);
 
@@ -503,7 +580,7 @@ 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. 
+// 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.
@@ -588,12 +665,12 @@ struct NVGparams {
 	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 (*renderViewport)(void* uptr, float width, float height, float devicePixelRatio);
 	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 (*renderFill)(void* uptr, NVGpaint* paint, NVGcompositeOperationState compositeOperation, NVGscissor* scissor, float fringe, const float* bounds, const NVGpath* paths, int npaths);
+	void (*renderStroke)(void* uptr, NVGpaint* paint, NVGcompositeOperationState compositeOperation, NVGscissor* scissor, float fringe, float strokeWidth, const NVGpath* paths, int npaths);
+	void (*renderTriangles)(void* uptr, NVGpaint* paint, NVGcompositeOperationState compositeOperation, NVGscissor* scissor, const NVGvertex* verts, int nverts, float fringe);
 	void (*renderDelete)(void* uptr);
 };
 typedef struct NVGparams NVGparams;
diff --git a/libs/nanovg/nanovg/nanovg_gl.h b/libs/nanovg/nanovg/nanovg_gl.h
index 45d9fdea8..7904426c5 100755
--- a/libs/nanovg/nanovg/nanovg_gl.h
+++ b/libs/nanovg/nanovg/nanovg_gl.h
@@ -60,7 +60,7 @@ NVGcontext* nvgCreateGL2(int flags);
 void nvgDeleteGL2(NVGcontext* ctx);
 
 int nvglCreateImageFromHandleGL2(NVGcontext* ctx, GLuint textureId, int w, int h, int flags);
-GLuint nvglImageFromHandleGL2(NVGcontext* ctx, int image);
+GLuint nvglImageHandleGL2(NVGcontext* ctx, int image);
 
 #endif
 
@@ -150,6 +150,15 @@ struct GLNVGtexture {
 };
 typedef struct GLNVGtexture GLNVGtexture;
 
+struct GLNVGblend
+{
+	GLenum srcRGB;
+	GLenum dstRGB;
+	GLenum srcAlpha;
+	GLenum dstAlpha;
+};
+typedef struct GLNVGblend GLNVGblend;
+
 enum GLNVGcallType {
 	GLNVG_NONE = 0,
 	GLNVG_FILL,
@@ -166,6 +175,7 @@ struct GLNVGcall {
 	int triangleOffset;
 	int triangleCount;
 	int uniformOffset;
+	GLNVGblend blendFunc;
 };
 typedef struct GLNVGcall GLNVGcall;
 
@@ -256,7 +266,10 @@ struct GLNVGcontext {
 	GLenum stencilFunc;
 	GLint stencilFuncRef;
 	GLuint stencilFuncMask;
+	GLNVGblend blendFunc;
 	#endif
+
+	int dummyTex;
 };
 typedef struct GLNVGcontext GLNVGcontext;
 
@@ -306,7 +319,7 @@ static void glnvg__stencilFunc(GLNVGcontext* gl, GLenum func, GLint ref, GLuint
 	if ((gl->stencilFunc != func) ||
 		(gl->stencilFuncRef != ref) ||
 		(gl->stencilFuncMask != mask)) {
-		
+
 		gl->stencilFunc = func;
 		gl->stencilFuncRef = ref;
 		gl->stencilFuncMask = mask;
@@ -316,6 +329,21 @@ static void glnvg__stencilFunc(GLNVGcontext* gl, GLenum func, GLint ref, GLuint
 	glStencilFunc(func, ref, mask);
 #endif
 }
+static void glnvg__blendFuncSeparate(GLNVGcontext* gl, const GLNVGblend* blend)
+{
+#if NANOVG_GL_USE_STATE_FILTER
+	if ((gl->blendFunc.srcRGB != blend->srcRGB) ||
+		(gl->blendFunc.dstRGB != blend->dstRGB) ||
+		(gl->blendFunc.srcAlpha != blend->srcAlpha) ||
+		(gl->blendFunc.dstAlpha != blend->dstAlpha)) {
+
+		gl->blendFunc = *blend;
+		glBlendFuncSeparate(blend->srcRGB, blend->dstRGB, blend->srcAlpha,blend->dstAlpha);
+	}
+#else
+	glBlendFuncSeparate(blend->srcRGB, blend->dstRGB, blend->srcAlpha,blend->dstAlpha);
+#endif
+}
 
 static GLNVGtexture* glnvg__allocTexture(GLNVGcontext* gl)
 {
@@ -339,10 +367,10 @@ static GLNVGtexture* glnvg__allocTexture(GLNVGcontext* gl)
 		}
 		tex = &gl->textures[gl->ntextures++];
 	}
-	
+
 	memset(tex, 0, sizeof(*tex));
 	tex->id = ++gl->textureId;
-	
+
 	return tex;
 }
 
@@ -474,6 +502,8 @@ static void glnvg__getUniforms(GLNVGshader* shader)
 #endif
 }
 
+static int glnvg__renderCreateTexture(void* uptr, int type, int w, int h, int imageFlags, const unsigned char* data);
+
 static int glnvg__renderCreate(void* uptr)
 {
 	GLNVGcontext* gl = (GLNVGcontext*)uptr;
@@ -522,7 +552,7 @@ static int glnvg__renderCreate(void* uptr)
 		"	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 = 
+	static const char* fillFragShader =
 		"#ifdef GL_ES\n"
 		"#if defined(GL_FRAGMENT_PRECISION_HIGH) || defined(NANOVG_GL3)\n"
 		" precision highp float;\n"
@@ -600,6 +630,7 @@ static int glnvg__renderCreate(void* uptr)
 		"	float scissor = scissorMask(fpos);\n"
 		"#ifdef EDGE_AA\n"
 		"	float strokeAlpha = strokeMask();\n"
+		"	if (strokeAlpha < strokeThr) discard;\n"
 		"#else\n"
 		"	float strokeAlpha = 1.0;\n"
 		"#endif\n"
@@ -639,9 +670,6 @@ static int glnvg__renderCreate(void* uptr)
 		"		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"
@@ -671,11 +699,15 @@ static int glnvg__renderCreate(void* uptr)
 #if NANOVG_GL_USE_UNIFORMBUFFER
 	// Create UBOs
 	glUniformBlockBinding(gl->shader.prog, gl->shader.loc[GLNVG_LOC_FRAG], GLNVG_FRAG_BINDING);
-	glGenBuffers(1, &gl->fragBuf); 
+	glGenBuffers(1, &gl->fragBuf);
 	glGetIntegerv(GL_UNIFORM_BUFFER_OFFSET_ALIGNMENT, &align);
 #endif
 	gl->fragSize = sizeof(GLNVGfragUniforms) + align - sizeof(GLNVGfragUniforms) % align;
 
+	// Some platforms does not allow to have samples to unset textures.
+	// Create empty one which is bound when there's no texture specified.
+	gl->dummyTex = glnvg__renderCreateTexture(gl, NVG_TEXTURE_ALPHA, 1, 1, 0, NULL);
+
 	glnvg__checkError(gl, "create done");
 
 	glFinish();
@@ -698,7 +730,7 @@ static int glnvg__renderCreateTexture(void* uptr, int type, int w, int h, int im
 			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. 
+		// 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;
@@ -734,7 +766,7 @@ static int glnvg__renderCreateTexture(void* uptr, int type, int w, int h, int im
 	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)
+#if defined(NANOVG_GLES2) || defined (NANOVG_GL2)
 		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);
@@ -743,11 +775,24 @@ static int glnvg__renderCreateTexture(void* uptr, int type, int w, int h, int im
 #endif
 
 	if (imageFlags & NVG_IMAGE_GENERATE_MIPMAPS) {
-		glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR_MIPMAP_LINEAR);
+		if (imageFlags & NVG_IMAGE_NEAREST) {
+			glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST_MIPMAP_NEAREST);
+		} else {
+			glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR_MIPMAP_LINEAR);
+		}
+	} else {
+		if (imageFlags & NVG_IMAGE_NEAREST) {
+			glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
+		} else {
+			glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
+		}
+	}
+
+	if (imageFlags & NVG_IMAGE_NEAREST) {
+		glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
 	} else {
-		glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
+		glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_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);
@@ -813,7 +858,7 @@ static int glnvg__renderUpdateTexture(void* uptr, int image, int x, int y, int w
 	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
+#if defined(NANOVG_GLES2) || defined(NANOVG_GL2)
 		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);
@@ -899,19 +944,30 @@ static int glnvg__convertPaint(GLNVGcontext* gl, GLNVGfragUniforms* frag, NVGpai
 		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);
+			float m1[6], m2[6];
+			nvgTransformTranslate(m1, 0.0f, frag->extent[1] * 0.5f);
+			nvgTransformMultiply(m1, paint->xform);
+			nvgTransformScale(m2, 1.0f, -1.0f);
+			nvgTransformMultiply(m2, m1);
+			nvgTransformTranslate(m1, 0.0f, -frag->extent[1] * 0.5f);
+			nvgTransformMultiply(m1, m2);
+			nvgTransformInverse(invxform, m1);
 		} else {
 			nvgTransformInverse(invxform, paint->xform);
 		}
 		frag->type = NSVG_SHADER_FILLIMG;
 
+		#if NANOVG_GL_USE_UNIFORMBUFFER
 		if (tex->type == NVG_TEXTURE_RGBA)
 			frag->texType = (tex->flags & NVG_IMAGE_PREMULTIPLIED) ? 0 : 1;
 		else
 			frag->texType = 2;
+		#else
+		if (tex->type == NVG_TEXTURE_RGBA)
+			frag->texType = (tex->flags & NVG_IMAGE_PREMULTIPLIED) ? 0.0f : 1.0f;
+		else
+			frag->texType = 2.0f;
+		#endif
 //		printf("frag->texType = %d\n", frag->texType);
 	} else {
 		frag->type = NSVG_SHADER_FILLGRAD;
@@ -929,6 +985,7 @@ static GLNVGfragUniforms* nvg__fragUniformPtr(GLNVGcontext* gl, int i);
 
 static void glnvg__setUniforms(GLNVGcontext* gl, int uniformOffset, int image)
 {
+	GLNVGtexture* tex = NULL;
 #if NANOVG_GL_USE_UNIFORMBUFFER
 	glBindBufferRange(GL_UNIFORM_BUFFER, GLNVG_FRAG_BINDING, gl->fragBuf, uniformOffset, sizeof(GLNVGfragUniforms));
 #else
@@ -937,19 +994,22 @@ static void glnvg__setUniforms(GLNVGcontext* gl, int uniformOffset, int image)
 #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);
+		tex = glnvg__findTexture(gl, image);
 	}
+	// If no image is set, use empty texture
+	if (tex == NULL) {
+		tex = glnvg__findTexture(gl, gl->dummyTex);
+	}
+	glnvg__bindTexture(gl, tex != NULL ? tex->tex : 0);
+	glnvg__checkError(gl, "tex paint tex");
 }
 
-static void glnvg__renderViewport(void* uptr, int width, int height)
+static void glnvg__renderViewport(void* uptr, float width, float height, float devicePixelRatio)
 {
+	NVG_NOTUSED(devicePixelRatio);
 	GLNVGcontext* gl = (GLNVGcontext*)uptr;
-	gl->view[0] = (float)width;
-	gl->view[1] = (float)height;
+	gl->view[0] = width;
+	gl->view[1] = height;
 }
 
 static void glnvg__fill(GLNVGcontext* gl, GLNVGcall* call)
@@ -991,7 +1051,7 @@ static void glnvg__fill(GLNVGcontext* gl, GLNVGcall* call)
 	// Draw fill
 	glnvg__stencilFunc(gl, GL_NOTEQUAL, 0x0, 0xff);
 	glStencilOp(GL_ZERO, GL_ZERO, GL_ZERO);
-	glDrawArrays(GL_TRIANGLES, call->triangleOffset, call->triangleCount);
+	glDrawArrays(GL_TRIANGLE_STRIP, call->triangleOffset, call->triangleCount);
 
 	glDisable(GL_STENCIL_TEST);
 }
@@ -1004,12 +1064,12 @@ static void glnvg__convexFill(GLNVGcontext* gl, GLNVGcall* call)
 	glnvg__setUniforms(gl, call->uniformOffset, call->image);
 	glnvg__checkError(gl, "convex fill");
 
-	for (i = 0; i < npaths; i++)
+	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++)
+		if (paths[i].strokeCount > 0) {
 			glDrawArrays(GL_TRIANGLE_STRIP, paths[i].strokeOffset, paths[i].strokeCount);
+		}
 	}
 }
 
@@ -1038,7 +1098,7 @@ static void glnvg__stroke(GLNVGcontext* gl, GLNVGcall* call)
 		for (i = 0; i < npaths; i++)
 			glDrawArrays(GL_TRIANGLE_STRIP, paths[i].strokeOffset, paths[i].strokeCount);
 
-		// Clear stencil buffer.		
+		// 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);
@@ -1076,6 +1136,50 @@ static void glnvg__renderCancel(void* uptr) {
 	gl->nuniforms = 0;
 }
 
+static GLenum glnvg_convertBlendFuncFactor(int factor)
+{
+	if (factor == NVG_ZERO)
+		return GL_ZERO;
+	if (factor == NVG_ONE)
+		return GL_ONE;
+	if (factor == NVG_SRC_COLOR)
+		return GL_SRC_COLOR;
+	if (factor == NVG_ONE_MINUS_SRC_COLOR)
+		return GL_ONE_MINUS_SRC_COLOR;
+	if (factor == NVG_DST_COLOR)
+		return GL_DST_COLOR;
+	if (factor == NVG_ONE_MINUS_DST_COLOR)
+		return GL_ONE_MINUS_DST_COLOR;
+	if (factor == NVG_SRC_ALPHA)
+		return GL_SRC_ALPHA;
+	if (factor == NVG_ONE_MINUS_SRC_ALPHA)
+		return GL_ONE_MINUS_SRC_ALPHA;
+	if (factor == NVG_DST_ALPHA)
+		return GL_DST_ALPHA;
+	if (factor == NVG_ONE_MINUS_DST_ALPHA)
+		return GL_ONE_MINUS_DST_ALPHA;
+	if (factor == NVG_SRC_ALPHA_SATURATE)
+		return GL_SRC_ALPHA_SATURATE;
+	return GL_INVALID_ENUM;
+}
+
+static GLNVGblend glnvg__blendCompositeOperation(NVGcompositeOperationState op)
+{
+	GLNVGblend blend;
+	blend.srcRGB = glnvg_convertBlendFuncFactor(op.srcRGB);
+	blend.dstRGB = glnvg_convertBlendFuncFactor(op.dstRGB);
+	blend.srcAlpha = glnvg_convertBlendFuncFactor(op.srcAlpha);
+	blend.dstAlpha = glnvg_convertBlendFuncFactor(op.dstAlpha);
+	if (blend.srcRGB == GL_INVALID_ENUM || blend.dstRGB == GL_INVALID_ENUM || blend.srcAlpha == GL_INVALID_ENUM || blend.dstAlpha == GL_INVALID_ENUM)
+	{
+		blend.srcRGB = GL_ONE;
+		blend.dstRGB = GL_ONE_MINUS_SRC_ALPHA;
+		blend.srcAlpha = GL_ONE;
+		blend.dstAlpha = GL_ONE_MINUS_SRC_ALPHA;
+	}
+	return blend;
+}
+
 static void glnvg__renderFlush(void* uptr)
 {
 	GLNVGcontext* gl = (GLNVGcontext*)uptr;
@@ -1086,7 +1190,6 @@ static void glnvg__renderFlush(void* uptr)
 		// 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);
@@ -1105,6 +1208,10 @@ static void glnvg__renderFlush(void* uptr)
 		gl->stencilFunc = GL_ALWAYS;
 		gl->stencilFuncRef = 0;
 		gl->stencilFuncMask = 0xffffffff;
+		gl->blendFunc.srcRGB = GL_INVALID_ENUM;
+		gl->blendFunc.srcAlpha = GL_INVALID_ENUM;
+		gl->blendFunc.dstRGB = GL_INVALID_ENUM;
+		gl->blendFunc.dstAlpha = GL_INVALID_ENUM;
 		#endif
 
 #if NANOVG_GL_USE_UNIFORMBUFFER
@@ -1134,6 +1241,7 @@ static void glnvg__renderFlush(void* uptr)
 
 		for (i = 0; i < gl->ncalls; i++) {
 			GLNVGcall* call = &gl->calls[i];
+			glnvg__blendFuncSeparate(gl,&call->blendFunc);
 			if (call->type == GLNVG_FILL)
 				glnvg__fill(gl, call);
 			else if (call->type == GLNVG_CONVEXFILL)
@@ -1148,7 +1256,7 @@ static void glnvg__renderFlush(void* uptr)
 		glDisableVertexAttribArray(1);
 #if defined NANOVG_GL3
 		glBindVertexArray(0);
-#endif	
+#endif
 		glDisable(GL_CULL_FACE);
 			glBindBuffer(GL_ARRAY_BUFFER, 0);
 		glUseProgram(0);
@@ -1249,7 +1357,7 @@ static void glnvg__vset(NVGvertex* vtx, float x, float y, float u, float v)
 	vtx->v = v;
 }
 
-static void glnvg__renderFill(void* uptr, NVGpaint* paint, NVGscissor* scissor, float fringe,
+static void glnvg__renderFill(void* uptr, NVGpaint* paint, NVGcompositeOperationState compositeOperation, NVGscissor* scissor, float fringe,
 							  const float* bounds, const NVGpath* paths, int npaths)
 {
 	GLNVGcontext* gl = (GLNVGcontext*)uptr;
@@ -1261,16 +1369,21 @@ static void glnvg__renderFill(void* uptr, NVGpaint* paint, NVGscissor* scissor,
 	if (call == NULL) return;
 
 	call->type = GLNVG_FILL;
+	call->triangleCount = 4;
 	call->pathOffset = glnvg__allocPaths(gl, npaths);
 	if (call->pathOffset == -1) goto error;
 	call->pathCount = npaths;
 	call->image = paint->image;
+	call->blendFunc = glnvg__blendCompositeOperation(compositeOperation);
 
 	if (npaths == 1 && paths[0].convex)
+	{
 		call->type = GLNVG_CONVEXFILL;
+		call->triangleCount = 0;	// Bounding box fill quad not needed for convex fill
+	}
 
 	// Allocate vertices for all the paths.
-	maxverts = glnvg__maxVertCount(paths, npaths) + 6;
+	maxverts = glnvg__maxVertCount(paths, npaths) + call->triangleCount;
 	offset = glnvg__allocVerts(gl, maxverts);
 	if (offset == -1) goto error;
 
@@ -1292,20 +1405,16 @@ static void glnvg__renderFill(void* uptr, NVGpaint* paint, NVGscissor* scissor,
 		}
 	}
 
-	// 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) {
+		// Quad
+		call->triangleOffset = offset;
+		quad = &gl->verts[call->triangleOffset];
+		glnvg__vset(&quad[0], bounds[2], bounds[3], 0.5f, 1.0f);
+		glnvg__vset(&quad[1], bounds[2], bounds[1], 0.5f, 1.0f);
+		glnvg__vset(&quad[2], bounds[0], bounds[3], 0.5f, 1.0f);
+		glnvg__vset(&quad[3], bounds[0], bounds[1], 0.5f, 1.0f);
+
 		call->uniformOffset = glnvg__allocFragUniforms(gl, 2);
 		if (call->uniformOffset == -1) goto error;
 		// Simple shader for stencil
@@ -1330,7 +1439,7 @@ static void glnvg__renderFill(void* uptr, NVGpaint* paint, NVGscissor* scissor,
 	if (gl->ncalls > 0) gl->ncalls--;
 }
 
-static void glnvg__renderStroke(void* uptr, NVGpaint* paint, NVGscissor* scissor, float fringe,
+static void glnvg__renderStroke(void* uptr, NVGpaint* paint, NVGcompositeOperationState compositeOperation, NVGscissor* scissor, float fringe,
 								float strokeWidth, const NVGpath* paths, int npaths)
 {
 	GLNVGcontext* gl = (GLNVGcontext*)uptr;
@@ -1344,6 +1453,7 @@ static void glnvg__renderStroke(void* uptr, NVGpaint* paint, NVGscissor* scissor
 	if (call->pathOffset == -1) goto error;
 	call->pathCount = npaths;
 	call->image = paint->image;
+	call->blendFunc = glnvg__blendCompositeOperation(compositeOperation);
 
 	// Allocate vertices for all the paths.
 	maxverts = glnvg__maxVertCount(paths, npaths);
@@ -1385,8 +1495,8 @@ static void glnvg__renderStroke(void* uptr, NVGpaint* paint, NVGscissor* scissor
 	if (gl->ncalls > 0) gl->ncalls--;
 }
 
-static void glnvg__renderTriangles(void* uptr, NVGpaint* paint, NVGscissor* scissor,
-								   const NVGvertex* verts, int nverts)
+static void glnvg__renderTriangles(void* uptr, NVGpaint* paint, NVGcompositeOperationState compositeOperation, NVGscissor* scissor,
+								   const NVGvertex* verts, int nverts, float fringe)
 {
 	GLNVGcontext* gl = (GLNVGcontext*)uptr;
 	GLNVGcall* call = glnvg__allocCall(gl);
@@ -1396,6 +1506,7 @@ static void glnvg__renderTriangles(void* uptr, NVGpaint* paint, NVGscissor* scis
 
 	call->type = GLNVG_TRIANGLES;
 	call->image = paint->image;
+	call->blendFunc = glnvg__blendCompositeOperation(compositeOperation);
 
 	// Allocate vertices for all the paths.
 	call->triangleOffset = glnvg__allocVerts(gl, nverts);
@@ -1408,7 +1519,7 @@ static void glnvg__renderTriangles(void* uptr, NVGpaint* paint, NVGscissor* scis
 	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);
+	glnvg__convertPaint(gl, frag, paint, scissor, 1.0f, fringe, -1.0f);
 	frag->type = NSVG_SHADER_IMG;
 
 	return;
diff --git a/libs/nanovg/nanovg/nanovg_gl_utils.h b/libs/nanovg/nanovg/nanovg_gl_utils.h
index 1323e90c6..f7384d803 100755
--- a/libs/nanovg/nanovg/nanovg_gl_utils.h
+++ b/libs/nanovg/nanovg/nanovg_gl_utils.h
@@ -90,7 +90,18 @@ NVGLUframebuffer* nvgluCreateFramebuffer(NVGcontext* ctx, int w, int h, int imag
 	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;
+	if (glCheckFramebufferStatus(GL_FRAMEBUFFER) != GL_FRAMEBUFFER_COMPLETE) {
+#ifdef GL_DEPTH24_STENCIL8
+		// If GL_STENCIL_INDEX8 is not supported, try GL_DEPTH24_STENCIL8 as a fallback.
+		// Some graphics cards require a depth buffer along with a stencil.
+		glRenderbufferStorage(GL_RENDERBUFFER, GL_DEPTH24_STENCIL8, w, h);
+		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)
+#endif // GL_DEPTH24_STENCIL8
+			goto error;
+	}
 
 	glBindFramebuffer(GL_FRAMEBUFFER, defaultFBO);
 	glBindRenderbuffer(GL_RENDERBUFFER, defaultRBO);
diff --git a/libs/nanovg/nanovg/stb_image.h b/libs/nanovg/nanovg/stb_image.h
index e5382d7c5..a59fd06a2 100755
--- a/libs/nanovg/nanovg/stb_image.h
+++ b/libs/nanovg/nanovg/stb_image.h
@@ -1,5 +1,5 @@
-/* stb_image - v2.10 - public domain image loader - http://nothings.org/stb_image.h
-                                     no warranty implied; use at your own risk
+/* stb_image - v2.29 - public domain image loader - http://nothings.org/stb
+                                  no warranty implied; use at your own risk
 
    Do this:
       #define STB_IMAGE_IMPLEMENTATION
@@ -21,7 +21,7 @@
           avoid problematic images and only need the trivial interface
 
       JPEG baseline & progressive (12 bpc/arithmetic not supported, same as stock IJG lib)
-      PNG 1/2/4/8-bit-per-channel (16 bpc not supported)
+      PNG 1/2/4/8/16-bit-per-channel
 
       TGA (not sure what subset, if a subset)
       BMP non-1bpp, non-RLE
@@ -42,135 +42,36 @@
    Full documentation under "DOCUMENTATION" below.
 
 
-   Revision 2.00 release notes:
-
-      - Progressive JPEG is now supported.
-
-      - PPM and PGM binary formats are now supported, thanks to Ken Miller.
-
-      - x86 platforms now make use of SSE2 SIMD instructions for
-        JPEG decoding, and ARM platforms can use NEON SIMD if requested.
-        This work was done by Fabian "ryg" Giesen. SSE2 is used by
-        default, but NEON must be enabled explicitly; see docs.
-
-        With other JPEG optimizations included in this version, we see
-        2x speedup on a JPEG on an x86 machine, and a 1.5x speedup
-        on a JPEG on an ARM machine, relative to previous versions of this
-        library. The same results will not obtain for all JPGs and for all
-        x86/ARM machines. (Note that progressive JPEGs are significantly
-        slower to decode than regular JPEGs.) This doesn't mean that this
-        is the fastest JPEG decoder in the land; rather, it brings it
-        closer to parity with standard libraries. If you want the fastest
-        decode, look elsewhere. (See "Philosophy" section of docs below.)
-
-        See final bullet items below for more info on SIMD.
-
-      - Added STBI_MALLOC, STBI_REALLOC, and STBI_FREE macros for replacing
-        the memory allocator. Unlike other STBI libraries, these macros don't
-        support a context parameter, so if you need to pass a context in to
-        the allocator, you'll have to store it in a global or a thread-local
-        variable.
-
-      - Split existing STBI_NO_HDR flag into two flags, STBI_NO_HDR and
-        STBI_NO_LINEAR.
-            STBI_NO_HDR:     suppress implementation of .hdr reader format
-            STBI_NO_LINEAR:  suppress high-dynamic-range light-linear float API
-
-      - You can suppress implementation of any of the decoders to reduce
-        your code footprint by #defining one or more of the following
-        symbols before creating the implementation.
-
-            STBI_NO_JPEG
-            STBI_NO_PNG
-            STBI_NO_BMP
-            STBI_NO_PSD
-            STBI_NO_TGA
-            STBI_NO_GIF
-            STBI_NO_HDR
-            STBI_NO_PIC
-            STBI_NO_PNM   (.ppm and .pgm)
-
-      - You can request *only* certain decoders and suppress all other ones
-        (this will be more forward-compatible, as addition of new decoders
-        doesn't require you to disable them explicitly):
-
-            STBI_ONLY_JPEG
-            STBI_ONLY_PNG
-            STBI_ONLY_BMP
-            STBI_ONLY_PSD
-            STBI_ONLY_TGA
-            STBI_ONLY_GIF
-            STBI_ONLY_HDR
-            STBI_ONLY_PIC
-            STBI_ONLY_PNM   (.ppm and .pgm)
-
-         Note that you can define multiples of these, and you will get all
-         of them ("only x" and "only y" is interpreted to mean "only x&y").
-
-       - If you use STBI_NO_PNG (or _ONLY_ without PNG), and you still
-         want the zlib decoder to be available, #define STBI_SUPPORT_ZLIB
-
-      - Compilation of all SIMD code can be suppressed with
-            #define STBI_NO_SIMD
-        It should not be necessary to disable SIMD unless you have issues
-        compiling (e.g. using an x86 compiler which doesn't support SSE
-        intrinsics or that doesn't support the method used to detect
-        SSE2 support at run-time), and even those can be reported as
-        bugs so I can refine the built-in compile-time checking to be
-        smarter.
-
-      - The old STBI_SIMD system which allowed installing a user-defined
-        IDCT etc. has been removed. If you need this, don't upgrade. My
-        assumption is that almost nobody was doing this, and those who
-        were will find the built-in SIMD more satisfactory anyway.
-
-      - RGB values computed for JPEG images are slightly different from
-        previous versions of stb_image. (This is due to using less
-        integer precision in SIMD.) The C code has been adjusted so
-        that the same RGB values will be computed regardless of whether
-        SIMD support is available, so your app should always produce
-        consistent results. But these results are slightly different from
-        previous versions. (Specifically, about 3% of available YCbCr values
-        will compute different RGB results from pre-1.49 versions by +-1;
-        most of the deviating values are one smaller in the G channel.)
-
-      - If you must produce consistent results with previous versions of
-        stb_image, #define STBI_JPEG_OLD and you will get the same results
-        you used to; however, you will not get the SIMD speedups for
-        the YCbCr-to-RGB conversion step (although you should still see
-        significant JPEG speedup from the other changes).
-
-        Please note that STBI_JPEG_OLD is a temporary feature; it will be
-        removed in future versions of the library. It is only intended for
-        near-term back-compatibility use.
-
-
-   Latest revision history:
+LICENSE
+
+  See end of file for license information.
+
+RECENT REVISION HISTORY:
+
+      2.29  (2023-05-xx) optimizations
+      2.28  (2023-01-29) many error fixes, security errors, just tons of stuff
+      2.27  (2021-07-11) document stbi_info better, 16-bit PNM support, bug fixes
+      2.26  (2020-07-13) many minor fixes
+      2.25  (2020-02-02) fix warnings
+      2.24  (2020-02-02) fix warnings; thread-local failure_reason and flip_vertically
+      2.23  (2019-08-11) fix clang static analysis warning
+      2.22  (2019-03-04) gif fixes, fix warnings
+      2.21  (2019-02-25) fix typo in comment
+      2.20  (2019-02-07) support utf8 filenames in Windows; fix warnings and platform ifdefs
+      2.19  (2018-02-11) fix warning
+      2.18  (2018-01-30) fix warnings
+      2.17  (2018-01-29) bugfix, 1-bit BMP, 16-bitness query, fix warnings
+      2.16  (2017-07-23) all functions have 16-bit variants; optimizations; bugfixes
+      2.15  (2017-03-18) fix png-1,2,4; all Imagenet JPGs; no runtime SSE detection on GCC
+      2.14  (2017-03-03) remove deprecated STBI_JPEG_OLD; fixes for Imagenet JPGs
+      2.13  (2016-12-04) experimental 16-bit API, only for PNG so far; fixes
+      2.12  (2016-04-02) fix typo in 2.11 PSD fix that caused crashes
+      2.11  (2016-04-02) 16-bit PNGS; enable SSE2 in non-gcc x64
+                         RGB-format JPEG; remove white matting in PSD;
+                         allocate large structures on the stack;
+                         correct channel count for PNG & BMP
       2.10  (2016-01-22) avoid warning introduced in 2.09
       2.09  (2016-01-16) 16-bit TGA; comments in PNM files; STBI_REALLOC_SIZED
-      2.08  (2015-09-13) fix to 2.07 cleanup, reading RGB PSD as RGBA
-      2.07  (2015-09-13) partial animated GIF support
-                         limited 16-bit PSD support
-                         minor bugs, code cleanup, and compiler warnings
-      2.06  (2015-04-19) fix bug where PSD returns wrong '*comp' value
-      2.05  (2015-04-19) fix bug in progressive JPEG handling, fix warning
-      2.04  (2015-04-15) try to re-enable SIMD on MinGW 64-bit
-      2.03  (2015-04-12) additional corruption checking
-                         stbi_set_flip_vertically_on_load
-                         fix NEON support; fix mingw support
-      2.02  (2015-01-19) fix incorrect assert, fix warning
-      2.01  (2015-01-17) fix various warnings
-      2.00b (2014-12-25) fix STBI_MALLOC in progressive JPEG
-      2.00  (2014-12-25) optimize JPEG, including x86 SSE2 & ARM NEON SIMD
-                         progressive JPEG
-                         PGM/PPM support
-                         STBI_MALLOC,STBI_REALLOC,STBI_FREE
-                         STBI_NO_*, STBI_ONLY_*
-                         GIF bugfix
-      1.48  (2014-12-14) fix incorrectly-named assert()
-      1.47  (2014-12-14) 1/2/4-bit PNG support (both grayscale and paletted)
-                         optimize PNG
-                         fix bug in interlaced PNG with user-specified channel count
 
    See end of file for full revision history.
 
@@ -185,34 +86,43 @@
     Tom Seddon (pic)                       Omar Cornut (1/2/4-bit PNG)
     Thatcher Ulrich (psd)                  Nicolas Guillemot (vertical flip)
     Ken Miller (pgm, ppm)                  Richard Mitton (16-bit PSD)
-    urraka@github (animated gif)           Junggon Kim (PNM comments)
-                                           Daniel Gibson (16-bit TGA)
-
- Optimizations & bugfixes
-    Fabian "ryg" Giesen
-    Arseny Kapoulkine
+    github:urraka (animated gif)           Junggon Kim (PNM comments)
+    Christopher Forseth (animated gif)     Daniel Gibson (16-bit TGA)
+                                           socks-the-fox (16-bit PNG)
+                                           Jeremy Sawicki (handle all ImageNet JPGs)
+ Optimizations & bugfixes                  Mikhail Morozov (1-bit BMP)
+    Fabian "ryg" Giesen                    Anael Seghezzi (is-16-bit query)
+    Arseny Kapoulkine                      Simon Breuss (16-bit PNM)
+    John-Mark Allen
+    Carmelo J Fdez-Aguera
 
  Bug & warning fixes
-    Marc LeBlanc            David Woo          Guillaume George   Martins Mozeiko
-    Christpher Lloyd        Martin Golini      Jerry Jansson      Joseph Thomson
-    Dave Moore              Roy Eltham         Hayaki Saito       Phil Jordan
-    Won Chun                Luke Graham        Johan Duparc       Nathan Reed
-    the Horde3D community   Thomas Ruf         Ronny Chevalier    Nick Verigakis
-    Janez Zemva             John Bartholomew   Michal Cichon      svdijk@github
-    Jonathan Blow           Ken Hamada         Tero Hanninen      Baldur Karlsson
-    Laurent Gomila          Cort Stratton      Sergio Gonzalez    romigrou@github
-    Aruelien Pocheville     Thibault Reuille   Cass Everitt
-    Ryamond Barbiero        Paul Du Bois       Engin Manap
-    Blazej Dariusz Roszkowski
-    Michaelangel007@github
-
-
-LICENSE
-
-This software is in the public domain. Where that dedication is not
-recognized, you are granted a perpetual, irrevocable license to copy,
-distribute, and modify this file as you see fit.
-
+    Marc LeBlanc            David Woo          Guillaume George     Martins Mozeiko
+    Christpher Lloyd        Jerry Jansson      Joseph Thomson       Blazej Dariusz Roszkowski
+    Phil Jordan                                Dave Moore           Roy Eltham
+    Hayaki Saito            Nathan Reed        Won Chun
+    Luke Graham             Johan Duparc       Nick Verigakis       the Horde3D community
+    Thomas Ruf              Ronny Chevalier                         github:rlyeh
+    Janez Zemva             John Bartholomew   Michal Cichon        github:romigrou
+    Jonathan Blow           Ken Hamada         Tero Hanninen        github:svdijk
+    Eugene Golushkov        Laurent Gomila     Cort Stratton        github:snagar
+    Aruelien Pocheville     Sergio Gonzalez    Thibault Reuille     github:Zelex
+    Cass Everitt            Ryamond Barbiero                        github:grim210
+    Paul Du Bois            Engin Manap        Aldo Culquicondor    github:sammyhw
+    Philipp Wiesemann       Dale Weiler        Oriol Ferrer Mesia   github:phprus
+    Josh Tobin              Neil Bickford      Matthew Gregan       github:poppolopoppo
+    Julian Raschke          Gregory Mullen     Christian Floisand   github:darealshinji
+    Baldur Karlsson         Kevin Schmidt      JR Smith             github:Michaelangel007
+                            Brad Weinberger    Matvey Cherevko      github:mosra
+    Luca Sas                Alexander Veselov  Zack Middleton       [reserved]
+    Ryan C. Gordon          [reserved]                              [reserved]
+                     DO NOT ADD YOUR NAME HERE
+
+                     Jacko Dirks
+
+  To add your name to the credits, pick a random blank space in the middle and fill it.
+  80% of merge conflicts on stb PRs are due to people adding their name at the end
+  of the credits.
 */
 
 #ifndef STBI_INCLUDE_STB_IMAGE_H
@@ -221,10 +131,8 @@ distribute, and modify this file as you see fit.
 // DOCUMENTATION
 //
 // Limitations:
-//    - no 16-bit-per-channel PNG
 //    - no 12-bit-per-channel JPEG
 //    - no JPEGs with arithmetic coding
-//    - no 1-bit BMP
 //    - GIF always returns *comp=4
 //
 // Basic usage (see HDR discussion below for HDR usage):
@@ -234,13 +142,13 @@ distribute, and modify this file as you see fit.
 //    // ... 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)
+//    stbi_image_free(data);
 //
 // Standard parameters:
-//    int *x       -- outputs image width in pixels
-//    int *y       -- outputs image height in pixels
-//    int *comp    -- outputs # of image components in image file
-//    int req_comp -- if non-zero, # of image components requested in result
+//    int *x                 -- outputs image width in pixels
+//    int *y                 -- outputs image height in pixels
+//    int *channels_in_file  -- outputs # of image components in image file
+//    int desired_channels   -- if non-zero, # of image components requested in result
 //
 // The return value from an image loader is an 'unsigned char *' which points
 // to the pixel data, or NULL on an allocation failure or if the image is
@@ -248,11 +156,12 @@ distribute, and modify this file as you see fit.
 // with each pixel consisting of N interleaved 8-bit components; the first
 // pixel pointed to is top-left-most in the image. There is no padding between
 // image scanlines or between pixels, regardless of format. The number of
-// components N is 'req_comp' if req_comp is non-zero, or *comp otherwise.
-// If req_comp is non-zero, *comp has the number of components that _would_
-// have been output otherwise. E.g. if you set req_comp to 4, you will always
-// get RGBA output, but you can check *comp to see if it's trivially opaque
-// because e.g. there were only 3 channels in the source image.
+// components N is 'desired_channels' if desired_channels is non-zero, or
+// *channels_in_file otherwise. If desired_channels is non-zero,
+// *channels_in_file has the number of components that _would_ have been
+// output otherwise. E.g. if you set desired_channels to 4, you will always
+// get RGBA output, but you can check *channels_in_file to see if it's trivially
+// opaque because e.g. there were only 3 channels in the source image.
 //
 // An output image with N components has the following components interleaved
 // in this order in each pixel:
@@ -264,14 +173,50 @@ distribute, and modify this file as you see fit.
 //       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
+// and *x, *y, *channels_in_file will be unchanged. The function
+// stbi_failure_reason() can be queried for an extremely brief, end-user
+// unfriendly explanation of why the load failed. Define STBI_NO_FAILURE_STRINGS
+// to avoid compiling these strings at all, and STBI_FAILURE_USERMSG to get slightly
 // more user-friendly ones.
 //
 // Paletted PNG, BMP, GIF, and PIC images are automatically depalettized.
 //
+// To query the width, height and component count of an image without having to
+// decode the full file, you can use the stbi_info family of functions:
+//
+//   int x,y,n,ok;
+//   ok = stbi_info(filename, &x, &y, &n);
+//   // returns ok=1 and sets x, y, n if image is a supported format,
+//   // 0 otherwise.
+//
+// Note that stb_image pervasively uses ints in its public API for sizes,
+// including sizes of memory buffers. This is now part of the API and thus
+// hard to change without causing breakage. As a result, the various image
+// loaders all have certain limits on image size; these differ somewhat
+// by format but generally boil down to either just under 2GB or just under
+// 1GB. When the decoded image would be larger than this, stb_image decoding
+// will fail.
+//
+// Additionally, stb_image will reject image files that have any of their
+// dimensions set to a larger value than the configurable STBI_MAX_DIMENSIONS,
+// which defaults to 2**24 = 16777216 pixels. Due to the above memory limit,
+// the only way to have an image with such dimensions load correctly
+// is for it to have a rather extreme aspect ratio. Either way, the
+// assumption here is that such larger images are likely to be malformed
+// or malicious. If you do need to load an image with individual dimensions
+// larger than that, and it still fits in the overall size limit, you can
+// #define STBI_MAX_DIMENSIONS on your own to be something larger.
+//
+// ===========================================================================
+//
+// UNICODE:
+//
+//   If compiling for Windows and you wish to use Unicode filenames, compile
+//   with
+//       #define STBI_WINDOWS_UTF8
+//   and pass utf8-encoded filenames. Call stbi_convert_wchar_to_utf8 to convert
+//   Windows wchar_t filenames to utf8.
+//
 // ===========================================================================
 //
 // Philosophy
@@ -284,15 +229,15 @@ distribute, and modify this file as you see fit.
 //
 // Sometimes I let "good performance" creep up in priority over "easy to maintain",
 // and for best performance I may provide less-easy-to-use APIs that give higher
-// performance, in addition to the easy to use ones. Nevertheless, it's important
+// performance, in addition to the easy-to-use ones. Nevertheless, it's important
 // to keep in mind that from the standpoint of you, a client of this library,
-// all you care about is #1 and #3, and stb libraries do not emphasize #3 above all.
+// all you care about is #1 and #3, and stb libraries DO NOT emphasize #3 above all.
 //
 // Some secondary priorities arise directly from the first two, some of which
-// make more explicit reasons why performance can't be emphasized.
+// provide more explicit reasons why performance can't be emphasized.
 //
 //    - Portable ("ease of use")
-//    - Small footprint ("easy to maintain")
+//    - Small source code footprint ("easy to maintain")
 //    - No dependencies ("ease of use")
 //
 // ===========================================================================
@@ -324,13 +269,6 @@ distribute, and modify this file as you see fit.
 // (at least this is true for iOS and Android). Therefore, the NEON support is
 // toggled by a build flag: define STBI_NEON to get NEON loops.
 //
-// The output of the JPEG decoder is slightly different from versions where
-// SIMD support was introduced (that is, for versions before 1.49). The
-// difference is only +-1 in the 8-bit RGB channels, and only on a small
-// fraction of pixels. You can force the pre-1.49 behavior by defining
-// STBI_JPEG_OLD, but this will disable some of the SIMD decoding path
-// and hence cost some performance.
-//
 // If for some reason you do not want to use any of SIMD code, or if
 // you have issues compiling it, you can disable it entirely by
 // defining STBI_NO_SIMD.
@@ -339,11 +277,10 @@ distribute, and modify this file as you see fit.
 //
 // 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;
+// stb_image supports loading HDR images in general, and currently the Radiance
+// .HDR file format specifically. You can still load any file through the existing
+// interface; if you attempt to load an HDR file, it will be automatically remapped
+// to LDR, assuming gamma 2.2 and an arbitrary scale factor defaulting to 1;
 // both of these constants can be reconfigured through this interface:
 //
 //     stbi_hdr_to_ldr_gamma(2.2f);
@@ -375,18 +312,59 @@ distribute, and modify this file as you see fit.
 //
 // iPhone PNG support:
 //
-// By default we convert iphone-formatted PNGs back to RGB, even though
-// they are internally encoded differently. You can disable this conversion
-// by by calling stbi_convert_iphone_png_to_rgb(0), in which case
-// you will always just get the native iphone "format" through (which
-// is BGR stored in RGB).
+// We optionally support converting iPhone-formatted PNGs (which store
+// premultiplied BGRA) back to RGB, even though they're internally encoded
+// differently. To enable this conversion, call
+// stbi_convert_iphone_png_to_rgb(1).
 //
 // Call stbi_set_unpremultiply_on_load(1) as well to force a divide per
 // pixel to remove any premultiplied alpha *only* if the image file explicitly
 // says there's premultiplied data (currently only happens in iPhone images,
 // and only if iPhone convert-to-rgb processing is on).
 //
-
+// ===========================================================================
+//
+// ADDITIONAL CONFIGURATION
+//
+//  - You can suppress implementation of any of the decoders to reduce
+//    your code footprint by #defining one or more of the following
+//    symbols before creating the implementation.
+//
+//        STBI_NO_JPEG
+//        STBI_NO_PNG
+//        STBI_NO_BMP
+//        STBI_NO_PSD
+//        STBI_NO_TGA
+//        STBI_NO_GIF
+//        STBI_NO_HDR
+//        STBI_NO_PIC
+//        STBI_NO_PNM   (.ppm and .pgm)
+//
+//  - You can request *only* certain decoders and suppress all other ones
+//    (this will be more forward-compatible, as addition of new decoders
+//    doesn't require you to disable them explicitly):
+//
+//        STBI_ONLY_JPEG
+//        STBI_ONLY_PNG
+//        STBI_ONLY_BMP
+//        STBI_ONLY_PSD
+//        STBI_ONLY_TGA
+//        STBI_ONLY_GIF
+//        STBI_ONLY_HDR
+//        STBI_ONLY_PIC
+//        STBI_ONLY_PNM   (.ppm and .pgm)
+//
+//   - If you use STBI_NO_PNG (or _ONLY_ without PNG), and you still
+//     want the zlib decoder to be available, #define STBI_SUPPORT_ZLIB
+//
+//  - If you define STBI_MAX_DIMENSIONS, stb_image will reject images greater
+//    than that size (in either width or height) without further processing.
+//    This is to let programs in the wild set an upper bound to prevent
+//    denial-of-service attacks on untrusted data, as one could generate a
+//    valid image of gigantic dimensions and force stb_image to allocate a
+//    huge block of memory and spend disproportionate time decoding it. By
+//    default this is set to (1 << 24), which is 16777216, but that's still
+//    very big.
 
 #ifndef STBI_NO_STDIO
 #include <stdio.h>
@@ -396,25 +374,29 @@ distribute, and modify this file as you see fit.
 
 enum
 {
-   STBI_default = 0, // only used for req_comp
+   STBI_default = 0, // only used for desired_channels
 
-   STBI_grey       = 1,
+   STBI_grey = 1,
    STBI_grey_alpha = 2,
-   STBI_rgb        = 3,
-   STBI_rgb_alpha  = 4
+   STBI_rgb = 3,
+   STBI_rgb_alpha = 4
 };
 
+#include <stdlib.h>
 typedef unsigned char stbi_uc;
+typedef unsigned short stbi_us;
 
 #ifdef __cplusplus
 extern "C" {
 #endif
 
+#ifndef STBIDEF
 #ifdef STB_IMAGE_STATIC
 #define STBIDEF static
 #else
 #define STBIDEF extern
 #endif
+#endif
 
 //////////////////////////////////////////////////////////////////////////////
 //
@@ -427,68 +409,100 @@ extern "C" {
 
 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
+   int (*read)(void* user, char* data, int size); // fill 'data' with 'size' bytes.  return number of bytes actually read
+   void (*skip)(void* user, int n); // skip the next 'n' bytes, or 'unget' the last -n bytes if negative
+   int (*eof)(void* user); // returns nonzero if we are at end of file/data
 } stbi_io_callbacks;
 
-STBIDEF stbi_uc *stbi_load               (char              const *filename,           int *x, int *y, int *comp, int req_comp);
-STBIDEF stbi_uc *stbi_load_from_memory   (stbi_uc           const *buffer, int len   , int *x, int *y, int *comp, int req_comp);
-STBIDEF stbi_uc *stbi_load_from_callbacks(stbi_io_callbacks const *clbk  , void *user, int *x, int *y, int *comp, int req_comp);
+////////////////////////////////////
+//
+// 8-bits-per-channel interface
+//
+
+STBIDEF stbi_uc* stbi_load_from_memory(stbi_uc const* buffer, int len, int* x, int* y, int* channels_in_file, int desired_channels);
+STBIDEF stbi_uc* stbi_load_from_callbacks(stbi_io_callbacks const* clbk, void* user, int* x, int* y, int* channels_in_file, int desired_channels);
 
 #ifndef STBI_NO_STDIO
-STBIDEF stbi_uc *stbi_load_from_file  (FILE *f,                  int *x, int *y, int *comp, int req_comp);
+STBIDEF stbi_uc* stbi_load(char const* filename, int* x, int* y, int* channels_in_file, int desired_channels);
+STBIDEF stbi_uc* stbi_load_from_file(FILE* f, int* x, int* y, int* channels_in_file, int desired_channels);
 // for stbi_load_from_file, file pointer is left pointing immediately after image
 #endif
 
+#ifndef STBI_NO_GIF
+STBIDEF stbi_uc* stbi_load_gif_from_memory(stbi_uc const* buffer, int len, int** delays, int* x, int* y, int* z, int* comp, int req_comp);
+#endif
+
+#ifdef STBI_WINDOWS_UTF8
+STBIDEF int stbi_convert_wchar_to_utf8(char* buffer, size_t bufferlen, const wchar_t* input);
+#endif
+
+////////////////////////////////////
+//
+// 16-bits-per-channel interface
+//
+
+STBIDEF stbi_us* stbi_load_16_from_memory(stbi_uc const* buffer, int len, int* x, int* y, int* channels_in_file, int desired_channels);
+STBIDEF stbi_us* stbi_load_16_from_callbacks(stbi_io_callbacks const* clbk, void* user, int* x, int* y, int* channels_in_file, int desired_channels);
+
+#ifndef STBI_NO_STDIO
+STBIDEF stbi_us* stbi_load_16(char const* filename, int* x, int* y, int* channels_in_file, int desired_channels);
+STBIDEF stbi_us* stbi_load_from_file_16(FILE* f, int* x, int* y, int* channels_in_file, int desired_channels);
+#endif
+
+////////////////////////////////////
+//
+// float-per-channel interface
+//
 #ifndef STBI_NO_LINEAR
-   STBIDEF float *stbi_loadf                 (char const *filename,           int *x, int *y, int *comp, int req_comp);
-   STBIDEF float *stbi_loadf_from_memory     (stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp);
-   STBIDEF float *stbi_loadf_from_callbacks  (stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp, int req_comp);
+STBIDEF float* stbi_loadf_from_memory(stbi_uc const* buffer, int len, int* x, int* y, int* channels_in_file, int desired_channels);
+STBIDEF float* stbi_loadf_from_callbacks(stbi_io_callbacks const* clbk, void* user, int* x, int* y, int* channels_in_file, int desired_channels);
 
-   #ifndef STBI_NO_STDIO
-   STBIDEF float *stbi_loadf_from_file  (FILE *f,                int *x, int *y, int *comp, int req_comp);
-   #endif
+#ifndef STBI_NO_STDIO
+STBIDEF float* stbi_loadf(char const* filename, int* x, int* y, int* channels_in_file, int desired_channels);
+STBIDEF float* stbi_loadf_from_file(FILE* f, int* x, int* y, int* channels_in_file, int desired_channels);
+#endif
 #endif
 
 #ifndef STBI_NO_HDR
-   STBIDEF void   stbi_hdr_to_ldr_gamma(float gamma);
-   STBIDEF void   stbi_hdr_to_ldr_scale(float scale);
+STBIDEF void stbi_hdr_to_ldr_gamma(float gamma);
+STBIDEF void stbi_hdr_to_ldr_scale(float scale);
 #endif // STBI_NO_HDR
 
 #ifndef STBI_NO_LINEAR
-   STBIDEF void   stbi_ldr_to_hdr_gamma(float gamma);
-   STBIDEF void   stbi_ldr_to_hdr_scale(float scale);
+STBIDEF void stbi_ldr_to_hdr_gamma(float gamma);
+STBIDEF void stbi_ldr_to_hdr_scale(float scale);
 #endif // STBI_NO_LINEAR
 
 // stbi_is_hdr is always defined, but always returns false if STBI_NO_HDR
-STBIDEF int    stbi_is_hdr_from_callbacks(stbi_io_callbacks const *clbk, void *user);
-STBIDEF int    stbi_is_hdr_from_memory(stbi_uc const *buffer, int len);
+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);
+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);
+// on most compilers (and ALL modern mainstream compilers) this is threadsafe
+STBIDEF const char* stbi_failure_reason(void);
 
 // free the loaded image -- this is just free()
-STBIDEF void     stbi_image_free      (void *retval_from_stbi_load);
+STBIDEF void stbi_image_free(void* retval_from_stbi_load);
 
 // get image dimensions & components without fully decoding
-STBIDEF int      stbi_info_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp);
-STBIDEF int      stbi_info_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp);
+STBIDEF int stbi_info_from_memory(stbi_uc const* buffer, int len, int* x, int* y, int* comp);
+STBIDEF int stbi_info_from_callbacks(stbi_io_callbacks const* clbk, void* user, int* x, int* y, int* comp);
+STBIDEF int stbi_is_16_bit_from_memory(stbi_uc const* buffer, int len);
+STBIDEF int stbi_is_16_bit_from_callbacks(stbi_io_callbacks const* clbk, void* user);
 
 #ifndef STBI_NO_STDIO
-STBIDEF int      stbi_info            (char const *filename,     int *x, int *y, int *comp);
-STBIDEF int      stbi_info_from_file  (FILE *f,                  int *x, int *y, int *comp);
-
+STBIDEF int stbi_info(char const* filename, int* x, int* y, int* comp);
+STBIDEF int stbi_info_from_file(FILE* f, int* x, int* y, int* comp);
+STBIDEF int stbi_is_16_bit(char const* filename);
+STBIDEF int stbi_is_16_bit_from_file(FILE* f);
 #endif
 
 
-
 // for image formats that explicitly notate that they have premultiplied alpha,
 // we just return the colors as stored in the file. set this flag to force
 // unpremultiplication. results are undefined if the unpremultiply overflow.
@@ -501,15 +515,22 @@ STBIDEF void stbi_convert_iphone_png_to_rgb(int flag_true_if_should_convert);
 // flip the image vertically, so the first pixel in the output array is the bottom left
 STBIDEF void stbi_set_flip_vertically_on_load(int flag_true_if_should_flip);
 
+// as above, but only applies to images loaded on the thread that calls the function
+// this function is only available if your compiler supports thread-local variables;
+// calling it will fail to link if your compiler doesn't
+STBIDEF void stbi_set_unpremultiply_on_load_thread(int flag_true_if_should_unpremultiply);
+STBIDEF void stbi_convert_iphone_png_to_rgb_thread(int flag_true_if_should_convert);
+STBIDEF void stbi_set_flip_vertically_on_load_thread(int flag_true_if_should_flip);
+
 // ZLIB client - used by PNG, available for other purposes
 
-STBIDEF char *stbi_zlib_decode_malloc_guesssize(const char *buffer, int len, int initial_size, int *outlen);
-STBIDEF char *stbi_zlib_decode_malloc_guesssize_headerflag(const char *buffer, int len, int initial_size, int *outlen, int parse_header);
-STBIDEF char *stbi_zlib_decode_malloc(const char *buffer, int len, int *outlen);
-STBIDEF int   stbi_zlib_decode_buffer(char *obuffer, int olen, const char *ibuffer, int ilen);
+STBIDEF char* stbi_zlib_decode_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);
+STBIDEF char* stbi_zlib_decode_noheader_malloc(const char* buffer, int len, int* outlen);
+STBIDEF int stbi_zlib_decode_noheader_buffer(char* obuffer, int olen, const char* ibuffer, int ilen);
 
 
 #ifdef __cplusplus
@@ -523,37 +544,34 @@ STBIDEF int   stbi_zlib_decode_noheader_buffer(char *obuffer, int olen, const ch
 
 #ifdef STB_IMAGE_IMPLEMENTATION
 
-#if defined(STBI_ONLY_JPEG) || defined(STBI_ONLY_PNG) || defined(STBI_ONLY_BMP) \
-  || defined(STBI_ONLY_TGA) || defined(STBI_ONLY_GIF) || defined(STBI_ONLY_PSD) \
-  || defined(STBI_ONLY_HDR) || defined(STBI_ONLY_PIC) || defined(STBI_ONLY_PNM) \
-  || defined(STBI_ONLY_ZLIB)
-   #ifndef STBI_ONLY_JPEG
-   #define STBI_NO_JPEG
-   #endif
-   #ifndef STBI_ONLY_PNG
-   #define STBI_NO_PNG
-   #endif
-   #ifndef STBI_ONLY_BMP
-   #define STBI_NO_BMP
-   #endif
-   #ifndef STBI_ONLY_PSD
-   #define STBI_NO_PSD
-   #endif
-   #ifndef STBI_ONLY_TGA
-   #define STBI_NO_TGA
-   #endif
-   #ifndef STBI_ONLY_GIF
-   #define STBI_NO_GIF
-   #endif
-   #ifndef STBI_ONLY_HDR
-   #define STBI_NO_HDR
-   #endif
-   #ifndef STBI_ONLY_PIC
-   #define STBI_NO_PIC
-   #endif
-   #ifndef STBI_ONLY_PNM
-   #define STBI_NO_PNM
-   #endif
+#if defined(STBI_ONLY_JPEG) || defined(STBI_ONLY_PNG) || defined(STBI_ONLY_BMP) || defined(STBI_ONLY_TGA) || defined(STBI_ONLY_GIF) || defined(STBI_ONLY_PSD) || defined(STBI_ONLY_HDR) || defined(STBI_ONLY_PIC) || defined(STBI_ONLY_PNM) || defined(STBI_ONLY_ZLIB)
+#ifndef STBI_ONLY_JPEG
+#define STBI_NO_JPEG
+#endif
+#ifndef STBI_ONLY_PNG
+#define STBI_NO_PNG
+#endif
+#ifndef STBI_ONLY_BMP
+#define STBI_NO_BMP
+#endif
+#ifndef STBI_ONLY_PSD
+#define STBI_NO_PSD
+#endif
+#ifndef STBI_ONLY_TGA
+#define STBI_NO_TGA
+#endif
+#ifndef STBI_ONLY_GIF
+#define STBI_NO_GIF
+#endif
+#ifndef STBI_ONLY_HDR
+#define STBI_NO_HDR
+#endif
+#ifndef STBI_ONLY_PIC
+#define STBI_NO_PIC
+#endif
+#ifndef STBI_ONLY_PNM
+#define STBI_NO_PNM
+#endif
 #endif
 
 #if defined(STBI_NO_PNG) && !defined(STBI_SUPPORT_ZLIB) && !defined(STBI_NO_ZLIB)
@@ -565,9 +583,10 @@ STBIDEF int   stbi_zlib_decode_noheader_buffer(char *obuffer, int olen, const ch
 #include <stddef.h> // ptrdiff_t on osx
 #include <stdlib.h>
 #include <string.h>
+#include <limits.h>
 
 #if !defined(STBI_NO_LINEAR) || !defined(STBI_NO_HDR)
-#include <math.h>  // ldexp
+#include <math.h> // ldexp, pow
 #endif
 
 #ifndef STBI_NO_STDIO
@@ -579,38 +598,61 @@ STBIDEF int   stbi_zlib_decode_noheader_buffer(char *obuffer, int olen, const ch
 #define STBI_ASSERT(x) assert(x)
 #endif
 
+#ifdef __cplusplus
+#define STBI_EXTERN extern "C"
+#else
+#define STBI_EXTERN extern
+#endif
+
 
 #ifndef _MSC_VER
-   #ifdef __cplusplus
-   #define stbi_inline inline
-   #else
-   #define stbi_inline
-   #endif
+#ifdef __cplusplus
+#define stbi_inline inline
+#else
+#define stbi_inline
+#endif
 #else
-   #define stbi_inline __forceinline
+#define stbi_inline __forceinline
 #endif
 
+#ifndef STBI_NO_THREAD_LOCALS
+#if defined(__cplusplus) && __cplusplus >= 201103L
+#define STBI_THREAD_LOCAL thread_local
+#elif defined(__GNUC__) && __GNUC__ < 5
+#define STBI_THREAD_LOCAL __thread
+#elif defined(_MSC_VER)
+#define STBI_THREAD_LOCAL __declspec(thread)
+#elif defined(__STDC_VERSION__) && __STDC_VERSION__ >= 201112L && !defined(__STDC_NO_THREADS__)
+#define STBI_THREAD_LOCAL _Thread_local
+#endif
 
-#ifdef _MSC_VER
+#ifndef STBI_THREAD_LOCAL
+#if defined(__GNUC__)
+#define STBI_THREAD_LOCAL __thread
+#endif
+#endif
+#endif
+
+#if defined(_MSC_VER) || defined(__SYMBIAN32__)
 typedef unsigned short stbi__uint16;
-typedef   signed short stbi__int16;
-typedef unsigned int   stbi__uint32;
-typedef   signed int   stbi__int32;
+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 int16_t stbi__int16;
 typedef uint32_t stbi__uint32;
-typedef int32_t  stbi__int32;
+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];
+typedef unsigned char validate_uint32[sizeof(stbi__uint32) == 4 ? 1 : -1];
 
 #ifdef _MSC_VER
-#define STBI_NOTUSED(v)  (void)(v)
+#define STBI_NOTUSED(v) (void)(v)
 #else
-#define STBI_NOTUSED(v)  (void)sizeof(v)
+#define STBI_NOTUSED(v) (void)sizeof(v)
 #endif
 
 #ifdef _MSC_VER
@@ -618,9 +660,9 @@ typedef unsigned char validate_uint32[sizeof(stbi__uint32)==4 ? 1 : -1];
 #endif
 
 #ifdef STBI_HAS_LROTL
-   #define stbi_lrot(x,y)  _lrotl(x,y)
+#define stbi_lrot(x, y) _lrotl(x, y)
 #else
-   #define stbi_lrot(x,y)  (((x) << (y)) | ((x) >> (32 - (y))))
+#define stbi_lrot(x, y) (((x) << (y)) | ((x) >> (-(y)&31)))
 #endif
 
 #if defined(STBI_MALLOC) && defined(STBI_FREE) && (defined(STBI_REALLOC) || defined(STBI_REALLOC_SIZED))
@@ -632,13 +674,13 @@ typedef unsigned char validate_uint32[sizeof(stbi__uint32)==4 ? 1 : -1];
 #endif
 
 #ifndef STBI_MALLOC
-#define STBI_MALLOC(sz)           malloc(sz)
-#define STBI_REALLOC(p,newsz)     realloc(p,newsz)
-#define STBI_FREE(p)              free(p)
+#define STBI_MALLOC(sz) malloc(sz)
+#define STBI_REALLOC(p, newsz) realloc(p, newsz)
+#define STBI_FREE(p) free(p)
 #endif
 
 #ifndef STBI_REALLOC_SIZED
-#define STBI_REALLOC_SIZED(p,oldsz,newsz) STBI_REALLOC(p,newsz)
+#define STBI_REALLOC_SIZED(p, oldsz, newsz) STBI_REALLOC(p, newsz)
 #endif
 
 // x86/x64 detection
@@ -648,12 +690,14 @@ typedef unsigned char validate_uint32[sizeof(stbi__uint32)==4 ? 1 : -1];
 #define STBI__X86_TARGET
 #endif
 
-#if defined(__GNUC__) && (defined(STBI__X86_TARGET) || defined(STBI__X64_TARGET)) && !defined(__SSE2__) && !defined(STBI_NO_SIMD)
-// NOTE: not clear do we actually need this for the 64-bit path?
+#if defined(__GNUC__) && defined(STBI__X86_TARGET) && !defined(__SSE2__) && !defined(STBI_NO_SIMD)
 // gcc doesn't support sse2 intrinsics unless you compile with -msse2,
-// (but compiling with -msse2 allows the compiler to use SSE2 everywhere;
-// this is just broken and gcc are jerks for not fixing it properly
-// http://www.virtualdub.org/blog/pivot/entry.php?id=363 )
+// which in turn means it gets to use SSE2 everywhere. This is unfortunate,
+// but previous attempts to provide the SSE2 functions with runtime
+// detection caused numerous issues. The way architecture extensions are
+// exposed in GCC/Clang is, sadly, not really suited for one-file libs.
+// New behavior: if compiled with -msse2, we use SSE2 without any
+// detection; if not, we don't use it at all.
 #define STBI_NO_SIMD
 #endif
 
@@ -672,18 +716,18 @@ typedef unsigned char validate_uint32[sizeof(stbi__uint32)==4 ? 1 : -1];
 #define STBI_NO_SIMD
 #endif
 
-#if !defined(STBI_NO_SIMD) && defined(STBI__X86_TARGET)
+#if !defined(STBI_NO_SIMD) && (defined(STBI__X86_TARGET) || defined(STBI__X64_TARGET))
 #define STBI_SSE2
 #include <emmintrin.h>
 
 #ifdef _MSC_VER
 
-#if _MSC_VER >= 1400  // not VC6
+#if _MSC_VER >= 1400 // not VC6
 #include <intrin.h> // __cpuid
 static int stbi__cpuid3(void)
 {
    int info[4];
-   __cpuid(info,1);
+   __cpuid(info, 1);
    return info[3];
 }
 #else
@@ -701,25 +745,27 @@ static int stbi__cpuid3(void)
 
 #define STBI_SIMD_ALIGN(type, name) __declspec(align(16)) type name
 
-static int stbi__sse2_available()
+#if !defined(STBI_NO_JPEG) && defined(STBI_SSE2)
+static int stbi__sse2_available(void)
 {
    int info3 = stbi__cpuid3();
    return ((info3 >> 26) & 1) != 0;
 }
+#endif
+
 #else // assume GCC-style if not VC++
 #define STBI_SIMD_ALIGN(type, name) type name __attribute__((aligned(16)))
 
-static int stbi__sse2_available()
+#if !defined(STBI_NO_JPEG) && defined(STBI_SSE2)
+static int stbi__sse2_available(void)
 {
-#if defined(__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__) >= 408 // GCC 4.8 or later
-   // GCC 4.8+ has a nice way to do this
-   return __builtin_cpu_supports("sse2");
-#else
-   // portable way to do this, preferably without using GCC inline ASM?
-   // just bail for now.
-   return 0;
-#endif
+   // If we're even attempting to compile this on GCC/Clang, that means
+   // -msse2 is on, which means the compiler is allowed to use SSE2
+   // instructions at will, and so are we.
+   return 1;
 }
+#endif
+
 #endif
 #endif
 
@@ -730,14 +776,21 @@ static int stbi__sse2_available()
 
 #ifdef STBI_NEON
 #include <arm_neon.h>
-// assume GCC or Clang on ARM targets
+#ifdef _MSC_VER
+#define STBI_SIMD_ALIGN(type, name) __declspec(align(16)) type name
+#else
 #define STBI_SIMD_ALIGN(type, name) type name __attribute__((aligned(16)))
 #endif
+#endif
 
 #ifndef STBI_SIMD_ALIGN
 #define STBI_SIMD_ALIGN(type, name) type name
 #endif
 
+#ifndef STBI_MAX_DIMENSIONS
+#define STBI_MAX_DIMENSIONS (1 << 24)
+#endif
+
 ///////////////////////////////////////////////
 //
 //  stbi__context struct and start_xxx functions
@@ -750,74 +803,82 @@ typedef struct
    int img_n, img_out_n;
 
    stbi_io_callbacks io;
-   void *io_user_data;
+   void* io_user_data;
 
    int read_from_callbacks;
    int buflen;
    stbi_uc buffer_start[128];
+   int callback_already_read;
 
    stbi_uc *img_buffer, *img_buffer_end;
    stbi_uc *img_buffer_original, *img_buffer_original_end;
 } stbi__context;
 
 
-static void stbi__refill_buffer(stbi__context *s);
+static void stbi__refill_buffer(stbi__context* s);
 
 // initialize a memory-decode context
-static void stbi__start_mem(stbi__context *s, stbi_uc const *buffer, int len)
+static void stbi__start_mem(stbi__context* s, stbi_uc const* buffer, int len)
 {
    s->io.read = NULL;
    s->read_from_callbacks = 0;
-   s->img_buffer = s->img_buffer_original = (stbi_uc *) buffer;
-   s->img_buffer_end = s->img_buffer_original_end = (stbi_uc *) buffer+len;
+   s->callback_already_read = 0;
+   s->img_buffer = s->img_buffer_original = (stbi_uc*)buffer;
+   s->img_buffer_end = s->img_buffer_original_end = (stbi_uc*)buffer + len;
 }
 
 // initialize a callback-based context
-static void stbi__start_callbacks(stbi__context *s, stbi_io_callbacks *c, void *user)
+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;
+   s->callback_already_read = 0;
+   s->img_buffer = s->img_buffer_original = s->buffer_start;
    stbi__refill_buffer(s);
    s->img_buffer_original_end = s->img_buffer_end;
 }
 
 #ifndef STBI_NO_STDIO
 
-static int stbi__stdio_read(void *user, char *data, int size)
+static int stbi__stdio_read(void* user, char* data, int size)
 {
-   return (int) fread(data,1,size,(FILE*) user);
+   return (int)fread(data, 1, size, (FILE*)user);
 }
 
-static void stbi__stdio_skip(void *user, int n)
+static void stbi__stdio_skip(void* user, int n)
 {
-   fseek((FILE*) user, n, SEEK_CUR);
+   int ch;
+   fseek((FILE*)user, n, SEEK_CUR);
+   ch = fgetc((FILE*)user); /* have to read a byte to reset feof()'s flag */
+   if (ch != EOF)
+   {
+      ungetc(ch, (FILE*)user); /* push byte back onto stream if valid. */
+   }
 }
 
-static int stbi__stdio_eof(void *user)
+static int stbi__stdio_eof(void* user)
 {
-   return feof((FILE*) user);
+   return feof((FILE*)user) || ferror((FILE*)user);
 }
 
-static stbi_io_callbacks stbi__stdio_callbacks =
-{
+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)
+static void stbi__start_file(stbi__context* s, FILE* f)
 {
-   stbi__start_callbacks(s, &stbi__stdio_callbacks, (void *) 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)
+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
@@ -826,77 +887,206 @@ static void stbi__rewind(stbi__context *s)
    s->img_buffer_end = s->img_buffer_original_end;
 }
 
+enum
+{
+   STBI_ORDER_RGB,
+   STBI_ORDER_BGR
+};
+
+typedef struct
+{
+   int bits_per_channel;
+   int num_channels;
+   int channel_order;
+} stbi__result_info;
+
 #ifndef STBI_NO_JPEG
-static int      stbi__jpeg_test(stbi__context *s);
-static 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__jpeg_test(stbi__context* s);
+static void* stbi__jpeg_load(stbi__context* s, int* x, int* y, int* comp, int req_comp, stbi__result_info* ri);
+static int stbi__jpeg_info(stbi__context* s, int* x, int* y, int* comp);
 #endif
 
 #ifndef STBI_NO_PNG
-static int      stbi__png_test(stbi__context *s);
-static 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__png_test(stbi__context* s);
+static void* stbi__png_load(stbi__context* s, int* x, int* y, int* comp, int req_comp, stbi__result_info* ri);
+static int stbi__png_info(stbi__context* s, int* x, int* y, int* comp);
+static int stbi__png_is16(stbi__context* s);
 #endif
 
 #ifndef STBI_NO_BMP
-static int      stbi__bmp_test(stbi__context *s);
-static stbi_uc *stbi__bmp_load(stbi__context *s, int *x, int *y, int *comp, int req_comp);
-static int      stbi__bmp_info(stbi__context *s, int *x, int *y, int *comp);
+static int stbi__bmp_test(stbi__context* s);
+static void* stbi__bmp_load(stbi__context* s, int* x, int* y, int* comp, int req_comp, stbi__result_info* ri);
+static int stbi__bmp_info(stbi__context* s, int* x, int* y, int* comp);
 #endif
 
 #ifndef STBI_NO_TGA
-static int      stbi__tga_test(stbi__context *s);
-static 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__tga_test(stbi__context* s);
+static void* stbi__tga_load(stbi__context* s, int* x, int* y, int* comp, int req_comp, stbi__result_info* ri);
+static int stbi__tga_info(stbi__context* s, int* x, int* y, int* comp);
 #endif
 
 #ifndef STBI_NO_PSD
-static int      stbi__psd_test(stbi__context *s);
-static stbi_uc *stbi__psd_load(stbi__context *s, int *x, int *y, int *comp, int req_comp);
-static int      stbi__psd_info(stbi__context *s, int *x, int *y, int *comp);
+static int stbi__psd_test(stbi__context* s);
+static void* stbi__psd_load(stbi__context* s, int* x, int* y, int* comp, int req_comp, stbi__result_info* ri, int bpc);
+static int stbi__psd_info(stbi__context* s, int* x, int* y, int* comp);
+static int stbi__psd_is16(stbi__context* s);
 #endif
 
 #ifndef STBI_NO_HDR
-static int      stbi__hdr_test(stbi__context *s);
-static float   *stbi__hdr_load(stbi__context *s, int *x, int *y, int *comp, int req_comp);
-static int      stbi__hdr_info(stbi__context *s, int *x, int *y, int *comp);
+static int stbi__hdr_test(stbi__context* s);
+static float* stbi__hdr_load(stbi__context* s, int* x, int* y, int* comp, int req_comp, stbi__result_info* ri);
+static int stbi__hdr_info(stbi__context* s, int* x, int* y, int* comp);
 #endif
 
 #ifndef STBI_NO_PIC
-static int      stbi__pic_test(stbi__context *s);
-static stbi_uc *stbi__pic_load(stbi__context *s, int *x, int *y, int *comp, int req_comp);
-static int      stbi__pic_info(stbi__context *s, int *x, int *y, int *comp);
+static int stbi__pic_test(stbi__context* s);
+static void* stbi__pic_load(stbi__context* s, int* x, int* y, int* comp, int req_comp, stbi__result_info* ri);
+static int stbi__pic_info(stbi__context* s, int* x, int* y, int* comp);
 #endif
 
 #ifndef STBI_NO_GIF
-static int      stbi__gif_test(stbi__context *s);
-static 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);
+static int stbi__gif_test(stbi__context* s);
+static void* stbi__gif_load(stbi__context* s, int* x, int* y, int* comp, int req_comp, stbi__result_info* ri);
+static void* stbi__load_gif_main(stbi__context* s, int** delays, int* x, int* y, int* z, int* comp, int req_comp);
+static int stbi__gif_info(stbi__context* s, int* x, int* y, int* comp);
 #endif
 
 #ifndef STBI_NO_PNM
-static int      stbi__pnm_test(stbi__context *s);
-static stbi_uc *stbi__pnm_load(stbi__context *s, int *x, int *y, int *comp, int req_comp);
-static int      stbi__pnm_info(stbi__context *s, int *x, int *y, int *comp);
+static int stbi__pnm_test(stbi__context* s);
+static void* stbi__pnm_load(stbi__context* s, int* x, int* y, int* comp, int req_comp, stbi__result_info* ri);
+static int stbi__pnm_info(stbi__context* s, int* x, int* y, int* comp);
+static int stbi__pnm_is16(stbi__context* s);
 #endif
 
-// this is not threadsafe
-static const char *stbi__g_failure_reason;
+static
+#ifdef STBI_THREAD_LOCAL
+STBI_THREAD_LOCAL
+#endif
+const char* stbi__g_failure_reason;
 
-STBIDEF const char *stbi_failure_reason(void)
+STBIDEF const char* stbi_failure_reason(void)
 {
    return stbi__g_failure_reason;
 }
 
-static int stbi__err(const char *str)
+#ifndef STBI_NO_FAILURE_STRINGS
+static int stbi__err(const char* str)
 {
    stbi__g_failure_reason = str;
    return 0;
 }
+#endif
+
+static void* stbi__malloc(size_t size)
+{
+   return STBI_MALLOC(size);
+}
+
+// stb_image uses ints pervasively, including for offset calculations.
+// therefore the largest decoded image size we can support with the
+// current code, even on 64-bit targets, is INT_MAX. this is not a
+// significant limitation for the intended use case.
+//
+// we do, however, need to make sure our size calculations don't
+// overflow. hence a few helper functions for size calculations that
+// multiply integers together, making sure that they're non-negative
+// and no overflow occurs.
+
+// return 1 if the sum is valid, 0 on overflow.
+// negative terms are considered invalid.
+static int stbi__addsizes_valid(int a, int b)
+{
+   if (b < 0)
+      return 0;
+   // now 0 <= b <= INT_MAX, hence also
+   // 0 <= INT_MAX - b <= INTMAX.
+   // And "a + b <= INT_MAX" (which might overflow) is the
+   // same as a <= INT_MAX - b (no overflow)
+   return a <= INT_MAX - b;
+}
+
+// returns 1 if the product is valid, 0 on overflow.
+// negative factors are considered invalid.
+static int stbi__mul2sizes_valid(int a, int b)
+{
+   if (a < 0 || b < 0)
+      return 0;
+   if (b == 0)
+      return 1; // mul-by-0 is always safe
+   // portable way to check for no overflows in a*b
+   return a <= INT_MAX / b;
+}
+
+#if !defined(STBI_NO_JPEG) || !defined(STBI_NO_PNG) || !defined(STBI_NO_TGA) || !defined(STBI_NO_HDR)
+// returns 1 if "a*b + add" has no negative terms/factors and doesn't overflow
+static int stbi__mad2sizes_valid(int a, int b, int add)
+{
+   return stbi__mul2sizes_valid(a, b) && stbi__addsizes_valid(a * b, add);
+}
+#endif
+
+// returns 1 if "a*b*c + add" has no negative terms/factors and doesn't overflow
+static int stbi__mad3sizes_valid(int a, int b, int c, int add)
+{
+   return stbi__mul2sizes_valid(a, b) && stbi__mul2sizes_valid(a * b, c) &&
+          stbi__addsizes_valid(a * b * c, add);
+}
 
-static void *stbi__malloc(size_t size)
+// returns 1 if "a*b*c*d + add" has no negative terms/factors and doesn't overflow
+#if !defined(STBI_NO_LINEAR) || !defined(STBI_NO_HDR) || !defined(STBI_NO_PNM)
+static int stbi__mad4sizes_valid(int a, int b, int c, int d, int add)
 {
-    return STBI_MALLOC(size);
+   return stbi__mul2sizes_valid(a, b) && stbi__mul2sizes_valid(a * b, c) &&
+          stbi__mul2sizes_valid(a * b * c, d) && stbi__addsizes_valid(a * b * c * d, add);
+}
+#endif
+
+#if !defined(STBI_NO_JPEG) || !defined(STBI_NO_PNG) || !defined(STBI_NO_TGA) || !defined(STBI_NO_HDR)
+// mallocs with size overflow checking
+static void* stbi__malloc_mad2(int a, int b, int add)
+{
+   if (!stbi__mad2sizes_valid(a, b, add))
+      return NULL;
+   return stbi__malloc(a * b + add);
+}
+#endif
+
+static void* stbi__malloc_mad3(int a, int b, int c, int add)
+{
+   if (!stbi__mad3sizes_valid(a, b, c, add))
+      return NULL;
+   return stbi__malloc(a * b * c + add);
+}
+
+#if !defined(STBI_NO_LINEAR) || !defined(STBI_NO_HDR) || !defined(STBI_NO_PNM)
+static void* stbi__malloc_mad4(int a, int b, int c, int d, int add)
+{
+   if (!stbi__mad4sizes_valid(a, b, c, d, add))
+      return NULL;
+   return stbi__malloc(a * b * c * d + add);
+}
+#endif
+
+// returns 1 if the sum of two signed ints is valid (between -2^31 and 2^31-1 inclusive), 0 on overflow.
+static int stbi__addints_valid(int a, int b)
+{
+   if ((a >= 0) != (b >= 0))
+      return 1; // a and b have different signs, so no overflow
+   if (a < 0 && b < 0)
+      return a >= INT_MIN - b; // same as a + b >= INT_MIN; INT_MIN - b cannot overflow since b < 0.
+   return a <= INT_MAX - b;
+}
+
+// returns 1 if the product of two ints fits in a signed short, 0 on overflow.
+static int stbi__mul2shorts_valid(int a, int b)
+{
+   if (b == 0 || b == -1)
+      return 1; // multiplication by 0 is always 0; check for -1 so SHRT_MIN/b doesn't overflow
+   if ((a >= 0) == (b >= 0))
+      return a <= SHRT_MAX / b; // product is positive, so similar to mul2sizes_valid
+   if (b < 0)
+      return a <= SHRT_MIN / b; // same as a * b >= SHRT_MIN
+   return a >= SHRT_MIN / b;
 }
 
 // stbi__err - error
@@ -904,132 +1094,292 @@ static void *stbi__malloc(size_t size)
 // stbi__errpuc - error returning pointer to unsigned char
 
 #ifdef STBI_NO_FAILURE_STRINGS
-   #define stbi__err(x,y)  0
+#define stbi__err(x, y) 0
 #elif defined(STBI_FAILURE_USERMSG)
-   #define stbi__err(x,y)  stbi__err(y)
+#define stbi__err(x, y) stbi__err(y)
 #else
-   #define stbi__err(x,y)  stbi__err(x)
+#define stbi__err(x, y) stbi__err(x)
 #endif
 
-#define stbi__errpf(x,y)   ((float *)(size_t) (stbi__err(x,y)?NULL:NULL))
-#define stbi__errpuc(x,y)  ((unsigned char *)(size_t) (stbi__err(x,y)?NULL:NULL))
+#define stbi__errpf(x, y) ((float*)(size_t)(stbi__err(x, y) ? NULL : NULL))
+#define stbi__errpuc(x, y) ((unsigned char*)(size_t)(stbi__err(x, y) ? NULL : NULL))
 
-STBIDEF void stbi_image_free(void *retval_from_stbi_load)
+STBIDEF void stbi_image_free(void* retval_from_stbi_load)
 {
    STBI_FREE(retval_from_stbi_load);
 }
 
 #ifndef STBI_NO_LINEAR
-static float   *stbi__ldr_to_hdr(stbi_uc *data, int x, int y, int comp);
+static float* stbi__ldr_to_hdr(stbi_uc* data, int x, int y, int comp);
 #endif
 
 #ifndef STBI_NO_HDR
-static stbi_uc *stbi__hdr_to_ldr(float   *data, int x, int y, int comp);
+static stbi_uc* stbi__hdr_to_ldr(float* data, int x, int y, int comp);
 #endif
 
-static int stbi__vertically_flip_on_load = 0;
+static int stbi__vertically_flip_on_load_global = 0;
 
 STBIDEF void stbi_set_flip_vertically_on_load(int flag_true_if_should_flip)
 {
-    stbi__vertically_flip_on_load = flag_true_if_should_flip;
-}
-
-static unsigned char *stbi__load_main(stbi__context *s, int *x, int *y, int *comp, int req_comp)
-{
-   #ifndef STBI_NO_JPEG
-   if (stbi__jpeg_test(s)) return stbi__jpeg_load(s,x,y,comp,req_comp);
-   #endif
-   #ifndef STBI_NO_PNG
-   if (stbi__png_test(s))  return stbi__png_load(s,x,y,comp,req_comp);
-   #endif
-   #ifndef STBI_NO_BMP
-   if (stbi__bmp_test(s))  return stbi__bmp_load(s,x,y,comp,req_comp);
-   #endif
-   #ifndef STBI_NO_GIF
-   if (stbi__gif_test(s))  return stbi__gif_load(s,x,y,comp,req_comp);
-   #endif
-   #ifndef STBI_NO_PSD
-   if (stbi__psd_test(s))  return stbi__psd_load(s,x,y,comp,req_comp);
-   #endif
-   #ifndef STBI_NO_PIC
-   if (stbi__pic_test(s))  return stbi__pic_load(s,x,y,comp,req_comp);
-   #endif
-   #ifndef STBI_NO_PNM
-   if (stbi__pnm_test(s))  return stbi__pnm_load(s,x,y,comp,req_comp);
-   #endif
-
-   #ifndef STBI_NO_HDR
-   if (stbi__hdr_test(s)) {
-      float *hdr = stbi__hdr_load(s, x,y,comp,req_comp);
+   stbi__vertically_flip_on_load_global = flag_true_if_should_flip;
+}
+
+#ifndef STBI_THREAD_LOCAL
+#define stbi__vertically_flip_on_load stbi__vertically_flip_on_load_global
+#else
+static STBI_THREAD_LOCAL int stbi__vertically_flip_on_load_local, stbi__vertically_flip_on_load_set;
+
+STBIDEF void stbi_set_flip_vertically_on_load_thread(int flag_true_if_should_flip)
+{
+   stbi__vertically_flip_on_load_local = flag_true_if_should_flip;
+   stbi__vertically_flip_on_load_set = 1;
+}
+
+#define stbi__vertically_flip_on_load (stbi__vertically_flip_on_load_set     \
+                                       ? stbi__vertically_flip_on_load_local \
+                                       : stbi__vertically_flip_on_load_global)
+#endif // STBI_THREAD_LOCAL
+
+static void* stbi__load_main(stbi__context* s, int* x, int* y, int* comp, int req_comp, stbi__result_info* ri, int bpc)
+{
+   memset(ri, 0, sizeof(*ri)); // make sure it's initialized if we add new fields
+   ri->bits_per_channel = 8; // default is 8 so most paths don't have to be changed
+   ri->channel_order = STBI_ORDER_RGB; // all current input & output are this, but this is here so we can add BGR order
+   ri->num_channels = 0;
+
+// test the formats with a very explicit header first (at least a FOURCC
+// or distinctive magic number first)
+#ifndef STBI_NO_PNG
+   if (stbi__png_test(s))
+      return stbi__png_load(s, x, y, comp, req_comp, ri);
+#endif
+#ifndef STBI_NO_BMP
+   if (stbi__bmp_test(s))
+      return stbi__bmp_load(s, x, y, comp, req_comp, ri);
+#endif
+#ifndef STBI_NO_GIF
+   if (stbi__gif_test(s))
+      return stbi__gif_load(s, x, y, comp, req_comp, ri);
+#endif
+#ifndef STBI_NO_PSD
+   if (stbi__psd_test(s))
+      return stbi__psd_load(s, x, y, comp, req_comp, ri, bpc);
+#else
+   STBI_NOTUSED(bpc);
+#endif
+#ifndef STBI_NO_PIC
+   if (stbi__pic_test(s))
+      return stbi__pic_load(s, x, y, comp, req_comp, ri);
+#endif
+
+// then the formats that can end up attempting to load with just 1 or 2
+// bytes matching expectations; these are prone to false positives, so
+// try them later
+#ifndef STBI_NO_JPEG
+   if (stbi__jpeg_test(s))
+      return stbi__jpeg_load(s, x, y, comp, req_comp, ri);
+#endif
+#ifndef STBI_NO_PNM
+   if (stbi__pnm_test(s))
+      return stbi__pnm_load(s, x, y, comp, req_comp, ri);
+#endif
+
+#ifndef STBI_NO_HDR
+   if (stbi__hdr_test(s))
+   {
+      float* hdr = stbi__hdr_load(s, x, y, comp, req_comp, ri);
       return stbi__hdr_to_ldr(hdr, *x, *y, req_comp ? req_comp : *comp);
    }
-   #endif
+#endif
 
-   #ifndef STBI_NO_TGA
+#ifndef STBI_NO_TGA
    // test tga last because it's a crappy test!
    if (stbi__tga_test(s))
-      return stbi__tga_load(s,x,y,comp,req_comp);
-   #endif
+      return stbi__tga_load(s, x, y, comp, req_comp, ri);
+#endif
 
    return stbi__errpuc("unknown image type", "Image not of any known type, or corrupt");
 }
 
-static unsigned char *stbi__load_flip(stbi__context *s, int *x, int *y, int *comp, int req_comp)
+static stbi_uc* stbi__convert_16_to_8(stbi__uint16* orig, int w, int h, int channels)
 {
-   unsigned char *result = stbi__load_main(s, x, y, comp, req_comp);
+   int i;
+   int img_len = w * h * channels;
+   stbi_uc* reduced;
 
-   if (stbi__vertically_flip_on_load && result != NULL) {
-      int w = *x, h = *y;
-      int depth = req_comp ? req_comp : *comp;
-      int row,col,z;
-      stbi_uc temp;
+   reduced = (stbi_uc*)stbi__malloc(img_len);
+   if (reduced == NULL)
+      return stbi__errpuc("outofmem", "Out of memory");
 
-      // @OPTIMIZE: use a bigger temp buffer and memcpy multiple pixels at once
-      for (row = 0; row < (h>>1); row++) {
-         for (col = 0; col < w; col++) {
-            for (z = 0; z < depth; z++) {
-               temp = result[(row * w + col) * depth + z];
-               result[(row * w + col) * depth + z] = result[((h - row - 1) * w + col) * depth + z];
-               result[((h - row - 1) * w + col) * depth + z] = temp;
-            }
-         }
+   for (i = 0; i < img_len; ++i)
+      reduced[i] = (stbi_uc)((orig[i] >> 8) & 0xFF); // top half of each byte is sufficient approx of 16->8 bit scaling
+
+   STBI_FREE(orig);
+   return reduced;
+}
+
+static stbi__uint16* stbi__convert_8_to_16(stbi_uc* orig, int w, int h, int channels)
+{
+   int i;
+   int img_len = w * h * channels;
+   stbi__uint16* enlarged;
+
+   enlarged = (stbi__uint16*)stbi__malloc(img_len * 2);
+   if (enlarged == NULL)
+      return (stbi__uint16*)stbi__errpuc("outofmem", "Out of memory");
+
+   for (i = 0; i < img_len; ++i)
+      enlarged[i] = (stbi__uint16)((orig[i] << 8) + orig[i]); // replicate to high and low byte, maps 0->0, 255->0xffff
+
+   STBI_FREE(orig);
+   return enlarged;
+}
+
+static void stbi__vertical_flip(void* image, int w, int h, int bytes_per_pixel)
+{
+   int row;
+   size_t bytes_per_row = (size_t)w * bytes_per_pixel;
+   stbi_uc temp[2048];
+   stbi_uc* bytes = (stbi_uc*)image;
+
+   for (row = 0; row < (h >> 1); row++)
+   {
+      stbi_uc* row0 = bytes + row * bytes_per_row;
+      stbi_uc* row1 = bytes + (h - row - 1) * bytes_per_row;
+      // swap row0 with row1
+      size_t bytes_left = bytes_per_row;
+      while (bytes_left)
+      {
+         size_t bytes_copy = (bytes_left < sizeof(temp)) ? bytes_left : sizeof(temp);
+         memcpy(temp, row0, bytes_copy);
+         memcpy(row0, row1, bytes_copy);
+         memcpy(row1, temp, bytes_copy);
+         row0 += bytes_copy;
+         row1 += bytes_copy;
+         bytes_left -= bytes_copy;
       }
    }
+}
 
-   return result;
+#ifndef STBI_NO_GIF
+static void stbi__vertical_flip_slices(void* image, int w, int h, int z, int bytes_per_pixel)
+{
+   int slice;
+   int slice_size = w * h * bytes_per_pixel;
+
+   stbi_uc* bytes = (stbi_uc*)image;
+   for (slice = 0; slice < z; ++slice)
+   {
+      stbi__vertical_flip(bytes, w, h, bytes_per_pixel);
+      bytes += slice_size;
+   }
 }
+#endif
 
-#ifndef STBI_NO_HDR
-static void stbi__float_postprocess(float *result, int *x, int *y, int *comp, int req_comp)
-{
-   if (stbi__vertically_flip_on_load && result != NULL) {
-      int w = *x, h = *y;
-      int depth = req_comp ? req_comp : *comp;
-      int row,col,z;
-      float temp;
-
-      // @OPTIMIZE: use a bigger temp buffer and memcpy multiple pixels at once
-      for (row = 0; row < (h>>1); row++) {
-         for (col = 0; col < w; col++) {
-            for (z = 0; z < depth; z++) {
-               temp = result[(row * w + col) * depth + z];
-               result[(row * w + col) * depth + z] = result[((h - row - 1) * w + col) * depth + z];
-               result[((h - row - 1) * w + col) * depth + z] = temp;
-            }
-         }
-      }
+static unsigned char* stbi__load_and_postprocess_8bit(stbi__context* s, int* x, int* y, int* comp, int req_comp)
+{
+   stbi__result_info ri;
+   void* result = stbi__load_main(s, x, y, comp, req_comp, &ri, 8);
+
+   if (result == NULL)
+      return NULL;
+
+   // it is the responsibility of the loaders to make sure we get either 8 or 16 bit.
+   STBI_ASSERT(ri.bits_per_channel == 8 || ri.bits_per_channel == 16);
+
+   if (ri.bits_per_channel != 8)
+   {
+      result = stbi__convert_16_to_8((stbi__uint16*)result, *x, *y, req_comp == 0 ? *comp : req_comp);
+      ri.bits_per_channel = 8;
+   }
+
+   // @TODO: move stbi__convert_format to here
+
+   if (stbi__vertically_flip_on_load)
+   {
+      int channels = req_comp ? req_comp : *comp;
+      stbi__vertical_flip(result, *x, *y, channels * sizeof(stbi_uc));
+   }
+
+   return (unsigned char*)result;
+}
+
+static stbi__uint16* stbi__load_and_postprocess_16bit(stbi__context* s, int* x, int* y, int* comp, int req_comp)
+{
+   stbi__result_info ri;
+   void* result = stbi__load_main(s, x, y, comp, req_comp, &ri, 16);
+
+   if (result == NULL)
+      return NULL;
+
+   // it is the responsibility of the loaders to make sure we get either 8 or 16 bit.
+   STBI_ASSERT(ri.bits_per_channel == 8 || ri.bits_per_channel == 16);
+
+   if (ri.bits_per_channel != 16)
+   {
+      result = stbi__convert_8_to_16((stbi_uc*)result, *x, *y, req_comp == 0 ? *comp : req_comp);
+      ri.bits_per_channel = 16;
+   }
+
+   // @TODO: move stbi__convert_format16 to here
+   // @TODO: special case RGB-to-Y (and RGBA-to-YA) for 8-bit-to-16-bit case to keep more precision
+
+   if (stbi__vertically_flip_on_load)
+   {
+      int channels = req_comp ? req_comp : *comp;
+      stbi__vertical_flip(result, *x, *y, channels * sizeof(stbi__uint16));
+   }
+
+   return (stbi__uint16*)result;
+}
+
+#if !defined(STBI_NO_HDR) && !defined(STBI_NO_LINEAR)
+static void stbi__float_postprocess(float* result, int* x, int* y, int* comp, int req_comp)
+{
+   if (stbi__vertically_flip_on_load && result != NULL)
+   {
+      int channels = req_comp ? req_comp : *comp;
+      stbi__vertical_flip(result, *x, *y, channels * sizeof(float));
    }
 }
 #endif
 
 #ifndef STBI_NO_STDIO
 
-static FILE *stbi__fopen(char const *filename, char const *mode)
+#if defined(_WIN32) && defined(STBI_WINDOWS_UTF8)
+STBI_EXTERN __declspec(dllimport) int __stdcall MultiByteToWideChar(unsigned int cp, unsigned long flags, const char* str, int cbmb, wchar_t* widestr, int cchwide);
+STBI_EXTERN __declspec(dllimport) int __stdcall WideCharToMultiByte(unsigned int cp, unsigned long flags, const wchar_t* widestr, int cchwide, char* str, int cbmb, const char* defchar, int* used_default);
+#endif
+
+#if defined(_WIN32) && defined(STBI_WINDOWS_UTF8)
+STBIDEF int stbi_convert_wchar_to_utf8(char* buffer, size_t bufferlen, const wchar_t* input)
+{
+   return WideCharToMultiByte(65001 /* UTF8 */, 0, input, -1, buffer, (int)bufferlen, NULL, NULL);
+}
+#endif
+
+static FILE* stbi__fopen(char const* filename, char const* mode)
 {
-   FILE *f;
+   FILE* f;
+#if defined(_WIN32) && defined(STBI_WINDOWS_UTF8)
+   wchar_t wMode[64];
+   wchar_t wFilename[1024];
+   if (0 == MultiByteToWideChar(65001 /* UTF8 */, 0, filename, -1, wFilename, sizeof(wFilename) / sizeof(*wFilename)))
+      return 0;
+
+   if (0 == MultiByteToWideChar(65001 /* UTF8 */, 0, mode, -1, wMode, sizeof(wMode) / sizeof(*wMode)))
+      return 0;
+
 #if defined(_MSC_VER) && _MSC_VER >= 1400
+   if (0 != _wfopen_s(&f, wFilename, wMode))
+      f = 0;
+#else
+   f = _wfopen(wFilename, wMode);
+#endif
+
+#elif defined(_MSC_VER) && _MSC_VER >= 1400
    if (0 != fopen_s(&f, filename, mode))
-      f=0;
+      f = 0;
 #else
    f = fopen(filename, mode);
 #endif
@@ -1037,92 +1387,155 @@ static FILE *stbi__fopen(char const *filename, char const *mode)
 }
 
 
-STBIDEF stbi_uc *stbi_load(char const *filename, int *x, int *y, int *comp, int req_comp)
+STBIDEF stbi_uc* stbi_load(char const* filename, int* x, int* y, int* comp, int req_comp)
 {
-   FILE *f = stbi__fopen(filename, "rb");
-   unsigned char *result;
-   if (!f) return stbi__errpuc("can't fopen", "Unable to open file");
-   result = stbi_load_from_file(f,x,y,comp,req_comp);
+   FILE* f = stbi__fopen(filename, "rb");
+   unsigned char* result;
+   if (!f)
+      return stbi__errpuc("can't fopen", "Unable to open file");
+   result = stbi_load_from_file(f, x, y, comp, req_comp);
    fclose(f);
    return result;
 }
 
-STBIDEF stbi_uc *stbi_load_from_file(FILE *f, int *x, int *y, int *comp, int req_comp)
+STBIDEF stbi_uc* stbi_load_from_file(FILE* f, int* x, int* y, int* comp, int req_comp)
 {
-   unsigned char *result;
+   unsigned char* result;
    stbi__context s;
-   stbi__start_file(&s,f);
-   result = stbi__load_flip(&s,x,y,comp,req_comp);
-   if (result) {
+   stbi__start_file(&s, f);
+   result = stbi__load_and_postprocess_8bit(&s, x, y, comp, req_comp);
+   if (result)
+   {
       // need to 'unget' all the characters in the IO buffer
-      fseek(f, - (int) (s.img_buffer_end - s.img_buffer), SEEK_CUR);
+      fseek(f, -(int)(s.img_buffer_end - s.img_buffer), SEEK_CUR);
    }
    return result;
 }
+
+STBIDEF stbi__uint16* stbi_load_from_file_16(FILE* f, int* x, int* y, int* comp, int req_comp)
+{
+   stbi__uint16* result;
+   stbi__context s;
+   stbi__start_file(&s, f);
+   result = stbi__load_and_postprocess_16bit(&s, x, y, comp, req_comp);
+   if (result)
+   {
+      // need to 'unget' all the characters in the IO buffer
+      fseek(f, -(int)(s.img_buffer_end - s.img_buffer), SEEK_CUR);
+   }
+   return result;
+}
+
+STBIDEF stbi_us* stbi_load_16(char const* filename, int* x, int* y, int* comp, int req_comp)
+{
+   FILE* f = stbi__fopen(filename, "rb");
+   stbi__uint16* result;
+   if (!f)
+      return (stbi_us*)stbi__errpuc("can't fopen", "Unable to open file");
+   result = stbi_load_from_file_16(f, x, y, comp, req_comp);
+   fclose(f);
+   return result;
+}
+
+
 #endif //!STBI_NO_STDIO
 
-STBIDEF stbi_uc *stbi_load_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp)
+STBIDEF stbi_us* stbi_load_16_from_memory(stbi_uc const* buffer, int len, int* x, int* y, int* channels_in_file, int desired_channels)
 {
    stbi__context s;
-   stbi__start_mem(&s,buffer,len);
-   return stbi__load_flip(&s,x,y,comp,req_comp);
+   stbi__start_mem(&s, buffer, len);
+   return stbi__load_and_postprocess_16bit(&s, x, y, channels_in_file, desired_channels);
 }
 
-STBIDEF stbi_uc *stbi_load_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp, int req_comp)
+STBIDEF stbi_us* stbi_load_16_from_callbacks(stbi_io_callbacks const* clbk, void* user, int* x, int* y, int* channels_in_file, int desired_channels)
 {
    stbi__context s;
-   stbi__start_callbacks(&s, (stbi_io_callbacks *) clbk, user);
-   return stbi__load_flip(&s,x,y,comp,req_comp);
+   stbi__start_callbacks(&s, (stbi_io_callbacks*)clbk, user);
+   return stbi__load_and_postprocess_16bit(&s, x, y, channels_in_file, desired_channels);
+}
+
+STBIDEF stbi_uc* stbi_load_from_memory(stbi_uc const* buffer, int len, int* x, int* y, int* comp, int req_comp)
+{
+   stbi__context s;
+   stbi__start_mem(&s, buffer, len);
+   return stbi__load_and_postprocess_8bit(&s, x, y, comp, req_comp);
+}
+
+STBIDEF stbi_uc* stbi_load_from_callbacks(stbi_io_callbacks const* clbk, void* user, int* x, int* y, int* comp, int req_comp)
+{
+   stbi__context s;
+   stbi__start_callbacks(&s, (stbi_io_callbacks*)clbk, user);
+   return stbi__load_and_postprocess_8bit(&s, x, y, comp, req_comp);
+}
+
+#ifndef STBI_NO_GIF
+STBIDEF stbi_uc* stbi_load_gif_from_memory(stbi_uc const* buffer, int len, int** delays, int* x, int* y, int* z, int* comp, int req_comp)
+{
+   unsigned char* result;
+   stbi__context s;
+   stbi__start_mem(&s, buffer, len);
+
+   result = (unsigned char*)stbi__load_gif_main(&s, delays, x, y, z, comp, req_comp);
+   if (stbi__vertically_flip_on_load)
+   {
+      stbi__vertical_flip_slices(result, *x, *y, *z, *comp);
+   }
+
+   return result;
 }
+#endif
 
 #ifndef STBI_NO_LINEAR
-static float *stbi__loadf_main(stbi__context *s, int *x, int *y, int *comp, int req_comp)
+static float* stbi__loadf_main(stbi__context* s, int* x, int* y, int* comp, int req_comp)
 {
-   unsigned char *data;
-   #ifndef STBI_NO_HDR
-   if (stbi__hdr_test(s)) {
-      float *hdr_data = stbi__hdr_load(s,x,y,comp,req_comp);
+   unsigned char* data;
+#ifndef STBI_NO_HDR
+   if (stbi__hdr_test(s))
+   {
+      stbi__result_info ri;
+      float* hdr_data = stbi__hdr_load(s, x, y, comp, req_comp, &ri);
       if (hdr_data)
-         stbi__float_postprocess(hdr_data,x,y,comp,req_comp);
+         stbi__float_postprocess(hdr_data, x, y, comp, req_comp);
       return hdr_data;
    }
-   #endif
-   data = stbi__load_flip(s, x, y, comp, req_comp);
+#endif
+   data = stbi__load_and_postprocess_8bit(s, x, y, comp, req_comp);
    if (data)
       return stbi__ldr_to_hdr(data, *x, *y, req_comp ? req_comp : *comp);
    return stbi__errpf("unknown image type", "Image not of any known type, or corrupt");
 }
 
-STBIDEF float *stbi_loadf_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp)
+STBIDEF float* stbi_loadf_from_memory(stbi_uc const* buffer, int len, int* x, int* y, int* comp, int req_comp)
 {
    stbi__context s;
-   stbi__start_mem(&s,buffer,len);
-   return stbi__loadf_main(&s,x,y,comp,req_comp);
+   stbi__start_mem(&s, buffer, len);
+   return stbi__loadf_main(&s, x, y, comp, req_comp);
 }
 
-STBIDEF float *stbi_loadf_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp, int req_comp)
+STBIDEF float* stbi_loadf_from_callbacks(stbi_io_callbacks const* clbk, void* user, int* x, int* y, int* comp, int req_comp)
 {
    stbi__context s;
-   stbi__start_callbacks(&s, (stbi_io_callbacks *) clbk, user);
-   return stbi__loadf_main(&s,x,y,comp,req_comp);
+   stbi__start_callbacks(&s, (stbi_io_callbacks*)clbk, user);
+   return stbi__loadf_main(&s, x, y, comp, req_comp);
 }
 
 #ifndef STBI_NO_STDIO
-STBIDEF float *stbi_loadf(char const *filename, int *x, int *y, int *comp, int req_comp)
+STBIDEF float* stbi_loadf(char const* filename, int* x, int* y, int* comp, int req_comp)
 {
-   float *result;
-   FILE *f = stbi__fopen(filename, "rb");
-   if (!f) return stbi__errpf("can't fopen", "Unable to open file");
-   result = stbi_loadf_from_file(f,x,y,comp,req_comp);
+   float* result;
+   FILE* f = stbi__fopen(filename, "rb");
+   if (!f)
+      return stbi__errpf("can't fopen", "Unable to open file");
+   result = stbi_loadf_from_file(f, x, y, comp, req_comp);
    fclose(f);
    return result;
 }
 
-STBIDEF float *stbi_loadf_from_file(FILE *f, int *x, int *y, int *comp, int req_comp)
+STBIDEF float* stbi_loadf_from_file(FILE* f, int* x, int* y, int* comp, int req_comp)
 {
    stbi__context s;
-   stbi__start_file(&s,f);
-   return stbi__loadf_main(&s,x,y,comp,req_comp);
+   stbi__start_file(&s, f);
+   return stbi__loadf_main(&s, x, y, comp, req_comp);
 }
 #endif // !STBI_NO_STDIO
 
@@ -1132,68 +1545,73 @@ STBIDEF float *stbi_loadf_from_file(FILE *f, int *x, int *y, int *comp, int req_
 // defined, for API simplicity; if STBI_NO_LINEAR is defined, it always
 // reports false!
 
-STBIDEF int stbi_is_hdr_from_memory(stbi_uc const *buffer, int len)
+STBIDEF int stbi_is_hdr_from_memory(stbi_uc const* buffer, int len)
 {
-   #ifndef STBI_NO_HDR
+#ifndef STBI_NO_HDR
    stbi__context s;
-   stbi__start_mem(&s,buffer,len);
+   stbi__start_mem(&s, buffer, len);
    return stbi__hdr_test(&s);
-   #else
+#else
    STBI_NOTUSED(buffer);
    STBI_NOTUSED(len);
    return 0;
-   #endif
+#endif
 }
 
 #ifndef STBI_NO_STDIO
-STBIDEF int      stbi_is_hdr          (char const *filename)
+STBIDEF int stbi_is_hdr(char const* filename)
 {
-   FILE *f = stbi__fopen(filename, "rb");
-   int result=0;
-   if (f) {
+   FILE* f = stbi__fopen(filename, "rb");
+   int result = 0;
+   if (f)
+   {
       result = stbi_is_hdr_from_file(f);
       fclose(f);
    }
    return result;
 }
 
-STBIDEF int      stbi_is_hdr_from_file(FILE *f)
+STBIDEF int stbi_is_hdr_from_file(FILE* f)
 {
-   #ifndef STBI_NO_HDR
+#ifndef STBI_NO_HDR
+   long pos = ftell(f);
+   int res;
    stbi__context s;
-   stbi__start_file(&s,f);
-   return stbi__hdr_test(&s);
-   #else
+   stbi__start_file(&s, f);
+   res = stbi__hdr_test(&s);
+   fseek(f, pos, SEEK_SET);
+   return res;
+#else
    STBI_NOTUSED(f);
    return 0;
-   #endif
+#endif
 }
 #endif // !STBI_NO_STDIO
 
-STBIDEF int      stbi_is_hdr_from_callbacks(stbi_io_callbacks const *clbk, void *user)
+STBIDEF int stbi_is_hdr_from_callbacks(stbi_io_callbacks const* clbk, void* user)
 {
-   #ifndef STBI_NO_HDR
+#ifndef STBI_NO_HDR
    stbi__context s;
-   stbi__start_callbacks(&s, (stbi_io_callbacks *) clbk, user);
+   stbi__start_callbacks(&s, (stbi_io_callbacks*)clbk, user);
    return stbi__hdr_test(&s);
-   #else
+#else
    STBI_NOTUSED(clbk);
    STBI_NOTUSED(user);
    return 0;
-   #endif
+#endif
 }
 
 #ifndef STBI_NO_LINEAR
-static float stbi__l2h_gamma=2.2f, stbi__l2h_scale=1.0f;
+static float stbi__l2h_gamma = 2.2f, stbi__l2h_scale = 1.0f;
 
-STBIDEF void   stbi_ldr_to_hdr_gamma(float gamma) { stbi__l2h_gamma = gamma; }
-STBIDEF void   stbi_ldr_to_hdr_scale(float scale) { stbi__l2h_scale = scale; }
+STBIDEF void stbi_ldr_to_hdr_gamma(float gamma) { stbi__l2h_gamma = gamma; }
+STBIDEF void stbi_ldr_to_hdr_scale(float scale) { stbi__l2h_scale = scale; }
 #endif
 
-static float stbi__h2l_gamma_i=1.0f/2.2f, stbi__h2l_scale_i=1.0f;
+static float stbi__h2l_gamma_i = 1.0f / 2.2f, stbi__h2l_scale_i = 1.0f;
 
-STBIDEF void   stbi_hdr_to_ldr_gamma(float gamma) { stbi__h2l_gamma_i = 1/gamma; }
-STBIDEF void   stbi_hdr_to_ldr_scale(float scale) { stbi__h2l_scale_i = 1/scale; }
+STBIDEF void stbi_hdr_to_ldr_gamma(float gamma) { stbi__h2l_gamma_i = 1 / gamma; }
+STBIDEF void stbi_hdr_to_ldr_scale(float scale) { stbi__h2l_scale_i = 1 / scale; }
 
 
 //////////////////////////////////////////////////////////////////////////////
@@ -1203,59 +1621,79 @@ STBIDEF void   stbi_hdr_to_ldr_scale(float scale) { stbi__h2l_scale_i = 1/scale;
 
 enum
 {
-   STBI__SCAN_load=0,
+   STBI__SCAN_load = 0,
    STBI__SCAN_type,
    STBI__SCAN_header
 };
 
-static void stbi__refill_buffer(stbi__context *s)
+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) {
+   int n = (s->io.read)(s->io_user_data, (char*)s->buffer_start, s->buflen);
+   s->callback_already_read += (int)(s->img_buffer - s->img_buffer_original);
+   if (n == 0)
+   {
       // at end of file, treat same as if from memory, but need to handle case
       // where s->img_buffer isn't pointing to safe memory, e.g. 0-byte file
       s->read_from_callbacks = 0;
       s->img_buffer = s->buffer_start;
-      s->img_buffer_end = s->buffer_start+1;
+      s->img_buffer_end = s->buffer_start + 1;
       *s->img_buffer = 0;
-   } else {
+   }
+   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)
+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) {
+   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 defined(STBI_NO_JPEG) && defined(STBI_NO_HDR) && defined(STBI_NO_PIC) && defined(STBI_NO_PNM)
+// nothing
+#else
+stbi_inline static int stbi__at_eof(stbi__context* s)
 {
-   if (s->io.read) {
-      if (!(s->io.eof)(s->io_user_data)) return 0;
+   if (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;
+      if (s->read_from_callbacks == 0)
+         return 1;
    }
 
    return s->img_buffer >= s->img_buffer_end;
 }
+#endif
 
-static void stbi__skip(stbi__context *s, int n)
+#if defined(STBI_NO_JPEG) && defined(STBI_NO_PNG) && defined(STBI_NO_BMP) && defined(STBI_NO_PSD) && defined(STBI_NO_TGA) && defined(STBI_NO_GIF) && defined(STBI_NO_PIC)
+// nothing
+#else
+static void stbi__skip(stbi__context* s, int n)
 {
-   if (n < 0) {
+   if (n == 0)
+      return; // already there!
+   if (n < 0)
+   {
       s->img_buffer = s->img_buffer_end;
       return;
    }
-   if (s->io.read) {
-      int blen = (int) (s->img_buffer_end - s->img_buffer);
-      if (blen < n) {
+   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;
@@ -1263,47 +1701,64 @@ static void stbi__skip(stbi__context *s, int n)
    }
    s->img_buffer += n;
 }
+#endif
 
-static int stbi__getn(stbi__context *s, stbi_uc *buffer, int n)
+#if defined(STBI_NO_PNG) && defined(STBI_NO_TGA) && defined(STBI_NO_HDR) && defined(STBI_NO_PNM)
+// nothing
+#else
+static int stbi__getn(stbi__context* s, stbi_uc* buffer, int n)
 {
-   if (s->io.read) {
-      int blen = (int) (s->img_buffer_end - s->img_buffer);
-      if (blen < n) {
+   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));
+         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) {
+   if (s->img_buffer + n <= s->img_buffer_end)
+   {
       memcpy(buffer, s->img_buffer, n);
       s->img_buffer += n;
       return 1;
-   } else
+   }
+   else
       return 0;
 }
+#endif
 
-static int stbi__get16be(stbi__context *s)
+#if defined(STBI_NO_JPEG) && defined(STBI_NO_PNG) && defined(STBI_NO_PSD) && defined(STBI_NO_PIC)
+// nothing
+#else
+static int stbi__get16be(stbi__context* s)
 {
    int z = stbi__get8(s);
    return (z << 8) + stbi__get8(s);
 }
+#endif
 
-static stbi__uint32 stbi__get32be(stbi__context *s)
+#if defined(STBI_NO_PNG) && defined(STBI_NO_PSD) && defined(STBI_NO_PIC)
+// nothing
+#else
+static stbi__uint32 stbi__get32be(stbi__context* s)
 {
    stbi__uint32 z = stbi__get16be(s);
    return (z << 16) + stbi__get16be(s);
 }
+#endif
 
 #if defined(STBI_NO_BMP) && defined(STBI_NO_TGA) && defined(STBI_NO_GIF)
 // nothing
 #else
-static int stbi__get16le(stbi__context *s)
+static int stbi__get16le(stbi__context* s)
 {
    int z = stbi__get8(s);
    return z + (stbi__get8(s) << 8);
@@ -1311,16 +1766,19 @@ static int stbi__get16le(stbi__context *s)
 #endif
 
 #ifndef STBI_NO_BMP
-static stbi__uint32 stbi__get32le(stbi__context *s)
+static stbi__uint32 stbi__get32le(stbi__context* s)
 {
    stbi__uint32 z = stbi__get16le(s);
-   return z + (stbi__get16le(s) << 16);
+   z += (stbi__uint32)stbi__get16le(s) << 16;
+   return z;
 }
 #endif
 
-#define STBI__BYTECAST(x)  ((stbi_uc) ((x) & 255))  // truncate int to byte without warnings
-
+#define STBI__BYTECAST(x) ((stbi_uc)((x)&255)) // truncate int to byte without warnings
 
+#if defined(STBI_NO_JPEG) && defined(STBI_NO_PNG) && defined(STBI_NO_BMP) && defined(STBI_NO_PSD) && defined(STBI_NO_TGA) && defined(STBI_NO_GIF) && defined(STBI_NO_PIC) && defined(STBI_NO_PNM)
+// nothing
+#else
 //////////////////////////////////////////////////////////////////////////////
 //
 //  generic converter from built-in img_n to req_comp
@@ -1334,66 +1792,252 @@ static stbi__uint32 stbi__get32le(stbi__context *s)
 
 static stbi_uc stbi__compute_y(int r, int g, int b)
 {
-   return (stbi_uc) (((r*77) + (g*150) +  (29*b)) >> 8);
+   return (stbi_uc)(((r * 77) + (g * 150) + (29 * b)) >> 8);
 }
+#endif
 
-static unsigned char *stbi__convert_format(unsigned char *data, int img_n, int req_comp, unsigned int x, unsigned int y)
+#if defined(STBI_NO_PNG) && defined(STBI_NO_BMP) && defined(STBI_NO_PSD) && defined(STBI_NO_TGA) && defined(STBI_NO_GIF) && defined(STBI_NO_PIC) && defined(STBI_NO_PNM)
+// nothing
+#else
+static unsigned char* stbi__convert_format(unsigned char* data, int img_n, int req_comp, unsigned int x, unsigned int y)
 {
-   int i,j;
-   unsigned char *good;
+   int i, j;
+   unsigned char* good;
 
-   if (req_comp == img_n) return data;
+   if (req_comp == img_n)
+      return data;
    STBI_ASSERT(req_comp >= 1 && req_comp <= 4);
 
-   good = (unsigned char *) stbi__malloc(req_comp * x * y);
-   if (good == NULL) {
+   good = (unsigned char*)stbi__malloc_mad3(req_comp, x, y, 0);
+   if (good == NULL)
+   {
       STBI_FREE(data);
       return stbi__errpuc("outofmem", "Out of memory");
    }
 
-   for (j=0; j < (int) y; ++j) {
-      unsigned char *src  = data + j * x * img_n   ;
-      unsigned char *dest = good + j * x * req_comp;
+   for (j = 0; j < (int)y; ++j)
+   {
+      unsigned char* src = data + j * x * img_n;
+      unsigned char* dest = good + j * x * req_comp;
+
+#define STBI__COMBO(a, b) ((a)*8 + (b))
+#define STBI__CASE(a, b)   \
+   case STBI__COMBO(a, b): \
+      for (i = x - 1; i >= 0; --i, src += a, dest += b)
+      // convert source image with img_n components to one with req_comp components;
+      // avoid switch per pixel, so use switch per scanline and massive macros
+      switch (STBI__COMBO(img_n, req_comp))
+      {
+         STBI__CASE(1, 2)
+         {
+            dest[0] = src[0];
+            dest[1] = 255;
+         }
+         break;
+         STBI__CASE(1, 3) { dest[0] = dest[1] = dest[2] = src[0]; }
+         break;
+         STBI__CASE(1, 4)
+         {
+            dest[0] = dest[1] = dest[2] = src[0];
+            dest[3] = 255;
+         }
+         break;
+         STBI__CASE(2, 1) { dest[0] = src[0]; }
+         break;
+         STBI__CASE(2, 3) { dest[0] = dest[1] = dest[2] = src[0]; }
+         break;
+         STBI__CASE(2, 4)
+         {
+            dest[0] = dest[1] = dest[2] = src[0];
+            dest[3] = src[1];
+         }
+         break;
+         STBI__CASE(3, 4)
+         {
+            dest[0] = src[0];
+            dest[1] = src[1];
+            dest[2] = src[2];
+            dest[3] = 255;
+         }
+         break;
+         STBI__CASE(3, 1) { dest[0] = stbi__compute_y(src[0], src[1], src[2]); }
+         break;
+         STBI__CASE(3, 2)
+         {
+            dest[0] = stbi__compute_y(src[0], src[1], src[2]);
+            dest[1] = 255;
+         }
+         break;
+         STBI__CASE(4, 1) { dest[0] = stbi__compute_y(src[0], src[1], src[2]); }
+         break;
+         STBI__CASE(4, 2)
+         {
+            dest[0] = stbi__compute_y(src[0], src[1], src[2]);
+            dest[1] = src[3];
+         }
+         break;
+         STBI__CASE(4, 3)
+         {
+            dest[0] = src[0];
+            dest[1] = src[1];
+            dest[2] = src[2];
+         }
+         break;
+         default:
+            STBI_ASSERT(0);
+            STBI_FREE(data);
+            STBI_FREE(good);
+            return stbi__errpuc("unsupported", "Unsupported format conversion");
+      }
+#undef STBI__CASE
+   }
+
+   STBI_FREE(data);
+   return good;
+}
+#endif
+
+#if defined(STBI_NO_PNG) && defined(STBI_NO_PSD)
+// nothing
+#else
+static stbi__uint16 stbi__compute_y_16(int r, int g, int b)
+{
+   return (stbi__uint16)(((r * 77) + (g * 150) + (29 * b)) >> 8);
+}
+#endif
+
+#if defined(STBI_NO_PNG) && defined(STBI_NO_PSD)
+// nothing
+#else
+static stbi__uint16* stbi__convert_format16(stbi__uint16* data, int img_n, int req_comp, unsigned int x, unsigned int y)
+{
+   int i, j;
+   stbi__uint16* good;
+
+   if (req_comp == img_n)
+      return data;
+   STBI_ASSERT(req_comp >= 1 && req_comp <= 4);
+
+   good = (stbi__uint16*)stbi__malloc(req_comp * x * y * 2);
+   if (good == NULL)
+   {
+      STBI_FREE(data);
+      return (stbi__uint16*)stbi__errpuc("outofmem", "Out of memory");
+   }
+
+   for (j = 0; j < (int)y; ++j)
+   {
+      stbi__uint16* src = data + j * x * img_n;
+      stbi__uint16* dest = good + j * x * req_comp;
 
-      #define 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) {
+#define STBI__COMBO(a, b) ((a)*8 + (b))
+#define STBI__CASE(a, b)   \
+   case STBI__COMBO(a, b): \
+      for (i = x - 1; i >= 0; --i, src += a, dest += b)
       // convert source image with img_n components to one with req_comp components;
       // avoid switch per pixel, so use switch per scanline and massive macros
-      switch (COMBO(img_n, req_comp)) {
-         CASE(1,2) dest[0]=src[0]; dest[1]=255; } break;
-         CASE(1,3) dest[0]=dest[1]=dest[2]=src[0]; } break;
-   CASE(1,4) dest[0]=dest[1]=dest[2]=src[0]; dest[3]=255; } break;
-         CASE(2,1) dest[0]=src[0]; } break;
-         CASE(2,3) dest[0]=dest[1]=dest[2]=src[0]; } break;
-CASE(2,4) dest[0]=dest[1]=dest[2]=src[0]; dest[3]=src[1]; } break;
-CASE(3,4) dest[0]=src[0];dest[1]=src[1];dest[2]=src[2];dest[3]=255; } break;
-         CASE(3,1) dest[0]=stbi__compute_y(src[0],src[1],src[2]); } break;
-CASE(3,2) dest[0]=stbi__compute_y(src[0],src[1],src[2]); dest[1] = 255; } break;
-         CASE(4,1) dest[0]=stbi__compute_y(src[0],src[1],src[2]); } break;
-CASE(4,2) dest[0]=stbi__compute_y(src[0],src[1],src[2]); dest[1] = src[3]; } break;
-CASE(4,3) dest[0]=src[0];dest[1]=src[1];dest[2]=src[2]; } break;
-         default: STBI_ASSERT(0);
-      }
-      #undef CASE
+      switch (STBI__COMBO(img_n, req_comp))
+      {
+         STBI__CASE(1, 2)
+         {
+            dest[0] = src[0];
+            dest[1] = 0xffff;
+         }
+         break;
+         STBI__CASE(1, 3) { dest[0] = dest[1] = dest[2] = src[0]; }
+         break;
+         STBI__CASE(1, 4)
+         {
+            dest[0] = dest[1] = dest[2] = src[0];
+            dest[3] = 0xffff;
+         }
+         break;
+         STBI__CASE(2, 1) { dest[0] = src[0]; }
+         break;
+         STBI__CASE(2, 3) { dest[0] = dest[1] = dest[2] = src[0]; }
+         break;
+         STBI__CASE(2, 4)
+         {
+            dest[0] = dest[1] = dest[2] = src[0];
+            dest[3] = src[1];
+         }
+         break;
+         STBI__CASE(3, 4)
+         {
+            dest[0] = src[0];
+            dest[1] = src[1];
+            dest[2] = src[2];
+            dest[3] = 0xffff;
+         }
+         break;
+         STBI__CASE(3, 1) { dest[0] = stbi__compute_y_16(src[0], src[1], src[2]); }
+         break;
+         STBI__CASE(3, 2)
+         {
+            dest[0] = stbi__compute_y_16(src[0], src[1], src[2]);
+            dest[1] = 0xffff;
+         }
+         break;
+         STBI__CASE(4, 1) { dest[0] = stbi__compute_y_16(src[0], src[1], src[2]); }
+         break;
+         STBI__CASE(4, 2)
+         {
+            dest[0] = stbi__compute_y_16(src[0], src[1], src[2]);
+            dest[1] = src[3];
+         }
+         break;
+         STBI__CASE(4, 3)
+         {
+            dest[0] = src[0];
+            dest[1] = src[1];
+            dest[2] = src[2];
+         }
+         break;
+         default:
+            STBI_ASSERT(0);
+            STBI_FREE(data);
+            STBI_FREE(good);
+            return (stbi__uint16*)stbi__errpuc("unsupported", "Unsupported format conversion");
+      }
+#undef STBI__CASE
    }
 
    STBI_FREE(data);
    return good;
 }
+#endif
 
 #ifndef STBI_NO_LINEAR
-static float   *stbi__ldr_to_hdr(stbi_uc *data, int x, int y, int comp)
+static float* stbi__ldr_to_hdr(stbi_uc* data, int x, int y, int comp)
 {
-   int i,k,n;
-   float *output = (float *) stbi__malloc(x * y * comp * sizeof(float));
-   if (output == NULL) { STBI_FREE(data); return stbi__errpf("outofmem", "Out of memory"); }
+   int i, k, n;
+   float* output;
+   if (!data)
+      return NULL;
+   output = (float*)stbi__malloc_mad4(x, y, comp, sizeof(float), 0);
+   if (output == NULL)
+   {
+      STBI_FREE(data);
+      return stbi__errpf("outofmem", "Out of memory");
+   }
    // compute number of non-alpha components
-   if (comp & 1) n = comp; else n = comp-1;
-   for (i=0; i < x*y; ++i) {
-      for (k=0; k < n; ++k) {
-         output[i*comp + k] = (float) (pow(data[i*comp+k]/255.0f, stbi__l2h_gamma) * stbi__l2h_scale);
+   if (comp & 1)
+      n = comp;
+   else
+      n = comp - 1;
+   for (i = 0; i < x * y; ++i)
+   {
+      for (k = 0; k < n; ++k)
+      {
+         output[i * comp + k] = (float)(pow(data[i * comp + k] / 255.0f, stbi__l2h_gamma) * stbi__l2h_scale);
+      }
+   }
+   if (n < comp)
+   {
+      for (i = 0; i < x * y; ++i)
+      {
+         output[i * comp + n] = data[i * comp + n] / 255.0f;
       }
-      if (k < comp) output[i*comp + k] = data[i*comp+k]/255.0f;
    }
    STBI_FREE(data);
    return output;
@@ -1401,26 +2045,43 @@ static float   *stbi__ldr_to_hdr(stbi_uc *data, int x, int y, int comp)
 #endif
 
 #ifndef STBI_NO_HDR
-#define stbi__float2int(x)   ((int) (x))
-static stbi_uc *stbi__hdr_to_ldr(float   *data, int x, int y, int comp)
+#define stbi__float2int(x) ((int)(x))
+static stbi_uc* stbi__hdr_to_ldr(float* data, int x, int y, int comp)
 {
-   int i,k,n;
-   stbi_uc *output = (stbi_uc *) stbi__malloc(x * y * comp);
-   if (output == NULL) { STBI_FREE(data); return stbi__errpuc("outofmem", "Out of memory"); }
+   int i, k, n;
+   stbi_uc* output;
+   if (!data)
+      return NULL;
+   output = (stbi_uc*)stbi__malloc_mad3(x, y, comp, 0);
+   if (output == NULL)
+   {
+      STBI_FREE(data);
+      return stbi__errpuc("outofmem", "Out of memory");
+   }
    // compute number of non-alpha components
-   if (comp & 1) n = comp; else n = comp-1;
-   for (i=0; i < x*y; ++i) {
-      for (k=0; k < n; ++k) {
-         float z = (float) pow(data[i*comp+k]*stbi__h2l_scale_i, stbi__h2l_gamma_i) * 255 + 0.5f;
-         if (z < 0) z = 0;
-         if (z > 255) z = 255;
-         output[i*comp + k] = (stbi_uc) stbi__float2int(z);
+   if (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);
+      if (k < comp)
+      {
+         float z = data[i * comp + k] * 255 + 0.5f;
+         if (z < 0)
+            z = 0;
+         if (z > 255)
+            z = 255;
+         output[i * comp + k] = (stbi_uc)stbi__float2int(z);
       }
    }
    STBI_FREE(data);
@@ -1452,105 +2113,121 @@ static stbi_uc *stbi__hdr_to_ldr(float   *data, int x, int y, int comp)
 #ifndef STBI_NO_JPEG
 
 // huffman decoding acceleration
-#define FAST_BITS   9  // larger handles more cases; smaller stomps less cache
+#define FAST_BITS 9 // larger handles more cases; smaller stomps less cache
 
 typedef struct
 {
-   stbi_uc  fast[1 << FAST_BITS];
+   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];
+   stbi_uc values[256];
+   stbi_uc size[257];
    unsigned int maxcode[18];
-   int    delta[17];   // old 'firstsymbol' - old 'firstcode'
+   int delta[17]; // old 'firstsymbol' - old 'firstcode'
 } stbi__huffman;
 
 typedef struct
 {
-   stbi__context *s;
+   stbi__context* s;
    stbi__huffman huff_dc[4];
    stbi__huffman huff_ac[4];
-   stbi_uc dequant[4][64];
+   stbi__uint16 dequant[4][64];
    stbi__int16 fast_ac[4][1 << FAST_BITS];
 
-// sizes for components, interleaved MCUs
+   // 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
+   // definition of jpeg image component
    struct
    {
       int id;
-      int h,v;
+      int h, v;
       int tq;
-      int hd,ha;
+      int hd, ha;
       int dc_pred;
 
-      int x,y,w2,h2;
-      stbi_uc *data;
+      int x, y, w2, h2;
+      stbi_uc* data;
       void *raw_data, *raw_coeff;
-      stbi_uc *linebuf;
-      short   *coeff;   // progressive only
-      int      coeff_w, coeff_h; // number of 8x8 coefficient blocks
+      stbi_uc* linebuf;
+      short* coeff; // progressive only
+      int coeff_w, coeff_h; // number of 8x8 coefficient blocks
    } img_comp[4];
 
-   stbi__uint32   code_buffer; // jpeg entropy-coded buffer
-   int            code_bits;   // number of valid bits
-   unsigned char  marker;      // marker seen while filling entropy buffer
-   int            nomore;      // flag if we saw a marker so must stop
-
-   int            progressive;
-   int            spec_start;
-   int            spec_end;
-   int            succ_high;
-   int            succ_low;
-   int            eob_run;
+   stbi__uint32 code_buffer; // jpeg entropy-coded buffer
+   int code_bits; // number of valid bits
+   unsigned char marker; // marker seen while filling entropy buffer
+   int nomore; // flag if we saw a marker so must stop
+
+   int progressive;
+   int spec_start;
+   int spec_end;
+   int succ_high;
+   int succ_low;
+   int eob_run;
+   int jfif;
+   int app14_color_transform; // Adobe APP14 tag
+   int rgb;
 
    int scan_n, order[4];
    int restart_interval, todo;
 
-// kernels
-   void (*idct_block_kernel)(stbi_uc *out, int out_stride, short data[64]);
-   void (*YCbCr_to_RGB_kernel)(stbi_uc *out, const stbi_uc *y, const stbi_uc *pcb, const stbi_uc *pcr, int count, int step);
-   stbi_uc *(*resample_row_hv_2_kernel)(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs);
+   // kernels
+   void (*idct_block_kernel)(stbi_uc* out, int out_stride, short data[64]);
+   void (*YCbCr_to_RGB_kernel)(stbi_uc* out, const stbi_uc* y, const stbi_uc* pcb, const stbi_uc* pcr, int count, int step);
+   stbi_uc* (*resample_row_hv_2_kernel)(stbi_uc* out, stbi_uc* in_near, stbi_uc* in_far, int w, int hs);
 } stbi__jpeg;
 
-static int stbi__build_huffman(stbi__huffman *h, int *count)
+static int stbi__build_huffman(stbi__huffman* h, int* count)
 {
-   int i,j,k=0,code;
+   int i, j, k = 0;
+   unsigned int code;
    // build size list for each symbol (from JPEG spec)
-   for (i=0; i < 16; ++i)
-      for (j=0; j < count[i]; ++j)
-         h->size[k++] = (stbi_uc) (i+1);
+   for (i = 0; i < 16; ++i)
+   {
+      for (j = 0; j < count[i]; ++j)
+      {
+         h->size[k++] = (stbi_uc)(i + 1);
+         if (k >= 257)
+            return stbi__err("bad size list", "Corrupt JPEG");
+      }
+   }
    h->size[k] = 0;
 
    // compute actual symbols (from jpeg spec)
    code = 0;
    k = 0;
-   for(j=1; j <= 16; ++j) {
+   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) {
+      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");
+            h->code[k++] = (stbi__uint16)(code++);
+         if (code - 1 >= (1u << j))
+            return stbi__err("bad code lengths", "Corrupt JPEG");
       }
       // compute largest code + 1 for this size, preshifted as needed later
-      h->maxcode[j] = code << (16-j);
+      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) {
+   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;
+      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;
          }
       }
    }
@@ -1559,39 +2236,48 @@ static int stbi__build_huffman(stbi__huffman *h, int *count)
 
 // build a table that decodes both magnitude and value of small ACs in
 // one go.
-static void stbi__build_fast_ac(stbi__int16 *fast_ac, stbi__huffman *h)
+static void stbi__build_fast_ac(stbi__int16* fast_ac, stbi__huffman* h)
 {
    int i;
-   for (i=0; i < (1 << FAST_BITS); ++i) {
+   for (i = 0; i < (1 << FAST_BITS); ++i)
+   {
       stbi_uc fast = h->fast[i];
       fast_ac[i] = 0;
-      if (fast < 255) {
+      if (fast < 255)
+      {
          int rs = h->values[fast];
          int run = (rs >> 4) & 15;
          int magbits = rs & 15;
          int len = h->size[fast];
 
-         if (magbits && len + magbits <= FAST_BITS) {
+         if (magbits && len + magbits <= FAST_BITS)
+         {
             // magnitude code followed by receive_extend code
             int k = ((i << len) & ((1 << FAST_BITS) - 1)) >> (FAST_BITS - magbits);
             int m = 1 << (magbits - 1);
-            if (k < m) k += (-1 << magbits) + 1;
+            if (k < m)
+               k += (~0U << magbits) + 1;
             // if the result is small enough, we can fit it in fast_ac table
             if (k >= -128 && k <= 127)
-               fast_ac[i] = (stbi__int16) ((k << 8) + (run << 4) + (len + magbits));
+               fast_ac[i] = (stbi__int16)((k * 256) + (run * 16) + (len + magbits));
          }
       }
    }
 }
 
-static void stbi__grow_buffer_unsafe(stbi__jpeg *j)
+static void stbi__grow_buffer_unsafe(stbi__jpeg* j)
 {
-   do {
-      int b = j->nomore ? 0 : stbi__get8(j->s);
-      if (b == 0xff) {
+   do
+   {
+      unsigned 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;
+         while (c == 0xff)
+            c = stbi__get8(j->s); // consume fill bytes
+         if (c != 0)
+         {
+            j->marker = (unsigned char)c;
             j->nomore = 1;
             return;
          }
@@ -1602,21 +2288,23 @@ static void stbi__grow_buffer_unsafe(stbi__jpeg *j)
 }
 
 // (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};
+static const stbi__uint32 stbi__bmask[17] = { 0, 1, 3, 7, 15, 31, 63, 127, 255, 511, 1023, 2047, 4095, 8191, 16383, 32767, 65535 };
 
 // decode a jpeg huffman value from the bitstream
-stbi_inline static int stbi__jpeg_huff_decode(stbi__jpeg *j, stbi__huffman *h)
+stbi_inline static int stbi__jpeg_huff_decode(stbi__jpeg* j, stbi__huffman* h)
 {
    unsigned int temp;
-   int c,k;
+   int c, k;
 
-   if (j->code_bits < 16) stbi__grow_buffer_unsafe(j);
+   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);
+   c = (j->code_buffer >> (32 - FAST_BITS)) & ((1 << FAST_BITS) - 1);
    k = h->fast[c];
-   if (k < 255) {
+   if (k < 255)
+   {
       int s = h->size[k];
       if (s > j->code_bits)
          return -1;
@@ -1632,10 +2320,11 @@ stbi_inline static int stbi__jpeg_huff_decode(stbi__jpeg *j, stbi__huffman *h)
    // 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)
+   for (k = FAST_BITS + 1;; ++k)
       if (temp < h->maxcode[k])
          break;
-   if (k == 17) {
+   if (k == 17)
+   {
       // error! code not found
       j->code_bits -= 16;
       return -1;
@@ -1646,6 +2335,8 @@ stbi_inline static int stbi__jpeg_huff_decode(stbi__jpeg *j, stbi__huffman *h)
 
    // convert the huffman code to the symbol id
    c = ((j->code_buffer >> (32 - k)) & stbi__bmask[k]) + h->delta[k];
+   if (c < 0 || c >= 256) // symbol id out of bounds!
+      return -1;
    STBI_ASSERT((((j->code_buffer) >> (32 - h->size[c])) & stbi__bmask[h->size[c]]) == h->code[c]);
 
    // convert the id to a symbol
@@ -1655,30 +2346,35 @@ stbi_inline static int stbi__jpeg_huff_decode(stbi__jpeg *j, stbi__huffman *h)
 }
 
 // bias[n] = (-1<<n) + 1
-static int const stbi__jbias[16] = {0,-1,-3,-7,-15,-31,-63,-127,-255,-511,-1023,-2047,-4095,-8191,-16383,-32767};
+static const int stbi__jbias[16] = { 0, -1, -3, -7, -15, -31, -63, -127, -255, -511, -1023, -2047, -4095, -8191, -16383, -32767 };
 
 // combined JPEG 'receive' and JPEG 'extend', since baseline
 // always extends everything it receives.
-stbi_inline static int stbi__extend_receive(stbi__jpeg *j, int n)
+stbi_inline static int stbi__extend_receive(stbi__jpeg* j, int n)
 {
    unsigned int k;
    int sgn;
-   if (j->code_bits < n) stbi__grow_buffer_unsafe(j);
+   if (j->code_bits < n)
+      stbi__grow_buffer_unsafe(j);
+   if (j->code_bits < n)
+      return 0; // ran out of bits from stream, return 0s intead of continuing
 
-   sgn = (stbi__int32)j->code_buffer >> 31; // sign bit is always in MSB
+   sgn = j->code_buffer >> 31; // sign bit always in MSB; 0 if MSB clear (positive), 1 if MSB set (negative)
    k = stbi_lrot(j->code_buffer, n);
-   STBI_ASSERT(n >= 0 && n < (int) (sizeof(stbi__bmask)/sizeof(*stbi__bmask)));
    j->code_buffer = k & ~stbi__bmask[n];
    k &= stbi__bmask[n];
    j->code_bits -= n;
-   return k + (stbi__jbias[n] & ~sgn);
+   return k + (stbi__jbias[n] & (sgn - 1));
 }
 
 // get some unsigned bits
-stbi_inline static int stbi__jpeg_get_bits(stbi__jpeg *j, int n)
+stbi_inline static int stbi__jpeg_get_bits(stbi__jpeg* j, int n)
 {
    unsigned int k;
-   if (j->code_bits < n) stbi__grow_buffer_unsafe(j);
+   if (j->code_bits < n)
+      stbi__grow_buffer_unsafe(j);
+   if (j->code_bits < n)
+      return 0; // ran out of bits from stream, return 0s intead of continuing
    k = stbi_lrot(j->code_buffer, n);
    j->code_buffer = k & ~stbi__bmask[n];
    k &= stbi__bmask[n];
@@ -1686,10 +2382,13 @@ stbi_inline static int stbi__jpeg_get_bits(stbi__jpeg *j, int n)
    return k;
 }
 
-stbi_inline static int stbi__jpeg_get_bit(stbi__jpeg *j)
+stbi_inline static int stbi__jpeg_get_bit(stbi__jpeg* j)
 {
    unsigned int k;
-   if (j->code_bits < 1) stbi__grow_buffer_unsafe(j);
+   if (j->code_bits < 1)
+      stbi__grow_buffer_unsafe(j);
+   if (j->code_bits < 1)
+      return 0; // ran out of bits from stream, return 0s intead of continuing
    k = j->code_buffer;
    j->code_buffer <<= 1;
    --j->code_bits;
@@ -1698,12 +2397,11 @@ stbi_inline static int stbi__jpeg_get_bit(stbi__jpeg *j)
 
 // 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,
+static const stbi_uc stbi__jpeg_dezigzag[64 + 15] = {
+   0, 1, 8, 16, 9, 2, 3, 10,
+   17, 24, 32, 25, 18, 11, 4, 5,
    12, 19, 26, 33, 40, 48, 41, 34,
-   27, 20, 13,  6,  7, 14, 21, 28,
+   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,
@@ -1714,119 +2412,161 @@ static stbi_uc stbi__jpeg_dezigzag[64+15] =
 };
 
 // decode one 64-entry block--
-static int stbi__jpeg_decode_block(stbi__jpeg *j, short data[64], stbi__huffman *hdc, stbi__huffman *hac, stbi__int16 *fac, int b, stbi_uc *dequant)
+static int stbi__jpeg_decode_block(stbi__jpeg* j, short data[64], stbi__huffman* hdc, stbi__huffman* hac, stbi__int16* fac, int b, stbi__uint16* dequant)
 {
-   int diff,dc,k;
+   int diff, dc, k;
    int t;
 
-   if (j->code_bits < 16) stbi__grow_buffer_unsafe(j);
+   if (j->code_bits < 16)
+      stbi__grow_buffer_unsafe(j);
    t = stbi__jpeg_huff_decode(j, hdc);
-   if (t < 0) return stbi__err("bad huffman code","Corrupt JPEG");
+   if (t < 0 || t > 15)
+      return stbi__err("bad huffman code", "Corrupt JPEG");
 
    // 0 all the ac values now so we can do it 32-bits at a time
-   memset(data,0,64*sizeof(data[0]));
+   memset(data, 0, 64 * sizeof(data[0]));
 
    diff = t ? stbi__extend_receive(j, t) : 0;
+   if (!stbi__addints_valid(j->img_comp[b].dc_pred, diff))
+      return stbi__err("bad delta", "Corrupt JPEG");
    dc = j->img_comp[b].dc_pred + diff;
    j->img_comp[b].dc_pred = dc;
-   data[0] = (short) (dc * dequant[0]);
+   if (!stbi__mul2shorts_valid(dc, dequant[0]))
+      return stbi__err("can't merge dc and ac", "Corrupt JPEG");
+   data[0] = (short)(dc * dequant[0]);
 
    // decode AC components, see JPEG spec
    k = 1;
-   do {
+   do
+   {
       unsigned int zig;
-      int c,r,s;
-      if (j->code_bits < 16) stbi__grow_buffer_unsafe(j);
-      c = (j->code_buffer >> (32 - FAST_BITS)) & ((1 << FAST_BITS)-1);
+      int c, r, s;
+      if (j->code_bits < 16)
+         stbi__grow_buffer_unsafe(j);
+      c = (j->code_buffer >> (32 - FAST_BITS)) & ((1 << FAST_BITS) - 1);
       r = fac[c];
-      if (r) { // fast-AC path
+      if (r)
+      { // fast-AC path
          k += (r >> 4) & 15; // run
          s = r & 15; // combined length
+         if (s > j->code_bits)
+            return stbi__err("bad huffman code", "Combined length longer than code bits available");
          j->code_buffer <<= s;
          j->code_bits -= s;
          // decode into unzigzag'd location
          zig = stbi__jpeg_dezigzag[k++];
-         data[zig] = (short) ((r >> 8) * dequant[zig]);
-      } else {
+         data[zig] = (short)((r >> 8) * dequant[zig]);
+      }
+      else
+      {
          int rs = stbi__jpeg_huff_decode(j, hac);
-         if (rs < 0) return stbi__err("bad huffman code","Corrupt JPEG");
+         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
+         if (s == 0)
+         {
+            if (rs != 0xf0)
+               break; // end block
             k += 16;
-         } else {
+         }
+         else
+         {
             k += r;
             // decode into unzigzag'd location
             zig = stbi__jpeg_dezigzag[k++];
-            data[zig] = (short) (stbi__extend_receive(j,s) * dequant[zig]);
+            data[zig] = (short)(stbi__extend_receive(j, s) * dequant[zig]);
          }
       }
    } while (k < 64);
    return 1;
 }
 
-static int stbi__jpeg_decode_block_prog_dc(stbi__jpeg *j, short data[64], stbi__huffman *hdc, int b)
+static int stbi__jpeg_decode_block_prog_dc(stbi__jpeg* j, short data[64], stbi__huffman* hdc, int b)
 {
-   int diff,dc;
+   int diff, dc;
    int t;
-   if (j->spec_end != 0) return stbi__err("can't merge dc and ac", "Corrupt JPEG");
+   if (j->spec_end != 0)
+      return stbi__err("can't merge dc and ac", "Corrupt JPEG");
 
-   if (j->code_bits < 16) stbi__grow_buffer_unsafe(j);
+   if (j->code_bits < 16)
+      stbi__grow_buffer_unsafe(j);
 
-   if (j->succ_high == 0) {
+   if (j->succ_high == 0)
+   {
       // first scan for DC coefficient, must be first
-      memset(data,0,64*sizeof(data[0])); // 0 all the ac values now
+      memset(data, 0, 64 * sizeof(data[0])); // 0 all the ac values now
       t = stbi__jpeg_huff_decode(j, hdc);
+      if (t < 0 || t > 15)
+         return stbi__err("can't merge dc and ac", "Corrupt JPEG");
       diff = t ? stbi__extend_receive(j, t) : 0;
 
+      if (!stbi__addints_valid(j->img_comp[b].dc_pred, diff))
+         return stbi__err("bad delta", "Corrupt JPEG");
       dc = j->img_comp[b].dc_pred + diff;
       j->img_comp[b].dc_pred = dc;
-      data[0] = (short) (dc << j->succ_low);
-   } else {
+      if (!stbi__mul2shorts_valid(dc, 1 << j->succ_low))
+         return stbi__err("can't merge dc and ac", "Corrupt JPEG");
+      data[0] = (short)(dc * (1 << j->succ_low));
+   }
+   else
+   {
       // refinement scan for DC coefficient
       if (stbi__jpeg_get_bit(j))
-         data[0] += (short) (1 << j->succ_low);
+         data[0] += (short)(1 << j->succ_low);
    }
    return 1;
 }
 
 // @OPTIMIZE: store non-zigzagged during the decode passes,
 // and only de-zigzag when dequantizing
-static int stbi__jpeg_decode_block_prog_ac(stbi__jpeg *j, short data[64], stbi__huffman *hac, stbi__int16 *fac)
+static int stbi__jpeg_decode_block_prog_ac(stbi__jpeg* j, short data[64], stbi__huffman* hac, stbi__int16* fac)
 {
    int k;
-   if (j->spec_start == 0) return stbi__err("can't merge dc and ac", "Corrupt JPEG");
+   if (j->spec_start == 0)
+      return stbi__err("can't merge dc and ac", "Corrupt JPEG");
 
-   if (j->succ_high == 0) {
+   if (j->succ_high == 0)
+   {
       int shift = j->succ_low;
 
-      if (j->eob_run) {
+      if (j->eob_run)
+      {
          --j->eob_run;
          return 1;
       }
 
       k = j->spec_start;
-      do {
+      do
+      {
          unsigned int zig;
-         int c,r,s;
-         if (j->code_bits < 16) stbi__grow_buffer_unsafe(j);
-         c = (j->code_buffer >> (32 - FAST_BITS)) & ((1 << FAST_BITS)-1);
+         int c, r, s;
+         if (j->code_bits < 16)
+            stbi__grow_buffer_unsafe(j);
+         c = (j->code_buffer >> (32 - FAST_BITS)) & ((1 << FAST_BITS) - 1);
          r = fac[c];
-         if (r) { // fast-AC path
+         if (r)
+         { // fast-AC path
             k += (r >> 4) & 15; // run
             s = r & 15; // combined length
+            if (s > j->code_bits)
+               return stbi__err("bad huffman code", "Combined length longer than code bits available");
             j->code_buffer <<= s;
             j->code_bits -= s;
             zig = stbi__jpeg_dezigzag[k++];
-            data[zig] = (short) ((r >> 8) << shift);
-         } else {
+            data[zig] = (short)((r >> 8) * (1 << shift));
+         }
+         else
+         {
             int rs = stbi__jpeg_huff_decode(j, hac);
-            if (rs < 0) return stbi__err("bad huffman code","Corrupt JPEG");
+            if (rs < 0)
+               return stbi__err("bad huffman code", "Corrupt JPEG");
             s = rs & 15;
             r = rs >> 4;
-            if (s == 0) {
-               if (r < 15) {
+            if (s == 0)
+            {
+               if (r < 15)
+               {
                   j->eob_run = (1 << r);
                   if (r)
                      j->eob_run += stbi__jpeg_get_bits(j, r);
@@ -1834,73 +2574,97 @@ static int stbi__jpeg_decode_block_prog_ac(stbi__jpeg *j, short data[64], stbi__
                   break;
                }
                k += 16;
-            } else {
+            }
+            else
+            {
                k += r;
                zig = stbi__jpeg_dezigzag[k++];
-               data[zig] = (short) (stbi__extend_receive(j,s) << shift);
+               data[zig] = (short)(stbi__extend_receive(j, s) * (1 << shift));
             }
          }
       } while (k <= j->spec_end);
-   } else {
+   }
+   else
+   {
       // refinement scan for these AC coefficients
 
-      short bit = (short) (1 << j->succ_low);
+      short bit = (short)(1 << j->succ_low);
 
-      if (j->eob_run) {
+      if (j->eob_run)
+      {
          --j->eob_run;
-         for (k = j->spec_start; k <= j->spec_end; ++k) {
-            short *p = &data[stbi__jpeg_dezigzag[k]];
+         for (k = j->spec_start; k <= j->spec_end; ++k)
+         {
+            short* p = &data[stbi__jpeg_dezigzag[k]];
             if (*p != 0)
                if (stbi__jpeg_get_bit(j))
-                  if ((*p & bit)==0) {
+                  if ((*p & bit) == 0)
+                  {
                      if (*p > 0)
                         *p += bit;
                      else
                         *p -= bit;
                   }
          }
-      } else {
+      }
+      else
+      {
          k = j->spec_start;
-         do {
-            int r,s;
+         do
+         {
+            int r, s;
             int rs = stbi__jpeg_huff_decode(j, hac); // @OPTIMIZE see if we can use the fast path here, advance-by-r is so slow, eh
-            if (rs < 0) return stbi__err("bad huffman code","Corrupt JPEG");
+            if (rs < 0)
+               return stbi__err("bad huffman code", "Corrupt JPEG");
             s = rs & 15;
             r = rs >> 4;
-            if (s == 0) {
-               if (r < 15) {
+            if (s == 0)
+            {
+               if (r < 15)
+               {
                   j->eob_run = (1 << r) - 1;
                   if (r)
                      j->eob_run += stbi__jpeg_get_bits(j, r);
                   r = 64; // force end of block
-               } else {
+               }
+               else
+               {
                   // r=15 s=0 should write 16 0s, so we just do
                   // a run of 15 0s and then write s (which is 0),
                   // so we don't have to do anything special here
                }
-            } else {
-               if (s != 1) return stbi__err("bad huffman code", "Corrupt JPEG");
-               // sign bit
-               if (stbi__jpeg_get_bit(j))
+            }
+            else
+            {
+               if (s != 1)
+                  return stbi__err("bad huffman code", "Corrupt JPEG");
+               // sign bit
+               if (stbi__jpeg_get_bit(j))
                   s = bit;
                else
                   s = -bit;
             }
 
             // advance by r
-            while (k <= j->spec_end) {
-               short *p = &data[stbi__jpeg_dezigzag[k++]];
-               if (*p != 0) {
+            while (k <= j->spec_end)
+            {
+               short* p = &data[stbi__jpeg_dezigzag[k++]];
+               if (*p != 0)
+               {
                   if (stbi__jpeg_get_bit(j))
-                     if ((*p & bit)==0) {
+                     if ((*p & bit) == 0)
+                     {
                         if (*p > 0)
                            *p += bit;
                         else
                            *p -= bit;
                      }
-               } else {
-                  if (r == 0) {
-                     *p = (short) s;
+               }
+               else
+               {
+                  if (r == 0)
+                  {
+                     *p = (short)s;
                      break;
                   }
                   --r;
@@ -1916,110 +2680,120 @@ static int stbi__jpeg_decode_block_prog_ac(stbi__jpeg *j, short data[64], stbi__
 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;
+   if ((unsigned int)x > 255)
+   {
+      if (x < 0)
+         return 0;
+      if (x > 255)
+         return 255;
    }
-   return (stbi_uc) x;
+   return (stbi_uc)x;
 }
 
-#define stbi__f2f(x)  ((int) (((x) * 4096 + 0.5)))
-#define stbi__fsh(x)  ((x) << 12)
+#define stbi__f2f(x) ((int)(((x)*4096 + 0.5)))
+#define stbi__fsh(x) ((x)*4096)
 
 // derived from jidctint -- DCT_ISLOW
-#define STBI__IDCT_1D(s0,s1,s2,s3,s4,s5,s6,s7) \
-   int t0,t1,t2,t3,p1,p2,p3,p4,p5,x0,x1,x2,x3; \
-   p2 = s2;                                    \
-   p3 = s6;                                    \
-   p1 = (p2+p3) * stbi__f2f(0.5411961f);       \
-   t2 = p1 + p3*stbi__f2f(-1.847759065f);      \
-   t3 = p1 + p2*stbi__f2f( 0.765366865f);      \
-   p2 = s0;                                    \
-   p3 = s4;                                    \
-   t0 = stbi__fsh(p2+p3);                      \
-   t1 = stbi__fsh(p2-p3);                      \
-   x0 = t0+t3;                                 \
-   x3 = t0-t3;                                 \
-   x1 = t1+t2;                                 \
-   x2 = t1-t2;                                 \
-   t0 = s7;                                    \
-   t1 = s5;                                    \
-   t2 = s3;                                    \
-   t3 = s1;                                    \
-   p3 = t0+t2;                                 \
-   p4 = t1+t3;                                 \
-   p1 = t0+t3;                                 \
-   p2 = t1+t2;                                 \
-   p5 = (p3+p4)*stbi__f2f( 1.175875602f);      \
-   t0 = t0*stbi__f2f( 0.298631336f);           \
-   t1 = t1*stbi__f2f( 2.053119869f);           \
-   t2 = t2*stbi__f2f( 3.072711026f);           \
-   t3 = t3*stbi__f2f( 1.501321110f);           \
-   p1 = p5 + p1*stbi__f2f(-0.899976223f);      \
-   p2 = p5 + p2*stbi__f2f(-2.562915447f);      \
-   p3 = p3*stbi__f2f(-1.961570560f);           \
-   p4 = p4*stbi__f2f(-0.390180644f);           \
-   t3 += p1+p4;                                \
-   t2 += p2+p3;                                \
-   t1 += p2+p4;                                \
-   t0 += p1+p3;
-
-static void stbi__idct_block(stbi_uc *out, int out_stride, short data[64])
-{
-   int i,val[64],*v=val;
-   stbi_uc *o;
-   short *d = data;
+#define STBI__IDCT_1D(s0, s1, s2, s3, s4, s5, s6, s7)      \
+   int t0, t1, t2, t3, p1, p2, p3, p4, p5, x0, x1, x2, x3; \
+   p2 = s2;                                                \
+   p3 = s6;                                                \
+   p1 = (p2 + p3) * stbi__f2f(0.5411961f);                 \
+   t2 = p1 + p3 * stbi__f2f(-1.847759065f);                \
+   t3 = p1 + p2 * stbi__f2f(0.765366865f);                 \
+   p2 = s0;                                                \
+   p3 = s4;                                                \
+   t0 = stbi__fsh(p2 + p3);                                \
+   t1 = stbi__fsh(p2 - p3);                                \
+   x0 = t0 + t3;                                           \
+   x3 = t0 - t3;                                           \
+   x1 = t1 + t2;                                           \
+   x2 = t1 - t2;                                           \
+   t0 = s7;                                                \
+   t1 = s5;                                                \
+   t2 = s3;                                                \
+   t3 = s1;                                                \
+   p3 = t0 + t2;                                           \
+   p4 = t1 + t3;                                           \
+   p1 = t0 + t3;                                           \
+   p2 = t1 + t2;                                           \
+   p5 = (p3 + p4) * stbi__f2f(1.175875602f);               \
+   t0 = t0 * stbi__f2f(0.298631336f);                      \
+   t1 = t1 * stbi__f2f(2.053119869f);                      \
+   t2 = t2 * stbi__f2f(3.072711026f);                      \
+   t3 = t3 * stbi__f2f(1.501321110f);                      \
+   p1 = p5 + p1 * stbi__f2f(-0.899976223f);                \
+   p2 = p5 + p2 * stbi__f2f(-2.562915447f);                \
+   p3 = p3 * stbi__f2f(-1.961570560f);                     \
+   p4 = p4 * stbi__f2f(-0.390180644f);                     \
+   t3 += p1 + p4;                                          \
+   t2 += p2 + p3;                                          \
+   t1 += p2 + p4;                                          \
+   t0 += p1 + p3;
+
+static void stbi__idct_block(stbi_uc* out, int out_stride, short data[64])
+{
+   int i, val[64], *v = val;
+   stbi_uc* o;
+   short* d = data;
 
    // columns
-   for (i=0; i < 8; ++i,++d, ++v) {
+   for (i = 0; i < 8; ++i, ++d, ++v)
+   {
       // if all zeroes, shortcut -- this avoids dequantizing 0s and IDCTing
-      if (d[ 8]==0 && d[16]==0 && d[24]==0 && d[32]==0
-           && d[40]==0 && d[48]==0 && d[56]==0) {
+      if (d[8] == 0 && d[16] == 0 && d[24] == 0 && d[32] == 0 && d[40] == 0 && d[48] == 0 && d[56] == 0)
+      {
          //    no shortcut                 0     seconds
          //    (1|2|3|4|5|6|7)==0          0     seconds
          //    all separate               -0.047 seconds
          //    1 && 2|3 && 4|5 && 6|7:    -0.047 seconds
-         int dcterm = d[0] << 2;
+         int dcterm = d[0] * 4;
          v[0] = v[8] = v[16] = v[24] = v[32] = v[40] = v[48] = v[56] = dcterm;
-      } else {
-         STBI__IDCT_1D(d[ 0],d[ 8],d[16],d[24],d[32],d[40],d[48],d[56])
+      }
+      else
+      {
+         STBI__IDCT_1D(d[0], d[8], d[16], d[24], d[32], d[40], d[48], d[56])
          // constants scaled things up by 1<<12; let's bring them back
          // down, but keep 2 extra bits of precision
-         x0 += 512; x1 += 512; x2 += 512; x3 += 512;
-         v[ 0] = (x0+t3) >> 10;
-         v[56] = (x0-t3) >> 10;
-         v[ 8] = (x1+t2) >> 10;
-         v[48] = (x1-t2) >> 10;
-         v[16] = (x2+t1) >> 10;
-         v[40] = (x2-t1) >> 10;
-         v[24] = (x3+t0) >> 10;
-         v[32] = (x3-t0) >> 10;
+         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) {
+   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])
+      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);
+      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);
+      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);
    }
 }
 
@@ -2027,113 +2801,113 @@ static void stbi__idct_block(stbi_uc *out, int out_stride, short data[64])
 // sse2 integer IDCT. not the fastest possible implementation but it
 // produces bit-identical results to the generic C version so it's
 // fully "transparent".
-static void stbi__idct_simd(stbi_uc *out, int out_stride, short data[64])
+static void stbi__idct_simd(stbi_uc* out, int out_stride, short data[64])
 {
    // This is constructed to match our regular (generic) integer IDCT exactly.
    __m128i row0, row1, row2, row3, row4, row5, row6, row7;
    __m128i tmp;
 
-   // dot product constant: even elems=x, odd elems=y
-   #define dct_const(x,y)  _mm_setr_epi16((x),(y),(x),(y),(x),(y),(x),(y))
-
-   // out(0) = c0[even]*x + c0[odd]*y   (c0, x, y 16-bit, out 32-bit)
-   // out(1) = c1[even]*x + c1[odd]*y
-   #define dct_rot(out0,out1, x,y,c0,c1) \
-      __m128i c0##lo = _mm_unpacklo_epi16((x),(y)); \
-      __m128i c0##hi = _mm_unpackhi_epi16((x),(y)); \
-      __m128i out0##_l = _mm_madd_epi16(c0##lo, c0); \
-      __m128i out0##_h = _mm_madd_epi16(c0##hi, c0); \
-      __m128i out1##_l = _mm_madd_epi16(c0##lo, c1); \
-      __m128i out1##_h = _mm_madd_epi16(c0##hi, c1)
-
-   // out = in << 12  (in 16-bit, out 32-bit)
-   #define dct_widen(out, in) \
-      __m128i out##_l = _mm_srai_epi32(_mm_unpacklo_epi16(_mm_setzero_si128(), (in)), 4); \
-      __m128i out##_h = _mm_srai_epi32(_mm_unpackhi_epi16(_mm_setzero_si128(), (in)), 4)
-
-   // wide add
-   #define dct_wadd(out, a, b) \
-      __m128i out##_l = _mm_add_epi32(a##_l, b##_l); \
-      __m128i out##_h = _mm_add_epi32(a##_h, b##_h)
-
-   // wide sub
-   #define dct_wsub(out, a, b) \
-      __m128i out##_l = _mm_sub_epi32(a##_l, b##_l); \
-      __m128i out##_h = _mm_sub_epi32(a##_h, b##_h)
-
-   // butterfly a/b, add bias, then shift by "s" and pack
-   #define dct_bfly32o(out0, out1, a,b,bias,s) \
-      { \
-         __m128i abiased_l = _mm_add_epi32(a##_l, bias); \
-         __m128i abiased_h = _mm_add_epi32(a##_h, bias); \
-         dct_wadd(sum, abiased, b); \
-         dct_wsub(dif, abiased, b); \
-         out0 = _mm_packs_epi32(_mm_srai_epi32(sum_l, s), _mm_srai_epi32(sum_h, s)); \
-         out1 = _mm_packs_epi32(_mm_srai_epi32(dif_l, s), _mm_srai_epi32(dif_h, s)); \
-      }
-
-   // 8-bit interleave step (for transposes)
-   #define dct_interleave8(a, b) \
-      tmp = a; \
-      a = _mm_unpacklo_epi8(a, b); \
-      b = _mm_unpackhi_epi8(tmp, b)
-
-   // 16-bit interleave step (for transposes)
-   #define dct_interleave16(a, b) \
-      tmp = a; \
-      a = _mm_unpacklo_epi16(a, b); \
-      b = _mm_unpackhi_epi16(tmp, b)
-
-   #define dct_pass(bias,shift) \
-      { \
-         /* even part */ \
-         dct_rot(t2e,t3e, row2,row6, rot0_0,rot0_1); \
-         __m128i sum04 = _mm_add_epi16(row0, row4); \
-         __m128i dif04 = _mm_sub_epi16(row0, row4); \
-         dct_widen(t0e, sum04); \
-         dct_widen(t1e, dif04); \
-         dct_wadd(x0, t0e, t3e); \
-         dct_wsub(x3, t0e, t3e); \
-         dct_wadd(x1, t1e, t2e); \
-         dct_wsub(x2, t1e, t2e); \
-         /* odd part */ \
-         dct_rot(y0o,y2o, row7,row3, rot2_0,rot2_1); \
-         dct_rot(y1o,y3o, row5,row1, rot3_0,rot3_1); \
-         __m128i sum17 = _mm_add_epi16(row1, row7); \
-         __m128i sum35 = _mm_add_epi16(row3, row5); \
-         dct_rot(y4o,y5o, sum17,sum35, rot1_0,rot1_1); \
-         dct_wadd(x4, y0o, y4o); \
-         dct_wadd(x5, y1o, y5o); \
-         dct_wadd(x6, y2o, y5o); \
-         dct_wadd(x7, y3o, y4o); \
-         dct_bfly32o(row0,row7, x0,x7,bias,shift); \
-         dct_bfly32o(row1,row6, x1,x6,bias,shift); \
-         dct_bfly32o(row2,row5, x2,x5,bias,shift); \
-         dct_bfly32o(row3,row4, x3,x4,bias,shift); \
-      }
+// dot product constant: even elems=x, odd elems=y
+#define dct_const(x, y) _mm_setr_epi16((x), (y), (x), (y), (x), (y), (x), (y))
+
+// out(0) = c0[even]*x + c0[odd]*y   (c0, x, y 16-bit, out 32-bit)
+// out(1) = c1[even]*x + c1[odd]*y
+#define dct_rot(out0, out1, x, y, c0, c1)         \
+   __m128i c0##lo = _mm_unpacklo_epi16((x), (y)); \
+   __m128i c0##hi = _mm_unpackhi_epi16((x), (y)); \
+   __m128i out0##_l = _mm_madd_epi16(c0##lo, c0); \
+   __m128i out0##_h = _mm_madd_epi16(c0##hi, c0); \
+   __m128i out1##_l = _mm_madd_epi16(c0##lo, c1); \
+   __m128i out1##_h = _mm_madd_epi16(c0##hi, c1)
+
+// out = in << 12  (in 16-bit, out 32-bit)
+#define dct_widen(out, in)                                                             \
+   __m128i out##_l = _mm_srai_epi32(_mm_unpacklo_epi16(_mm_setzero_si128(), (in)), 4); \
+   __m128i out##_h = _mm_srai_epi32(_mm_unpackhi_epi16(_mm_setzero_si128(), (in)), 4)
+
+// wide add
+#define dct_wadd(out, a, b)                       \
+   __m128i out##_l = _mm_add_epi32(a##_l, b##_l); \
+   __m128i out##_h = _mm_add_epi32(a##_h, b##_h)
+
+// wide sub
+#define dct_wsub(out, a, b)                       \
+   __m128i out##_l = _mm_sub_epi32(a##_l, b##_l); \
+   __m128i out##_h = _mm_sub_epi32(a##_h, b##_h)
+
+// butterfly a/b, add bias, then shift by "s" and pack
+#define dct_bfly32o(out0, out1, a, b, bias, s)                                    \
+   {                                                                              \
+      __m128i abiased_l = _mm_add_epi32(a##_l, bias);                             \
+      __m128i abiased_h = _mm_add_epi32(a##_h, bias);                             \
+      dct_wadd(sum, abiased, b);                                                  \
+      dct_wsub(dif, abiased, b);                                                  \
+      out0 = _mm_packs_epi32(_mm_srai_epi32(sum_l, s), _mm_srai_epi32(sum_h, s)); \
+      out1 = _mm_packs_epi32(_mm_srai_epi32(dif_l, s), _mm_srai_epi32(dif_h, s)); \
+   }
+
+// 8-bit interleave step (for transposes)
+#define dct_interleave8(a, b)   \
+   tmp = a;                     \
+   a = _mm_unpacklo_epi8(a, b); \
+   b = _mm_unpackhi_epi8(tmp, b)
+
+// 16-bit interleave step (for transposes)
+#define dct_interleave16(a, b)   \
+   tmp = a;                      \
+   a = _mm_unpacklo_epi16(a, b); \
+   b = _mm_unpackhi_epi16(tmp, b)
+
+#define dct_pass(bias, shift)                          \
+   {                                                   \
+      /* even part */                                  \
+      dct_rot(t2e, t3e, row2, row6, rot0_0, rot0_1);   \
+      __m128i sum04 = _mm_add_epi16(row0, row4);       \
+      __m128i dif04 = _mm_sub_epi16(row0, row4);       \
+      dct_widen(t0e, sum04);                           \
+      dct_widen(t1e, dif04);                           \
+      dct_wadd(x0, t0e, t3e);                          \
+      dct_wsub(x3, t0e, t3e);                          \
+      dct_wadd(x1, t1e, t2e);                          \
+      dct_wsub(x2, t1e, t2e);                          \
+      /* odd part */                                   \
+      dct_rot(y0o, y2o, row7, row3, rot2_0, rot2_1);   \
+      dct_rot(y1o, y3o, row5, row1, rot3_0, rot3_1);   \
+      __m128i sum17 = _mm_add_epi16(row1, row7);       \
+      __m128i sum35 = _mm_add_epi16(row3, row5);       \
+      dct_rot(y4o, y5o, sum17, sum35, rot1_0, rot1_1); \
+      dct_wadd(x4, y0o, y4o);                          \
+      dct_wadd(x5, y1o, y5o);                          \
+      dct_wadd(x6, y2o, y5o);                          \
+      dct_wadd(x7, y3o, y4o);                          \
+      dct_bfly32o(row0, row7, x0, x7, bias, shift);    \
+      dct_bfly32o(row1, row6, x1, x6, bias, shift);    \
+      dct_bfly32o(row2, row5, x2, x5, bias, shift);    \
+      dct_bfly32o(row3, row4, x3, x4, bias, shift);    \
+   }
 
    __m128i rot0_0 = dct_const(stbi__f2f(0.5411961f), stbi__f2f(0.5411961f) + stbi__f2f(-1.847759065f));
-   __m128i rot0_1 = dct_const(stbi__f2f(0.5411961f) + stbi__f2f( 0.765366865f), stbi__f2f(0.5411961f));
+   __m128i rot0_1 = dct_const(stbi__f2f(0.5411961f) + stbi__f2f(0.765366865f), stbi__f2f(0.5411961f));
    __m128i rot1_0 = dct_const(stbi__f2f(1.175875602f) + stbi__f2f(-0.899976223f), stbi__f2f(1.175875602f));
    __m128i rot1_1 = dct_const(stbi__f2f(1.175875602f), stbi__f2f(1.175875602f) + stbi__f2f(-2.562915447f));
-   __m128i rot2_0 = dct_const(stbi__f2f(-1.961570560f) + stbi__f2f( 0.298631336f), stbi__f2f(-1.961570560f));
-   __m128i rot2_1 = dct_const(stbi__f2f(-1.961570560f), stbi__f2f(-1.961570560f) + stbi__f2f( 3.072711026f));
-   __m128i rot3_0 = dct_const(stbi__f2f(-0.390180644f) + stbi__f2f( 2.053119869f), stbi__f2f(-0.390180644f));
-   __m128i rot3_1 = dct_const(stbi__f2f(-0.390180644f), stbi__f2f(-0.390180644f) + stbi__f2f( 1.501321110f));
+   __m128i rot2_0 = dct_const(stbi__f2f(-1.961570560f) + stbi__f2f(0.298631336f), stbi__f2f(-1.961570560f));
+   __m128i rot2_1 = dct_const(stbi__f2f(-1.961570560f), stbi__f2f(-1.961570560f) + stbi__f2f(3.072711026f));
+   __m128i rot3_0 = dct_const(stbi__f2f(-0.390180644f) + stbi__f2f(2.053119869f), stbi__f2f(-0.390180644f));
+   __m128i rot3_1 = dct_const(stbi__f2f(-0.390180644f), stbi__f2f(-0.390180644f) + stbi__f2f(1.501321110f));
 
    // rounding biases in column/row passes, see stbi__idct_block for explanation.
    __m128i bias_0 = _mm_set1_epi32(512);
-   __m128i bias_1 = _mm_set1_epi32(65536 + (128<<17));
+   __m128i bias_1 = _mm_set1_epi32(65536 + (128 << 17));
 
    // load
-   row0 = _mm_load_si128((const __m128i *) (data + 0*8));
-   row1 = _mm_load_si128((const __m128i *) (data + 1*8));
-   row2 = _mm_load_si128((const __m128i *) (data + 2*8));
-   row3 = _mm_load_si128((const __m128i *) (data + 3*8));
-   row4 = _mm_load_si128((const __m128i *) (data + 4*8));
-   row5 = _mm_load_si128((const __m128i *) (data + 5*8));
-   row6 = _mm_load_si128((const __m128i *) (data + 6*8));
-   row7 = _mm_load_si128((const __m128i *) (data + 7*8));
+   row0 = _mm_load_si128((const __m128i*)(data + 0 * 8));
+   row1 = _mm_load_si128((const __m128i*)(data + 1 * 8));
+   row2 = _mm_load_si128((const __m128i*)(data + 2 * 8));
+   row3 = _mm_load_si128((const __m128i*)(data + 3 * 8));
+   row4 = _mm_load_si128((const __m128i*)(data + 4 * 8));
+   row5 = _mm_load_si128((const __m128i*)(data + 5 * 8));
+   row6 = _mm_load_si128((const __m128i*)(data + 6 * 8));
+   row7 = _mm_load_si128((const __m128i*)(data + 7 * 8));
 
    // column pass
    dct_pass(bias_0, 10);
@@ -2181,14 +2955,21 @@ static void stbi__idct_simd(stbi_uc *out, int out_stride, short data[64])
       dct_interleave8(p1, p3); // a4b4c4d4...
 
       // store
-      _mm_storel_epi64((__m128i *) out, p0); out += out_stride;
-      _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p0, 0x4e)); out += out_stride;
-      _mm_storel_epi64((__m128i *) out, p2); out += out_stride;
-      _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p2, 0x4e)); out += out_stride;
-      _mm_storel_epi64((__m128i *) out, p1); out += out_stride;
-      _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p1, 0x4e)); out += out_stride;
-      _mm_storel_epi64((__m128i *) out, p3); out += out_stride;
-      _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p3, 0x4e));
+      _mm_storel_epi64((__m128i*)out, p0);
+      out += out_stride;
+      _mm_storel_epi64((__m128i*)out, _mm_shuffle_epi32(p0, 0x4e));
+      out += out_stride;
+      _mm_storel_epi64((__m128i*)out, p2);
+      out += out_stride;
+      _mm_storel_epi64((__m128i*)out, _mm_shuffle_epi32(p2, 0x4e));
+      out += out_stride;
+      _mm_storel_epi64((__m128i*)out, p1);
+      out += out_stride;
+      _mm_storel_epi64((__m128i*)out, _mm_shuffle_epi32(p1, 0x4e));
+      out += out_stride;
+      _mm_storel_epi64((__m128i*)out, p3);
+      out += out_stride;
+      _mm_storel_epi64((__m128i*)out, _mm_shuffle_epi32(p3, 0x4e));
    }
 
 #undef dct_const
@@ -2208,103 +2989,103 @@ static void stbi__idct_simd(stbi_uc *out, int out_stride, short data[64])
 
 // NEON integer IDCT. should produce bit-identical
 // results to the generic C version.
-static void stbi__idct_simd(stbi_uc *out, int out_stride, short data[64])
+static void stbi__idct_simd(stbi_uc* out, int out_stride, short data[64])
 {
    int16x8_t row0, row1, row2, row3, row4, row5, row6, row7;
 
    int16x4_t rot0_0 = vdup_n_s16(stbi__f2f(0.5411961f));
    int16x4_t rot0_1 = vdup_n_s16(stbi__f2f(-1.847759065f));
-   int16x4_t rot0_2 = vdup_n_s16(stbi__f2f( 0.765366865f));
-   int16x4_t rot1_0 = vdup_n_s16(stbi__f2f( 1.175875602f));
+   int16x4_t rot0_2 = vdup_n_s16(stbi__f2f(0.765366865f));
+   int16x4_t rot1_0 = vdup_n_s16(stbi__f2f(1.175875602f));
    int16x4_t rot1_1 = vdup_n_s16(stbi__f2f(-0.899976223f));
    int16x4_t rot1_2 = vdup_n_s16(stbi__f2f(-2.562915447f));
    int16x4_t rot2_0 = vdup_n_s16(stbi__f2f(-1.961570560f));
    int16x4_t rot2_1 = vdup_n_s16(stbi__f2f(-0.390180644f));
-   int16x4_t rot3_0 = vdup_n_s16(stbi__f2f( 0.298631336f));
-   int16x4_t rot3_1 = vdup_n_s16(stbi__f2f( 2.053119869f));
-   int16x4_t rot3_2 = vdup_n_s16(stbi__f2f( 3.072711026f));
-   int16x4_t rot3_3 = vdup_n_s16(stbi__f2f( 1.501321110f));
+   int16x4_t rot3_0 = vdup_n_s16(stbi__f2f(0.298631336f));
+   int16x4_t rot3_1 = vdup_n_s16(stbi__f2f(2.053119869f));
+   int16x4_t rot3_2 = vdup_n_s16(stbi__f2f(3.072711026f));
+   int16x4_t rot3_3 = vdup_n_s16(stbi__f2f(1.501321110f));
 
-#define dct_long_mul(out, inq, coeff) \
+#define dct_long_mul(out, inq, coeff)                       \
    int32x4_t out##_l = vmull_s16(vget_low_s16(inq), coeff); \
    int32x4_t out##_h = vmull_s16(vget_high_s16(inq), coeff)
 
-#define dct_long_mac(out, acc, inq, coeff) \
+#define dct_long_mac(out, acc, inq, coeff)                           \
    int32x4_t out##_l = vmlal_s16(acc##_l, vget_low_s16(inq), coeff); \
    int32x4_t out##_h = vmlal_s16(acc##_h, vget_high_s16(inq), coeff)
 
-#define dct_widen(out, inq) \
+#define dct_widen(out, inq)                                \
    int32x4_t out##_l = vshll_n_s16(vget_low_s16(inq), 12); \
    int32x4_t out##_h = vshll_n_s16(vget_high_s16(inq), 12)
 
 // wide add
-#define dct_wadd(out, a, b) \
+#define dct_wadd(out, a, b)                     \
    int32x4_t out##_l = vaddq_s32(a##_l, b##_l); \
    int32x4_t out##_h = vaddq_s32(a##_h, b##_h)
 
 // wide sub
-#define dct_wsub(out, a, b) \
+#define dct_wsub(out, a, b)                     \
    int32x4_t out##_l = vsubq_s32(a##_l, b##_l); \
    int32x4_t out##_h = vsubq_s32(a##_h, b##_h)
 
 // butterfly a/b, then shift using "shiftop" by "s" and pack
-#define dct_bfly32o(out0,out1, a,b,shiftop,s) \
-   { \
-      dct_wadd(sum, a, b); \
-      dct_wsub(dif, a, b); \
+#define dct_bfly32o(out0, out1, a, b, shiftop, s)                \
+   {                                                             \
+      dct_wadd(sum, a, b);                                       \
+      dct_wsub(dif, a, b);                                       \
       out0 = vcombine_s16(shiftop(sum_l, s), shiftop(sum_h, s)); \
       out1 = vcombine_s16(shiftop(dif_l, s), shiftop(dif_h, s)); \
    }
 
-#define dct_pass(shiftop, shift) \
-   { \
-      /* even part */ \
-      int16x8_t sum26 = vaddq_s16(row2, row6); \
-      dct_long_mul(p1e, sum26, rot0_0); \
-      dct_long_mac(t2e, p1e, row6, rot0_1); \
-      dct_long_mac(t3e, p1e, row2, rot0_2); \
-      int16x8_t sum04 = vaddq_s16(row0, row4); \
-      int16x8_t dif04 = vsubq_s16(row0, row4); \
-      dct_widen(t0e, sum04); \
-      dct_widen(t1e, dif04); \
-      dct_wadd(x0, t0e, t3e); \
-      dct_wsub(x3, t0e, t3e); \
-      dct_wadd(x1, t1e, t2e); \
-      dct_wsub(x2, t1e, t2e); \
-      /* odd part */ \
-      int16x8_t sum15 = vaddq_s16(row1, row5); \
-      int16x8_t sum17 = vaddq_s16(row1, row7); \
-      int16x8_t sum35 = vaddq_s16(row3, row5); \
-      int16x8_t sum37 = vaddq_s16(row3, row7); \
-      int16x8_t sumodd = vaddq_s16(sum17, sum35); \
-      dct_long_mul(p5o, sumodd, rot1_0); \
-      dct_long_mac(p1o, p5o, sum17, rot1_1); \
-      dct_long_mac(p2o, p5o, sum35, rot1_2); \
-      dct_long_mul(p3o, sum37, rot2_0); \
-      dct_long_mul(p4o, sum15, rot2_1); \
-      dct_wadd(sump13o, p1o, p3o); \
-      dct_wadd(sump24o, p2o, p4o); \
-      dct_wadd(sump23o, p2o, p3o); \
-      dct_wadd(sump14o, p1o, p4o); \
-      dct_long_mac(x4, sump13o, row7, rot3_0); \
-      dct_long_mac(x5, sump24o, row5, rot3_1); \
-      dct_long_mac(x6, sump23o, row3, rot3_2); \
-      dct_long_mac(x7, sump14o, row1, rot3_3); \
-      dct_bfly32o(row0,row7, x0,x7,shiftop,shift); \
-      dct_bfly32o(row1,row6, x1,x6,shiftop,shift); \
-      dct_bfly32o(row2,row5, x2,x5,shiftop,shift); \
-      dct_bfly32o(row3,row4, x3,x4,shiftop,shift); \
+#define dct_pass(shiftop, shift)                       \
+   {                                                   \
+      /* even part */                                  \
+      int16x8_t sum26 = vaddq_s16(row2, row6);         \
+      dct_long_mul(p1e, sum26, rot0_0);                \
+      dct_long_mac(t2e, p1e, row6, rot0_1);            \
+      dct_long_mac(t3e, p1e, row2, rot0_2);            \
+      int16x8_t sum04 = vaddq_s16(row0, row4);         \
+      int16x8_t dif04 = vsubq_s16(row0, row4);         \
+      dct_widen(t0e, sum04);                           \
+      dct_widen(t1e, dif04);                           \
+      dct_wadd(x0, t0e, t3e);                          \
+      dct_wsub(x3, t0e, t3e);                          \
+      dct_wadd(x1, t1e, t2e);                          \
+      dct_wsub(x2, t1e, t2e);                          \
+      /* odd part */                                   \
+      int16x8_t sum15 = vaddq_s16(row1, row5);         \
+      int16x8_t sum17 = vaddq_s16(row1, row7);         \
+      int16x8_t sum35 = vaddq_s16(row3, row5);         \
+      int16x8_t sum37 = vaddq_s16(row3, row7);         \
+      int16x8_t sumodd = vaddq_s16(sum17, sum35);      \
+      dct_long_mul(p5o, sumodd, rot1_0);               \
+      dct_long_mac(p1o, p5o, sum17, rot1_1);           \
+      dct_long_mac(p2o, p5o, sum35, rot1_2);           \
+      dct_long_mul(p3o, sum37, rot2_0);                \
+      dct_long_mul(p4o, sum15, rot2_1);                \
+      dct_wadd(sump13o, p1o, p3o);                     \
+      dct_wadd(sump24o, p2o, p4o);                     \
+      dct_wadd(sump23o, p2o, p3o);                     \
+      dct_wadd(sump14o, p1o, p4o);                     \
+      dct_long_mac(x4, sump13o, row7, rot3_0);         \
+      dct_long_mac(x5, sump24o, row5, rot3_1);         \
+      dct_long_mac(x6, sump23o, row3, rot3_2);         \
+      dct_long_mac(x7, sump14o, row1, rot3_3);         \
+      dct_bfly32o(row0, row7, x0, x7, shiftop, shift); \
+      dct_bfly32o(row1, row6, x1, x6, shiftop, shift); \
+      dct_bfly32o(row2, row5, x2, x5, shiftop, shift); \
+      dct_bfly32o(row3, row4, x3, x4, shiftop, shift); \
    }
 
    // load
-   row0 = vld1q_s16(data + 0*8);
-   row1 = vld1q_s16(data + 1*8);
-   row2 = vld1q_s16(data + 2*8);
-   row3 = vld1q_s16(data + 3*8);
-   row4 = vld1q_s16(data + 4*8);
-   row5 = vld1q_s16(data + 5*8);
-   row6 = vld1q_s16(data + 6*8);
-   row7 = vld1q_s16(data + 7*8);
+   row0 = vld1q_s16(data + 0 * 8);
+   row1 = vld1q_s16(data + 1 * 8);
+   row2 = vld1q_s16(data + 2 * 8);
+   row3 = vld1q_s16(data + 3 * 8);
+   row4 = vld1q_s16(data + 4 * 8);
+   row5 = vld1q_s16(data + 5 * 8);
+   row6 = vld1q_s16(data + 6 * 8);
+   row7 = vld1q_s16(data + 7 * 8);
 
    // add DC bias
    row0 = vaddq_s16(row0, vsetq_lane_s16(1024, vdupq_n_s16(0), 0));
@@ -2316,9 +3097,25 @@ static void stbi__idct_simd(stbi_uc *out, int out_stride, short data[64])
    {
 // these three map to a single VTRN.16, VTRN.32, and VSWP, respectively.
 // whether compilers actually get this is another story, sadly.
-#define dct_trn16(x, y) { int16x8x2_t t = vtrnq_s16(x, y); x = t.val[0]; y = t.val[1]; }
-#define dct_trn32(x, y) { int32x4x2_t t = vtrnq_s32(vreinterpretq_s32_s16(x), vreinterpretq_s32_s16(y)); x = vreinterpretq_s16_s32(t.val[0]); y = vreinterpretq_s16_s32(t.val[1]); }
-#define dct_trn64(x, y) { int16x8_t x0 = x; int16x8_t y0 = y; x = vcombine_s16(vget_low_s16(x0), vget_low_s16(y0)); y = vcombine_s16(vget_high_s16(x0), vget_high_s16(y0)); }
+#define dct_trn16(x, y)                \
+   {                                   \
+      int16x8x2_t t = vtrnq_s16(x, y); \
+      x = t.val[0];                    \
+      y = t.val[1];                    \
+   }
+#define dct_trn32(x, y)                                                              \
+   {                                                                                 \
+      int32x4x2_t t = vtrnq_s32(vreinterpretq_s32_s16(x), vreinterpretq_s32_s16(y)); \
+      x = vreinterpretq_s16_s32(t.val[0]);                                           \
+      y = vreinterpretq_s16_s32(t.val[1]);                                           \
+   }
+#define dct_trn64(x, y)                                       \
+   {                                                          \
+      int16x8_t x0 = x;                                       \
+      int16x8_t y0 = y;                                       \
+      x = vcombine_s16(vget_low_s16(x0), vget_low_s16(y0));   \
+      y = vcombine_s16(vget_high_s16(x0), vget_high_s16(y0)); \
+   }
 
       // pass 1
       dct_trn16(row0, row1); // a0b0a2b2a4b4a6b6
@@ -2361,9 +3158,24 @@ static void stbi__idct_simd(stbi_uc *out, int out_stride, short data[64])
       uint8x8_t p7 = vqrshrun_n_s16(row7, 1);
 
       // again, these can translate into one instruction, but often don't.
-#define dct_trn8_8(x, y) { uint8x8x2_t t = vtrn_u8(x, y); x = t.val[0]; y = t.val[1]; }
-#define dct_trn8_16(x, y) { uint16x4x2_t t = vtrn_u16(vreinterpret_u16_u8(x), vreinterpret_u16_u8(y)); x = vreinterpret_u8_u16(t.val[0]); y = vreinterpret_u8_u16(t.val[1]); }
-#define dct_trn8_32(x, y) { uint32x2x2_t t = vtrn_u32(vreinterpret_u32_u8(x), vreinterpret_u32_u8(y)); x = vreinterpret_u8_u32(t.val[0]); y = vreinterpret_u8_u32(t.val[1]); }
+#define dct_trn8_8(x, y)             \
+   {                                 \
+      uint8x8x2_t t = vtrn_u8(x, y); \
+      x = t.val[0];                  \
+      y = t.val[1];                  \
+   }
+#define dct_trn8_16(x, y)                                                        \
+   {                                                                             \
+      uint16x4x2_t t = vtrn_u16(vreinterpret_u16_u8(x), vreinterpret_u16_u8(y)); \
+      x = vreinterpret_u8_u16(t.val[0]);                                         \
+      y = vreinterpret_u8_u16(t.val[1]);                                         \
+   }
+#define dct_trn8_32(x, y)                                                        \
+   {                                                                             \
+      uint32x2x2_t t = vtrn_u32(vreinterpret_u32_u8(x), vreinterpret_u32_u8(y)); \
+      x = vreinterpret_u8_u32(t.val[0]);                                         \
+      y = vreinterpret_u8_u32(t.val[1]);                                         \
+   }
 
       // sadly can't use interleaved stores here since we only write
       // 8 bytes to each scan line!
@@ -2387,13 +3199,20 @@ static void stbi__idct_simd(stbi_uc *out, int out_stride, short data[64])
       dct_trn8_32(p3, p7);
 
       // store
-      vst1_u8(out, p0); out += out_stride;
-      vst1_u8(out, p1); out += out_stride;
-      vst1_u8(out, p2); out += out_stride;
-      vst1_u8(out, p3); out += out_stride;
-      vst1_u8(out, p4); out += out_stride;
-      vst1_u8(out, p5); out += out_stride;
-      vst1_u8(out, p6); out += out_stride;
+      vst1_u8(out, p0);
+      out += out_stride;
+      vst1_u8(out, p1);
+      out += out_stride;
+      vst1_u8(out, p2);
+      out += out_stride;
+      vst1_u8(out, p3);
+      out += out_stride;
+      vst1_u8(out, p4);
+      out += out_stride;
+      vst1_u8(out, p5);
+      out += out_stride;
+      vst1_u8(out, p6);
+      out += out_stride;
       vst1_u8(out, p7);
 
 #undef dct_trn8_8
@@ -2412,33 +3231,39 @@ static void stbi__idct_simd(stbi_uc *out, int out_stride, short data[64])
 
 #endif // STBI_NEON
 
-#define STBI__MARKER_none  0xff
+#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)
+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; }
+   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;
+   if (x != 0xff)
+      return STBI__MARKER_none;
    while (x == 0xff)
-      x = stbi__get8(j->s);
+      x = stbi__get8(j->s); // consume repeated 0xff fill bytes
    return x;
 }
 
 // in each scan, we'll have scan_n components, and the order
 // of the components is specified by order[]
-#define STBI__RESTART(x)     ((x) >= 0xd0 && (x) <= 0xd7)
+#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)
+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->img_comp[0].dc_pred = j->img_comp[1].dc_pred = j->img_comp[2].dc_pred = j->img_comp[3].dc_pred = 0;
    j->marker = STBI__MARKER_none;
    j->todo = j->restart_interval ? j->restart_interval : 0x7fffffff;
    j->eob_run = 0;
@@ -2446,111 +3271,148 @@ static void stbi__jpeg_reset(stbi__jpeg *j)
    // since we don't even allow 1<<30 pixels
 }
 
-static int stbi__parse_entropy_coded_data(stbi__jpeg *z)
+static int stbi__parse_entropy_coded_data(stbi__jpeg* z)
 {
    stbi__jpeg_reset(z);
-   if (!z->progressive) {
-      if (z->scan_n == 1) {
-         int i,j;
+   if (!z->progressive)
+   {
+      if (z->scan_n == 1)
+      {
+         int i, j;
          STBI_SIMD_ALIGN(short, data[64]);
          int n = z->order[0];
          // non-interleaved data, we just need to process one block at a time,
          // in trivial scanline order
          // number of blocks to do just depends on how many actual "pixels" this
          // component has, independent of interleaved MCU blocking and such
-         int w = (z->img_comp[n].x+7) >> 3;
-         int h = (z->img_comp[n].y+7) >> 3;
-         for (j=0; j < h; ++j) {
-            for (i=0; i < w; ++i) {
+         int w = (z->img_comp[n].x + 7) >> 3;
+         int h = (z->img_comp[n].y + 7) >> 3;
+         for (j = 0; j < h; ++j)
+         {
+            for (i = 0; i < w; ++i)
+            {
                int ha = z->img_comp[n].ha;
-               if (!stbi__jpeg_decode_block(z, data, z->huff_dc+z->img_comp[n].hd, z->huff_ac+ha, z->fast_ac[ha], n, z->dequant[z->img_comp[n].tq])) return 0;
-               z->idct_block_kernel(z->img_comp[n].data+z->img_comp[n].w2*j*8+i*8, z->img_comp[n].w2, data);
+               if (!stbi__jpeg_decode_block(z, data, z->huff_dc + z->img_comp[n].hd, z->huff_ac + ha, z->fast_ac[ha], n, z->dequant[z->img_comp[n].tq]))
+                  return 0;
+               z->idct_block_kernel(z->img_comp[n].data + z->img_comp[n].w2 * j * 8 + i * 8, z->img_comp[n].w2, data);
                // every data block is an MCU, so countdown the restart interval
-               if (--z->todo <= 0) {
-                  if (z->code_bits < 24) stbi__grow_buffer_unsafe(z);
+               if (--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;
+                  if (!STBI__RESTART(z->marker))
+                     return 1;
                   stbi__jpeg_reset(z);
                }
             }
          }
          return 1;
-      } else { // interleaved
-         int i,j,k,x,y;
+      }
+      else
+      { // interleaved
+         int i, j, k, x, y;
          STBI_SIMD_ALIGN(short, data[64]);
-         for (j=0; j < z->img_mcu_y; ++j) {
-            for (i=0; i < z->img_mcu_x; ++i) {
+         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) {
+               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;
+                  for (y = 0; y < z->img_comp[n].v; ++y)
+                  {
+                     for (x = 0; x < z->img_comp[n].h; ++x)
+                     {
+                        int x2 = (i * z->img_comp[n].h + x) * 8;
+                        int y2 = (j * z->img_comp[n].v + y) * 8;
                         int ha = z->img_comp[n].ha;
-                        if (!stbi__jpeg_decode_block(z, data, z->huff_dc+z->img_comp[n].hd, z->huff_ac+ha, z->fast_ac[ha], n, z->dequant[z->img_comp[n].tq])) return 0;
-                        z->idct_block_kernel(z->img_comp[n].data+z->img_comp[n].w2*y2+x2, z->img_comp[n].w2, data);
+                        if (!stbi__jpeg_decode_block(z, data, z->huff_dc + z->img_comp[n].hd, z->huff_ac + ha, z->fast_ac[ha], n, z->dequant[z->img_comp[n].tq]))
+                           return 0;
+                        z->idct_block_kernel(z->img_comp[n].data + z->img_comp[n].w2 * y2 + x2, z->img_comp[n].w2, data);
                      }
                   }
                }
                // after all interleaved components, that's an interleaved MCU,
                // so now count down the restart interval
-               if (--z->todo <= 0) {
-                  if (z->code_bits < 24) stbi__grow_buffer_unsafe(z);
-                  if (!STBI__RESTART(z->marker)) return 1;
+               if (--z->todo <= 0)
+               {
+                  if (z->code_bits < 24)
+                     stbi__grow_buffer_unsafe(z);
+                  if (!STBI__RESTART(z->marker))
+                     return 1;
                   stbi__jpeg_reset(z);
                }
             }
          }
          return 1;
       }
-   } else {
-      if (z->scan_n == 1) {
-         int i,j;
+   }
+   else
+   {
+      if (z->scan_n == 1)
+      {
+         int i, j;
          int n = z->order[0];
          // non-interleaved data, we just need to process one block at a time,
          // in trivial scanline order
          // number of blocks to do just depends on how many actual "pixels" this
          // component has, independent of interleaved MCU blocking and such
-         int w = (z->img_comp[n].x+7) >> 3;
-         int h = (z->img_comp[n].y+7) >> 3;
-         for (j=0; j < h; ++j) {
-            for (i=0; i < w; ++i) {
-               short *data = z->img_comp[n].coeff + 64 * (i + j * z->img_comp[n].coeff_w);
-               if (z->spec_start == 0) {
+         int w = (z->img_comp[n].x + 7) >> 3;
+         int h = (z->img_comp[n].y + 7) >> 3;
+         for (j = 0; j < h; ++j)
+         {
+            for (i = 0; i < w; ++i)
+            {
+               short* data = z->img_comp[n].coeff + 64 * (i + j * z->img_comp[n].coeff_w);
+               if (z->spec_start == 0)
+               {
                   if (!stbi__jpeg_decode_block_prog_dc(z, data, &z->huff_dc[z->img_comp[n].hd], n))
                      return 0;
-               } else {
+               }
+               else
+               {
                   int ha = z->img_comp[n].ha;
                   if (!stbi__jpeg_decode_block_prog_ac(z, data, &z->huff_ac[ha], z->fast_ac[ha]))
                      return 0;
                }
                // every data block is an MCU, so countdown the restart interval
-               if (--z->todo <= 0) {
-                  if (z->code_bits < 24) stbi__grow_buffer_unsafe(z);
-                  if (!STBI__RESTART(z->marker)) return 1;
+               if (--z->todo <= 0)
+               {
+                  if (z->code_bits < 24)
+                     stbi__grow_buffer_unsafe(z);
+                  if (!STBI__RESTART(z->marker))
+                     return 1;
                   stbi__jpeg_reset(z);
                }
             }
          }
          return 1;
-      } else { // interleaved
-         int i,j,k,x,y;
-         for (j=0; j < z->img_mcu_y; ++j) {
-            for (i=0; i < z->img_mcu_x; ++i) {
+      }
+      else
+      { // interleaved
+         int i, j, k, x, y;
+         for (j = 0; j < z->img_mcu_y; ++j)
+         {
+            for (i = 0; i < z->img_mcu_x; ++i)
+            {
                // scan an interleaved mcu... process scan_n components in order
-               for (k=0; k < z->scan_n; ++k) {
+               for (k = 0; k < z->scan_n; ++k)
+               {
                   int n = z->order[k];
                   // scan out an mcu's worth of this component; that's just determined
                   // by the basic H and V specified for the component
-                  for (y=0; y < z->img_comp[n].v; ++y) {
-                     for (x=0; x < z->img_comp[n].h; ++x) {
-                        int x2 = (i*z->img_comp[n].h + x);
-                        int y2 = (j*z->img_comp[n].v + y);
-                        short *data = z->img_comp[n].coeff + 64 * (x2 + y2 * z->img_comp[n].coeff_w);
+                  for (y = 0; y < z->img_comp[n].v; ++y)
+                  {
+                     for (x = 0; x < z->img_comp[n].h; ++x)
+                     {
+                        int x2 = (i * z->img_comp[n].h + x);
+                        int y2 = (j * z->img_comp[n].v + y);
+                        short* data = z->img_comp[n].coeff + 64 * (x2 + y2 * z->img_comp[n].coeff_w);
                         if (!stbi__jpeg_decode_block_prog_dc(z, data, &z->huff_dc[z->img_comp[n].hd], n))
                            return 0;
                      }
@@ -2558,9 +3420,12 @@ static int stbi__parse_entropy_coded_data(stbi__jpeg *z)
                }
                // after all interleaved components, that's an interleaved MCU,
                // so now count down the restart interval
-               if (--z->todo <= 0) {
-                  if (z->code_bits < 24) stbi__grow_buffer_unsafe(z);
-                  if (!STBI__RESTART(z->marker)) return 1;
+               if (--z->todo <= 0)
+               {
+                  if (z->code_bits < 24)
+                     stbi__grow_buffer_unsafe(z);
+                  if (!STBI__RESTART(z->marker))
+                     return 1;
                   stbi__jpeg_reset(z);
                }
             }
@@ -2570,128 +3435,205 @@ static int stbi__parse_entropy_coded_data(stbi__jpeg *z)
    }
 }
 
-static void stbi__jpeg_dequantize(short *data, stbi_uc *dequant)
+static void stbi__jpeg_dequantize(short* data, stbi__uint16* dequant)
 {
    int i;
-   for (i=0; i < 64; ++i)
+   for (i = 0; i < 64; ++i)
       data[i] *= dequant[i];
 }
 
-static void stbi__jpeg_finish(stbi__jpeg *z)
+static void stbi__jpeg_finish(stbi__jpeg* z)
 {
-   if (z->progressive) {
+   if (z->progressive)
+   {
       // dequantize and idct the data
-      int i,j,n;
-      for (n=0; n < z->s->img_n; ++n) {
-         int w = (z->img_comp[n].x+7) >> 3;
-         int h = (z->img_comp[n].y+7) >> 3;
-         for (j=0; j < h; ++j) {
-            for (i=0; i < w; ++i) {
-               short *data = z->img_comp[n].coeff + 64 * (i + j * z->img_comp[n].coeff_w);
+      int i, j, n;
+      for (n = 0; n < z->s->img_n; ++n)
+      {
+         int w = (z->img_comp[n].x + 7) >> 3;
+         int h = (z->img_comp[n].y + 7) >> 3;
+         for (j = 0; j < h; ++j)
+         {
+            for (i = 0; i < w; ++i)
+            {
+               short* data = z->img_comp[n].coeff + 64 * (i + j * z->img_comp[n].coeff_w);
                stbi__jpeg_dequantize(data, z->dequant[z->img_comp[n].tq]);
-               z->idct_block_kernel(z->img_comp[n].data+z->img_comp[n].w2*j*8+i*8, z->img_comp[n].w2, data);
+               z->idct_block_kernel(z->img_comp[n].data + z->img_comp[n].w2 * j * 8 + i * 8, z->img_comp[n].w2, data);
             }
          }
       }
    }
 }
 
-static int stbi__process_marker(stbi__jpeg *z, int m)
+static int stbi__process_marker(stbi__jpeg* z, int m)
 {
    int L;
-   switch (m) {
+   switch (m)
+   {
       case STBI__MARKER_none: // no marker found
-         return stbi__err("expected marker","Corrupt JPEG");
+         return stbi__err("expected marker", "Corrupt JPEG");
 
       case 0xDD: // DRI - specify restart interval
-         if (stbi__get16be(z->s) != 4) return stbi__err("bad DRI len","Corrupt JPEG");
+         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) {
+         L = stbi__get16be(z->s) - 2;
+         while (L > 0)
+         {
             int q = stbi__get8(z->s);
-            int p = q >> 4;
-            int t = q & 15,i;
-            if (p != 0) return stbi__err("bad DQT type","Corrupt JPEG");
-            if (t > 3) return stbi__err("bad DQT table","Corrupt JPEG");
-            for (i=0; i < 64; ++i)
-               z->dequant[t][stbi__jpeg_dezigzag[i]] = stbi__get8(z->s);
-            L -= 65;
+            int p = q >> 4, sixteen = (p != 0);
+            int t = q & 15, i;
+            if (p != 0 && p != 1)
+               return stbi__err("bad DQT type", "Corrupt JPEG");
+            if (t > 3)
+               return stbi__err("bad DQT table", "Corrupt JPEG");
+
+            for (i = 0; i < 64; ++i)
+               z->dequant[t][stbi__jpeg_dezigzag[i]] = (stbi__uint16)(sixteen ? stbi__get16be(z->s) : stbi__get8(z->s));
+            L -= (sixteen ? 129 : 65);
          }
-         return L==0;
+         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;
+         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) {
+            if (tc > 1 || th > 3)
+               return stbi__err("bad DHT header", "Corrupt JPEG");
+            for (i = 0; i < 16; ++i)
+            {
                sizes[i] = stbi__get8(z->s);
                n += sizes[i];
             }
+            if (n > 256)
+               return stbi__err("bad DHT header", "Corrupt JPEG"); // Loop over i < n would write past end of values!
             L -= 17;
-            if (tc == 0) {
-               if (!stbi__build_huffman(z->huff_dc+th, sizes)) return 0;
+            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;
+            }
+            else
+            {
+               if (!stbi__build_huffman(z->huff_ac + th, sizes))
+                  return 0;
                v = z->huff_ac[th].values;
             }
-            for (i=0; i < n; ++i)
+            for (i = 0; i < n; ++i)
                v[i] = stbi__get8(z->s);
             if (tc != 0)
                stbi__build_fast_ac(z->fast_ac[th], z->huff_ac + th);
             L -= n;
          }
-         return L==0;
+         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);
+   if ((m >= 0xE0 && m <= 0xEF) || m == 0xFE)
+   {
+      L = stbi__get16be(z->s);
+      if (L < 2)
+      {
+         if (m == 0xFE)
+            return stbi__err("bad COM len", "Corrupt JPEG");
+         else
+            return stbi__err("bad APP len", "Corrupt JPEG");
+      }
+      L -= 2;
+
+      if (m == 0xE0 && L >= 5)
+      { // JFIF APP0 segment
+         static const unsigned char tag[5] = { 'J', 'F', 'I', 'F', '\0' };
+         int ok = 1;
+         int i;
+         for (i = 0; i < 5; ++i)
+            if (stbi__get8(z->s) != tag[i])
+               ok = 0;
+         L -= 5;
+         if (ok)
+            z->jfif = 1;
+      }
+      else if (m == 0xEE && L >= 12)
+      { // Adobe APP14 segment
+         static const unsigned char tag[6] = { 'A', 'd', 'o', 'b', 'e', '\0' };
+         int ok = 1;
+         int i;
+         for (i = 0; i < 6; ++i)
+            if (stbi__get8(z->s) != tag[i])
+               ok = 0;
+         L -= 6;
+         if (ok)
+         {
+            stbi__get8(z->s); // version
+            stbi__get16be(z->s); // flags0
+            stbi__get16be(z->s); // flags1
+            z->app14_color_transform = stbi__get8(z->s); // color transform
+            L -= 6;
+         }
+      }
+
+      stbi__skip(z->s, L);
       return 1;
    }
-   return 0;
+
+   return stbi__err("unknown marker", "Corrupt JPEG");
 }
 
 // after we see SOS
-static int stbi__process_scan_header(stbi__jpeg *z)
+static int stbi__process_scan_header(stbi__jpeg* z)
 {
    int i;
    int Ls = stbi__get16be(z->s);
    z->scan_n = stbi__get8(z->s);
-   if (z->scan_n < 1 || z->scan_n > 4 || z->scan_n > (int) z->s->img_n) return stbi__err("bad SOS component count","Corrupt JPEG");
-   if (Ls != 6+2*z->scan_n) return stbi__err("bad SOS len","Corrupt JPEG");
-   for (i=0; i < z->scan_n; ++i) {
+   if (z->scan_n < 1 || z->scan_n > 4 || z->scan_n > (int)z->s->img_n)
+      return stbi__err("bad SOS component count", "Corrupt JPEG");
+   if (Ls != 6 + 2 * z->scan_n)
+      return stbi__err("bad SOS len", "Corrupt JPEG");
+   for (i = 0; i < z->scan_n; ++i)
+   {
       int id = stbi__get8(z->s), which;
       int q = stbi__get8(z->s);
       for (which = 0; which < z->s->img_n; ++which)
          if (z->img_comp[which].id == id)
             break;
-      if (which == z->s->img_n) return 0; // no match
-      z->img_comp[which].hd = q >> 4;   if (z->img_comp[which].hd > 3) return stbi__err("bad DC huff","Corrupt JPEG");
-      z->img_comp[which].ha = q & 15;   if (z->img_comp[which].ha > 3) return stbi__err("bad AC huff","Corrupt JPEG");
+      if (which == z->s->img_n)
+         return 0; // no match
+      z->img_comp[which].hd = q >> 4;
+      if (z->img_comp[which].hd > 3)
+         return stbi__err("bad DC huff", "Corrupt JPEG");
+      z->img_comp[which].ha = q & 15;
+      if (z->img_comp[which].ha > 3)
+         return stbi__err("bad AC huff", "Corrupt JPEG");
       z->order[i] = which;
    }
 
    {
       int aa;
       z->spec_start = stbi__get8(z->s);
-      z->spec_end   = stbi__get8(z->s); // should be 63, but might be 0
+      z->spec_end = stbi__get8(z->s); // should be 63, but might be 0
       aa = stbi__get8(z->s);
       z->succ_high = (aa >> 4);
-      z->succ_low  = (aa & 15);
-      if (z->progressive) {
-         if (z->spec_start > 63 || z->spec_end > 63  || z->spec_start > z->spec_end || z->succ_high > 13 || z->succ_low > 13)
+      z->succ_low = (aa & 15);
+      if (z->progressive)
+      {
+         if (z->spec_start > 63 || z->spec_end > 63 || z->spec_start > z->spec_end || z->succ_high > 13 || z->succ_low > 13)
+            return stbi__err("bad SOS", "Corrupt JPEG");
+      }
+      else
+      {
+         if (z->spec_start != 0)
+            return stbi__err("bad SOS", "Corrupt JPEG");
+         if (z->succ_high != 0 || z->succ_low != 0)
             return stbi__err("bad SOS", "Corrupt JPEG");
-      } else {
-         if (z->spec_start != 0) return stbi__err("bad SOS","Corrupt JPEG");
-         if (z->succ_high != 0 || z->succ_low != 0) return stbi__err("bad SOS","Corrupt JPEG");
          z->spec_end = 63;
       }
    }
@@ -2699,42 +3641,106 @@ static int stbi__process_scan_header(stbi__jpeg *z)
    return 1;
 }
 
-static int stbi__process_frame_header(stbi__jpeg *z, int scan)
+static int stbi__free_jpeg_components(stbi__jpeg* z, int ncomp, int why)
 {
-   stbi__context *s = z->s;
-   int Lf,p,i,q, h_max=1,v_max=1,c;
-   Lf = stbi__get16be(s);         if (Lf < 11) return stbi__err("bad SOF len","Corrupt JPEG"); // JPEG
-   p  = stbi__get8(s);            if (p != 8) return stbi__err("only 8-bit","JPEG format not supported: 8-bit only"); // JPEG baseline
-   s->img_y = stbi__get16be(s);   if (s->img_y == 0) return stbi__err("no header height", "JPEG format not supported: delayed height"); // Legal, but we don't handle it--but neither does IJG
-   s->img_x = stbi__get16be(s);   if (s->img_x == 0) return stbi__err("0 width","Corrupt JPEG"); // JPEG requires
+   int i;
+   for (i = 0; i < ncomp; ++i)
+   {
+      if (z->img_comp[i].raw_data)
+      {
+         STBI_FREE(z->img_comp[i].raw_data);
+         z->img_comp[i].raw_data = NULL;
+         z->img_comp[i].data = NULL;
+      }
+      if (z->img_comp[i].raw_coeff)
+      {
+         STBI_FREE(z->img_comp[i].raw_coeff);
+         z->img_comp[i].raw_coeff = 0;
+         z->img_comp[i].coeff = 0;
+      }
+      if (z->img_comp[i].linebuf)
+      {
+         STBI_FREE(z->img_comp[i].linebuf);
+         z->img_comp[i].linebuf = NULL;
+      }
+   }
+   return why;
+}
+
+static int stbi__process_frame_header(stbi__jpeg* z, int scan)
+{
+   stbi__context* s = z->s;
+   int Lf, p, i, q, h_max = 1, v_max = 1, c;
+   Lf = stbi__get16be(s);
+   if (Lf < 11)
+      return stbi__err("bad SOF len", "Corrupt JPEG"); // JPEG
+   p = stbi__get8(s);
+   if (p != 8)
+      return stbi__err("only 8-bit", "JPEG format not supported: 8-bit only"); // JPEG baseline
+   s->img_y = stbi__get16be(s);
+   if (s->img_y == 0)
+      return stbi__err("no header height", "JPEG format not supported: delayed height"); // Legal, but we don't handle it--but neither does IJG
+   s->img_x = stbi__get16be(s);
+   if (s->img_x == 0)
+      return stbi__err("0 width", "Corrupt JPEG"); // JPEG requires
+   if (s->img_y > STBI_MAX_DIMENSIONS)
+      return stbi__err("too large", "Very large image (corrupt?)");
+   if (s->img_x > STBI_MAX_DIMENSIONS)
+      return stbi__err("too large", "Very large image (corrupt?)");
    c = stbi__get8(s);
-   if (c != 3 && c != 1) return stbi__err("bad component count","Corrupt JPEG");    // JFIF requires
+   if (c != 3 && c != 1 && c != 4)
+      return stbi__err("bad component count", "Corrupt JPEG");
    s->img_n = c;
-   for (i=0; i < c; ++i) {
+   for (i = 0; i < c; ++i)
+   {
       z->img_comp[i].data = NULL;
       z->img_comp[i].linebuf = NULL;
    }
 
-   if (Lf != 8+3*s->img_n) return stbi__err("bad SOF len","Corrupt JPEG");
+   if (Lf != 8 + 3 * s->img_n)
+      return stbi__err("bad SOF len", "Corrupt JPEG");
 
-   for (i=0; i < s->img_n; ++i) {
+   z->rgb = 0;
+   for (i = 0; i < s->img_n; ++i)
+   {
+      static const unsigned char rgb[3] = { 'R', 'G', 'B' };
       z->img_comp[i].id = stbi__get8(s);
-      if (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");
+      if (s->img_n == 3 && z->img_comp[i].id == rgb[i])
+         ++z->rgb;
       q = stbi__get8(s);
-      z->img_comp[i].h = (q >> 4);  if (!z->img_comp[i].h || z->img_comp[i].h > 4) return stbi__err("bad H","Corrupt JPEG");
-      z->img_comp[i].v = q & 15;    if (!z->img_comp[i].v || z->img_comp[i].v > 4) return stbi__err("bad V","Corrupt JPEG");
-      z->img_comp[i].tq = stbi__get8(s);  if (z->img_comp[i].tq > 3) return stbi__err("bad TQ","Corrupt JPEG");
-   }
+      z->img_comp[i].h = (q >> 4);
+      if (!z->img_comp[i].h || z->img_comp[i].h > 4)
+         return stbi__err("bad H", "Corrupt JPEG");
+      z->img_comp[i].v = q & 15;
+      if (!z->img_comp[i].v || z->img_comp[i].v > 4)
+         return stbi__err("bad V", "Corrupt JPEG");
+      z->img_comp[i].tq = stbi__get8(s);
+      if (z->img_comp[i].tq > 3)
+         return stbi__err("bad TQ", "Corrupt JPEG");
+   }
+
+   if (scan != STBI__SCAN_load)
+      return 1;
 
-   if (scan != STBI__SCAN_load) return 1;
+   if (!stbi__mad3sizes_valid(s->img_x, s->img_y, s->img_n, 0))
+      return stbi__err("too large", "Image too large to decode");
 
-   if ((1 << 30) / s->img_x / s->img_n < s->img_y) return stbi__err("too large", "Image too large to decode");
+   for (i = 0; i < s->img_n; ++i)
+   {
+      if (z->img_comp[i].h > h_max)
+         h_max = z->img_comp[i].h;
+      if (z->img_comp[i].v > v_max)
+         v_max = z->img_comp[i].v;
+   }
 
-   for (i=0; i < s->img_n; ++i) {
-      if (z->img_comp[i].h > h_max) h_max = z->img_comp[i].h;
-      if (z->img_comp[i].v > v_max) v_max = z->img_comp[i].v;
+   // check that plane subsampling factors are integer ratios; our resamplers can't deal with fractional ratios
+   // and I've never seen a non-corrupted JPEG file actually use them
+   for (i = 0; i < s->img_n; ++i)
+   {
+      if (h_max % z->img_comp[i].h != 0)
+         return stbi__err("bad H", "Corrupt JPEG");
+      if (v_max % z->img_comp[i].v != 0)
+         return stbi__err("bad V", "Corrupt JPEG");
    }
 
    // compute interleaved mcu info
@@ -2742,39 +3748,41 @@ static int stbi__process_frame_header(stbi__jpeg *z, int scan)
    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;
+   // these sizes can't be more than 17 bits
+   z->img_mcu_x = (s->img_x + z->img_mcu_w - 1) / z->img_mcu_w;
+   z->img_mcu_y = (s->img_y + z->img_mcu_h - 1) / z->img_mcu_h;
 
-   for (i=0; i < s->img_n; ++i) {
+   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;
+      z->img_comp[i].x = (s->img_x * z->img_comp[i].h + h_max - 1) / h_max;
+      z->img_comp[i].y = (s->img_y * z->img_comp[i].v + v_max - 1) / v_max;
       // to simplify generation, we'll allocate enough memory to decode
       // the bogus oversized data from using interleaved MCUs and their
       // big blocks (e.g. a 16x16 iMCU on an image of width 33); we won't
       // discard the extra data until colorspace conversion
+      //
+      // img_mcu_x, img_mcu_y: <=17 bits; comp[i].h and .v are <=4 (checked earlier)
+      // so these muls can't overflow with 32-bit ints (which we require)
       z->img_comp[i].w2 = z->img_mcu_x * z->img_comp[i].h * 8;
       z->img_comp[i].h2 = z->img_mcu_y * z->img_comp[i].v * 8;
-      z->img_comp[i].raw_data = stbi__malloc(z->img_comp[i].w2 * z->img_comp[i].h2+15);
-
-      if (z->img_comp[i].raw_data == NULL) {
-         for(--i; i >= 0; --i) {
-            STBI_FREE(z->img_comp[i].raw_data);
-            z->img_comp[i].raw_data = NULL;
-         }
-         return stbi__err("outofmem", "Out of memory");
-      }
-      // align blocks for idct using mmx/sse
-      z->img_comp[i].data = (stbi_uc*) (((size_t) z->img_comp[i].raw_data + 15) & ~15);
+      z->img_comp[i].coeff = 0;
+      z->img_comp[i].raw_coeff = 0;
       z->img_comp[i].linebuf = NULL;
-      if (z->progressive) {
-         z->img_comp[i].coeff_w = (z->img_comp[i].w2 + 7) >> 3;
-         z->img_comp[i].coeff_h = (z->img_comp[i].h2 + 7) >> 3;
-         z->img_comp[i].raw_coeff = STBI_MALLOC(z->img_comp[i].coeff_w * z->img_comp[i].coeff_h * 64 * sizeof(short) + 15);
-         z->img_comp[i].coeff = (short*) (((size_t) z->img_comp[i].raw_coeff + 15) & ~15);
-      } else {
-         z->img_comp[i].coeff = 0;
-         z->img_comp[i].raw_coeff = 0;
+      z->img_comp[i].raw_data = stbi__malloc_mad2(z->img_comp[i].w2, z->img_comp[i].h2, 15);
+      if (z->img_comp[i].raw_data == NULL)
+         return stbi__free_jpeg_components(z, i + 1, stbi__err("outofmem", "Out of memory"));
+      // align blocks for idct using mmx/sse
+      z->img_comp[i].data = (stbi_uc*)(((size_t)z->img_comp[i].raw_data + 15) & ~15);
+      if (z->progressive)
+      {
+         // w2, h2 are multiples of 8 (see above)
+         z->img_comp[i].coeff_w = z->img_comp[i].w2 / 8;
+         z->img_comp[i].coeff_h = z->img_comp[i].h2 / 8;
+         z->img_comp[i].raw_coeff = stbi__malloc_mad3(z->img_comp[i].w2, z->img_comp[i].h2, sizeof(short), 15);
+         if (z->img_comp[i].raw_coeff == NULL)
+            return stbi__free_jpeg_components(z, i + 1, stbi__err("outofmem", "Out of memory"));
+         z->img_comp[i].coeff = (short*)(((size_t)z->img_comp[i].raw_coeff + 15) & ~15);
       }
    }
 
@@ -2782,68 +3790,117 @@ static int stbi__process_frame_header(stbi__jpeg *z, int scan)
 }
 
 // use comparisons since in some cases we handle more than one case (e.g. SOF)
-#define stbi__DNL(x)         ((x) == 0xdc)
-#define stbi__SOI(x)         ((x) == 0xd8)
-#define stbi__EOI(x)         ((x) == 0xd9)
-#define stbi__SOF(x)         ((x) == 0xc0 || (x) == 0xc1 || (x) == 0xc2)
-#define stbi__SOS(x)         ((x) == 0xda)
+#define stbi__DNL(x) ((x) == 0xdc)
+#define stbi__SOI(x) ((x) == 0xd8)
+#define stbi__EOI(x) ((x) == 0xd9)
+#define stbi__SOF(x) ((x) == 0xc0 || (x) == 0xc1 || (x) == 0xc2)
+#define stbi__SOS(x) ((x) == 0xda)
 
-#define stbi__SOF_progressive(x)   ((x) == 0xc2)
+#define stbi__SOF_progressive(x) ((x) == 0xc2)
 
-static int stbi__decode_jpeg_header(stbi__jpeg *z, int scan)
+static int stbi__decode_jpeg_header(stbi__jpeg* z, int scan)
 {
    int m;
+   z->jfif = 0;
+   z->app14_color_transform = -1; // valid values are 0,1,2
    z->marker = STBI__MARKER_none; // initialize cached marker to empty
    m = stbi__get_marker(z);
-   if (!stbi__SOI(m)) return stbi__err("no SOI","Corrupt JPEG");
-   if (scan == STBI__SCAN_type) return 1;
+   if (!stbi__SOI(m))
+      return stbi__err("no SOI", "Corrupt JPEG");
+   if (scan == STBI__SCAN_type)
+      return 1;
    m = stbi__get_marker(z);
-   while (!stbi__SOF(m)) {
-      if (!stbi__process_marker(z,m)) return 0;
+   while (!stbi__SOF(m))
+   {
+      if (!stbi__process_marker(z, m))
+         return 0;
       m = stbi__get_marker(z);
-      while (m == STBI__MARKER_none) {
+      while (m == STBI__MARKER_none)
+      {
          // some files have extra padding after their blocks, so ok, we'll scan
-         if (stbi__at_eof(z->s)) return stbi__err("no SOF", "Corrupt JPEG");
+         if (stbi__at_eof(z->s))
+            return stbi__err("no SOF", "Corrupt JPEG");
          m = stbi__get_marker(z);
       }
    }
    z->progressive = stbi__SOF_progressive(m);
-   if (!stbi__process_frame_header(z, scan)) return 0;
+   if (!stbi__process_frame_header(z, scan))
+      return 0;
    return 1;
 }
 
+static stbi_uc stbi__skip_jpeg_junk_at_end(stbi__jpeg* j)
+{
+   // some JPEGs have junk at end, skip over it but if we find what looks
+   // like a valid marker, resume there
+   while (!stbi__at_eof(j->s))
+   {
+      stbi_uc x = stbi__get8(j->s);
+      while (x == 0xff)
+      { // might be a marker
+         if (stbi__at_eof(j->s))
+            return STBI__MARKER_none;
+         x = stbi__get8(j->s);
+         if (x != 0x00 && x != 0xff)
+         {
+            // not a stuffed zero or lead-in to another marker, looks
+            // like an actual marker, return it
+            return x;
+         }
+         // stuffed zero has x=0 now which ends the loop, meaning we go
+         // back to regular scan loop.
+         // repeated 0xff keeps trying to read the next byte of the marker.
+      }
+   }
+   return STBI__MARKER_none;
+}
+
 // decode image to YCbCr format
-static int stbi__decode_jpeg_image(stbi__jpeg *j)
+static int stbi__decode_jpeg_image(stbi__jpeg* j)
 {
    int m;
-   for (m = 0; m < 4; m++) {
+   for (m = 0; m < 4; m++)
+   {
       j->img_comp[m].raw_data = NULL;
       j->img_comp[m].raw_coeff = NULL;
    }
    j->restart_interval = 0;
-   if (!stbi__decode_jpeg_header(j, STBI__SCAN_load)) return 0;
+   if (!stbi__decode_jpeg_header(j, STBI__SCAN_load))
+      return 0;
    m = stbi__get_marker(j);
-   while (!stbi__EOI(m)) {
-      if (stbi__SOS(m)) {
-         if (!stbi__process_scan_header(j)) return 0;
-         if (!stbi__parse_entropy_coded_data(j)) return 0;
-         if (j->marker == STBI__MARKER_none ) {
-            // handle 0s at the end of image data from IP Kamera 9060
-            while (!stbi__at_eof(j->s)) {
-               int x = stbi__get8(j->s);
-               if (x == 255) {
-                  j->marker = stbi__get8(j->s);
-                  break;
-               } else if (x != 0) {
-                  return stbi__err("junk before marker", "Corrupt JPEG");
-               }
-            }
+   while (!stbi__EOI(m))
+   {
+      if (stbi__SOS(m))
+      {
+         if (!stbi__process_scan_header(j))
+            return 0;
+         if (!stbi__parse_entropy_coded_data(j))
+            return 0;
+         if (j->marker == STBI__MARKER_none)
+         {
+            j->marker = stbi__skip_jpeg_junk_at_end(j);
             // if we reach eof without hitting a marker, stbi__get_marker() below will fail and we'll eventually return 0
          }
-      } else {
-         if (!stbi__process_marker(j, m)) return 0;
+         m = stbi__get_marker(j);
+         if (STBI__RESTART(m))
+            m = stbi__get_marker(j);
+      }
+      else if (stbi__DNL(m))
+      {
+         int Ld = stbi__get16be(j->s);
+         stbi__uint32 NL = stbi__get16be(j->s);
+         if (Ld != 4)
+            return stbi__err("bad DNL len", "Corrupt JPEG");
+         if (NL != j->s->img_y)
+            return stbi__err("bad DNL height", "Corrupt JPEG");
+         m = stbi__get_marker(j);
+      }
+      else
+      {
+         if (!stbi__process_marker(j, m))
+            return 1;
+         m = stbi__get_marker(j);
       }
-      m = stbi__get_marker(j);
    }
    if (j->progressive)
       stbi__jpeg_finish(j);
@@ -2852,12 +3909,12 @@ static int stbi__decode_jpeg_image(stbi__jpeg *j)
 
 // 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);
+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))
+#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)
+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);
@@ -2866,37 +3923,39 @@ static stbi_uc *resample_row_1(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far,
    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)
+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);
+   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)
+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;
+   stbi_uc* input = in_near;
 
-   if (w == 1) {
+   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[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];
+   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);
@@ -2904,26 +3963,28 @@ static stbi_uc*  stbi__resample_row_h_2(stbi_uc *out, stbi_uc *in_near, stbi_uc
    return out;
 }
 
-#define stbi__div16(x) ((stbi_uc) ((x) >> 4))
+#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)
+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);
+   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) {
+   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);
+      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);
+   out[w * 2 - 1] = stbi__div4(t1 + 2);
 
    STBI_NOTUSED(hs);
 
@@ -2931,32 +3992,34 @@ static stbi_uc *stbi__resample_row_hv_2(stbi_uc *out, stbi_uc *in_near, stbi_uc
 }
 
 #if defined(STBI_SSE2) || defined(STBI_NEON)
-static stbi_uc *stbi__resample_row_hv_2_simd(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
+static stbi_uc* stbi__resample_row_hv_2_simd(stbi_uc* out, stbi_uc* in_near, stbi_uc* in_far, int w, int hs)
 {
    // need to generate 2x2 samples for every one in input
-   int i=0,t0,t1;
+   int i = 0, t0, t1;
 
-   if (w == 1) {
-      out[0] = out[1] = stbi__div4(3*in_near[0] + in_far[0] + 2);
+   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];
+   t1 = 3 * in_near[0] + in_far[0];
    // process groups of 8 pixels for as long as we can.
    // note we can't handle the last pixel in a row in this loop
    // because we need to handle the filter boundary conditions.
-   for (; i < ((w-1) & ~7); i += 8) {
+   for (; i < ((w - 1) & ~7); i += 8)
+   {
 #if defined(STBI_SSE2)
       // load and perform the vertical filtering pass
       // this uses 3*x + y = 4*x + (y - x)
-      __m128i zero  = _mm_setzero_si128();
-      __m128i farb  = _mm_loadl_epi64((__m128i *) (in_far + i));
-      __m128i nearb = _mm_loadl_epi64((__m128i *) (in_near + i));
-      __m128i farw  = _mm_unpacklo_epi8(farb, zero);
+      __m128i zero = _mm_setzero_si128();
+      __m128i farb = _mm_loadl_epi64((__m128i*)(in_far + i));
+      __m128i nearb = _mm_loadl_epi64((__m128i*)(in_near + i));
+      __m128i farw = _mm_unpacklo_epi8(farb, zero);
       __m128i nearw = _mm_unpacklo_epi8(nearb, zero);
-      __m128i diff  = _mm_sub_epi16(farw, nearw);
+      __m128i diff = _mm_sub_epi16(farw, nearw);
       __m128i nears = _mm_slli_epi16(nearw, 2);
-      __m128i curr  = _mm_add_epi16(nears, diff); // current row
+      __m128i curr = _mm_add_epi16(nears, diff); // current row
 
       // horizontal filter works the same based on shifted vers of current
       // row. "prev" is current row shifted right by 1 pixel; we need to
@@ -2966,37 +4029,37 @@ static stbi_uc *stbi__resample_row_hv_2_simd(stbi_uc *out, stbi_uc *in_near, stb
       __m128i prv0 = _mm_slli_si128(curr, 2);
       __m128i nxt0 = _mm_srli_si128(curr, 2);
       __m128i prev = _mm_insert_epi16(prv0, t1, 0);
-      __m128i next = _mm_insert_epi16(nxt0, 3*in_near[i+8] + in_far[i+8], 7);
+      __m128i next = _mm_insert_epi16(nxt0, 3 * in_near[i + 8] + in_far[i + 8], 7);
 
       // horizontal filter, polyphase implementation since it's convenient:
       // even pixels = 3*cur + prev = cur*4 + (prev - cur)
       // odd  pixels = 3*cur + next = cur*4 + (next - cur)
       // note the shared term.
-      __m128i bias  = _mm_set1_epi16(8);
+      __m128i bias = _mm_set1_epi16(8);
       __m128i curs = _mm_slli_epi16(curr, 2);
       __m128i prvd = _mm_sub_epi16(prev, curr);
       __m128i nxtd = _mm_sub_epi16(next, curr);
       __m128i curb = _mm_add_epi16(curs, bias);
       __m128i even = _mm_add_epi16(prvd, curb);
-      __m128i odd  = _mm_add_epi16(nxtd, curb);
+      __m128i odd = _mm_add_epi16(nxtd, curb);
 
       // interleave even and odd pixels, then undo scaling.
       __m128i int0 = _mm_unpacklo_epi16(even, odd);
       __m128i int1 = _mm_unpackhi_epi16(even, odd);
-      __m128i de0  = _mm_srli_epi16(int0, 4);
-      __m128i de1  = _mm_srli_epi16(int1, 4);
+      __m128i de0 = _mm_srli_epi16(int0, 4);
+      __m128i de1 = _mm_srli_epi16(int1, 4);
 
       // pack and write output
       __m128i outv = _mm_packus_epi16(de0, de1);
-      _mm_storeu_si128((__m128i *) (out + i*2), outv);
+      _mm_storeu_si128((__m128i*)(out + i * 2), outv);
 #elif defined(STBI_NEON)
       // load and perform the vertical filtering pass
       // this uses 3*x + y = 4*x + (y - x)
-      uint8x8_t farb  = vld1_u8(in_far + i);
+      uint8x8_t farb = vld1_u8(in_far + i);
       uint8x8_t nearb = vld1_u8(in_near + i);
-      int16x8_t diff  = vreinterpretq_s16_u16(vsubl_u8(farb, nearb));
+      int16x8_t diff = vreinterpretq_s16_u16(vsubl_u8(farb, nearb));
       int16x8_t nears = vreinterpretq_s16_u16(vshll_n_u8(nearb, 2));
-      int16x8_t curr  = vaddq_s16(nears, diff); // current row
+      int16x8_t curr = vaddq_s16(nears, diff); // current row
 
       // horizontal filter works the same based on shifted vers of current
       // row. "prev" is current row shifted right by 1 pixel; we need to
@@ -3006,7 +4069,7 @@ static stbi_uc *stbi__resample_row_hv_2_simd(stbi_uc *out, stbi_uc *in_near, stb
       int16x8_t prv0 = vextq_s16(curr, curr, 7);
       int16x8_t nxt0 = vextq_s16(curr, curr, 1);
       int16x8_t prev = vsetq_lane_s16(t1, prv0, 0);
-      int16x8_t next = vsetq_lane_s16(3*in_near[i+8] + in_far[i+8], nxt0, 7);
+      int16x8_t next = vsetq_lane_s16(3 * in_near[i + 8] + in_far[i + 8], nxt0, 7);
 
       // horizontal filter, polyphase implementation since it's convenient:
       // even pixels = 3*cur + prev = cur*4 + (prev - cur)
@@ -3016,30 +4079,31 @@ static stbi_uc *stbi__resample_row_hv_2_simd(stbi_uc *out, stbi_uc *in_near, stb
       int16x8_t prvd = vsubq_s16(prev, curr);
       int16x8_t nxtd = vsubq_s16(next, curr);
       int16x8_t even = vaddq_s16(curs, prvd);
-      int16x8_t odd  = vaddq_s16(curs, nxtd);
+      int16x8_t odd = vaddq_s16(curs, nxtd);
 
       // undo scaling and round, then store with even/odd phases interleaved
       uint8x8x2_t o;
       o.val[0] = vqrshrun_n_s16(even, 4);
-      o.val[1] = vqrshrun_n_s16(odd,  4);
-      vst2_u8(out + i*2, o);
+      o.val[1] = vqrshrun_n_s16(odd, 4);
+      vst2_u8(out + i * 2, o);
 #endif
 
       // "previous" value for next iter
-      t1 = 3*in_near[i+7] + in_far[i+7];
+      t1 = 3 * in_near[i + 7] + in_far[i + 7];
    }
 
    t0 = t1;
-   t1 = 3*in_near[i] + in_far[i];
-   out[i*2] = stbi__div16(3*t1 + t0 + 8);
+   t1 = 3 * in_near[i] + in_far[i];
+   out[i * 2] = stbi__div16(3 * t1 + t0 + 8);
 
-   for (++i; i < w; ++i) {
+   for (++i; i < w; ++i)
+   {
       t0 = t1;
-      t1 = 3*in_near[i]+in_far[i];
-      out[i*2-1] = stbi__div16(3*t0 + t1 + 8);
-      out[i*2  ] = stbi__div16(3*t1 + t0 + 8);
+      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);
+   out[w * 2 - 1] = stbi__div4(t1 + 2);
 
    STBI_NOTUSED(hs);
 
@@ -3047,66 +4111,56 @@ static stbi_uc *stbi__resample_row_hv_2_simd(stbi_uc *out, stbi_uc *in_near, stb
 }
 #endif
 
-static stbi_uc *stbi__resample_row_generic(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs)
+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;
+   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];
+   for (i = 0; i < w; ++i)
+      for (j = 0; j < hs; ++j)
+         out[i * hs + j] = in_near[i];
    return out;
 }
 
-#ifdef STBI_JPEG_OLD
-// this is the same YCbCr-to-RGB calculation that stb_image has used
-// historically before the algorithm changes in 1.49
-#define float2fixed(x)  ((int) ((x) * 65536 + 0.5))
-static void stbi__YCbCr_to_RGB_row(stbi_uc *out, const stbi_uc *y, const stbi_uc *pcb, const stbi_uc *pcr, int count, int step)
-{
-   int i;
-   for (i=0; i < count; ++i) {
-      int y_fixed = (y[i] << 16) + 32768; // rounding
-      int r,g,b;
-      int cr = pcr[i] - 128;
-      int cb = pcb[i] - 128;
-      r = y_fixed + cr*float2fixed(1.40200f);
-      g = y_fixed - cr*float2fixed(0.71414f) - cb*float2fixed(0.34414f);
-      b = y_fixed                            + cb*float2fixed(1.77200f);
-      r >>= 16;
-      g >>= 16;
-      b >>= 16;
-      if ((unsigned) r > 255) { if (r < 0) r = 0; else r = 255; }
-      if ((unsigned) g > 255) { if (g < 0) g = 0; else g = 255; }
-      if ((unsigned) b > 255) { if (b < 0) b = 0; else b = 255; }
-      out[0] = (stbi_uc)r;
-      out[1] = (stbi_uc)g;
-      out[2] = (stbi_uc)b;
-      out[3] = 255;
-      out += step;
-   }
-}
-#else
 // this is a reduced-precision calculation of YCbCr-to-RGB introduced
 // to make sure the code produces the same results in both SIMD and scalar
-#define float2fixed(x)  (((int) ((x) * 4096.0f + 0.5f)) << 8)
-static void stbi__YCbCr_to_RGB_row(stbi_uc *out, const stbi_uc *y, const stbi_uc *pcb, const stbi_uc *pcr, int count, int step)
+#define stbi__float2fixed(x) (((int)((x)*4096.0f + 0.5f)) << 8)
+static void stbi__YCbCr_to_RGB_row(stbi_uc* out, const stbi_uc* y, const stbi_uc* pcb, const stbi_uc* pcr, int count, int step)
 {
    int i;
-   for (i=0; i < count; ++i) {
-      int y_fixed = (y[i] << 20) + (1<<19); // rounding
-      int r,g,b;
+   for (i = 0; i < count; ++i)
+   {
+      int y_fixed = (y[i] << 20) + (1 << 19); // rounding
+      int r, g, b;
       int cr = pcr[i] - 128;
       int cb = pcb[i] - 128;
-      r = y_fixed +  cr* float2fixed(1.40200f);
-      g = y_fixed + (cr*-float2fixed(0.71414f)) + ((cb*-float2fixed(0.34414f)) & 0xffff0000);
-      b = y_fixed                               +   cb* float2fixed(1.77200f);
+      r = y_fixed + cr * stbi__float2fixed(1.40200f);
+      g = y_fixed + (cr * -stbi__float2fixed(0.71414f)) + ((cb * -stbi__float2fixed(0.34414f)) & 0xffff0000);
+      b = y_fixed + cb * stbi__float2fixed(1.77200f);
       r >>= 20;
       g >>= 20;
       b >>= 20;
-      if ((unsigned) r > 255) { if (r < 0) r = 0; else r = 255; }
-      if ((unsigned) g > 255) { if (g < 0) g = 0; else g = 255; }
-      if ((unsigned) b > 255) { if (b < 0) b = 0; else b = 255; }
+      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;
@@ -3114,10 +4168,9 @@ static void stbi__YCbCr_to_RGB_row(stbi_uc *out, const stbi_uc *y, const stbi_uc
       out += step;
    }
 }
-#endif
 
 #if defined(STBI_SSE2) || defined(STBI_NEON)
-static void stbi__YCbCr_to_RGB_simd(stbi_uc *out, stbi_uc const *y, stbi_uc const *pcb, stbi_uc const *pcr, int count, int step)
+static void stbi__YCbCr_to_RGB_simd(stbi_uc* out, stbi_uc const* y, stbi_uc const* pcb, stbi_uc const* pcr, int count, int step)
 {
    int i = 0;
 
@@ -3125,26 +4178,28 @@ static void stbi__YCbCr_to_RGB_simd(stbi_uc *out, stbi_uc const *y, stbi_uc cons
    // step == 3 is pretty ugly on the final interleave, and i'm not convinced
    // it's useful in practice (you wouldn't use it for textures, for example).
    // so just accelerate step == 4 case.
-   if (step == 4) {
+   if (step == 4)
+   {
       // this is a fairly straightforward implementation and not super-optimized.
-      __m128i signflip  = _mm_set1_epi8(-0x80);
-      __m128i cr_const0 = _mm_set1_epi16(   (short) ( 1.40200f*4096.0f+0.5f));
-      __m128i cr_const1 = _mm_set1_epi16( - (short) ( 0.71414f*4096.0f+0.5f));
-      __m128i cb_const0 = _mm_set1_epi16( - (short) ( 0.34414f*4096.0f+0.5f));
-      __m128i cb_const1 = _mm_set1_epi16(   (short) ( 1.77200f*4096.0f+0.5f));
-      __m128i y_bias = _mm_set1_epi8((char) (unsigned char) 128);
+      __m128i signflip = _mm_set1_epi8(-0x80);
+      __m128i cr_const0 = _mm_set1_epi16((short)(1.40200f * 4096.0f + 0.5f));
+      __m128i cr_const1 = _mm_set1_epi16(-(short)(0.71414f * 4096.0f + 0.5f));
+      __m128i cb_const0 = _mm_set1_epi16(-(short)(0.34414f * 4096.0f + 0.5f));
+      __m128i cb_const1 = _mm_set1_epi16((short)(1.77200f * 4096.0f + 0.5f));
+      __m128i y_bias = _mm_set1_epi8((char)(unsigned char)128);
       __m128i xw = _mm_set1_epi16(255); // alpha channel
 
-      for (; i+7 < count; i += 8) {
+      for (; i + 7 < count; i += 8)
+      {
          // load
-         __m128i y_bytes = _mm_loadl_epi64((__m128i *) (y+i));
-         __m128i cr_bytes = _mm_loadl_epi64((__m128i *) (pcr+i));
-         __m128i cb_bytes = _mm_loadl_epi64((__m128i *) (pcb+i));
+         __m128i y_bytes = _mm_loadl_epi64((__m128i*)(y + i));
+         __m128i cr_bytes = _mm_loadl_epi64((__m128i*)(pcr + i));
+         __m128i cb_bytes = _mm_loadl_epi64((__m128i*)(pcb + i));
          __m128i cr_biased = _mm_xor_si128(cr_bytes, signflip); // -128
          __m128i cb_biased = _mm_xor_si128(cb_bytes, signflip); // -128
 
          // unpack to short (and left-shift cr, cb by 8)
-         __m128i yw  = _mm_unpacklo_epi8(y_bias, y_bytes);
+         __m128i yw = _mm_unpacklo_epi8(y_bias, y_bytes);
          __m128i crw = _mm_unpacklo_epi8(_mm_setzero_si128(), cr_biased);
          __m128i cbw = _mm_unpacklo_epi8(_mm_setzero_si128(), cb_biased);
 
@@ -3175,8 +4230,8 @@ static void stbi__YCbCr_to_RGB_simd(stbi_uc *out, stbi_uc const *y, stbi_uc cons
          __m128i o1 = _mm_unpackhi_epi16(t0, t1);
 
          // store
-         _mm_storeu_si128((__m128i *) (out + 0), o0);
-         _mm_storeu_si128((__m128i *) (out + 16), o1);
+         _mm_storeu_si128((__m128i*)(out + 0), o0);
+         _mm_storeu_si128((__m128i*)(out + 16), o1);
          out += 32;
       }
    }
@@ -3184,17 +4239,19 @@ static void stbi__YCbCr_to_RGB_simd(stbi_uc *out, stbi_uc const *y, stbi_uc cons
 
 #ifdef STBI_NEON
    // in this version, step=3 support would be easy to add. but is there demand?
-   if (step == 4) {
+   if (step == 4)
+   {
       // this is a fairly straightforward implementation and not super-optimized.
       uint8x8_t signflip = vdup_n_u8(0x80);
-      int16x8_t cr_const0 = vdupq_n_s16(   (short) ( 1.40200f*4096.0f+0.5f));
-      int16x8_t cr_const1 = vdupq_n_s16( - (short) ( 0.71414f*4096.0f+0.5f));
-      int16x8_t cb_const0 = vdupq_n_s16( - (short) ( 0.34414f*4096.0f+0.5f));
-      int16x8_t cb_const1 = vdupq_n_s16(   (short) ( 1.77200f*4096.0f+0.5f));
+      int16x8_t cr_const0 = vdupq_n_s16((short)(1.40200f * 4096.0f + 0.5f));
+      int16x8_t cr_const1 = vdupq_n_s16(-(short)(0.71414f * 4096.0f + 0.5f));
+      int16x8_t cb_const0 = vdupq_n_s16(-(short)(0.34414f * 4096.0f + 0.5f));
+      int16x8_t cb_const1 = vdupq_n_s16((short)(1.77200f * 4096.0f + 0.5f));
 
-      for (; i+7 < count; i += 8) {
+      for (; i + 7 < count; i += 8)
+      {
          // load
-         uint8x8_t y_bytes  = vld1_u8(y + i);
+         uint8x8_t y_bytes = vld1_u8(y + i);
          uint8x8_t cr_bytes = vld1_u8(pcr + i);
          uint8x8_t cb_bytes = vld1_u8(pcb + i);
          int8x8_t cr_biased = vreinterpret_s8_u8(vsub_u8(cr_bytes, signflip));
@@ -3223,25 +4280,44 @@ static void stbi__YCbCr_to_RGB_simd(stbi_uc *out, stbi_uc const *y, stbi_uc cons
 
          // store, interleaving r/g/b/a
          vst4_u8(out, o);
-         out += 8*4;
+         out += 8 * 4;
       }
    }
 #endif
 
-   for (; i < count; ++i) {
-      int y_fixed = (y[i] << 20) + (1<<19); // rounding
-      int r,g,b;
+   for (; i < count; ++i)
+   {
+      int y_fixed = (y[i] << 20) + (1 << 19); // rounding
+      int r, g, b;
       int cr = pcr[i] - 128;
       int cb = pcb[i] - 128;
-      r = y_fixed + cr* float2fixed(1.40200f);
-      g = y_fixed + cr*-float2fixed(0.71414f) + ((cb*-float2fixed(0.34414f)) & 0xffff0000);
-      b = y_fixed                             +   cb* float2fixed(1.77200f);
+      r = y_fixed + cr * stbi__float2fixed(1.40200f);
+      g = y_fixed + cr * -stbi__float2fixed(0.71414f) + ((cb * -stbi__float2fixed(0.34414f)) & 0xffff0000);
+      b = y_fixed + cb * stbi__float2fixed(1.77200f);
       r >>= 20;
       g >>= 20;
       b >>= 20;
-      if ((unsigned) r > 255) { if (r < 0) r = 0; else r = 255; }
-      if ((unsigned) g > 255) { if (g < 0) g = 0; else g = 255; }
-      if ((unsigned) b > 255) { if (b < 0) b = 0; else b = 255; }
+      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;
@@ -3252,196 +4328,334 @@ static void stbi__YCbCr_to_RGB_simd(stbi_uc *out, stbi_uc const *y, stbi_uc cons
 #endif
 
 // set up the kernels
-static void stbi__setup_jpeg(stbi__jpeg *j)
+static void stbi__setup_jpeg(stbi__jpeg* j)
 {
    j->idct_block_kernel = stbi__idct_block;
    j->YCbCr_to_RGB_kernel = stbi__YCbCr_to_RGB_row;
    j->resample_row_hv_2_kernel = stbi__resample_row_hv_2;
 
 #ifdef STBI_SSE2
-   if (stbi__sse2_available()) {
+   if (stbi__sse2_available())
+   {
       j->idct_block_kernel = stbi__idct_simd;
-      #ifndef STBI_JPEG_OLD
       j->YCbCr_to_RGB_kernel = stbi__YCbCr_to_RGB_simd;
-      #endif
       j->resample_row_hv_2_kernel = stbi__resample_row_hv_2_simd;
    }
 #endif
 
 #ifdef STBI_NEON
    j->idct_block_kernel = stbi__idct_simd;
-   #ifndef STBI_JPEG_OLD
    j->YCbCr_to_RGB_kernel = stbi__YCbCr_to_RGB_simd;
-   #endif
    j->resample_row_hv_2_kernel = stbi__resample_row_hv_2_simd;
 #endif
 }
 
 // clean up the temporary component buffers
-static void stbi__cleanup_jpeg(stbi__jpeg *j)
+static void stbi__cleanup_jpeg(stbi__jpeg* j)
 {
-   int i;
-   for (i=0; i < j->s->img_n; ++i) {
-      if (j->img_comp[i].raw_data) {
-         STBI_FREE(j->img_comp[i].raw_data);
-         j->img_comp[i].raw_data = NULL;
-         j->img_comp[i].data = NULL;
-      }
-      if (j->img_comp[i].raw_coeff) {
-         STBI_FREE(j->img_comp[i].raw_coeff);
-         j->img_comp[i].raw_coeff = 0;
-         j->img_comp[i].coeff = 0;
-      }
-      if (j->img_comp[i].linebuf) {
-         STBI_FREE(j->img_comp[i].linebuf);
-         j->img_comp[i].linebuf = NULL;
-      }
-   }
+   stbi__free_jpeg_components(j, j->s->img_n, 0);
 }
 
 typedef struct
 {
    resample_row_func resample;
-   stbi_uc *line0,*line1;
-   int hs,vs;   // expansion factor in each axis
+   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
+   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)
+// fast 0..255 * 0..255 => 0..255 rounded multiplication
+static stbi_uc stbi__blinn_8x8(stbi_uc x, stbi_uc y)
+{
+   unsigned int t = x * y + 128;
+   return (stbi_uc)((t + (t >> 8)) >> 8);
+}
+
+static stbi_uc* load_jpeg_image(stbi__jpeg* z, int* out_x, int* out_y, int* comp, int req_comp)
 {
-   int n, decode_n;
+   int n, decode_n, is_rgb;
    z->s->img_n = 0; // make stbi__cleanup_jpeg safe
 
    // validate req_comp
-   if (req_comp < 0 || req_comp > 4) return stbi__errpuc("bad req_comp", "Internal error");
+   if (req_comp < 0 || req_comp > 4)
+      return stbi__errpuc("bad req_comp", "Internal error");
 
    // load a jpeg image from whichever source, but leave in YCbCr format
-   if (!stbi__decode_jpeg_image(z)) { stbi__cleanup_jpeg(z); return NULL; }
+   if (!stbi__decode_jpeg_image(z))
+   {
+      stbi__cleanup_jpeg(z);
+      return NULL;
+   }
 
    // determine actual number of components to generate
-   n = req_comp ? req_comp : z->s->img_n;
+   n = req_comp ? req_comp : z->s->img_n >= 3 ? 3
+                                              : 1;
+
+   is_rgb = z->s->img_n == 3 && (z->rgb == 3 || (z->app14_color_transform == 0 && !z->jfif));
 
-   if (z->s->img_n == 3 && n < 3)
+   if (z->s->img_n == 3 && n < 3 && !is_rgb)
       decode_n = 1;
    else
       decode_n = z->s->img_n;
 
+   // nothing to do if no components requested; check this now to avoid
+   // accessing uninitialized coutput[0] later
+   if (decode_n <= 0)
+   {
+      stbi__cleanup_jpeg(z);
+      return NULL;
+   }
+
    // resample and color-convert
    {
       int k;
-      unsigned int i,j;
-      stbi_uc *output;
-      stbi_uc *coutput[4];
+      unsigned int i, j;
+      stbi_uc* output;
+      stbi_uc* coutput[4] = { NULL, NULL, NULL, NULL };
 
       stbi__resample res_comp[4];
 
-      for (k=0; k < decode_n; ++k) {
-         stbi__resample *r = &res_comp[k];
+      for (k = 0; k < decode_n; ++k)
+      {
+         stbi__resample* r = &res_comp[k];
 
          // allocate line buffer big enough for upsampling off the edges
          // with upsample factor of 4
-         z->img_comp[k].linebuf = (stbi_uc *) stbi__malloc(z->s->img_x + 3);
-         if (!z->img_comp[k].linebuf) { stbi__cleanup_jpeg(z); return stbi__errpuc("outofmem", "Out of memory"); }
-
-         r->hs      = z->img_h_max / z->img_comp[k].h;
-         r->vs      = z->img_v_max / z->img_comp[k].v;
-         r->ystep   = r->vs >> 1;
-         r->w_lores = (z->s->img_x + r->hs-1) / r->hs;
-         r->ypos    = 0;
-         r->line0   = r->line1 = z->img_comp[k].data;
+         z->img_comp[k].linebuf = (stbi_uc*)stbi__malloc(z->s->img_x + 3);
+         if (!z->img_comp[k].linebuf)
+         {
+            stbi__cleanup_jpeg(z);
+            return stbi__errpuc("outofmem", "Out of memory");
+         }
 
-         if      (r->hs == 1 && r->vs == 1) r->resample = resample_row_1;
-         else if (r->hs == 1 && r->vs == 2) r->resample = stbi__resample_row_v_2;
-         else if (r->hs == 2 && r->vs == 1) r->resample = stbi__resample_row_h_2;
-         else if (r->hs == 2 && r->vs == 2) r->resample = z->resample_row_hv_2_kernel;
-         else                               r->resample = stbi__resample_row_generic;
+         r->hs = z->img_h_max / z->img_comp[k].h;
+         r->vs = z->img_v_max / z->img_comp[k].v;
+         r->ystep = r->vs >> 1;
+         r->w_lores = (z->s->img_x + r->hs - 1) / r->hs;
+         r->ypos = 0;
+         r->line0 = r->line1 = z->img_comp[k].data;
+
+         if (r->hs == 1 && r->vs == 1)
+            r->resample = resample_row_1;
+         else if (r->hs == 1 && r->vs == 2)
+            r->resample = stbi__resample_row_v_2;
+         else if (r->hs == 2 && r->vs == 1)
+            r->resample = stbi__resample_row_h_2;
+         else if (r->hs == 2 && r->vs == 2)
+            r->resample = z->resample_row_hv_2_kernel;
+         else
+            r->resample = stbi__resample_row_generic;
       }
 
       // can't error after this so, this is safe
-      output = (stbi_uc *) stbi__malloc(n * z->s->img_x * z->s->img_y + 1);
-      if (!output) { stbi__cleanup_jpeg(z); return stbi__errpuc("outofmem", "Out of memory"); }
+      output = (stbi_uc*)stbi__malloc_mad3(n, z->s->img_x, z->s->img_y, 1);
+      if (!output)
+      {
+         stbi__cleanup_jpeg(z);
+         return stbi__errpuc("outofmem", "Out of memory");
+      }
 
       // now go ahead and resample
-      for (j=0; j < z->s->img_y; ++j) {
-         stbi_uc *out = output + n * z->s->img_x * j;
-         for (k=0; k < decode_n; ++k) {
-            stbi__resample *r = &res_comp[k];
+      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) {
+            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) {
-               z->YCbCr_to_RGB_kernel(out, y, coutput[1], coutput[2], z->s->img_x, n);
-            } else
-               for (i=0; i < z->s->img_x; ++i) {
+         if (n >= 3)
+         {
+            stbi_uc* y = coutput[0];
+            if (z->s->img_n == 3)
+            {
+               if (is_rgb)
+               {
+                  for (i = 0; i < z->s->img_x; ++i)
+                  {
+                     out[0] = y[i];
+                     out[1] = coutput[1][i];
+                     out[2] = coutput[2][i];
+                     out[3] = 255;
+                     out += n;
+                  }
+               }
+               else
+               {
+                  z->YCbCr_to_RGB_kernel(out, y, coutput[1], coutput[2], z->s->img_x, n);
+               }
+            }
+            else if (z->s->img_n == 4)
+            {
+               if (z->app14_color_transform == 0)
+               { // CMYK
+                  for (i = 0; i < z->s->img_x; ++i)
+                  {
+                     stbi_uc m = coutput[3][i];
+                     out[0] = stbi__blinn_8x8(coutput[0][i], m);
+                     out[1] = stbi__blinn_8x8(coutput[1][i], m);
+                     out[2] = stbi__blinn_8x8(coutput[2][i], m);
+                     out[3] = 255;
+                     out += n;
+                  }
+               }
+               else if (z->app14_color_transform == 2)
+               { // YCCK
+                  z->YCbCr_to_RGB_kernel(out, y, coutput[1], coutput[2], z->s->img_x, n);
+                  for (i = 0; i < z->s->img_x; ++i)
+                  {
+                     stbi_uc m = coutput[3][i];
+                     out[0] = stbi__blinn_8x8(255 - out[0], m);
+                     out[1] = stbi__blinn_8x8(255 - out[1], m);
+                     out[2] = stbi__blinn_8x8(255 - out[2], m);
+                     out += n;
+                  }
+               }
+               else
+               { // YCbCr + alpha?  Ignore the fourth channel for now
+                  z->YCbCr_to_RGB_kernel(out, y, coutput[1], coutput[2], z->s->img_x, n);
+               }
+            }
+            else
+               for (i = 0; i < z->s->img_x; ++i)
+               {
                   out[0] = out[1] = out[2] = y[i];
                   out[3] = 255; // not used if n==3
                   out += n;
                }
-         } else {
-            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;
-   stbi__setup_jpeg(&j);
-   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;
-   stbi__setup_jpeg(&j);
-   r = stbi__decode_jpeg_header(&j, STBI__SCAN_type);
-   stbi__rewind(s);
-   return r;
-}
-
-static int stbi__jpeg_info_raw(stbi__jpeg *j, int *x, int *y, int *comp)
-{
-   if (!stbi__decode_jpeg_header(j, STBI__SCAN_header)) {
-      stbi__rewind( j->s );
-      return 0;
+         else
+         {
+            if (is_rgb)
+            {
+               if (n == 1)
+                  for (i = 0; i < z->s->img_x; ++i)
+                     *out++ = stbi__compute_y(coutput[0][i], coutput[1][i], coutput[2][i]);
+               else
+               {
+                  for (i = 0; i < z->s->img_x; ++i, out += 2)
+                  {
+                     out[0] = stbi__compute_y(coutput[0][i], coutput[1][i], coutput[2][i]);
+                     out[1] = 255;
+                  }
+               }
+            }
+            else if (z->s->img_n == 4 && z->app14_color_transform == 0)
+            {
+               for (i = 0; i < z->s->img_x; ++i)
+               {
+                  stbi_uc m = coutput[3][i];
+                  stbi_uc r = stbi__blinn_8x8(coutput[0][i], m);
+                  stbi_uc g = stbi__blinn_8x8(coutput[1][i], m);
+                  stbi_uc b = stbi__blinn_8x8(coutput[2][i], m);
+                  out[0] = stbi__compute_y(r, g, b);
+                  out[1] = 255;
+                  out += n;
+               }
+            }
+            else if (z->s->img_n == 4 && z->app14_color_transform == 2)
+            {
+               for (i = 0; i < z->s->img_x; ++i)
+               {
+                  out[0] = stbi__blinn_8x8(255 - coutput[0][i], coutput[3][i]);
+                  out[1] = 255;
+                  out += n;
+               }
+            }
+            else
+            {
+               stbi_uc* y = coutput[0];
+               if (n == 1)
+                  for (i = 0; i < z->s->img_x; ++i)
+                     out[i] = y[i];
+               else
+                  for (i = 0; i < z->s->img_x; ++i)
+                  {
+                     *out++ = y[i];
+                     *out++ = 255;
+                  }
+            }
+         }
+      }
+      stbi__cleanup_jpeg(z);
+      *out_x = z->s->img_x;
+      *out_y = z->s->img_y;
+      if (comp)
+         *comp = z->s->img_n >= 3 ? 3 : 1; // report original components, not output
+      return output;
    }
-   if (x) *x = j->s->img_x;
-   if (y) *y = j->s->img_y;
-   if (comp) *comp = j->s->img_n;
+}
+
+static void* stbi__jpeg_load(stbi__context* s, int* x, int* y, int* comp, int req_comp, stbi__result_info* ri)
+{
+   unsigned char* result;
+   stbi__jpeg* j = (stbi__jpeg*)stbi__malloc(sizeof(stbi__jpeg));
+   if (!j)
+      return stbi__errpuc("outofmem", "Out of memory");
+   memset(j, 0, sizeof(stbi__jpeg));
+   STBI_NOTUSED(ri);
+   j->s = s;
+   stbi__setup_jpeg(j);
+   result = load_jpeg_image(j, x, y, comp, req_comp);
+   STBI_FREE(j);
+   return result;
+}
+
+static int stbi__jpeg_test(stbi__context* s)
+{
+   int r;
+   stbi__jpeg* j = (stbi__jpeg*)stbi__malloc(sizeof(stbi__jpeg));
+   if (!j)
+      return stbi__err("outofmem", "Out of memory");
+   memset(j, 0, sizeof(stbi__jpeg));
+   j->s = s;
+   stbi__setup_jpeg(j);
+   r = stbi__decode_jpeg_header(j, STBI__SCAN_type);
+   stbi__rewind(s);
+   STBI_FREE(j);
+   return r;
+}
+
+static int stbi__jpeg_info_raw(stbi__jpeg* j, int* x, int* y, int* comp)
+{
+   if (!stbi__decode_jpeg_header(j, STBI__SCAN_header))
+   {
+      stbi__rewind(j->s);
+      return 0;
+   }
+   if (x)
+      *x = j->s->img_x;
+   if (y)
+      *y = j->s->img_y;
+   if (comp)
+      *comp = j->s->img_n >= 3 ? 3 : 1;
    return 1;
 }
 
-static int stbi__jpeg_info(stbi__context *s, int *x, int *y, int *comp)
+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);
+   int result;
+   stbi__jpeg* j = (stbi__jpeg*)(stbi__malloc(sizeof(stbi__jpeg)));
+   if (!j)
+      return stbi__err("outofmem", "Out of memory");
+   memset(j, 0, sizeof(stbi__jpeg));
+   j->s = s;
+   result = stbi__jpeg_info_raw(j, x, y, comp);
+   STBI_FREE(j);
+   return result;
 }
 #endif
 
@@ -3455,8 +4669,9 @@ static int stbi__jpeg_info(stbi__context *s, int *x, int *y, int *comp)
 #ifndef STBI_NO_ZLIB
 
 // fast-way is faster to check than jpeg huffman, but slow way is slower
-#define STBI__ZFAST_BITS  9 // accelerate all cases in default tables
-#define STBI__ZFAST_MASK  ((1 << STBI__ZFAST_BITS) - 1)
+#define STBI__ZFAST_BITS 9 // accelerate all cases in default tables
+#define STBI__ZFAST_MASK ((1 << STBI__ZFAST_BITS) - 1)
+#define STBI__ZNSYMS 288 // number of symbols in literal/length alphabet
 
 // zlib-style huffman encoding
 // (jpegs packs from left, zlib from right, so can't share code)
@@ -3466,17 +4681,17 @@ typedef struct
    stbi__uint16 firstcode[16];
    int maxcode[17];
    stbi__uint16 firstsymbol[16];
-   stbi_uc  size[288];
-   stbi__uint16 value[288];
+   stbi_uc size[STBI__ZNSYMS];
+   stbi__uint16 value[STBI__ZNSYMS];
 } stbi__zhuffman;
 
 stbi_inline static int stbi__bitreverse16(int n)
 {
-  n = ((n & 0xAAAA) >>  1) | ((n & 0x5555) << 1);
-  n = ((n & 0xCCCC) >>  2) | ((n & 0x3333) << 2);
-  n = ((n & 0xF0F0) >>  4) | ((n & 0x0F0F) << 4);
-  n = ((n & 0xFF00) >>  8) | ((n & 0x00FF) << 8);
-  return n;
+   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)
@@ -3484,46 +4699,52 @@ stbi_inline static int stbi__bit_reverse(int v, int bits)
    STBI_ASSERT(bits <= 16);
    // to bit reverse n bits, reverse 16 and shift
    // e.g. 11 bits, bit reverse and shift away 5
-   return stbi__bitreverse16(v) >> (16-bits);
+   return stbi__bitreverse16(v) >> (16 - bits);
 }
 
-static int stbi__zbuild_huffman(stbi__zhuffman *z, stbi_uc *sizelist, int num)
+static int stbi__zbuild_huffman(stbi__zhuffman* z, const stbi_uc* sizelist, int num)
 {
-   int i,k=0;
+   int i, k = 0;
    int code, next_code[16], sizes[17];
 
    // DEFLATE spec for generating codes
    memset(sizes, 0, sizeof(sizes));
    memset(z->fast, 0, sizeof(z->fast));
-   for (i=0; i < num; ++i)
+   for (i = 0; i < num; ++i)
       ++sizes[sizelist[i]];
    sizes[0] = 0;
-   for (i=1; i < 16; ++i)
+   for (i = 1; i < 16; ++i)
       if (sizes[i] > (1 << i))
          return stbi__err("bad sizes", "Corrupt PNG");
    code = 0;
-   for (i=1; i < 16; ++i) {
+   for (i = 1; i < 16; ++i)
+   {
       next_code[i] = code;
-      z->firstcode[i] = (stbi__uint16) code;
-      z->firstsymbol[i] = (stbi__uint16) k;
+      z->firstcode[i] = (stbi__uint16)code;
+      z->firstsymbol[i] = (stbi__uint16)k;
       code = (code + sizes[i]);
       if (sizes[i])
-         if (code-1 >= (1 << i)) return stbi__err("bad codelengths","Corrupt PNG");
-      z->maxcode[i] = code << (16-i); // preshift for inner loop
+         if (code - 1 >= (1 << i))
+            return stbi__err("bad codelengths", "Corrupt PNG");
+      z->maxcode[i] = code << (16 - i); // preshift for inner loop
       code <<= 1;
       k += sizes[i];
    }
    z->maxcode[16] = 0x10000; // sentinel
-   for (i=0; i < num; ++i) {
+   for (i = 0; i < num; ++i)
+   {
       int s = sizelist[i];
-      if (s) {
+      if (s)
+      {
          int c = next_code[s] - z->firstcode[s] + z->firstsymbol[s];
-         stbi__uint16 fastv = (stbi__uint16) ((s << 9) | i);
-         z->size [c] = (stbi_uc     ) s;
-         z->value[c] = (stbi__uint16) i;
-         if (s <= STBI__ZFAST_BITS) {
-            int j = stbi__bit_reverse(next_code[s],s);
-            while (j < (1 << STBI__ZFAST_BITS)) {
+         stbi__uint16 fastv = (stbi__uint16)((s << 9) | i);
+         z->size[c] = (stbi_uc)s;
+         z->value[c] = (stbi__uint16)i;
+         if (s <= STBI__ZFAST_BITS)
+         {
+            int j = stbi__bit_reverse(next_code[s], s);
+            while (j < (1 << STBI__ZFAST_BITS))
+            {
                z->fast[j] = fastv;
                j += (1 << s);
             }
@@ -3544,65 +4765,104 @@ typedef struct
 {
    stbi_uc *zbuffer, *zbuffer_end;
    int num_bits;
+   int hit_zeof_once;
    stbi__uint32 code_buffer;
 
-   char *zout;
-   char *zout_start;
-   char *zout_end;
-   int   z_expandable;
+   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)
+stbi_inline static int stbi__zeof(stbi__zbuf* z)
+{
+   return (z->zbuffer >= z->zbuffer_end);
+}
+
+stbi_inline static stbi_uc stbi__zget8(stbi__zbuf* z)
 {
-   if (z->zbuffer >= z->zbuffer_end) return 0;
-   return *z->zbuffer++;
+   return stbi__zeof(z) ? 0 : *z->zbuffer++;
 }
 
-static void stbi__fill_bits(stbi__zbuf *z)
+static void stbi__fill_bits(stbi__zbuf* z)
 {
-   do {
-      STBI_ASSERT(z->code_buffer < (1U << z->num_bits));
-      z->code_buffer |= (unsigned int) stbi__zget8(z) << z->num_bits;
+   do
+   {
+      if (z->code_buffer >= (1U << z->num_bits))
+      {
+         z->zbuffer = z->zbuffer_end; /* treat this as EOF so we fail. */
+         return;
+      }
+      z->code_buffer |= (unsigned int)stbi__zget8(z) << z->num_bits;
       z->num_bits += 8;
    } while (z->num_bits <= 24);
 }
 
-stbi_inline static unsigned int stbi__zreceive(stbi__zbuf *z, int n)
+stbi_inline static unsigned int stbi__zreceive(stbi__zbuf* z, int n)
 {
    unsigned int k;
-   if (z->num_bits < n) stbi__fill_bits(z);
+   if (z->num_bits < n)
+      stbi__fill_bits(z);
    k = z->code_buffer & ((1 << n) - 1);
    z->code_buffer >>= n;
    z->num_bits -= n;
    return k;
 }
 
-static int stbi__zhuffman_decode_slowpath(stbi__zbuf *a, stbi__zhuffman *z)
+static int stbi__zhuffman_decode_slowpath(stbi__zbuf* a, stbi__zhuffman* z)
 {
-   int b,s,k;
+   int b, s, k;
    // not resolved by fast table, so compute it the slow way
    // use jpeg approach, which requires MSbits at top
    k = stbi__bit_reverse(a->code_buffer, 16);
-   for (s=STBI__ZFAST_BITS+1; ; ++s)
+   for (s = STBI__ZFAST_BITS + 1;; ++s)
       if (k < z->maxcode[s])
          break;
-   if (s == 16) return -1; // invalid code!
+   if (s >= 16)
+      return -1; // invalid code!
    // code size is s, so:
-   b = (k >> (16-s)) - z->firstcode[s] + z->firstsymbol[s];
-   STBI_ASSERT(z->size[b] == s);
+   b = (k >> (16 - s)) - z->firstcode[s] + z->firstsymbol[s];
+   if (b >= STBI__ZNSYMS)
+      return -1; // some data was corrupt somewhere!
+   if (z->size[b] != s)
+      return -1; // was originally an assert, but report failure instead.
    a->code_buffer >>= s;
    a->num_bits -= s;
    return z->value[b];
 }
 
-stbi_inline static int stbi__zhuffman_decode(stbi__zbuf *a, stbi__zhuffman *z)
+stbi_inline static int stbi__zhuffman_decode(stbi__zbuf* a, stbi__zhuffman* z)
 {
-   int b,s;
-   if (a->num_bits < 16) stbi__fill_bits(a);
+   int b, s;
+   if (a->num_bits < 16)
+   {
+      if (stbi__zeof(a))
+      {
+         if (!a->hit_zeof_once)
+         {
+            // This is the first time we hit eof, insert 16 extra padding btis
+            // to allow us to keep going; if we actually consume any of them
+            // though, that is invalid data. This is caught later.
+            a->hit_zeof_once = 1;
+            a->num_bits += 16; // add 16 implicit zero bits
+         }
+         else
+         {
+            // We already inserted our extra 16 padding bits and are again
+            // out, this stream is actually prematurely terminated.
+            return -1;
+         }
+      }
+      else
+      {
+         stbi__fill_bits(a);
+      }
+   }
    b = z->fast[a->code_buffer & STBI__ZFAST_MASK];
-   if (b) {
+   if (b)
+   {
       s = b >> 9;
       a->code_buffer >>= s;
       a->num_bits -= s;
@@ -3611,172 +4871,260 @@ stbi_inline static int stbi__zhuffman_decode(stbi__zbuf *a, stbi__zhuffman *z)
    return stbi__zhuffman_decode_slowpath(a, z);
 }
 
-static int stbi__zexpand(stbi__zbuf *z, char *zout, int n)  // need to make room for n bytes
+static int stbi__zexpand(stbi__zbuf* z, char* zout, int n) // need to make room for n bytes
 {
-   char *q;
-   int cur, limit, old_limit;
+   char* q;
+   unsigned int cur, limit, old_limit;
    z->zout = zout;
-   if (!z->z_expandable) return stbi__err("output buffer limit","Corrupt PNG");
-   cur   = (int) (z->zout     - z->zout_start);
-   limit = old_limit = (int) (z->zout_end - z->zout_start);
+   if (!z->z_expandable)
+      return stbi__err("output buffer limit", "Corrupt PNG");
+   cur = (unsigned int)(z->zout - z->zout_start);
+   limit = old_limit = (unsigned)(z->zout_end - z->zout_start);
+   if (UINT_MAX - cur < (unsigned)n)
+      return stbi__err("outofmem", "Out of memory");
    while (cur + n > limit)
+   {
+      if (limit > UINT_MAX / 2)
+         return stbi__err("outofmem", "Out of memory");
       limit *= 2;
-   q = (char *) STBI_REALLOC_SIZED(z->zout_start, old_limit, limit);
+   }
+   q = (char*)STBI_REALLOC_SIZED(z->zout_start, old_limit, limit);
    STBI_NOTUSED(old_limit);
-   if (q == NULL) return stbi__err("outofmem", "Out of memory");
+   if (q == NULL)
+      return stbi__err("outofmem", "Out of memory");
    z->zout_start = q;
-   z->zout       = q + cur;
-   z->zout_end   = q + limit;
+   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 const int stbi__zlength_base[31] = {
+   3, 4, 5, 6, 7, 8, 9, 10, 11, 13,
+   15, 17, 19, 23, 27, 31, 35, 43, 51, 59,
+   67, 83, 99, 115, 131, 163, 195, 227, 258, 0, 0
+};
 
-static int stbi__zlength_extra[31]=
-{ 0,0,0,0,0,0,0,0,1,1,1,1,2,2,2,2,3,3,3,3,4,4,4,4,5,5,5,5,0,0,0 };
+static const int stbi__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 const int stbi__zdist_base[32] = { 1, 2, 3, 4, 5, 7, 9, 13, 17, 25, 33, 49, 65, 97, 129, 193,
+                                          257, 385, 513, 769, 1025, 1537, 2049, 3073, 4097, 6145, 8193, 12289, 16385, 24577, 0, 0 };
 
-static 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 const int stbi__zdist_extra[32] = { 0, 0, 0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12, 12, 13, 13 };
 
-static int stbi__parse_huffman_block(stbi__zbuf *a)
+static int stbi__parse_huffman_block(stbi__zbuf* a)
 {
-   char *zout = a->zout;
-   for(;;) {
+   char* zout = a->zout;
+   for (;;)
+   {
       int z = stbi__zhuffman_decode(a, &a->z_length);
-      if (z < 256) {
-         if (z < 0) return stbi__err("bad huffman code","Corrupt PNG"); // error in huffman codes
-         if (zout >= a->zout_end) {
-            if (!stbi__zexpand(a, zout, 1)) return 0;
+      if (z < 256)
+      {
+         if (z < 0)
+            return stbi__err("bad huffman code", "Corrupt PNG"); // error in huffman codes
+         if (zout >= a->zout_end)
+         {
+            if (!stbi__zexpand(a, zout, 1))
+               return 0;
             zout = a->zout;
          }
-         *zout++ = (char) z;
-      } else {
-         stbi_uc *p;
-         int len,dist;
-         if (z == 256) {
+         *zout++ = (char)z;
+      }
+      else
+      {
+         stbi_uc* p;
+         int len, dist;
+         if (z == 256)
+         {
             a->zout = zout;
+            if (a->hit_zeof_once && a->num_bits < 16)
+            {
+               // The first time we hit zeof, we inserted 16 extra zero bits into our bit
+               // buffer so the decoder can just do its speculative decoding. But if we
+               // actually consumed any of those bits (which is the case when num_bits < 16),
+               // the stream actually read past the end so it is malformed.
+               return stbi__err("unexpected end", "Corrupt PNG");
+            }
             return 1;
          }
+         if (z >= 286)
+            return stbi__err("bad huffman code", "Corrupt PNG"); // per DEFLATE, length codes 286 and 287 must not appear in compressed data
          z -= 257;
          len = stbi__zlength_base[z];
-         if (stbi__zlength_extra[z]) len += stbi__zreceive(a, stbi__zlength_extra[z]);
+         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");
+         if (z < 0 || z >= 30)
+            return stbi__err("bad huffman code", "Corrupt PNG"); // per DEFLATE, distance codes 30 and 31 must not appear in compressed data
          dist = stbi__zdist_base[z];
-         if (stbi__zdist_extra[z]) dist += stbi__zreceive(a, stbi__zdist_extra[z]);
-         if (zout - a->zout_start < dist) return stbi__err("bad dist","Corrupt PNG");
-         if (zout + len > a->zout_end) {
-            if (!stbi__zexpand(a, zout, len)) return 0;
+         if (stbi__zdist_extra[z])
+            dist += stbi__zreceive(a, stbi__zdist_extra[z]);
+         if (zout - a->zout_start < dist)
+            return stbi__err("bad dist", "Corrupt PNG");
+         if (len > a->zout_end - zout)
+         {
+            if (!stbi__zexpand(a, zout, len))
+               return 0;
             zout = a->zout;
          }
-         p = (stbi_uc *) (zout - dist);
-         if (dist == 1) { // run of one byte; common in images.
+         p = (stbi_uc*)(zout - dist);
+         if (dist == 1)
+         { // run of one byte; common in images.
             stbi_uc v = *p;
-            if (len) { do *zout++ = v; while (--len); }
-         } else {
-            if (len) { do *zout++ = *p++; while (--len); }
+            if (len)
+            {
+               do
+                  *zout++ = v;
+               while (--len);
+            }
+         }
+         else
+         {
+            if (len)
+            {
+               do
+                  *zout++ = *p++;
+               while (--len);
+            }
          }
       }
    }
 }
 
-static int stbi__compute_huffman_codes(stbi__zbuf *a)
+static 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 };
+   static const stbi_uc length_dezigzag[19] = { 16, 17, 18, 0, 8, 7, 9, 6, 10, 5, 11, 4, 12, 3, 13, 2, 14, 1, 15 };
    stbi__zhuffman z_codelength;
-   stbi_uc lencodes[286+32+137];//padding for maximum single op
+   stbi_uc lencodes[286 + 32 + 137]; //padding for maximum single op
    stbi_uc codelength_sizes[19];
-   int i,n;
+   int i, n;
 
-   int hlit  = stbi__zreceive(a,5) + 257;
-   int hdist = stbi__zreceive(a,5) + 1;
-   int hclen = stbi__zreceive(a,4) + 4;
+   int hlit = stbi__zreceive(a, 5) + 257;
+   int hdist = stbi__zreceive(a, 5) + 1;
+   int hclen = stbi__zreceive(a, 4) + 4;
+   int ntot = hlit + hdist;
 
    memset(codelength_sizes, 0, sizeof(codelength_sizes));
-   for (i=0; i < hclen; ++i) {
-      int s = stbi__zreceive(a,3);
-      codelength_sizes[length_dezigzag[i]] = (stbi_uc) s;
+   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;
+   if (!stbi__zbuild_huffman(&z_codelength, codelength_sizes, 19))
+      return 0;
 
    n = 0;
-   while (n < hlit + hdist) {
+   while (n < ntot)
+   {
       int c = stbi__zhuffman_decode(a, &z_codelength);
-      if (c < 0 || c >= 19) return stbi__err("bad codelengths", "Corrupt PNG");
+      if (c < 0 || c >= 19)
+         return stbi__err("bad codelengths", "Corrupt PNG");
       if (c < 16)
-         lencodes[n++] = (stbi_uc) c;
-      else if (c == 16) {
-         c = stbi__zreceive(a,2)+3;
-         memset(lencodes+n, lencodes[n-1], c);
-         n += c;
-      } else if (c == 17) {
-         c = stbi__zreceive(a,3)+3;
-         memset(lencodes+n, 0, c);
-         n += c;
-      } else {
-         STBI_ASSERT(c == 18);
-         c = stbi__zreceive(a,7)+11;
-         memset(lencodes+n, 0, c);
+         lencodes[n++] = (stbi_uc)c;
+      else
+      {
+         stbi_uc fill = 0;
+         if (c == 16)
+         {
+            c = stbi__zreceive(a, 2) + 3;
+            if (n == 0)
+               return stbi__err("bad codelengths", "Corrupt PNG");
+            fill = lencodes[n - 1];
+         }
+         else if (c == 17)
+         {
+            c = stbi__zreceive(a, 3) + 3;
+         }
+         else if (c == 18)
+         {
+            c = stbi__zreceive(a, 7) + 11;
+         }
+         else
+         {
+            return stbi__err("bad codelengths", "Corrupt PNG");
+         }
+         if (ntot - n < c)
+            return stbi__err("bad codelengths", "Corrupt PNG");
+         memset(lencodes + n, fill, c);
          n += c;
       }
    }
-   if (n != 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;
+   if (n != ntot)
+      return stbi__err("bad codelengths", "Corrupt PNG");
+   if (!stbi__zbuild_huffman(&a->z_length, lencodes, hlit))
+      return 0;
+   if (!stbi__zbuild_huffman(&a->z_distance, lencodes + hlit, hdist))
+      return 0;
    return 1;
 }
 
-static int stbi__parse_uncomperssed_block(stbi__zbuf *a)
+static int stbi__parse_uncompressed_block(stbi__zbuf* a)
 {
    stbi_uc header[4];
-   int len,nlen,k;
+   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
+   while (a->num_bits > 0)
+   {
+      header[k++] = (stbi_uc)(a->code_buffer & 255); // suppress MSVC run-time check
       a->code_buffer >>= 8;
       a->num_bits -= 8;
    }
-   STBI_ASSERT(a->num_bits == 0);
+   if (a->num_bits < 0)
+      return stbi__err("zlib corrupt", "Corrupt PNG");
    // now fill header the normal way
    while (k < 4)
       header[k++] = stbi__zget8(a);
-   len  = header[1] * 256 + header[0];
+   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 (nlen != (len ^ 0xffff))
+      return stbi__err("zlib corrupt", "Corrupt PNG");
+   if (a->zbuffer + len > a->zbuffer_end)
+      return stbi__err("read past buffer", "Corrupt PNG");
    if (a->zout + len > a->zout_end)
-      if (!stbi__zexpand(a, a->zout, len)) return 0;
+      if (!stbi__zexpand(a, a->zout, len))
+         return 0;
    memcpy(a->zout, a->zbuffer, len);
    a->zbuffer += len;
    a->zout += len;
    return 1;
 }
 
-static int stbi__parse_zlib_header(stbi__zbuf *a)
+static int stbi__parse_zlib_header(stbi__zbuf* a)
 {
-   int cmf   = stbi__zget8(a);
-   int cm    = cmf & 15;
+   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
+   int flg = stbi__zget8(a);
+   if (stbi__zeof(a))
+      return stbi__err("bad zlib header", "Corrupt PNG"); // zlib spec
+   if ((cmf * 256 + flg) % 31 != 0)
+      return stbi__err("bad zlib header", "Corrupt PNG"); // zlib spec
+   if (flg & 32)
+      return stbi__err("no preset dict", "Corrupt PNG"); // preset dictionary not allowed in png
+   if (cm != 8)
+      return stbi__err("bad compression", "Corrupt PNG"); // DEFLATE required for png
    // window = 1 << (8 + cinfo)... but who cares, we fully buffer output
    return 1;
 }
 
-// @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)
+static const stbi_uc stbi__zdefault_length[STBI__ZNSYMS] = {
+   8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
+   8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
+   8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
+   8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
+   8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
+   9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
+   9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
+   9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,
+   7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 8, 8
+};
+static const stbi_uc stbi__zdefault_distance[32] = {
+   5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5
+};
+/*
+Init algorithm:
 {
    int i;   // use <= to match clearly with spec
    for (i=0; i <= 143; ++i)     stbi__zdefault_length[i]   = 8;
@@ -3786,117 +5134,148 @@ static void stbi__init_zdefaults(void)
 
    for (i=0; i <=  31; ++i)     stbi__zdefault_distance[i] = 5;
 }
+*/
 
-static int stbi__parse_zlib(stbi__zbuf *a, int parse_header)
+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;
+      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) {
+   a->hit_zeof_once = 0;
+   do
+   {
+      final = stbi__zreceive(a, 1);
+      type = stbi__zreceive(a, 2);
+      if (type == 0)
+      {
+         if (!stbi__parse_uncompressed_block(a))
+            return 0;
+      }
+      else if (type == 3)
+      {
          return 0;
-      } else {
-         if (type == 1) {
+      }
+      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__zbuild_huffman(&a->z_length, stbi__zdefault_length, STBI__ZNSYMS))
+               return 0;
+            if (!stbi__zbuild_huffman(&a->z_distance, stbi__zdefault_distance, 32))
+               return 0;
          }
-         if (!stbi__parse_huffman_block(a)) 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)
+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->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)
+STBIDEF char* stbi_zlib_decode_malloc_guesssize(const char* buffer, int len, int initial_size, int* outlen)
 {
    stbi__zbuf a;
-   char *p = (char *) stbi__malloc(initial_size);
-   if (p == NULL) return NULL;
-   a.zbuffer = (stbi_uc *) buffer;
-   a.zbuffer_end = (stbi_uc *) buffer + len;
-   if (stbi__do_zlib(&a, p, initial_size, 1, 1)) {
-      if (outlen) *outlen = (int) (a.zout - a.zout_start);
+   char* p = (char*)stbi__malloc(initial_size);
+   if (p == NULL)
+      return NULL;
+   a.zbuffer = (stbi_uc*)buffer;
+   a.zbuffer_end = (stbi_uc*)buffer + len;
+   if (stbi__do_zlib(&a, p, initial_size, 1, 1))
+   {
+      if (outlen)
+         *outlen = (int)(a.zout - a.zout_start);
       return a.zout_start;
-   } else {
+   }
+   else
+   {
       STBI_FREE(a.zout_start);
       return NULL;
    }
 }
 
-STBIDEF char *stbi_zlib_decode_malloc(char const *buffer, int len, int *outlen)
+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)
+STBIDEF char* stbi_zlib_decode_malloc_guesssize_headerflag(const char* buffer, int len, int initial_size, int* outlen, int parse_header)
 {
    stbi__zbuf a;
-   char *p = (char *) stbi__malloc(initial_size);
-   if (p == NULL) return NULL;
-   a.zbuffer = (stbi_uc *) buffer;
-   a.zbuffer_end = (stbi_uc *) buffer + len;
-   if (stbi__do_zlib(&a, p, initial_size, 1, parse_header)) {
-      if (outlen) *outlen = (int) (a.zout - a.zout_start);
+   char* p = (char*)stbi__malloc(initial_size);
+   if (p == NULL)
+      return NULL;
+   a.zbuffer = (stbi_uc*)buffer;
+   a.zbuffer_end = (stbi_uc*)buffer + len;
+   if (stbi__do_zlib(&a, p, initial_size, 1, parse_header))
+   {
+      if (outlen)
+         *outlen = (int)(a.zout - a.zout_start);
       return a.zout_start;
-   } else {
+   }
+   else
+   {
       STBI_FREE(a.zout_start);
       return NULL;
    }
 }
 
-STBIDEF int stbi_zlib_decode_buffer(char *obuffer, int olen, char const *ibuffer, int ilen)
+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;
+   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);
+      return (int)(a.zout - a.zout_start);
    else
       return -1;
 }
 
-STBIDEF char *stbi_zlib_decode_noheader_malloc(char const *buffer, int len, int *outlen)
+STBIDEF char* stbi_zlib_decode_noheader_malloc(char const* buffer, int len, int* outlen)
 {
    stbi__zbuf a;
-   char *p = (char *) stbi__malloc(16384);
-   if (p == NULL) return NULL;
-   a.zbuffer = (stbi_uc *) buffer;
-   a.zbuffer_end = (stbi_uc *) buffer+len;
-   if (stbi__do_zlib(&a, p, 16384, 1, 0)) {
-      if (outlen) *outlen = (int) (a.zout - a.zout_start);
+   char* p = (char*)stbi__malloc(16384);
+   if (p == NULL)
+      return NULL;
+   a.zbuffer = (stbi_uc*)buffer;
+   a.zbuffer_end = (stbi_uc*)buffer + len;
+   if (stbi__do_zlib(&a, p, 16384, 1, 0))
+   {
+      if (outlen)
+         *outlen = (int)(a.zout - a.zout_start);
       return a.zout_start;
-   } else {
+   }
+   else
+   {
       STBI_FREE(a.zout_start);
       return NULL;
    }
 }
 
-STBIDEF int stbi_zlib_decode_noheader_buffer(char *obuffer, int olen, const char *ibuffer, int ilen)
+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;
+   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);
+      return (int)(a.zout - a.zout_start);
    else
       return -1;
 }
@@ -3919,272 +5298,334 @@ typedef struct
    stbi__uint32 type;
 } stbi__pngchunk;
 
-static stbi__pngchunk stbi__get_chunk_header(stbi__context *s)
+static stbi__pngchunk stbi__get_chunk_header(stbi__context* s)
 {
    stbi__pngchunk c;
    c.length = stbi__get32be(s);
-   c.type   = stbi__get32be(s);
+   c.type = stbi__get32be(s);
    return c;
 }
 
-static int stbi__check_png_header(stbi__context *s)
+static int stbi__check_png_header(stbi__context* s)
 {
-   static stbi_uc png_sig[8] = { 137,80,78,71,13,10,26,10 };
+   static const stbi_uc png_sig[8] = { 137, 80, 78, 71, 13, 10, 26, 10 };
    int i;
-   for (i=0; i < 8; ++i)
-      if (stbi__get8(s) != png_sig[i]) return stbi__err("bad png sig","Not a PNG");
+   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__context* s;
    stbi_uc *idata, *expanded, *out;
+   int depth;
 } stbi__png;
 
 
-enum {
-   STBI__F_none=0,
-   STBI__F_sub=1,
-   STBI__F_up=2,
-   STBI__F_avg=3,
-   STBI__F_paeth=4,
-   // synthetic filters used for first scanline to avoid needing a dummy row of 0s
-   STBI__F_avg_first,
-   STBI__F_paeth_first
+enum
+{
+   STBI__F_none = 0,
+   STBI__F_sub = 1,
+   STBI__F_up = 2,
+   STBI__F_avg = 3,
+   STBI__F_paeth = 4,
+   // synthetic filter used for first scanline to avoid needing a dummy row of 0s
+   STBI__F_avg_first
 };
 
-static stbi_uc first_row_filter[5] =
-{
+static stbi_uc first_row_filter[5] = {
    STBI__F_none,
    STBI__F_sub,
    STBI__F_none,
    STBI__F_avg_first,
-   STBI__F_paeth_first
+   STBI__F_sub // Paeth with b=c=0 turns out to be equivalent to sub
 };
 
 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;
+   // This formulation looks very different from the reference in the PNG spec, but is
+   // actually equivalent and has favorable data dependencies and admits straightforward
+   // generation of branch-free code, which helps performance significantly.
+   int thresh = c * 3 - (a + b);
+   int lo = a < b ? a : b;
+   int hi = a < b ? b : a;
+   int t0 = (hi <= thresh) ? lo : c;
+   int t1 = (thresh <= lo) ? hi : t0;
+   return t1;
 }
 
-static stbi_uc stbi__depth_scale_table[9] = { 0, 0xff, 0x55, 0, 0x11, 0,0,0, 0x01 };
+static const stbi_uc stbi__depth_scale_table[9] = { 0, 0xff, 0x55, 0, 0x11, 0, 0, 0, 0x01 };
+
+// adds an extra all-255 alpha channel
+// dest == src is legal
+// img_n must be 1 or 3
+static void stbi__create_png_alpha_expand8(stbi_uc* dest, stbi_uc* src, stbi__uint32 x, int img_n)
+{
+   int i;
+   // must process data backwards since we allow dest==src
+   if (img_n == 1)
+   {
+      for (i = x - 1; i >= 0; --i)
+      {
+         dest[i * 2 + 1] = 255;
+         dest[i * 2 + 0] = src[i];
+      }
+   }
+   else
+   {
+      STBI_ASSERT(img_n == 3);
+      for (i = x - 1; i >= 0; --i)
+      {
+         dest[i * 4 + 3] = 255;
+         dest[i * 4 + 2] = src[i * 3 + 2];
+         dest[i * 4 + 1] = src[i * 3 + 1];
+         dest[i * 4 + 0] = src[i * 3 + 0];
+      }
+   }
+}
 
 // create the png data from post-deflated data
-static int stbi__create_png_image_raw(stbi__png *a, stbi_uc *raw, stbi__uint32 raw_len, int out_n, stbi__uint32 x, stbi__uint32 y, int depth, int color)
+static int stbi__create_png_image_raw(stbi__png* a, stbi_uc* raw, stbi__uint32 raw_len, int out_n, stbi__uint32 x, stbi__uint32 y, int depth, int color)
 {
-   stbi__context *s = a->s;
-   stbi__uint32 i,j,stride = x*out_n;
+   int bytes = (depth == 16 ? 2 : 1);
+   stbi__context* s = a->s;
+   stbi__uint32 i, j, stride = x * out_n * bytes;
    stbi__uint32 img_len, img_width_bytes;
+   stbi_uc* filter_buf;
+   int all_ok = 1;
    int k;
    int img_n = s->img_n; // copy it into a local for later
 
-   STBI_ASSERT(out_n == s->img_n || out_n == s->img_n+1);
-   a->out = (stbi_uc *) stbi__malloc(x * y * out_n); // extra bytes to write off the end into
-   if (!a->out) return stbi__err("outofmem", "Out of memory");
+   int output_bytes = out_n * bytes;
+   int filter_bytes = img_n * bytes;
+   int width = x;
+
+   STBI_ASSERT(out_n == s->img_n || out_n == s->img_n + 1);
+   a->out = (stbi_uc*)stbi__malloc_mad3(x, y, output_bytes, 0); // extra bytes to write off the end into
+   if (!a->out)
+      return stbi__err("outofmem", "Out of memory");
 
+   // note: error exits here don't need to clean up a->out individually,
+   // stbi__do_png always does on error.
+   if (!stbi__mad3sizes_valid(img_n, x, depth, 7))
+      return stbi__err("too large", "Corrupt PNG");
    img_width_bytes = (((img_n * x * depth) + 7) >> 3);
+   if (!stbi__mad2sizes_valid(img_width_bytes, y, img_width_bytes))
+      return stbi__err("too large", "Corrupt PNG");
    img_len = (img_width_bytes + 1) * y;
-   if (s->img_x == x && s->img_y == y) {
-      if (raw_len != img_len) return stbi__err("not enough pixels","Corrupt PNG");
-   } else { // interlaced:
-      if (raw_len < img_len) return stbi__err("not enough pixels","Corrupt PNG");
+
+   // we used to check for exact match between raw_len and img_len on non-interlaced PNGs,
+   // but issue #276 reported a PNG in the wild that had extra data at the end (all zeros),
+   // so just check for raw_len < img_len always.
+   if (raw_len < img_len)
+      return stbi__err("not enough pixels", "Corrupt PNG");
+
+   // Allocate two scan lines worth of filter workspace buffer.
+   filter_buf = (stbi_uc*)stbi__malloc_mad2(img_width_bytes, 2, 0);
+   if (!filter_buf)
+      return stbi__err("outofmem", "Out of memory");
+
+   // Filtering for low-bit-depth images
+   if (depth < 8)
+   {
+      filter_bytes = 1;
+      width = img_width_bytes;
    }
 
-   for (j=0; j < y; ++j) {
-      stbi_uc *cur = a->out + stride*j;
-      stbi_uc *prior = cur - stride;
+   for (j = 0; j < y; ++j)
+   {
+      // cur/prior filter buffers alternate
+      stbi_uc* cur = filter_buf + (j & 1) * img_width_bytes;
+      stbi_uc* prior = filter_buf + (~j & 1) * img_width_bytes;
+      stbi_uc* dest = a->out + stride * j;
+      int nk = width * filter_bytes;
       int filter = *raw++;
-      int filter_bytes = img_n;
-      int width = x;
-      if (filter > 4)
-         return stbi__err("invalid filter","Corrupt PNG");
 
-      if (depth < 8) {
-         STBI_ASSERT(img_width_bytes <= x);
-         cur += x*out_n - img_width_bytes; // store output to the rightmost img_len bytes, so we can decode in place
-         filter_bytes = 1;
-         width = img_width_bytes;
+      // check filter type
+      if (filter > 4)
+      {
+         all_ok = stbi__err("invalid filter", "Corrupt PNG");
+         break;
       }
 
       // if first row, use special filter that doesn't sample previous row
-      if (j == 0) filter = first_row_filter[filter];
-
-      // handle first byte explicitly
-      for (k=0; k < filter_bytes; ++k) {
-         switch (filter) {
-            case STBI__F_none       : cur[k] = raw[k]; break;
-            case STBI__F_sub        : cur[k] = raw[k]; break;
-            case STBI__F_up         : cur[k] = STBI__BYTECAST(raw[k] + prior[k]); break;
-            case STBI__F_avg        : cur[k] = STBI__BYTECAST(raw[k] + (prior[k]>>1)); break;
-            case STBI__F_paeth      : cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(0,prior[k],0)); break;
-            case STBI__F_avg_first  : cur[k] = raw[k]; break;
-            case STBI__F_paeth_first: cur[k] = raw[k]; break;
-         }
-      }
+      if (j == 0)
+         filter = first_row_filter[filter];
 
-      if (depth == 8) {
-         if (img_n != out_n)
-            cur[img_n] = 255; // first pixel
-         raw += img_n;
-         cur += out_n;
-         prior += out_n;
-      } else {
-         raw += 1;
-         cur += 1;
-         prior += 1;
-      }
-
-      // this is a little gross, so that we don't switch per-pixel or per-component
-      if (depth < 8 || img_n == out_n) {
-         int nk = (width - 1)*img_n;
-         #define CASE(f) \
-             case f:     \
-                for (k=0; k < nk; ++k)
-         switch (filter) {
-            // "none" filter turns into a memcpy here; make that explicit.
-            case STBI__F_none:         memcpy(cur, raw, nk); break;
-            CASE(STBI__F_sub)          cur[k] = STBI__BYTECAST(raw[k] + cur[k-filter_bytes]); break;
-            CASE(STBI__F_up)           cur[k] = STBI__BYTECAST(raw[k] + prior[k]); break;
-            CASE(STBI__F_avg)          cur[k] = STBI__BYTECAST(raw[k] + ((prior[k] + cur[k-filter_bytes])>>1)); break;
-            CASE(STBI__F_paeth)        cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k-filter_bytes],prior[k],prior[k-filter_bytes])); break;
-            CASE(STBI__F_avg_first)    cur[k] = STBI__BYTECAST(raw[k] + (cur[k-filter_bytes] >> 1)); break;
-            CASE(STBI__F_paeth_first)  cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k-filter_bytes],0,0)); break;
-         }
-         #undef CASE
-         raw += nk;
-      } else {
-         STBI_ASSERT(img_n+1 == out_n);
-         #define CASE(f) \
-             case f:     \
-                for (i=x-1; i >= 1; --i, cur[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
+      // perform actual filtering
+      switch (filter)
+      {
+         case STBI__F_none:
+            memcpy(cur, raw, nk);
+            break;
+         case STBI__F_sub:
+            memcpy(cur, raw, filter_bytes);
+            for (k = filter_bytes; k < nk; ++k)
+               cur[k] = STBI__BYTECAST(raw[k] + cur[k - filter_bytes]);
+            break;
+         case STBI__F_up:
+            for (k = 0; k < nk; ++k)
+               cur[k] = STBI__BYTECAST(raw[k] + prior[k]);
+            break;
+         case STBI__F_avg:
+            for (k = 0; k < filter_bytes; ++k)
+               cur[k] = STBI__BYTECAST(raw[k] + (prior[k] >> 1));
+            for (k = filter_bytes; k < nk; ++k)
+               cur[k] = STBI__BYTECAST(raw[k] + ((prior[k] + cur[k - filter_bytes]) >> 1));
+            break;
+         case STBI__F_paeth:
+            for (k = 0; k < filter_bytes; ++k)
+               cur[k] = STBI__BYTECAST(raw[k] + prior[k]); // prior[k] == stbi__paeth(0,prior[k],0)
+            for (k = filter_bytes; k < nk; ++k)
+               cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k - filter_bytes], prior[k], prior[k - filter_bytes]));
+            break;
+         case STBI__F_avg_first:
+            memcpy(cur, raw, filter_bytes);
+            for (k = filter_bytes; k < nk; ++k)
+               cur[k] = STBI__BYTECAST(raw[k] + (cur[k - filter_bytes] >> 1));
+            break;
       }
-   }
 
-   // we make a separate pass to expand bits to pixels; for performance,
-   // this could run two scanlines behind the above code, so it won't
-   // intefere with filtering but will still be in the cache.
-   if (depth < 8) {
-      for (j=0; j < y; ++j) {
-         stbi_uc *cur = a->out + stride*j;
-         stbi_uc *in  = a->out + stride*j + x*out_n - img_width_bytes;
-         // unpack 1/2/4-bit into a 8-bit buffer. allows us to keep the common 8-bit path optimal at minimal cost for 1/2/4-bit
-         // png guarante byte alignment, if width is not multiple of 8/4/2 we'll decode dummy trailing data that will be skipped in the later loop
-         stbi_uc scale = (color == 0) ? stbi__depth_scale_table[depth] : 1; // scale grayscale values to 0..255 range
+      raw += nk;
 
-         // note that the final byte might overshoot and write more data than desired.
-         // we can allocate enough data that this never writes out of memory, but it
-         // could also overwrite the next scanline. can it overwrite non-empty data
-         // on the next scanline? yes, consider 1-pixel-wide scanlines with 1-bit-per-pixel.
-         // so we need to explicitly clamp the final ones
+      // expand decoded bits in cur to dest, also adding an extra alpha channel if desired
+      if (depth < 8)
+      {
+         stbi_uc scale = (color == 0) ? stbi__depth_scale_table[depth] : 1; // scale grayscale values to 0..255 range
+         stbi_uc* in = cur;
+         stbi_uc* out = dest;
+         stbi_uc inb = 0;
+         stbi__uint32 nsmp = x * img_n;
 
-         if (depth == 4) {
-            for (k=x*img_n; k >= 2; k-=2, ++in) {
-               *cur++ = scale * ((*in >> 4)       );
-               *cur++ = scale * ((*in     ) & 0x0f);
+         // expand bits to bytes first
+         if (depth == 4)
+         {
+            for (i = 0; i < nsmp; ++i)
+            {
+               if ((i & 1) == 0)
+                  inb = *in++;
+               *out++ = scale * (inb >> 4);
+               inb <<= 4;
             }
-            if (k > 0) *cur++ = scale * ((*in >> 4)       );
-         } else if (depth == 2) {
-            for (k=x*img_n; k >= 4; k-=4, ++in) {
-               *cur++ = scale * ((*in >> 6)       );
-               *cur++ = scale * ((*in >> 4) & 0x03);
-               *cur++ = scale * ((*in >> 2) & 0x03);
-               *cur++ = scale * ((*in     ) & 0x03);
+         }
+         else if (depth == 2)
+         {
+            for (i = 0; i < nsmp; ++i)
+            {
+               if ((i & 3) == 0)
+                  inb = *in++;
+               *out++ = scale * (inb >> 6);
+               inb <<= 2;
             }
-            if (k > 0) *cur++ = scale * ((*in >> 6)       );
-            if (k > 1) *cur++ = scale * ((*in >> 4) & 0x03);
-            if (k > 2) *cur++ = scale * ((*in >> 2) & 0x03);
-         } else if (depth == 1) {
-            for (k=x*img_n; k >= 8; k-=8, ++in) {
-               *cur++ = scale * ((*in >> 7)       );
-               *cur++ = scale * ((*in >> 6) & 0x01);
-               *cur++ = scale * ((*in >> 5) & 0x01);
-               *cur++ = scale * ((*in >> 4) & 0x01);
-               *cur++ = scale * ((*in >> 3) & 0x01);
-               *cur++ = scale * ((*in >> 2) & 0x01);
-               *cur++ = scale * ((*in >> 1) & 0x01);
-               *cur++ = scale * ((*in     ) & 0x01);
+         }
+         else
+         {
+            STBI_ASSERT(depth == 1);
+            for (i = 0; i < nsmp; ++i)
+            {
+               if ((i & 7) == 0)
+                  inb = *in++;
+               *out++ = scale * (inb >> 7);
+               inb <<= 1;
             }
-            if (k > 0) *cur++ = scale * ((*in >> 7)       );
-            if (k > 1) *cur++ = scale * ((*in >> 6) & 0x01);
-            if (k > 2) *cur++ = scale * ((*in >> 5) & 0x01);
-            if (k > 3) *cur++ = scale * ((*in >> 4) & 0x01);
-            if (k > 4) *cur++ = scale * ((*in >> 3) & 0x01);
-            if (k > 5) *cur++ = scale * ((*in >> 2) & 0x01);
-            if (k > 6) *cur++ = scale * ((*in >> 1) & 0x01);
          }
-         if (img_n != out_n) {
-            int q;
-            // insert alpha = 255
-            cur = a->out + stride*j;
-            if (img_n == 1) {
-               for (q=x-1; q >= 0; --q) {
-                  cur[q*2+1] = 255;
-                  cur[q*2+0] = cur[q];
+
+         // insert alpha=255 values if desired
+         if (img_n != out_n)
+            stbi__create_png_alpha_expand8(dest, dest, x, img_n);
+      }
+      else if (depth == 8)
+      {
+         if (img_n == out_n)
+            memcpy(dest, cur, x * img_n);
+         else
+            stbi__create_png_alpha_expand8(dest, cur, x, img_n);
+      }
+      else if (depth == 16)
+      {
+         // convert the image data from big-endian to platform-native
+         stbi__uint16* dest16 = (stbi__uint16*)dest;
+         stbi__uint32 nsmp = x * img_n;
+
+         if (img_n == out_n)
+         {
+            for (i = 0; i < nsmp; ++i, ++dest16, cur += 2)
+               *dest16 = (cur[0] << 8) | cur[1];
+         }
+         else
+         {
+            STBI_ASSERT(img_n + 1 == out_n);
+            if (img_n == 1)
+            {
+               for (i = 0; i < x; ++i, dest16 += 2, cur += 2)
+               {
+                  dest16[0] = (cur[0] << 8) | cur[1];
+                  dest16[1] = 0xffff;
                }
-            } else {
+            }
+            else
+            {
                STBI_ASSERT(img_n == 3);
-               for (q=x-1; q >= 0; --q) {
-                  cur[q*4+3] = 255;
-                  cur[q*4+2] = cur[q*3+2];
-                  cur[q*4+1] = cur[q*3+1];
-                  cur[q*4+0] = cur[q*3+0];
+               for (i = 0; i < x; ++i, dest16 += 4, cur += 6)
+               {
+                  dest16[0] = (cur[0] << 8) | cur[1];
+                  dest16[1] = (cur[2] << 8) | cur[3];
+                  dest16[2] = (cur[4] << 8) | cur[5];
+                  dest16[3] = 0xffff;
                }
             }
          }
       }
    }
 
+   STBI_FREE(filter_buf);
+   if (!all_ok)
+      return 0;
+
    return 1;
 }
 
-static int stbi__create_png_image(stbi__png *a, stbi_uc *image_data, stbi__uint32 image_data_len, int out_n, int depth, int color, int interlaced)
+static int stbi__create_png_image(stbi__png* a, stbi_uc* image_data, stbi__uint32 image_data_len, int out_n, int depth, int color, int interlaced)
 {
-   stbi_uc *final;
+   int bytes = (depth == 16 ? 2 : 1);
+   int out_bytes = out_n * bytes;
+   stbi_uc* final;
    int p;
    if (!interlaced)
       return stbi__create_png_image_raw(a, image_data, image_data_len, out_n, a->s->img_x, a->s->img_y, depth, color);
 
    // de-interlacing
-   final = (stbi_uc *) stbi__malloc(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;
+   final = (stbi_uc*)stbi__malloc_mad3(a->s->img_x, a->s->img_y, out_bytes, 0);
+   if (!final)
+      return stbi__err("outofmem", "Out of memory");
+   for (p = 0; p < 7; ++p)
+   {
+      int xorig[] = { 0, 4, 0, 2, 0, 1, 0 };
+      int yorig[] = { 0, 0, 4, 0, 2, 0, 1 };
+      int xspc[] = { 8, 8, 4, 4, 2, 2, 1 };
+      int yspc[] = { 8, 8, 8, 4, 4, 2, 2 };
+      int i, j, x, y;
       // pass1_x[4] = 0, pass1_x[5] = 1, pass1_x[12] = 1
-      x = (a->s->img_x - xorig[p] + xspc[p]-1) / xspc[p];
-      y = (a->s->img_y - yorig[p] + yspc[p]-1) / yspc[p];
-      if (x && y) {
+      x = (a->s->img_x - xorig[p] + xspc[p] - 1) / xspc[p];
+      y = (a->s->img_y - yorig[p] + yspc[p] - 1) / yspc[p];
+      if (x && y)
+      {
          stbi__uint32 img_len = ((((a->s->img_n * x * depth) + 7) >> 3) + 1) * y;
-         if (!stbi__create_png_image_raw(a, image_data, image_data_len, out_n, x, y, depth, color)) {
+         if (!stbi__create_png_image_raw(a, image_data, image_data_len, out_n, x, y, depth, color))
+         {
             STBI_FREE(final);
             return 0;
          }
-         for (j=0; j < y; ++j) {
-            for (i=0; i < x; ++i) {
-               int out_y = j*yspc[p]+yorig[p];
-               int out_x = i*xspc[p]+xorig[p];
-               memcpy(final + out_y*a->s->img_x*out_n + out_x*out_n,
-                      a->out + (j*x+i)*out_n, out_n);
+         for (j = 0; j < y; ++j)
+         {
+            for (i = 0; i < x; ++i)
+            {
+               int out_y = j * yspc[p] + yorig[p];
+               int out_x = i * xspc[p] + xorig[p];
+               memcpy(final + out_y * a->s->img_x * out_bytes + out_x * out_bytes,
+                      a->out + (j * x + i) * out_bytes, out_bytes);
             }
          }
          STBI_FREE(a->out);
@@ -4197,23 +5638,28 @@ static int stbi__create_png_image(stbi__png *a, stbi_uc *image_data, stbi__uint3
    return 1;
 }
 
-static int stbi__compute_transparency(stbi__png *z, stbi_uc tc[3], int out_n)
+static int stbi__compute_transparency(stbi__png* z, stbi_uc tc[3], int out_n)
 {
-   stbi__context *s = z->s;
+   stbi__context* s = z->s;
    stbi__uint32 i, pixel_count = s->img_x * s->img_y;
-   stbi_uc *p = z->out;
+   stbi_uc* p = z->out;
 
    // compute color-based transparency, assuming we've
    // already got 255 as the alpha value in the output
    STBI_ASSERT(out_n == 2 || out_n == 4);
 
-   if (out_n == 2) {
-      for (i=0; i < pixel_count; ++i) {
+   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) {
+   }
+   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;
@@ -4222,32 +5668,68 @@ static int stbi__compute_transparency(stbi__png *z, stbi_uc tc[3], int out_n)
    return 1;
 }
 
-static int stbi__expand_png_palette(stbi__png *a, stbi_uc *palette, int len, int pal_img_n)
+static int stbi__compute_transparency16(stbi__png* z, stbi__uint16 tc[3], int out_n)
+{
+   stbi__context* s = z->s;
+   stbi__uint32 i, pixel_count = s->img_x * s->img_y;
+   stbi__uint16* p = (stbi__uint16*)z->out;
+
+   // compute color-based transparency, assuming we've
+   // already got 65535 as the alpha value in the output
+   STBI_ASSERT(out_n == 2 || out_n == 4);
+
+   if (out_n == 2)
+   {
+      for (i = 0; i < pixel_count; ++i)
+      {
+         p[1] = (p[0] == tc[0] ? 0 : 65535);
+         p += 2;
+      }
+   }
+   else
+   {
+      for (i = 0; i < pixel_count; ++i)
+      {
+         if (p[0] == tc[0] && p[1] == tc[1] && p[2] == tc[2])
+            p[3] = 0;
+         p += 4;
+      }
+   }
+   return 1;
+}
+
+static int stbi__expand_png_palette(stbi__png* a, stbi_uc* palette, int len, int pal_img_n)
 {
    stbi__uint32 i, pixel_count = a->s->img_x * a->s->img_y;
    stbi_uc *p, *temp_out, *orig = a->out;
 
-   p = (stbi_uc *) stbi__malloc(pixel_count * pal_img_n);
-   if (p == NULL) return stbi__err("outofmem", "Out of memory");
+   p = (stbi_uc*)stbi__malloc_mad2(pixel_count, pal_img_n, 0);
+   if (p == NULL)
+      return stbi__err("outofmem", "Out of memory");
 
    // between here and free(out) below, exitting would leak
    temp_out = p;
 
-   if (pal_img_n == 3) {
-      for (i=0; i < pixel_count; ++i) {
-         int n = orig[i]*4;
-         p[0] = palette[n  ];
-         p[1] = palette[n+1];
-         p[2] = palette[n+2];
+   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];
+   }
+   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;
       }
    }
@@ -4259,52 +5741,92 @@ static int stbi__expand_png_palette(stbi__png *a, stbi_uc *palette, int len, int
    return 1;
 }
 
-static int stbi__unpremultiply_on_load = 0;
-static int stbi__de_iphone_flag = 0;
+static int stbi__unpremultiply_on_load_global = 0;
+static int stbi__de_iphone_flag_global = 0;
 
 STBIDEF void stbi_set_unpremultiply_on_load(int flag_true_if_should_unpremultiply)
 {
-   stbi__unpremultiply_on_load = flag_true_if_should_unpremultiply;
+   stbi__unpremultiply_on_load_global = flag_true_if_should_unpremultiply;
 }
 
 STBIDEF void stbi_convert_iphone_png_to_rgb(int flag_true_if_should_convert)
 {
-   stbi__de_iphone_flag = flag_true_if_should_convert;
+   stbi__de_iphone_flag_global = flag_true_if_should_convert;
+}
+
+#ifndef STBI_THREAD_LOCAL
+#define stbi__unpremultiply_on_load stbi__unpremultiply_on_load_global
+#define stbi__de_iphone_flag stbi__de_iphone_flag_global
+#else
+static STBI_THREAD_LOCAL int stbi__unpremultiply_on_load_local, stbi__unpremultiply_on_load_set;
+static STBI_THREAD_LOCAL int stbi__de_iphone_flag_local, stbi__de_iphone_flag_set;
+
+STBIDEF void stbi_set_unpremultiply_on_load_thread(int flag_true_if_should_unpremultiply)
+{
+   stbi__unpremultiply_on_load_local = flag_true_if_should_unpremultiply;
+   stbi__unpremultiply_on_load_set = 1;
+}
+
+STBIDEF void stbi_convert_iphone_png_to_rgb_thread(int flag_true_if_should_convert)
+{
+   stbi__de_iphone_flag_local = flag_true_if_should_convert;
+   stbi__de_iphone_flag_set = 1;
 }
 
-static void stbi__de_iphone(stbi__png *z)
+#define stbi__unpremultiply_on_load (stbi__unpremultiply_on_load_set     \
+                                     ? stbi__unpremultiply_on_load_local \
+                                     : stbi__unpremultiply_on_load_global)
+#define stbi__de_iphone_flag (stbi__de_iphone_flag_set     \
+                              ? stbi__de_iphone_flag_local \
+                              : stbi__de_iphone_flag_global)
+#endif // STBI_THREAD_LOCAL
+
+static void stbi__de_iphone(stbi__png* z)
 {
-   stbi__context *s = z->s;
+   stbi__context* s = z->s;
    stbi__uint32 i, pixel_count = s->img_x * s->img_y;
-   stbi_uc *p = z->out;
+   stbi_uc* p = z->out;
 
-   if (s->img_out_n == 3) {  // convert bgr to rgb
-      for (i=0; i < pixel_count; ++i) {
+   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 {
+   }
+   else
+   {
       STBI_ASSERT(s->img_out_n == 4);
-      if (stbi__unpremultiply_on_load) {
+      if (stbi__unpremultiply_on_load)
+      {
          // convert bgr to rgb and unpremultiply
-         for (i=0; i < pixel_count; ++i) {
+         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 {
+            if (a)
+            {
+               stbi_uc half = a / 2;
+               p[0] = (p[2] * 255 + half) / a;
+               p[1] = (p[1] * 255 + half) / a;
+               p[2] = (t * 255 + half) / a;
+            }
+            else
+            {
                p[0] = p[2];
                p[2] = t;
             }
             p += 4;
          }
-      } else {
+      }
+      else
+      {
          // convert bgr to rgb
-         for (i=0; i < pixel_count; ++i) {
+         for (i = 0; i < pixel_count; ++i)
+         {
             stbi_uc t = p[0];
             p[0] = p[2];
             p[2] = t;
@@ -4314,157 +5836,269 @@ static void stbi__de_iphone(stbi__png *z)
    }
 }
 
-#define STBI__PNG_TYPE(a,b,c,d)  (((a) << 24) + ((b) << 16) + ((c) << 8) + (d))
+#define STBI__PNG_TYPE(a, b, c, d) (((unsigned)(a) << 24) + ((unsigned)(b) << 16) + ((unsigned)(c) << 8) + (unsigned)(d))
 
-static int stbi__parse_png_file(stbi__png *z, int scan, int req_comp)
+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, color=0, depth=0, is_iphone=0;
-   stbi__context *s = z->s;
+   stbi_uc palette[1024], pal_img_n = 0;
+   stbi_uc has_trans = 0, tc[3] = { 0 };
+   stbi__uint16 tc16[3];
+   stbi__uint32 ioff = 0, idata_limit = 0, i, pal_len = 0;
+   int first = 1, k, interlace = 0, color = 0, is_iphone = 0;
+   stbi__context* s = z->s;
 
    z->expanded = NULL;
    z->idata = NULL;
    z->out = NULL;
 
-   if (!stbi__check_png_header(s)) return 0;
+   if (!stbi__check_png_header(s))
+      return 0;
 
-   if (scan == STBI__SCAN_type) return 1;
+   if (scan == STBI__SCAN_type)
+      return 1;
 
-   for (;;) {
+   for (;;)
+   {
       stbi__pngchunk c = stbi__get_chunk_header(s);
-      switch (c.type) {
-         case STBI__PNG_TYPE('C','g','B','I'):
+      switch (c.type)
+      {
+         case STBI__PNG_TYPE('C', 'g', 'B', 'I'):
             is_iphone = 1;
             stbi__skip(s, c.length);
             break;
-         case STBI__PNG_TYPE('I','H','D','R'): {
-            int comp,filter;
-            if (!first) return stbi__err("multiple IHDR","Corrupt PNG");
+         case STBI__PNG_TYPE('I', 'H', 'D', 'R'):
+         {
+            int comp, filter;
+            if (!first)
+               return stbi__err("multiple IHDR", "Corrupt PNG");
             first = 0;
-            if (c.length != 13) return stbi__err("bad IHDR len","Corrupt PNG");
-            s->img_x = stbi__get32be(s); if (s->img_x > (1 << 24)) return stbi__err("too large","Very large image (corrupt?)");
-            s->img_y = stbi__get32be(s); if (s->img_y > (1 << 24)) return stbi__err("too large","Very large image (corrupt?)");
-            depth = stbi__get8(s);  if (depth != 1 && depth != 2 && depth != 4 && depth != 8)  return stbi__err("1/2/4/8-bit only","PNG not supported: 1/2/4/8-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) {
+            if (c.length != 13)
+               return stbi__err("bad IHDR len", "Corrupt PNG");
+            s->img_x = stbi__get32be(s);
+            s->img_y = stbi__get32be(s);
+            if (s->img_y > STBI_MAX_DIMENSIONS)
+               return stbi__err("too large", "Very large image (corrupt?)");
+            if (s->img_x > STBI_MAX_DIMENSIONS)
+               return stbi__err("too large", "Very large image (corrupt?)");
+            z->depth = stbi__get8(s);
+            if (z->depth != 1 && z->depth != 2 && z->depth != 4 && z->depth != 8 && z->depth != 16)
+               return stbi__err("1/2/4/8/16-bit only", "PNG not supported: 1/2/4/8/16-bit only");
+            color = stbi__get8(s);
+            if (color > 6)
+               return stbi__err("bad ctype", "Corrupt PNG");
+            if (color == 3 && z->depth == 16)
+               return stbi__err("bad ctype", "Corrupt PNG");
+            if (color == 3)
+               pal_img_n = 3;
+            else if (color & 1)
+               return stbi__err("bad ctype", "Corrupt PNG");
+            comp = stbi__get8(s);
+            if (comp)
+               return stbi__err("bad comp method", "Corrupt PNG");
+            filter = stbi__get8(s);
+            if (filter)
+               return stbi__err("bad filter method", "Corrupt PNG");
+            interlace = stbi__get8(s);
+            if (interlace > 1)
+               return stbi__err("bad interlace method", "Corrupt PNG");
+            if (!s->img_x || !s->img_y)
+               return stbi__err("0-pixel image", "Corrupt PNG");
+            if (!pal_img_n)
+            {
                s->img_n = (color & 2 ? 3 : 1) + (color & 4 ? 1 : 0);
-               if ((1 << 30) / s->img_x / s->img_n < s->img_y) return stbi__err("too large", "Image too large to decode");
-               if (scan == STBI__SCAN_header) return 1;
-            } else {
+               if ((1 << 30) / s->img_x / s->img_n < s->img_y)
+                  return stbi__err("too large", "Image too large to decode");
+            }
+            else
+            {
                // if paletted, then pal_n is our final components, and
                // img_n is # components to decompress/filter.
                s->img_n = 1;
-               if ((1 << 30) / s->img_x / 4 < s->img_y) return stbi__err("too large","Corrupt PNG");
-               // if SCAN_header, have to scan to see if we have a tRNS
+               if ((1 << 30) / s->img_x / 4 < s->img_y)
+                  return stbi__err("too large", "Corrupt PNG");
             }
+            // even with SCAN_header, have to scan to see if we have a tRNS
             break;
          }
 
-         case STBI__PNG_TYPE('P','L','T','E'):  {
-            if (first) return stbi__err("first not IHDR", "Corrupt PNG");
-            if (c.length > 256*3) return stbi__err("invalid PLTE","Corrupt PNG");
+         case STBI__PNG_TYPE('P', 'L', 'T', 'E'):
+         {
+            if (first)
+               return stbi__err("first not IHDR", "Corrupt PNG");
+            if (c.length > 256 * 3)
+               return stbi__err("invalid PLTE", "Corrupt PNG");
             pal_len = c.length / 3;
-            if (pal_len * 3 != c.length) return stbi__err("invalid PLTE","Corrupt PNG");
-            for (i=0; i < pal_len; ++i) {
-               palette[i*4+0] = stbi__get8(s);
-               palette[i*4+1] = stbi__get8(s);
-               palette[i*4+2] = stbi__get8(s);
-               palette[i*4+3] = 255;
+            if (pal_len * 3 != c.length)
+               return stbi__err("invalid PLTE", "Corrupt PNG");
+            for (i = 0; i < pal_len; ++i)
+            {
+               palette[i * 4 + 0] = stbi__get8(s);
+               palette[i * 4 + 1] = stbi__get8(s);
+               palette[i * 4 + 2] = stbi__get8(s);
+               palette[i * 4 + 3] = 255;
             }
             break;
          }
 
-         case STBI__PNG_TYPE('t','R','N','S'): {
-            if (first) return stbi__err("first not IHDR", "Corrupt PNG");
-            if (z->idata) return stbi__err("tRNS after IDAT","Corrupt PNG");
-            if (pal_img_n) {
-               if (scan == STBI__SCAN_header) { s->img_n = 4; return 1; }
-               if (pal_len == 0) return stbi__err("tRNS before PLTE","Corrupt PNG");
-               if (c.length > pal_len) return stbi__err("bad tRNS len","Corrupt PNG");
+         case STBI__PNG_TYPE('t', 'R', 'N', 'S'):
+         {
+            if (first)
+               return stbi__err("first not IHDR", "Corrupt PNG");
+            if (z->idata)
+               return stbi__err("tRNS after IDAT", "Corrupt PNG");
+            if (pal_img_n)
+            {
+               if (scan == STBI__SCAN_header)
+               {
+                  s->img_n = 4;
+                  return 1;
+               }
+               if (pal_len == 0)
+                  return stbi__err("tRNS before PLTE", "Corrupt PNG");
+               if (c.length > pal_len)
+                  return stbi__err("bad tRNS len", "Corrupt PNG");
                pal_img_n = 4;
-               for (i=0; i < c.length; ++i)
-                  palette[i*4+3] = stbi__get8(s);
-            } else {
-               if (!(s->img_n & 1)) return stbi__err("tRNS with alpha","Corrupt PNG");
-               if (c.length != (stbi__uint32) s->img_n*2) return stbi__err("bad tRNS len","Corrupt PNG");
+               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) * stbi__depth_scale_table[depth]; // non 8-bit images will be larger
+               // non-paletted with tRNS = constant alpha. if header-scanning, we can stop now.
+               if (scan == STBI__SCAN_header)
+               {
+                  ++s->img_n;
+                  return 1;
+               }
+               if (z->depth == 16)
+               {
+                  for (k = 0; k < s->img_n; ++k)
+                     tc16[k] = (stbi__uint16)stbi__get16be(s); // copy the values as-is
+               }
+               else
+               {
+                  for (k = 0; k < s->img_n; ++k)
+                     tc[k] = (stbi_uc)(stbi__get16be(s) & 255) * stbi__depth_scale_table[z->depth]; // non 8-bit images will be larger
+               }
             }
             break;
          }
 
-         case STBI__PNG_TYPE('I','D','A','T'): {
-            if (first) return stbi__err("first not IHDR", "Corrupt PNG");
-            if (pal_img_n && !pal_len) return stbi__err("no PLTE","Corrupt PNG");
-            if (scan == STBI__SCAN_header) { s->img_n = pal_img_n; return 1; }
-            if ((int)(ioff + c.length) < (int)ioff) return 0;
-            if (ioff + c.length > idata_limit) {
+         case STBI__PNG_TYPE('I', 'D', 'A', 'T'):
+         {
+            if (first)
+               return stbi__err("first not IHDR", "Corrupt PNG");
+            if (pal_img_n && !pal_len)
+               return stbi__err("no PLTE", "Corrupt PNG");
+            if (scan == STBI__SCAN_header)
+            {
+               // header scan definitely stops at first IDAT
+               if (pal_img_n)
+                  s->img_n = pal_img_n;
+               return 1;
+            }
+            if (c.length > (1u << 30))
+               return stbi__err("IDAT size limit", "IDAT section larger than 2^30 bytes");
+            if ((int)(ioff + c.length) < (int)ioff)
+               return 0;
+            if (ioff + c.length > idata_limit)
+            {
                stbi__uint32 idata_limit_old = idata_limit;
-               stbi_uc *p;
-               if (idata_limit == 0) idata_limit = c.length > 4096 ? c.length : 4096;
+               stbi_uc* p;
+               if (idata_limit == 0)
+                  idata_limit = c.length > 4096 ? c.length : 4096;
                while (ioff + c.length > idata_limit)
                   idata_limit *= 2;
                STBI_NOTUSED(idata_limit_old);
-               p = (stbi_uc *) STBI_REALLOC_SIZED(z->idata, idata_limit_old, idata_limit); if (p == NULL) return stbi__err("outofmem", "Out of memory");
+               p = (stbi_uc*)STBI_REALLOC_SIZED(z->idata, idata_limit_old, idata_limit);
+               if (p == NULL)
+                  return stbi__err("outofmem", "Out of memory");
                z->idata = p;
             }
-            if (!stbi__getn(s, z->idata+ioff,c.length)) return stbi__err("outofdata","Corrupt PNG");
+            if (!stbi__getn(s, z->idata + ioff, c.length))
+               return stbi__err("outofdata", "Corrupt PNG");
             ioff += c.length;
             break;
          }
 
-         case STBI__PNG_TYPE('I','E','N','D'): {
+         case STBI__PNG_TYPE('I', 'E', 'N', 'D'):
+         {
             stbi__uint32 raw_len, bpl;
-            if (first) return stbi__err("first not IHDR", "Corrupt PNG");
-            if (scan != STBI__SCAN_load) return 1;
-            if (z->idata == NULL) return stbi__err("no IDAT","Corrupt PNG");
+            if (first)
+               return stbi__err("first not IHDR", "Corrupt PNG");
+            if (scan != STBI__SCAN_load)
+               return 1;
+            if (z->idata == NULL)
+               return stbi__err("no IDAT", "Corrupt PNG");
             // initial guess for decoded data size to avoid unnecessary reallocs
-            bpl = (s->img_x * depth + 7) / 8; // bytes per line, per component
+            bpl = (s->img_x * z->depth + 7) / 8; // bytes per line, per component
             raw_len = bpl * s->img_y * s->img_n /* pixels */ + s->img_y /* filter mode per row */;
-            z->expanded = (stbi_uc *) stbi_zlib_decode_malloc_guesssize_headerflag((char *) z->idata, ioff, raw_len, (int *) &raw_len, !is_iphone);
-            if (z->expanded == NULL) return 0; // zlib should set error
-            STBI_FREE(z->idata); z->idata = NULL;
-            if ((req_comp == s->img_n+1 && req_comp != 3 && !pal_img_n) || has_trans)
-               s->img_out_n = s->img_n+1;
+            z->expanded = (stbi_uc*)stbi_zlib_decode_malloc_guesssize_headerflag((char*)z->idata, ioff, raw_len, (int*)&raw_len, !is_iphone);
+            if (z->expanded == NULL)
+               return 0; // zlib should set error
+            STBI_FREE(z->idata);
+            z->idata = NULL;
+            if ((req_comp == s->img_n + 1 && req_comp != 3 && !pal_img_n) || has_trans)
+               s->img_out_n = s->img_n + 1;
             else
                s->img_out_n = s->img_n;
-            if (!stbi__create_png_image(z, z->expanded, raw_len, s->img_out_n, depth, color, interlace)) return 0;
+            if (!stbi__create_png_image(z, z->expanded, raw_len, s->img_out_n, z->depth, color, interlace))
+               return 0;
             if (has_trans)
-               if (!stbi__compute_transparency(z, tc, s->img_out_n)) return 0;
+            {
+               if (z->depth == 16)
+               {
+                  if (!stbi__compute_transparency16(z, tc16, s->img_out_n))
+                     return 0;
+               }
+               else
+               {
+                  if (!stbi__compute_transparency(z, tc, s->img_out_n))
+                     return 0;
+               }
+            }
             if (is_iphone && stbi__de_iphone_flag && s->img_out_n > 2)
                stbi__de_iphone(z);
-            if (pal_img_n) {
+            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 (req_comp >= 3)
+                  s->img_out_n = req_comp;
                if (!stbi__expand_png_palette(z, palette, pal_len, s->img_out_n))
                   return 0;
             }
-            STBI_FREE(z->expanded); z->expanded = NULL;
+            else if (has_trans)
+            {
+               // non-paletted image with tRNS -> source image has (constant) alpha
+               ++s->img_n;
+            }
+            STBI_FREE(z->expanded);
+            z->expanded = NULL;
+            // end of PNG chunk, read and skip CRC
+            stbi__get32be(s);
             return 1;
          }
 
          default:
             // if critical, fail
-            if (first) return stbi__err("first not IHDR", "Corrupt PNG");
-            if ((c.type & (1 << 29)) == 0) {
-               #ifndef STBI_NO_FAILURE_STRINGS
+            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
+               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);
@@ -4475,37 +6109,54 @@ static int stbi__parse_png_file(stbi__png *z, int scan, int req_comp)
    }
 }
 
-static unsigned char *stbi__do_png(stbi__png *p, int *x, int *y, int *n, int req_comp)
+static void* stbi__do_png(stbi__png* p, int* x, int* y, int* n, int req_comp, stbi__result_info* ri)
 {
-   unsigned char *result=NULL;
-   if (req_comp < 0 || req_comp > 4) return stbi__errpuc("bad req_comp", "Internal error");
-   if (stbi__parse_png_file(p, STBI__SCAN_load, req_comp)) {
+   void* result = NULL;
+   if (req_comp < 0 || req_comp > 4)
+      return stbi__errpuc("bad req_comp", "Internal error");
+   if (stbi__parse_png_file(p, STBI__SCAN_load, req_comp))
+   {
+      if (p->depth <= 8)
+         ri->bits_per_channel = 8;
+      else if (p->depth == 16)
+         ri->bits_per_channel = 16;
+      else
+         return stbi__errpuc("bad bits_per_channel", "PNG not supported: unsupported color depth");
       result = p->out;
       p->out = NULL;
-      if (req_comp && req_comp != p->s->img_out_n) {
-         result = stbi__convert_format(result, p->s->img_out_n, req_comp, p->s->img_x, p->s->img_y);
+      if (req_comp && req_comp != p->s->img_out_n)
+      {
+         if (ri->bits_per_channel == 8)
+            result = stbi__convert_format((unsigned char*)result, p->s->img_out_n, req_comp, p->s->img_x, p->s->img_y);
+         else
+            result = stbi__convert_format16((stbi__uint16*)result, p->s->img_out_n, req_comp, p->s->img_x, p->s->img_y);
          p->s->img_out_n = req_comp;
-         if (result == NULL) return result;
+         if (result == NULL)
+            return result;
       }
       *x = p->s->img_x;
       *y = p->s->img_y;
-      if (n) *n = p->s->img_out_n;
+      if (n)
+         *n = p->s->img_n;
    }
-   STBI_FREE(p->out);      p->out      = NULL;
-   STBI_FREE(p->expanded); p->expanded = NULL;
-   STBI_FREE(p->idata);    p->idata    = NULL;
+   STBI_FREE(p->out);
+   p->out = NULL;
+   STBI_FREE(p->expanded);
+   p->expanded = NULL;
+   STBI_FREE(p->idata);
+   p->idata = NULL;
 
    return result;
 }
 
-static unsigned char *stbi__png_load(stbi__context *s, int *x, int *y, int *comp, int req_comp)
+static void* stbi__png_load(stbi__context* s, int* x, int* y, int* comp, int req_comp, stbi__result_info* ri)
 {
    stbi__png p;
    p.s = s;
-   return stbi__do_png(&p, x,y,comp,req_comp);
+   return stbi__do_png(&p, x, y, comp, req_comp, ri);
 }
 
-static int stbi__png_test(stbi__context *s)
+static int stbi__png_test(stbi__context* s)
 {
    int r;
    r = stbi__check_png_header(s);
@@ -4513,35 +6164,55 @@ static int stbi__png_test(stbi__context *s)
    return r;
 }
 
-static int stbi__png_info_raw(stbi__png *p, int *x, int *y, int *comp)
+static int stbi__png_info_raw(stbi__png* p, int* x, int* y, int* comp)
 {
-   if (!stbi__parse_png_file(p, STBI__SCAN_header, 0)) {
-      stbi__rewind( p->s );
+   if (!stbi__parse_png_file(p, STBI__SCAN_header, 0))
+   {
+      stbi__rewind(p->s);
       return 0;
    }
-   if (x) *x = p->s->img_x;
-   if (y) *y = p->s->img_y;
-   if (comp) *comp = p->s->img_n;
+   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)
+static int stbi__png_info(stbi__context* s, int* x, int* y, int* comp)
 {
    stbi__png p;
    p.s = s;
    return stbi__png_info_raw(&p, x, y, comp);
 }
+
+static int stbi__png_is16(stbi__context* s)
+{
+   stbi__png p;
+   p.s = s;
+   if (!stbi__png_info_raw(&p, NULL, NULL, NULL))
+      return 0;
+   if (p.depth != 16)
+   {
+      stbi__rewind(p.s);
+      return 0;
+   }
+   return 1;
+}
 #endif
 
 // Microsoft/Windows BMP image
 
 #ifndef STBI_NO_BMP
-static int stbi__bmp_test_raw(stbi__context *s)
+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;
+   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
@@ -4551,7 +6222,7 @@ static int stbi__bmp_test_raw(stbi__context *s)
    return r;
 }
 
-static int stbi__bmp_test(stbi__context *s)
+static int stbi__bmp_test(stbi__context* s)
 {
    int r = stbi__bmp_test_raw(s);
    stbi__rewind(s);
@@ -4562,112 +6233,202 @@ static int stbi__bmp_test(stbi__context *s)
 // 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; }
+   int n = 0;
+   if (z == 0)
+      return -1;
+   if (z >= 0x10000)
+   {
+      n += 16;
+      z >>= 16;
+   }
+   if (z >= 0x00100)
+   {
+      n += 8;
+      z >>= 8;
+   }
+   if (z >= 0x00010)
+   {
+      n += 4;
+      z >>= 4;
+   }
+   if (z >= 0x00004)
+   {
+      n += 2;
+      z >>= 2;
+   }
+   if (z >= 0x00002)
+   {
+      n += 1; /* >>=  1;*/
+   }
    return n;
 }
 
 static int stbi__bitcount(unsigned int a)
 {
-   a = (a & 0x55555555) + ((a >>  1) & 0x55555555); // max 2
-   a = (a & 0x33333333) + ((a >>  2) & 0x33333333); // max 4
+   a = (a & 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;
+// extract an arbitrarily-aligned N-bit value (N=bits)
+// from v, and then make it 8-bits long and fractionally
+// extend it to full full range.
+static int stbi__shiftsigned(unsigned int v, int shift, int bits)
+{
+   static unsigned int mul_table[9] = {
+      0,
+      0xff /*0b11111111*/,
+      0x55 /*0b01010101*/,
+      0x49 /*0b01001001*/,
+      0x11 /*0b00010001*/,
+      0x21 /*0b00100001*/,
+      0x41 /*0b01000001*/,
+      0x81 /*0b10000001*/,
+      0x01 /*0b00000001*/,
+   };
+   static unsigned int shift_table[9] = {
+      0,
+      0,
+      0,
+      1,
+      0,
+      2,
+      4,
+      6,
+      0,
+   };
+   if (shift < 0)
+      v <<= -shift;
+   else
+      v >>= shift;
+   STBI_ASSERT(v < 256);
+   v >>= (8 - bits);
+   STBI_ASSERT(bits >= 0 && bits <= 8);
+   return (int)((unsigned)v * mul_table[bits]) >> shift_table[bits];
 }
 
 typedef struct
 {
    int bpp, offset, hsz;
-   unsigned int mr,mg,mb,ma, all_a;
+   unsigned int mr, mg, mb, ma, all_a;
+   int extra_read;
 } stbi__bmp_data;
 
-static void *stbi__bmp_parse_header(stbi__context *s, stbi__bmp_data *info)
+static int stbi__bmp_set_mask_defaults(stbi__bmp_data* info, int compress)
+{
+   // BI_BITFIELDS specifies masks explicitly, don't override
+   if (compress == 3)
+      return 1;
+
+   if (compress == 0)
+   {
+      if (info->bpp == 16)
+      {
+         info->mr = 31u << 10;
+         info->mg = 31u << 5;
+         info->mb = 31u << 0;
+      }
+      else if (info->bpp == 32)
+      {
+         info->mr = 0xffu << 16;
+         info->mg = 0xffu << 8;
+         info->mb = 0xffu << 0;
+         info->ma = 0xffu << 24;
+         info->all_a = 0; // if all_a is 0 at end, then we loaded alpha channel but it was all 0
+      }
+      else
+      {
+         // otherwise, use defaults, which is all-0
+         info->mr = info->mg = info->mb = info->ma = 0;
+      }
+      return 1;
+   }
+   return 0; // error
+}
+
+static void* stbi__bmp_parse_header(stbi__context* s, stbi__bmp_data* info)
 {
    int hsz;
-   if (stbi__get8(s) != 'B' || stbi__get8(s) != 'M') return stbi__errpuc("not BMP", "Corrupt BMP");
+   if (stbi__get8(s) != 'B' || stbi__get8(s) != 'M')
+      return stbi__errpuc("not BMP", "Corrupt BMP");
    stbi__get32le(s); // discard filesize
    stbi__get16le(s); // discard reserved
    stbi__get16le(s); // discard reserved
    info->offset = stbi__get32le(s);
    info->hsz = hsz = stbi__get32le(s);
-   
-   if (hsz != 12 && hsz != 40 && hsz != 56 && hsz != 108 && hsz != 124) return stbi__errpuc("unknown BMP", "BMP type not supported: unknown");
-   if (hsz == 12) {
+   info->mr = info->mg = info->mb = info->ma = 0;
+   info->extra_read = 14;
+
+   if (info->offset < 0)
+      return stbi__errpuc("bad BMP", "bad BMP");
+
+   if (hsz != 12 && hsz != 40 && hsz != 56 && hsz != 108 && hsz != 124)
+      return stbi__errpuc("unknown BMP", "BMP type not supported: unknown");
+   if (hsz == 12)
+   {
       s->img_x = stbi__get16le(s);
       s->img_y = stbi__get16le(s);
-   } else {
+   }
+   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");
+   if (stbi__get16le(s) != 1)
+      return stbi__errpuc("bad BMP", "bad BMP");
    info->bpp = stbi__get16le(s);
-   if (info->bpp == 1) return stbi__errpuc("monochrome", "BMP type not supported: 1-bit");
-   if (hsz != 12) {
+   if (hsz != 12)
+   {
       int compress = stbi__get32le(s);
-      if (compress == 1 || compress == 2) return stbi__errpuc("BMP RLE", "BMP type not supported: RLE");
+      if (compress == 1 || compress == 2)
+         return stbi__errpuc("BMP RLE", "BMP type not supported: RLE");
+      if (compress >= 4)
+         return stbi__errpuc("BMP JPEG/PNG", "BMP type not supported: unsupported compression"); // this includes PNG/JPEG modes
+      if (compress == 3 && info->bpp != 16 && info->bpp != 32)
+         return stbi__errpuc("bad BMP", "bad BMP"); // bitfields requires 16 or 32 bits/pixel
       stbi__get32le(s); // discard sizeof
       stbi__get32le(s); // discard hres
       stbi__get32le(s); // discard vres
       stbi__get32le(s); // discard colorsused
       stbi__get32le(s); // discard max important
-      if (hsz == 40 || hsz == 56) {
-         if (hsz == 56) {
+      if (hsz == 40 || hsz == 56)
+      {
+         if (hsz == 56)
+         {
             stbi__get32le(s);
             stbi__get32le(s);
             stbi__get32le(s);
             stbi__get32le(s);
          }
-         if (info->bpp == 16 || info->bpp == 32) {
-            info->mr = info->mg = info->mb = 0;
-            if (compress == 0) {
-               if (info->bpp == 32) {
-                  info->mr = 0xffu << 16;
-                  info->mg = 0xffu <<  8;
-                  info->mb = 0xffu <<  0;
-                  info->ma = 0xffu << 24;
-                  info->all_a = 0; // if all_a is 0 at end, then we loaded alpha channel but it was all 0
-               } else {
-                  info->mr = 31u << 10;
-                  info->mg = 31u <<  5;
-                  info->mb = 31u <<  0;
-               }
-            } else if (compress == 3) {
+         if (info->bpp == 16 || info->bpp == 32)
+         {
+            if (compress == 0)
+            {
+               stbi__bmp_set_mask_defaults(info, compress);
+            }
+            else if (compress == 3)
+            {
                info->mr = stbi__get32le(s);
                info->mg = stbi__get32le(s);
                info->mb = stbi__get32le(s);
+               info->extra_read += 12;
                // not documented, but generated by photoshop and handled by mspaint
-               if (info->mr == info->mg && info->mg == info->mb) {
+               if (info->mr == info->mg && info->mg == info->mb)
+               {
                   // ?!?!?
                   return stbi__errpuc("bad BMP", "bad BMP");
                }
-            } else
+            }
+            else
                return stbi__errpuc("bad BMP", "bad BMP");
          }
-      } else {
+      }
+      else
+      {
+         // V4/V5 header
          int i;
          if (hsz != 108 && hsz != 124)
             return stbi__errpuc("bad BMP", "bad BMP");
@@ -4675,10 +6436,13 @@ static void *stbi__bmp_parse_header(stbi__context *s, stbi__bmp_data *info)
          info->mg = stbi__get32le(s);
          info->mb = stbi__get32le(s);
          info->ma = stbi__get32le(s);
+         if (compress != 3) // override mr/mg/mb unless in BI_BITFIELDS mode, as per docs
+            stbi__bmp_set_mask_defaults(info, compress);
          stbi__get32le(s); // discard color space
-         for (i=0; i < 12; ++i)
+         for (i = 0; i < 12; ++i)
             stbi__get32le(s); // discard color space parameters
-         if (hsz == 124) {
+         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
@@ -4686,25 +6450,31 @@ static void *stbi__bmp_parse_header(stbi__context *s, stbi__bmp_data *info)
          }
       }
    }
-   return (void *) 1;
+   return (void*)1;
 }
 
 
-static stbi_uc *stbi__bmp_load(stbi__context *s, int *x, int *y, int *comp, int req_comp)
+static void* stbi__bmp_load(stbi__context* s, int* x, int* y, int* comp, int req_comp, stbi__result_info* ri)
 {
-   stbi_uc *out;
-   unsigned int mr=0,mg=0,mb=0,ma=0, all_a;
+   stbi_uc* out;
+   unsigned int mr = 0, mg = 0, mb = 0, ma = 0, all_a;
    stbi_uc pal[256][4];
-   int psize=0,i,j,width;
+   int psize = 0, i, j, width;
    int flip_vertically, pad, target;
    stbi__bmp_data info;
+   STBI_NOTUSED(ri);
 
-   info.all_a = 255;   
+   info.all_a = 255;
    if (stbi__bmp_parse_header(s, &info) == NULL)
       return NULL; // error code already set
 
-   flip_vertically = ((int) s->img_y) > 0;
-   s->img_y = abs((int) s->img_y);
+   flip_vertically = ((int)s->img_y) > 0;
+   s->img_y = abs((int)s->img_y);
+
+   if (s->img_y > STBI_MAX_DIMENSIONS)
+      return stbi__errpuc("too large", "Very large image (corrupt?)");
+   if (s->img_x > STBI_MAX_DIMENSIONS)
+      return stbi__errpuc("too large", "Very large image (corrupt?)");
 
    mr = info.mr;
    mg = info.mg;
@@ -4712,133 +6482,254 @@ static stbi_uc *stbi__bmp_load(stbi__context *s, int *x, int *y, int *comp, int
    ma = info.ma;
    all_a = info.all_a;
 
-   if (info.hsz == 12) {
+   if (info.hsz == 12)
+   {
       if (info.bpp < 24)
-         psize = (info.offset - 14 - 24) / 3;
-   } else {
+         psize = (info.offset - info.extra_read - 24) / 3;
+   }
+   else
+   {
       if (info.bpp < 16)
-         psize = (info.offset - 14 - info.hsz) >> 2;
+         psize = (info.offset - info.extra_read - info.hsz) >> 2;
+   }
+   if (psize == 0)
+   {
+      // accept some number of extra bytes after the header, but if the offset points either to before
+      // the header ends or implies a large amount of extra data, reject the file as malformed
+      int bytes_read_so_far = s->callback_already_read + (int)(s->img_buffer - s->img_buffer_original);
+      int header_limit = 1024; // max we actually read is below 256 bytes currently.
+      int extra_data_limit = 256 * 4; // what ordinarily goes here is a palette; 256 entries*4 bytes is its max size.
+      if (bytes_read_so_far <= 0 || bytes_read_so_far > header_limit)
+      {
+         return stbi__errpuc("bad header", "Corrupt BMP");
+      }
+      // we established that bytes_read_so_far is positive and sensible.
+      // the first half of this test rejects offsets that are either too small positives, or
+      // negative, and guarantees that info.offset >= bytes_read_so_far > 0. this in turn
+      // ensures the number computed in the second half of the test can't overflow.
+      if (info.offset < bytes_read_so_far || info.offset - bytes_read_so_far > extra_data_limit)
+      {
+         return stbi__errpuc("bad offset", "Corrupt BMP");
+      }
+      else
+      {
+         stbi__skip(s, info.offset - bytes_read_so_far);
+      }
    }
 
-   s->img_n = ma ? 4 : 3;
+   if (info.bpp == 24 && ma == 0xff000000)
+      s->img_n = 3;
+   else
+      s->img_n = ma ? 4 : 3;
    if (req_comp && req_comp >= 3) // we can directly decode 3 or 4
       target = req_comp;
    else
       target = s->img_n; // if they want monochrome, we'll post-convert
 
-   out = (stbi_uc *) stbi__malloc(target * s->img_x * s->img_y);
-   if (!out) return stbi__errpuc("outofmem", "Out of memory");
-   if (info.bpp < 16) {
-      int z=0;
-      if (psize == 0 || psize > 256) { STBI_FREE(out); return stbi__errpuc("invalid", "Corrupt BMP"); }
-      for (i=0; i < psize; ++i) {
+   // sanity-check size
+   if (!stbi__mad3sizes_valid(target, s->img_x, s->img_y, 0))
+      return stbi__errpuc("too large", "Corrupt BMP");
+
+   out = (stbi_uc*)stbi__malloc_mad3(target, s->img_x, s->img_y, 0);
+   if (!out)
+      return stbi__errpuc("outofmem", "Out of memory");
+   if (info.bpp < 16)
+   {
+      int z = 0;
+      if (psize == 0 || psize > 256)
+      {
+         STBI_FREE(out);
+         return stbi__errpuc("invalid", "Corrupt BMP");
+      }
+      for (i = 0; i < psize; ++i)
+      {
          pal[i][2] = stbi__get8(s);
          pal[i][1] = stbi__get8(s);
          pal[i][0] = stbi__get8(s);
-         if (info.hsz != 12) stbi__get8(s);
+         if (info.hsz != 12)
+            stbi__get8(s);
          pal[i][3] = 255;
       }
-      stbi__skip(s, info.offset - 14 - info.hsz - psize * (info.hsz == 12 ? 3 : 4));
-      if (info.bpp == 4) width = (s->img_x + 1) >> 1;
-      else if (info.bpp == 8) width = s->img_x;
-      else { STBI_FREE(out); return stbi__errpuc("bad bpp", "Corrupt BMP"); }
-      pad = (-width)&3;
-      for (j=0; j < (int) s->img_y; ++j) {
-         for (i=0; i < (int) s->img_x; i += 2) {
-            int v=stbi__get8(s),v2=0;
-            if (info.bpp == 4) {
-               v2 = v & 15;
-               v >>= 4;
+      stbi__skip(s, info.offset - info.extra_read - info.hsz - psize * (info.hsz == 12 ? 3 : 4));
+      if (info.bpp == 1)
+         width = (s->img_x + 7) >> 3;
+      else if (info.bpp == 4)
+         width = (s->img_x + 1) >> 1;
+      else if (info.bpp == 8)
+         width = s->img_x;
+      else
+      {
+         STBI_FREE(out);
+         return stbi__errpuc("bad bpp", "Corrupt BMP");
+      }
+      pad = (-width) & 3;
+      if (info.bpp == 1)
+      {
+         for (j = 0; j < (int)s->img_y; ++j)
+         {
+            int bit_offset = 7, v = stbi__get8(s);
+            for (i = 0; i < (int)s->img_x; ++i)
+            {
+               int color = (v >> bit_offset) & 0x1;
+               out[z++] = pal[color][0];
+               out[z++] = pal[color][1];
+               out[z++] = pal[color][2];
+               if (target == 4)
+                  out[z++] = 255;
+               if (i + 1 == (int)s->img_x)
+                  break;
+               if ((--bit_offset) < 0)
+               {
+                  bit_offset = 7;
+                  v = stbi__get8(s);
+               }
             }
-            out[z++] = pal[v][0];
-            out[z++] = pal[v][1];
-            out[z++] = pal[v][2];
-            if (target == 4) out[z++] = 255;
-            if (i+1 == (int) s->img_x) break;
-            v = (info.bpp == 8) ? stbi__get8(s) : v2;
-            out[z++] = pal[v][0];
-            out[z++] = pal[v][1];
-            out[z++] = pal[v][2];
-            if (target == 4) out[z++] = 255;
+            stbi__skip(s, pad);
          }
-         stbi__skip(s, pad);
       }
-   } else {
-      int rshift=0,gshift=0,bshift=0,ashift=0,rcount=0,gcount=0,bcount=0,acount=0;
+      else
+      {
+         for (j = 0; j < (int)s->img_y; ++j)
+         {
+            for (i = 0; i < (int)s->img_x; i += 2)
+            {
+               int v = stbi__get8(s), v2 = 0;
+               if (info.bpp == 4)
+               {
+                  v2 = v & 15;
+                  v >>= 4;
+               }
+               out[z++] = pal[v][0];
+               out[z++] = pal[v][1];
+               out[z++] = pal[v][2];
+               if (target == 4)
+                  out[z++] = 255;
+               if (i + 1 == (int)s->img_x)
+                  break;
+               v = (info.bpp == 8) ? stbi__get8(s) : v2;
+               out[z++] = pal[v][0];
+               out[z++] = pal[v][1];
+               out[z++] = pal[v][2];
+               if (target == 4)
+                  out[z++] = 255;
+            }
+            stbi__skip(s, pad);
+         }
+      }
+   }
+   else
+   {
+      int rshift = 0, gshift = 0, bshift = 0, ashift = 0, rcount = 0, gcount = 0, bcount = 0, acount = 0;
       int z = 0;
-      int easy=0;
-      stbi__skip(s, info.offset - 14 - info.hsz);
-      if (info.bpp == 24) width = 3 * s->img_x;
-      else if (info.bpp == 16) width = 2*s->img_x;
-      else /* bpp = 32 and pad = 0 */ width=0;
+      int easy = 0;
+      stbi__skip(s, info.offset - info.extra_read - info.hsz);
+      if (info.bpp == 24)
+         width = 3 * s->img_x;
+      else if (info.bpp == 16)
+         width = 2 * s->img_x;
+      else /* bpp = 32 and pad = 0 */
+         width = 0;
       pad = (-width) & 3;
-      if (info.bpp == 24) {
+      if (info.bpp == 24)
+      {
          easy = 1;
-      } else if (info.bpp == 32) {
+      }
+      else if (info.bpp == 32)
+      {
          if (mb == 0xff && mg == 0xff00 && mr == 0x00ff0000 && ma == 0xff000000)
             easy = 2;
       }
-      if (!easy) {
-         if (!mr || !mg || !mb) { STBI_FREE(out); return stbi__errpuc("bad masks", "Corrupt BMP"); }
+      if (!easy)
+      {
+         if (!mr || !mg || !mb)
+         {
+            STBI_FREE(out);
+            return stbi__errpuc("bad masks", "Corrupt BMP");
+         }
          // right shift amt to put high bit in position #7
-         rshift = stbi__high_bit(mr)-7; rcount = stbi__bitcount(mr);
-         gshift = stbi__high_bit(mg)-7; gcount = stbi__bitcount(mg);
-         bshift = stbi__high_bit(mb)-7; bcount = stbi__bitcount(mb);
-         ashift = stbi__high_bit(ma)-7; acount = stbi__bitcount(ma);
-      }
-      for (j=0; j < (int) s->img_y; ++j) {
-         if (easy) {
-            for (i=0; i < (int) s->img_x; ++i) {
+         rshift = stbi__high_bit(mr) - 7;
+         rcount = stbi__bitcount(mr);
+         gshift = stbi__high_bit(mg) - 7;
+         gcount = stbi__bitcount(mg);
+         bshift = stbi__high_bit(mb) - 7;
+         bcount = stbi__bitcount(mb);
+         ashift = stbi__high_bit(ma) - 7;
+         acount = stbi__bitcount(ma);
+         if (rcount > 8 || gcount > 8 || bcount > 8 || acount > 8)
+         {
+            STBI_FREE(out);
+            return stbi__errpuc("bad masks", "Corrupt BMP");
+         }
+      }
+      for (j = 0; j < (int)s->img_y; ++j)
+      {
+         if (easy)
+         {
+            for (i = 0; i < (int)s->img_x; ++i)
+            {
                unsigned char a;
-               out[z+2] = stbi__get8(s);
-               out[z+1] = stbi__get8(s);
-               out[z+0] = stbi__get8(s);
+               out[z + 2] = stbi__get8(s);
+               out[z + 1] = stbi__get8(s);
+               out[z + 0] = stbi__get8(s);
                z += 3;
                a = (easy == 2 ? stbi__get8(s) : 255);
                all_a |= a;
-               if (target == 4) out[z++] = a;
+               if (target == 4)
+                  out[z++] = a;
             }
-         } else {
+         }
+         else
+         {
             int bpp = info.bpp;
-            for (i=0; i < (int) s->img_x; ++i) {
-               stbi__uint32 v = (bpp == 16 ? (stbi__uint32) stbi__get16le(s) : stbi__get32le(s));
-               int a;
+            for (i = 0; i < (int)s->img_x; ++i)
+            {
+               stbi__uint32 v = (bpp == 16 ? (stbi__uint32)stbi__get16le(s) : stbi__get32le(s));
+               unsigned int a;
                out[z++] = STBI__BYTECAST(stbi__shiftsigned(v & mr, rshift, rcount));
                out[z++] = STBI__BYTECAST(stbi__shiftsigned(v & mg, gshift, gcount));
                out[z++] = STBI__BYTECAST(stbi__shiftsigned(v & mb, bshift, bcount));
                a = (ma ? stbi__shiftsigned(v & ma, ashift, acount) : 255);
                all_a |= a;
-               if (target == 4) out[z++] = STBI__BYTECAST(a);
+               if (target == 4)
+                  out[z++] = STBI__BYTECAST(a);
             }
          }
          stbi__skip(s, pad);
       }
    }
-   
+
    // if alpha channel is all 0s, replace with all 255s
    if (target == 4 && all_a == 0)
-      for (i=4*s->img_x*s->img_y-1; i >= 0; i -= 4)
+      for (i = 4 * s->img_x * s->img_y - 1; i >= 0; i -= 4)
          out[i] = 255;
 
-   if (flip_vertically) {
+   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;
+      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) {
+   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
+      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;
+   if (comp)
+      *comp = s->img_n;
    return out;
 }
 #endif
@@ -4850,107 +6741,141 @@ static stbi_uc *stbi__bmp_load(stbi__context *s, int *x, int *y, int *comp, int
 static int stbi__tga_get_comp(int bits_per_pixel, int is_grey, int* is_rgb16)
 {
    // only RGB or RGBA (incl. 16bit) or grey allowed
-   if(is_rgb16) *is_rgb16 = 0;
-   switch(bits_per_pixel) {
-      case 8:  return STBI_grey;
-      case 16: if(is_grey) return STBI_grey_alpha;
-            // else: fall-through
-      case 15: if(is_rgb16) *is_rgb16 = 1;
-            return STBI_rgb;
-      case 24: // fall-through
-      case 32: return bits_per_pixel/8;
+   if (is_rgb16)
+      *is_rgb16 = 0;
+   switch (bits_per_pixel)
+   {
+      case 8: return STBI_grey;
+      case 16:
+         if (is_grey)
+            return STBI_grey_alpha;
+         // fallthrough
+      case 15:
+         if (is_rgb16)
+            *is_rgb16 = 1;
+         return STBI_rgb;
+      case 24: // fallthrough
+      case 32: return bits_per_pixel / 8;
       default: return 0;
    }
 }
 
-static int stbi__tga_info(stbi__context *s, int *x, int *y, int *comp)
+static int stbi__tga_info(stbi__context* s, int* x, int* y, int* comp)
 {
-    int tga_w, tga_h, tga_comp, tga_image_type, tga_bits_per_pixel, tga_colormap_bpp;
-    int sz, tga_colormap_type;
-    stbi__get8(s);                   // discard Offset
-    tga_colormap_type = stbi__get8(s); // colormap type
-    if( tga_colormap_type > 1 ) {
-        stbi__rewind(s);
-        return 0;      // only RGB or indexed allowed
-    }
-    tga_image_type = stbi__get8(s); // image type
-    if ( tga_colormap_type == 1 ) { // colormapped (paletted) image
-        if (tga_image_type != 1 && tga_image_type != 9) {
-            stbi__rewind(s);
-            return 0;
-        }
-        stbi__skip(s,4);       // skip index of first colormap entry and number of entries
-        sz = stbi__get8(s);    //   check bits per palette color entry
-        if ( (sz != 8) && (sz != 15) && (sz != 16) && (sz != 24) && (sz != 32) ) {
-            stbi__rewind(s);
-            return 0;
-        }
-        stbi__skip(s,4);       // skip image x and y origin
-        tga_colormap_bpp = sz;
-    } else { // "normal" image w/o colormap - only RGB or grey allowed, +/- RLE
-        if ( (tga_image_type != 2) && (tga_image_type != 3) && (tga_image_type != 10) && (tga_image_type != 11) ) {
-            stbi__rewind(s);
-            return 0; // only RGB or grey allowed, +/- RLE
-        }
-        stbi__skip(s,9); // skip colormap specification and image x/y origin
-        tga_colormap_bpp = 0;
-    }
-    tga_w = stbi__get16le(s);
-    if( tga_w < 1 ) {
-        stbi__rewind(s);
-        return 0;   // test width
-    }
-    tga_h = stbi__get16le(s);
-    if( tga_h < 1 ) {
-        stbi__rewind(s);
-        return 0;   // test height
-    }
-    tga_bits_per_pixel = stbi__get8(s); // bits per pixel
-    stbi__get8(s); // ignore alpha bits
-    if (tga_colormap_bpp != 0) {
-        if((tga_bits_per_pixel != 8) && (tga_bits_per_pixel != 16)) {
-            // when using a colormap, tga_bits_per_pixel is the size of the indexes
-            // I don't think anything but 8 or 16bit indexes makes sense
-            stbi__rewind(s);
-            return 0;
-        }
-        tga_comp = stbi__tga_get_comp(tga_colormap_bpp, 0, NULL);
-    } else {
-        tga_comp = stbi__tga_get_comp(tga_bits_per_pixel, (tga_image_type == 3) || (tga_image_type == 11), NULL);
-    }
-    if(!tga_comp) {
+   int tga_w, tga_h, tga_comp, tga_image_type, tga_bits_per_pixel, tga_colormap_bpp;
+   int sz, tga_colormap_type;
+   stbi__get8(s); // discard Offset
+   tga_colormap_type = stbi__get8(s); // colormap type
+   if (tga_colormap_type > 1)
+   {
+      stbi__rewind(s);
+      return 0; // only RGB or indexed allowed
+   }
+   tga_image_type = stbi__get8(s); // image type
+   if (tga_colormap_type == 1)
+   { // colormapped (paletted) image
+      if (tga_image_type != 1 && tga_image_type != 9)
+      {
+         stbi__rewind(s);
+         return 0;
+      }
+      stbi__skip(s, 4); // skip index of first colormap entry and number of entries
+      sz = stbi__get8(s); //   check bits per palette color entry
+      if ((sz != 8) && (sz != 15) && (sz != 16) && (sz != 24) && (sz != 32))
+      {
+         stbi__rewind(s);
+         return 0;
+      }
+      stbi__skip(s, 4); // skip image x and y origin
+      tga_colormap_bpp = sz;
+   }
+   else
+   { // "normal" image w/o colormap - only RGB or grey allowed, +/- RLE
+      if ((tga_image_type != 2) && (tga_image_type != 3) && (tga_image_type != 10) && (tga_image_type != 11))
+      {
+         stbi__rewind(s);
+         return 0; // only RGB or grey allowed, +/- RLE
+      }
+      stbi__skip(s, 9); // skip colormap specification and image x/y origin
+      tga_colormap_bpp = 0;
+   }
+   tga_w = stbi__get16le(s);
+   if (tga_w < 1)
+   {
+      stbi__rewind(s);
+      return 0; // test width
+   }
+   tga_h = stbi__get16le(s);
+   if (tga_h < 1)
+   {
+      stbi__rewind(s);
+      return 0; // test height
+   }
+   tga_bits_per_pixel = stbi__get8(s); // bits per pixel
+   stbi__get8(s); // ignore alpha bits
+   if (tga_colormap_bpp != 0)
+   {
+      if ((tga_bits_per_pixel != 8) && (tga_bits_per_pixel != 16))
+      {
+         // when using a colormap, tga_bits_per_pixel is the size of the indexes
+         // I don't think anything but 8 or 16bit indexes makes sense
+         stbi__rewind(s);
+         return 0;
+      }
+      tga_comp = stbi__tga_get_comp(tga_colormap_bpp, 0, NULL);
+   }
+   else
+   {
+      tga_comp = stbi__tga_get_comp(tga_bits_per_pixel, (tga_image_type == 3) || (tga_image_type == 11), NULL);
+   }
+   if (!tga_comp)
+   {
       stbi__rewind(s);
       return 0;
-    }
-    if (x) *x = tga_w;
-    if (y) *y = tga_h;
-    if (comp) *comp = tga_comp;
-    return 1;                   // seems to have passed everything
+   }
+   if (x)
+      *x = tga_w;
+   if (y)
+      *y = tga_h;
+   if (comp)
+      *comp = tga_comp;
+   return 1; // seems to have passed everything
 }
 
-static int stbi__tga_test(stbi__context *s)
+static int stbi__tga_test(stbi__context* s)
 {
    int res = 0;
    int sz, tga_color_type;
-   stbi__get8(s);      //   discard Offset
-   tga_color_type = stbi__get8(s);   //   color type
-   if ( tga_color_type > 1 ) goto errorEnd;   //   only RGB or indexed allowed
-   sz = stbi__get8(s);   //   image type
-   if ( tga_color_type == 1 ) { // colormapped (paletted) image
-      if (sz != 1 && sz != 9) goto errorEnd; // colortype 1 demands image type 1 or 9
-      stbi__skip(s,4);       // skip index of first colormap entry and number of entries
-      sz = stbi__get8(s);    //   check bits per palette color entry
-      if ( (sz != 8) && (sz != 15) && (sz != 16) && (sz != 24) && (sz != 32) ) goto errorEnd;
-      stbi__skip(s,4);       // skip image x and y origin
-   } else { // "normal" image w/o colormap
-      if ( (sz != 2) && (sz != 3) && (sz != 10) && (sz != 11) ) goto errorEnd; // only RGB or grey allowed, +/- RLE
-      stbi__skip(s,9); // skip colormap specification and image x/y origin
-   }
-   if ( stbi__get16le(s) < 1 ) goto errorEnd;      //   test width
-   if ( stbi__get16le(s) < 1 ) goto errorEnd;      //   test height
-   sz = stbi__get8(s);   //   bits per pixel
-   if ( (tga_color_type == 1) && (sz != 8) && (sz != 16) ) goto errorEnd; // for colormapped images, bpp is size of an index
-   if ( (sz != 8) && (sz != 15) && (sz != 16) && (sz != 24) && (sz != 32) ) goto errorEnd;
+   stbi__get8(s); //   discard Offset
+   tga_color_type = stbi__get8(s); //   color type
+   if (tga_color_type > 1)
+      goto errorEnd; //   only RGB or indexed allowed
+   sz = stbi__get8(s); //   image type
+   if (tga_color_type == 1)
+   { // colormapped (paletted) image
+      if (sz != 1 && sz != 9)
+         goto errorEnd; // colortype 1 demands image type 1 or 9
+      stbi__skip(s, 4); // skip index of first colormap entry and number of entries
+      sz = stbi__get8(s); //   check bits per palette color entry
+      if ((sz != 8) && (sz != 15) && (sz != 16) && (sz != 24) && (sz != 32))
+         goto errorEnd;
+      stbi__skip(s, 4); // skip image x and y origin
+   }
+   else
+   { // "normal" image w/o colormap
+      if ((sz != 2) && (sz != 3) && (sz != 10) && (sz != 11))
+         goto errorEnd; // only RGB or grey allowed, +/- RLE
+      stbi__skip(s, 9); // skip colormap specification and image x/y origin
+   }
+   if (stbi__get16le(s) < 1)
+      goto errorEnd; //   test width
+   if (stbi__get16le(s) < 1)
+      goto errorEnd; //   test height
+   sz = stbi__get8(s); //   bits per pixel
+   if ((tga_color_type == 1) && (sz != 8) && (sz != 16))
+      goto errorEnd; // for colormapped images, bpp is size of an index
+   if ((sz != 8) && (sz != 15) && (sz != 16) && (sz != 24) && (sz != 32))
+      goto errorEnd;
 
    res = 1; // if we got this far, everything's good and we can return 1 instead of 0
 
@@ -4960,18 +6885,18 @@ static int stbi__tga_test(stbi__context *s)
 }
 
 // read 16bit value and convert to 24bit RGB
-void stbi__tga_read_rgb16(stbi__context *s, stbi_uc* out)
+static void stbi__tga_read_rgb16(stbi__context* s, stbi_uc* out)
 {
-   stbi__uint16 px = stbi__get16le(s);
+   stbi__uint16 px = (stbi__uint16)stbi__get16le(s);
    stbi__uint16 fiveBitMask = 31;
    // we have 3 channels with 5bits each
    int r = (px >> 10) & fiveBitMask;
    int g = (px >> 5) & fiveBitMask;
    int b = px & fiveBitMask;
    // Note that this saves the data in RGB(A) order, so it doesn't need to be swapped later
-   out[0] = (r * 255)/31;
-   out[1] = (g * 255)/31;
-   out[2] = (b * 255)/31;
+   out[0] = (stbi_uc)((r * 255) / 31);
+   out[1] = (stbi_uc)((g * 255) / 31);
+   out[2] = (stbi_uc)((b * 255) / 31);
 
    // some people claim that the most significant bit might be used for alpha
    // (possibly if an alpha-bit is set in the "image descriptor byte")
@@ -4979,7 +6904,7 @@ void stbi__tga_read_rgb16(stbi__context *s, stbi_uc* out)
    // so let's treat all 15 and 16bit TGAs as RGB with no alpha.
 }
 
-static stbi_uc *stbi__tga_load(stbi__context *s, int *x, int *y, int *comp, int req_comp)
+static void* stbi__tga_load(stbi__context* s, int* x, int* y, int* comp, int req_comp, stbi__result_info* ri)
 {
    //   read in the TGA header stuff
    int tga_offset = stbi__get8(s);
@@ -4994,20 +6919,28 @@ static stbi_uc *stbi__tga_load(stbi__context *s, int *x, int *y, int *comp, int
    int tga_width = stbi__get16le(s);
    int tga_height = stbi__get16le(s);
    int tga_bits_per_pixel = stbi__get8(s);
-   int tga_comp, tga_rgb16=0;
+   int tga_comp, tga_rgb16 = 0;
    int tga_inverted = stbi__get8(s);
    // int tga_alpha_bits = tga_inverted & 15; // the 4 lowest bits - unused (useless?)
    //   image data
-   unsigned char *tga_data;
-   unsigned char *tga_palette = NULL;
+   unsigned char* tga_data;
+   unsigned char* tga_palette = NULL;
    int i, j;
-   unsigned char raw_data[4];
+   unsigned char raw_data[4] = { 0 };
    int RLE_count = 0;
    int RLE_repeating = 0;
    int read_next_pixel = 1;
+   STBI_NOTUSED(ri);
+   STBI_NOTUSED(tga_x_origin); // @TODO
+   STBI_NOTUSED(tga_y_origin); // @TODO
+
+   if (tga_height > STBI_MAX_DIMENSIONS)
+      return stbi__errpuc("too large", "Very large image (corrupt?)");
+   if (tga_width > STBI_MAX_DIMENSIONS)
+      return stbi__errpuc("too large", "Very large image (corrupt?)");
 
    //   do a tiny bit of precessing
-   if ( tga_image_type >= 8 )
+   if (tga_image_type >= 8)
    {
       tga_image_type -= 8;
       tga_is_RLE = 1;
@@ -5015,97 +6948,128 @@ static stbi_uc *stbi__tga_load(stbi__context *s, int *x, int *y, int *comp, int
    tga_inverted = 1 - ((tga_inverted >> 5) & 1);
 
    //   If I'm paletted, then I'll use the number of bits from the palette
-   if ( tga_indexed ) tga_comp = stbi__tga_get_comp(tga_palette_bits, 0, &tga_rgb16);
-   else tga_comp = stbi__tga_get_comp(tga_bits_per_pixel, (tga_image_type == 3), &tga_rgb16);
+   if (tga_indexed)
+      tga_comp = stbi__tga_get_comp(tga_palette_bits, 0, &tga_rgb16);
+   else
+      tga_comp = stbi__tga_get_comp(tga_bits_per_pixel, (tga_image_type == 3), &tga_rgb16);
 
-   if(!tga_comp) // shouldn't really happen, stbi__tga_test() should have ensured basic consistency
+   if (!tga_comp) // shouldn't really happen, stbi__tga_test() should have ensured basic consistency
       return stbi__errpuc("bad format", "Can't find out TGA pixelformat");
 
    //   tga info
    *x = tga_width;
    *y = tga_height;
-   if (comp) *comp = tga_comp;
+   if (comp)
+      *comp = tga_comp;
+
+   if (!stbi__mad3sizes_valid(tga_width, tga_height, tga_comp, 0))
+      return stbi__errpuc("too large", "Corrupt TGA");
 
-   tga_data = (unsigned char*)stbi__malloc( (size_t)tga_width * tga_height * tga_comp );
-   if (!tga_data) return stbi__errpuc("outofmem", "Out of memory");
+   tga_data = (unsigned char*)stbi__malloc_mad3(tga_width, tga_height, tga_comp, 0);
+   if (!tga_data)
+      return stbi__errpuc("outofmem", "Out of memory");
 
    // skip to the data's starting position (offset usually = 0)
-   stbi__skip(s, tga_offset );
+   stbi__skip(s, tga_offset);
 
-   if ( !tga_indexed && !tga_is_RLE && !tga_rgb16 ) {
-      for (i=0; i < tga_height; ++i) {
-         int row = tga_inverted ? tga_height -i - 1 : i;
-         stbi_uc *tga_row = tga_data + row*tga_width*tga_comp;
+   if (!tga_indexed && !tga_is_RLE && !tga_rgb16)
+   {
+      for (i = 0; i < tga_height; ++i)
+      {
+         int row = tga_inverted ? tga_height - i - 1 : i;
+         stbi_uc* tga_row = tga_data + row * tga_width * tga_comp;
          stbi__getn(s, tga_row, tga_width * tga_comp);
       }
-   } else  {
+   }
+   else
+   {
       //   do I need to load a palette?
-      if ( tga_indexed)
+      if (tga_indexed)
       {
+         if (tga_palette_len == 0)
+         { /* you have to have at least one entry! */
+            STBI_FREE(tga_data);
+            return stbi__errpuc("bad palette", "Corrupt TGA");
+         }
+
          //   any data to skip? (offset usually = 0)
-         stbi__skip(s, tga_palette_start );
+         stbi__skip(s, tga_palette_start);
          //   load the palette
-         tga_palette = (unsigned char*)stbi__malloc( tga_palette_len * tga_comp );
-         if (!tga_palette) {
+         tga_palette = (unsigned char*)stbi__malloc_mad2(tga_palette_len, tga_comp, 0);
+         if (!tga_palette)
+         {
             STBI_FREE(tga_data);
             return stbi__errpuc("outofmem", "Out of memory");
          }
-         if (tga_rgb16) {
-            stbi_uc *pal_entry = tga_palette;
+         if (tga_rgb16)
+         {
+            stbi_uc* pal_entry = tga_palette;
             STBI_ASSERT(tga_comp == STBI_rgb);
-            for (i=0; i < tga_palette_len; ++i) {
+            for (i = 0; i < tga_palette_len; ++i)
+            {
                stbi__tga_read_rgb16(s, pal_entry);
                pal_entry += tga_comp;
             }
-         } else if (!stbi__getn(s, tga_palette, tga_palette_len * tga_comp)) {
-               STBI_FREE(tga_data);
-               STBI_FREE(tga_palette);
-               return stbi__errpuc("bad palette", "Corrupt TGA");
+         }
+         else if (!stbi__getn(s, tga_palette, tga_palette_len * tga_comp))
+         {
+            STBI_FREE(tga_data);
+            STBI_FREE(tga_palette);
+            return stbi__errpuc("bad palette", "Corrupt TGA");
          }
       }
       //   load the data
-      for (i=0; i < tga_width * tga_height; ++i)
+      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 (tga_is_RLE)
          {
-            if ( RLE_count == 0 )
+            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 )
+            }
+            else if (!RLE_repeating)
             {
                read_next_pixel = 1;
             }
-         } else
+         }
+         else
          {
             read_next_pixel = 1;
          }
          //   OK, if I need to read a pixel, do it now
-         if ( read_next_pixel )
+         if (read_next_pixel)
          {
             //   load however much data we did have
-            if ( tga_indexed )
+            if (tga_indexed)
             {
                // read in index, then perform the lookup
                int pal_idx = (tga_bits_per_pixel == 8) ? stbi__get8(s) : stbi__get16le(s);
-               if ( pal_idx >= tga_palette_len ) {
+               if (pal_idx >= tga_palette_len)
+               {
                   // invalid index
                   pal_idx = 0;
                }
                pal_idx *= tga_comp;
-               for (j = 0; j < tga_comp; ++j) {
-                  raw_data[j] = tga_palette[pal_idx+j];
+               for (j = 0; j < tga_comp; ++j)
+               {
+                  raw_data[j] = tga_palette[pal_idx + j];
                }
-            } else if(tga_rgb16) {
+            }
+            else if (tga_rgb16)
+            {
                STBI_ASSERT(tga_comp == STBI_rgb);
                stbi__tga_read_rgb16(s, raw_data);
-            } else {
+            }
+            else
+            {
                //   read in the data raw
-               for (j = 0; j < tga_comp; ++j) {
+               for (j = 0; j < tga_comp; ++j)
+               {
                   raw_data[j] = stbi__get8(s);
                }
             }
@@ -5115,15 +7079,15 @@ static stbi_uc *stbi__tga_load(stbi__context *s, int *x, int *y, int *comp, int
 
          // copy data
          for (j = 0; j < tga_comp; ++j)
-           tga_data[i*tga_comp+j] = raw_data[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 )
+      if (tga_inverted)
       {
-         for (j = 0; j*2 < tga_height; ++j)
+         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;
@@ -5138,9 +7102,9 @@ static stbi_uc *stbi__tga_load(stbi__context *s, int *x, int *y, int *comp, int
          }
       }
       //   clear my palette, if I had one
-      if ( tga_palette != NULL )
+      if (tga_palette != NULL)
       {
-         STBI_FREE( tga_palette );
+         STBI_FREE(tga_palette);
       }
    }
 
@@ -5148,7 +7112,7 @@ static stbi_uc *stbi__tga_load(stbi__context *s, int *x, int *y, int *comp, int
    if (tga_comp >= 3 && !tga_rgb16)
    {
       unsigned char* tga_pixel = tga_data;
-      for (i=0; i < tga_width * tga_height; ++i)
+      for (i = 0; i < tga_width * tga_height; ++i)
       {
          unsigned char temp = tga_pixel[0];
          tga_pixel[0] = tga_pixel[2];
@@ -5164,7 +7128,8 @@ static stbi_uc *stbi__tga_load(stbi__context *s, int *x, int *y, int *comp, int
    //   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;
+   tga_x_origin = tga_y_origin = 0;
+   STBI_NOTUSED(tga_palette_start);
    //   OK, done
    return tga_data;
 }
@@ -5174,24 +7139,73 @@ static stbi_uc *stbi__tga_load(stbi__context *s, int *x, int *y, int *comp, int
 // Photoshop PSD loader -- PD by Thatcher Ulrich, integration by Nicolas Schulz, tweaked by STB
 
 #ifndef STBI_NO_PSD
-static int stbi__psd_test(stbi__context *s)
+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)
+static int stbi__psd_decode_rle(stbi__context* s, stbi_uc* p, int pixelCount)
+{
+   int count, nleft, len;
+
+   count = 0;
+   while ((nleft = pixelCount - count) > 0)
+   {
+      len = stbi__get8(s);
+      if (len == 128)
+      {
+         // No-op.
+      }
+      else if (len < 128)
+      {
+         // Copy next len+1 bytes literally.
+         len++;
+         if (len > nleft)
+            return 0; // corrupt data
+         count += len;
+         while (len)
+         {
+            *p = stbi__get8(s);
+            p += 4;
+            len--;
+         }
+      }
+      else if (len > 128)
+      {
+         stbi_uc val;
+         // Next -len+1 bytes in the dest are replicated from next source byte.
+         // (Interpret len as a negative 8-bit int.)
+         len = 257 - len;
+         if (len > nleft)
+            return 0; // corrupt data
+         val = stbi__get8(s);
+         count += len;
+         while (len)
+         {
+            *p = val;
+            p += 4;
+            len--;
+         }
+      }
+   }
+
+   return 1;
+}
+
+static void* stbi__psd_load(stbi__context* s, int* x, int* y, int* comp, int req_comp, stbi__result_info* ri, int bpc)
 {
-   int   pixelCount;
+   int pixelCount;
    int channelCount, compression;
-   int channel, i, count, len;
+   int channel, i;
    int bitdepth;
-   int w,h;
-   stbi_uc *out;
+   int w, h;
+   stbi_uc* out;
+   STBI_NOTUSED(ri);
 
    // Check identifier
-   if (stbi__get32be(s) != 0x38425053)   // "8BPS"
+   if (stbi__get32be(s) != 0x38425053) // "8BPS"
       return stbi__errpuc("not PSD", "Corrupt PSD image");
 
    // Check file type version.
@@ -5199,7 +7213,7 @@ static stbi_uc *stbi__psd_load(stbi__context *s, int *x, int *y, int *comp, int
       return stbi__errpuc("wrong version", "Unsupported version of PSD image");
 
    // Skip 6 reserved bytes.
-   stbi__skip(s, 6 );
+   stbi__skip(s, 6);
 
    // Read the number of channels (R, G, B, A, etc).
    channelCount = stbi__get16be(s);
@@ -5210,6 +7224,11 @@ static stbi_uc *stbi__psd_load(stbi__context *s, int *x, int *y, int *comp, int
    h = stbi__get32be(s);
    w = stbi__get32be(s);
 
+   if (h > STBI_MAX_DIMENSIONS)
+      return stbi__errpuc("too large", "Very large image (corrupt?)");
+   if (w > STBI_MAX_DIMENSIONS)
+      return stbi__errpuc("too large", "Very large image (corrupt?)");
+
    // Make sure the depth is 8 bits.
    bitdepth = stbi__get16be(s);
    if (bitdepth != 8 && bitdepth != 16)
@@ -5229,13 +7248,13 @@ static stbi_uc *stbi__psd_load(stbi__context *s, int *x, int *y, int *comp, int
       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) );
+   stbi__skip(s, stbi__get32be(s));
 
    // Skip the image resources.  (resolution, pen tool paths, etc)
-   stbi__skip(s, stbi__get32be(s) );
+   stbi__skip(s, stbi__get32be(s));
 
    // Skip the reserved data.
-   stbi__skip(s, stbi__get32be(s) );
+   stbi__skip(s, stbi__get32be(s));
 
    // Find out if the data is compressed.
    // Known values:
@@ -5245,16 +7264,30 @@ static stbi_uc *stbi__psd_load(stbi__context *s, int *x, int *y, int *comp, int
    if (compression > 1)
       return stbi__errpuc("bad compression", "PSD has an unknown compression format");
 
+   // Check size
+   if (!stbi__mad3sizes_valid(4, w, h, 0))
+      return stbi__errpuc("too large", "Corrupt PSD");
+
    // Create the destination image.
-   out = (stbi_uc *) stbi__malloc(4 * w*h);
-   if (!out) return stbi__errpuc("outofmem", "Out of memory");
-   pixelCount = w*h;
+
+   if (!compression && bitdepth == 16 && bpc == 16)
+   {
+      out = (stbi_uc*)stbi__malloc_mad3(8, w, h, 0);
+      ri->bits_per_channel = 16;
+   }
+   else
+      out = (stbi_uc*)stbi__malloc(4 * w * h);
+
+   if (!out)
+      return stbi__errpuc("outofmem", "Out of memory");
+   pixelCount = w * h;
 
    // Initialize the data to zero.
    //memset( out, 0, pixelCount * 4 );
 
    // Finally, the image data.
-   if (compression) {
+   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.
@@ -5263,86 +7296,135 @@ static stbi_uc *stbi__psd_load(stbi__context *s, int *x, int *y, int *comp, int
       //     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,
+      // The RLE-compressed data is preceded by a 2-byte data count for each row in the data,
       // which we're going to just skip.
-      stbi__skip(s, h * channelCount * 2 );
+      stbi__skip(s, h * channelCount * 2);
 
       // Read the RLE data by channel.
-      for (channel = 0; channel < 4; channel++) {
-         stbi_uc *p;
+      for (channel = 0; channel < 4; channel++)
+      {
+         stbi_uc* p;
 
-         p = out+channel;
-         if (channel >= channelCount) {
+         p = out + channel;
+         if (channel >= channelCount)
+         {
             // Fill this channel with default data.
             for (i = 0; i < pixelCount; i++, p += 4)
                *p = (channel == 3 ? 255 : 0);
-         } else {
+         }
+         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--;
-                  }
-               }
+            if (!stbi__psd_decode_rle(s, p, pixelCount))
+            {
+               STBI_FREE(out);
+               return stbi__errpuc("corrupt", "bad RLE data");
             }
          }
       }
-
-   } else {
+   }
+   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.
+      // where each channel consists of an 8-bit (or 16-bit) value for each pixel in the image.
 
       // Read the data by channel.
-      for (channel = 0; channel < 4; channel++) {
-         stbi_uc *p;
-
-         p = out + channel;
-         if (channel >= channelCount) {
+      for (channel = 0; channel < 4; channel++)
+      {
+         if (channel >= channelCount)
+         {
             // Fill this channel with default data.
-            stbi_uc val = channel == 3 ? 255 : 0;
-            for (i = 0; i < pixelCount; i++, p += 4)
-               *p = val;
-         } else {
-            // Read the data.
-            if (bitdepth == 16) {
-               for (i = 0; i < pixelCount; i++, p += 4)
-                  *p = (stbi_uc) (stbi__get16be(s) >> 8);
-            } else {
+            if (bitdepth == 16 && bpc == 16)
+            {
+               stbi__uint16* q = ((stbi__uint16*)out) + channel;
+               stbi__uint16 val = channel == 3 ? 65535 : 0;
+               for (i = 0; i < pixelCount; i++, q += 4)
+                  *q = val;
+            }
+            else
+            {
+               stbi_uc* p = out + channel;
+               stbi_uc val = channel == 3 ? 255 : 0;
                for (i = 0; i < pixelCount; i++, p += 4)
-                  *p = stbi__get8(s);
+                  *p = val;
+            }
+         }
+         else
+         {
+            if (ri->bits_per_channel == 16)
+            { // output bpc
+               stbi__uint16* q = ((stbi__uint16*)out) + channel;
+               for (i = 0; i < pixelCount; i++, q += 4)
+                  *q = (stbi__uint16)stbi__get16be(s);
+            }
+            else
+            {
+               stbi_uc* p = out + channel;
+               if (bitdepth == 16)
+               { // input bpc
+                  for (i = 0; i < pixelCount; i++, p += 4)
+                     *p = (stbi_uc)(stbi__get16be(s) >> 8);
+               }
+               else
+               {
+                  for (i = 0; i < pixelCount; i++, p += 4)
+                     *p = stbi__get8(s);
+               }
             }
          }
       }
    }
 
-   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
+   // remove weird white matte from PSD
+   if (channelCount >= 4)
+   {
+      if (ri->bits_per_channel == 16)
+      {
+         for (i = 0; i < w * h; ++i)
+         {
+            stbi__uint16* pixel = (stbi__uint16*)out + 4 * i;
+            if (pixel[3] != 0 && pixel[3] != 65535)
+            {
+               float a = pixel[3] / 65535.0f;
+               float ra = 1.0f / a;
+               float inv_a = 65535.0f * (1 - ra);
+               pixel[0] = (stbi__uint16)(pixel[0] * ra + inv_a);
+               pixel[1] = (stbi__uint16)(pixel[1] * ra + inv_a);
+               pixel[2] = (stbi__uint16)(pixel[2] * ra + inv_a);
+            }
+         }
+      }
+      else
+      {
+         for (i = 0; i < w * h; ++i)
+         {
+            unsigned char* pixel = out + 4 * i;
+            if (pixel[3] != 0 && pixel[3] != 255)
+            {
+               float a = pixel[3] / 255.0f;
+               float ra = 1.0f / a;
+               float inv_a = 255.0f * (1 - ra);
+               pixel[0] = (unsigned char)(pixel[0] * ra + inv_a);
+               pixel[1] = (unsigned char)(pixel[1] * ra + inv_a);
+               pixel[2] = (unsigned char)(pixel[2] * ra + inv_a);
+            }
+         }
+      }
+   }
+
+   // convert to desired output format
+   if (req_comp && req_comp != 4)
+   {
+      if (ri->bits_per_channel == 16)
+         out = (stbi_uc*)stbi__convert_format16((stbi__uint16*)out, 4, req_comp, w, h);
+      else
+         out = stbi__convert_format(out, 4, req_comp, w, h);
+      if (out == NULL)
+         return out; // stbi__convert_format frees input on failure
    }
 
-   if (comp) *comp = 4;
+   if (comp)
+      *comp = 4;
    *y = h;
    *x = w;
 
@@ -5358,27 +7440,27 @@ static stbi_uc *stbi__psd_load(stbi__context *s, int *x, int *y, int *comp, int
 // See http://ozviz.wasp.uwa.edu.au/~pbourke/dataformats/softimagepic/
 
 #ifndef STBI_NO_PIC
-static int stbi__pic_is4(stbi__context *s,const char *str)
+static int stbi__pic_is4(stbi__context* s, const char* str)
 {
    int i;
-   for (i=0; i<4; ++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)
+static int stbi__pic_test_core(stbi__context* s)
 {
    int i;
 
-   if (!stbi__pic_is4(s,"\x53\x80\xF6\x34"))
+   if (!stbi__pic_is4(s, "\x53\x80\xF6\x34"))
       return 0;
 
-   for(i=0;i<84;++i)
+   for (i = 0; i < 84; ++i)
       stbi__get8(s);
 
-   if (!stbi__pic_is4(s,"PICT"))
+   if (!stbi__pic_is4(s, "PICT"))
       return 0;
 
    return 1;
@@ -5386,132 +7468,152 @@ static int stbi__pic_test_core(stbi__context *s)
 
 typedef struct
 {
-   stbi_uc size,type,channel;
+   stbi_uc size, type, channel;
 } stbi__pic_packet;
 
-static stbi_uc *stbi__readval(stbi__context *s, int channel, stbi_uc *dest)
+static stbi_uc* stbi__readval(stbi__context* s, int channel, stbi_uc* dest)
 {
-   int mask=0x80, i;
+   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);
+   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)
+static void stbi__copyval(int channel, stbi_uc* dest, const stbi_uc* src)
 {
-   int mask=0x80,i;
+   int mask = 0x80, i;
 
-   for (i=0;i<4; ++i, mask>>=1)
-      if (channel&mask)
-         dest[i]=src[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)
+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;
+   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;
+   // 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");
+      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->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");
+      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) {
+   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;
+      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) {
+         switch (packet->type)
+         {
             default:
-               return stbi__errpuc("bad format","packet has bad compression type");
+               return stbi__errpuc("bad format", "packet has bad compression type");
 
-            case 0: {//uncompressed
+            case 0:
+            { //uncompressed
                int x;
 
-               for(x=0;x<width;++x, dest+=4)
-                  if (!stbi__readval(s,packet->channel,dest))
+               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;
+            case 1: //Pure RLE
+            {
+               int left = width, i;
 
-                  while (left>0) {
-                     stbi_uc count,value[4];
+               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)");
+                  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 (count > left)
+                     count = (stbi_uc)left;
 
-                     if (!stbi__readval(s,packet->channel,value))  return 0;
+                  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;
-                  }
+                  for (i = 0; i < count; ++i, dest += 4)
+                     stbi__copyval(packet->channel, dest, value);
+                  left -= count;
                }
-               break;
+            }
+            break;
 
-            case 2: {//Mixed RLE
-               int left=width;
-               while (left>0) {
+            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 (stbi__at_eof(s))
+                     return stbi__errpuc("bad file", "file too short (mixed read count)");
 
-                  if (count >= 128) { // Repeated
+                  if (count >= 128)
+                  { // Repeated
                      stbi_uc value[4];
 
-                     if (count==128)
+                     if (count == 128)
                         count = stbi__get16be(s);
                      else
                         count -= 127;
                      if (count > left)
-                        return stbi__errpuc("bad file","scanline overrun");
+                        return stbi__errpuc("bad file", "scanline overrun");
 
-                     if (!stbi__readval(s,packet->channel,value))
+                     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
+                     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");
+                     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))
+                     for (i = 0; i < count; ++i, dest += 4)
+                        if (!stbi__readval(s, packet->channel, dest))
                            return 0;
                   }
-                  left-=count;
+                  left -= count;
                }
                break;
             }
@@ -5522,40 +7624,56 @@ static stbi_uc *stbi__pic_load_core(stbi__context *s,int width,int height,int *c
    return result;
 }
 
-static stbi_uc *stbi__pic_load(stbi__context *s,int *px,int *py,int *comp,int req_comp)
+static void* stbi__pic_load(stbi__context* s, int* px, int* py, int* comp, int req_comp, stbi__result_info* ri)
 {
-   stbi_uc *result;
-   int i, x,y;
+   stbi_uc* result;
+   int i, x, y, internal_comp;
+   STBI_NOTUSED(ri);
 
-   for (i=0; i<92; ++i)
+   if (!comp)
+      comp = &internal_comp;
+
+   for (i = 0; i < 92; ++i)
       stbi__get8(s);
 
    x = stbi__get16be(s);
    y = stbi__get16be(s);
-   if (stbi__at_eof(s))  return stbi__errpuc("bad file","file too short (pic header)");
-   if ((1 << 28) / x < y) return stbi__errpuc("too large", "Image too large to decode");
+
+   if (y > STBI_MAX_DIMENSIONS)
+      return stbi__errpuc("too large", "Very large image (corrupt?)");
+   if (x > STBI_MAX_DIMENSIONS)
+      return stbi__errpuc("too large", "Very large image (corrupt?)");
+
+   if (stbi__at_eof(s))
+      return stbi__errpuc("bad file", "file too short (pic header)");
+   if (!stbi__mad3sizes_valid(x, y, 4, 0))
+      return stbi__errpuc("too large", "PIC image too large to decode");
 
    stbi__get32be(s); //skip `ratio'
    stbi__get16be(s); //skip `fields'
    stbi__get16be(s); //skip `pad'
 
    // intermediate buffer is RGBA
-   result = (stbi_uc *) stbi__malloc(x*y*4);
-   memset(result, 0xff, x*y*4);
+   result = (stbi_uc*)stbi__malloc_mad3(x, y, 4, 0);
+   if (!result)
+      return stbi__errpuc("outofmem", "Out of memory");
+   memset(result, 0xff, x * y * 4);
 
-   if (!stbi__pic_load_core(s,x,y,comp, result)) {
+   if (!stbi__pic_load_core(s, x, y, comp, result))
+   {
       STBI_FREE(result);
-      result=0;
+      result = 0;
    }
    *px = x;
    *py = y;
-   if (req_comp == 0) req_comp = *comp;
-   result=stbi__convert_format(result,4,req_comp,x,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)
+static int stbi__pic_test(stbi__context* s)
 {
    int r = stbi__pic_test_core(s);
    stbi__rewind(s);
@@ -5576,42 +7694,49 @@ typedef struct
 
 typedef struct
 {
-   int w,h;
-   stbi_uc *out, *old_out;             // output buffer (always 4 components)
-   int flags, bgindex, ratio, transparent, eflags, delay;
-   stbi_uc  pal[256][4];
+   int w, h;
+   stbi_uc* out; // output buffer (always 4 components)
+   stbi_uc* background; // The current "background" as far as a gif is concerned
+   stbi_uc* history;
+   int flags, bgindex, ratio, transparent, eflags;
+   stbi_uc pal[256][4];
    stbi_uc lpal[256][4];
-   stbi__gif_lzw codes[4096];
-   stbi_uc *color_table;
+   stbi__gif_lzw codes[8192];
+   stbi_uc* color_table;
    int parse, step;
    int lflags;
    int start_x, start_y;
    int max_x, max_y;
    int cur_x, cur_y;
    int line_size;
+   int delay;
 } stbi__gif;
 
-static int stbi__gif_test_raw(stbi__context *s)
+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;
+   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;
+   if (sz != '9' && sz != '7')
+      return 0;
+   if (stbi__get8(s) != 'a')
+      return 0;
    return 1;
 }
 
-static int stbi__gif_test(stbi__context *s)
+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)
+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) {
+   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);
@@ -5619,15 +7744,17 @@ static void stbi__gif_parse_colortable(stbi__context *s, stbi_uc pal[256][4], in
    }
 }
 
-static int stbi__gif_header(stbi__context *s, stbi__gif *g, int *comp, int is_info)
+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");
+   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);
@@ -5637,43 +7764,62 @@ static int stbi__gif_header(stbi__context *s, stbi__gif *g, int *comp, int is_in
    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 (g->w > STBI_MAX_DIMENSIONS)
+      return stbi__err("too large", "Very large image (corrupt?)");
+   if (g->h > STBI_MAX_DIMENSIONS)
+      return stbi__err("too large", "Very large image (corrupt?)");
+
+   if (comp != 0)
+      *comp = 4; // can't actually tell whether it's 3 or 4 until we parse the comments
 
-   if (is_info) return 1;
+   if (is_info)
+      return 1;
 
    if (g->flags & 0x80)
-      stbi__gif_parse_colortable(s,g->pal, 2 << (g->flags & 7), -1);
+      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)
+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 );
+   stbi__gif* g = (stbi__gif*)stbi__malloc(sizeof(stbi__gif));
+   if (!g)
+      return stbi__err("outofmem", "Out of memory");
+   if (!stbi__gif_header(s, g, comp, 1))
+   {
+      STBI_FREE(g);
+      stbi__rewind(s);
       return 0;
    }
-   if (x) *x = g.w;
-   if (y) *y = g.h;
+   if (x)
+      *x = g->w;
+   if (y)
+      *y = g->h;
+   STBI_FREE(g);
    return 1;
 }
 
-static void stbi__out_gif_code(stbi__gif *g, stbi__uint16 code)
+static void stbi__out_gif_code(stbi__gif* g, stbi__uint16 code)
 {
    stbi_uc *p, *c;
+   int idx;
 
    // recurse to decode the prefixes, since the linked-list is backwards,
    // and working backwards through an interleaved image would be nasty
    if (g->codes[code].prefix >= 0)
       stbi__out_gif_code(g, g->codes[code].prefix);
 
-   if (g->cur_y >= g->max_y) return;
+   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];
+   idx = g->cur_x + g->cur_y;
+   p = &g->out[idx];
+   g->history[idx / 4] = 1;
 
-   if (c[3] >= 128) {
+   c = &g->color_table[g->codes[code].suffix * 4];
+   if (c[3] > 128)
+   { // don't render transparent pixels;
       p[0] = c[2];
       p[1] = c[1];
       p[2] = c[0];
@@ -5681,11 +7827,13 @@ static void stbi__out_gif_code(stbi__gif *g, stbi__uint16 code)
    }
    g->cur_x += 4;
 
-   if (g->cur_x >= g->max_x) {
+   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) {
+      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;
@@ -5693,141 +7841,199 @@ static void stbi__out_gif_code(stbi__gif *g, stbi__uint16 code)
    }
 }
 
-static stbi_uc *stbi__process_gif_raster(stbi__context *s, stbi__gif *g)
+static stbi_uc* stbi__process_gif_raster(stbi__context* s, stbi__gif* g)
 {
    stbi_uc lzw_cs;
    stbi__int32 len, init_code;
    stbi__uint32 first;
    stbi__int32 codesize, codemask, avail, oldcode, bits, valid_bits, clear;
-   stbi__gif_lzw *p;
+   stbi__gif_lzw* p;
 
    lzw_cs = stbi__get8(s);
-   if (lzw_cs > 12) return NULL;
+   if (lzw_cs > 12)
+      return NULL;
    clear = 1 << lzw_cs;
    first = 1;
    codesize = lzw_cs + 1;
    codemask = (1 << codesize) - 1;
    bits = 0;
    valid_bits = 0;
-   for (init_code = 0; init_code < clear; init_code++) {
+   for (init_code = 0; init_code < clear; init_code++)
+   {
       g->codes[init_code].prefix = -1;
-      g->codes[init_code].first = (stbi_uc) init_code;
-      g->codes[init_code].suffix = (stbi_uc) init_code;
+      g->codes[init_code].first = (stbi_uc)init_code;
+      g->codes[init_code].suffix = (stbi_uc)init_code;
    }
 
    // support no starting clear code
-   avail = clear+2;
+   avail = clear + 2;
    oldcode = -1;
 
    len = 0;
-   for(;;) {
-      if (valid_bits < codesize) {
-         if (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;
+         bits |= (stbi__int32)stbi__get8(s) << valid_bits;
          valid_bits += 8;
-      } else {
+      }
+      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
+         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
+         }
+         else if (code == clear + 1)
+         { // end of stream code
             stbi__skip(s, len);
             while ((len = stbi__get8(s)) > 0)
-               stbi__skip(s,len);
+               stbi__skip(s, len);
             return g->out;
-         } else if (code <= avail) {
-            if (first) return stbi__errpuc("no clear code", "Corrupt GIF");
+         }
+         else if (code <= avail)
+         {
+            if (first)
+            {
+               return stbi__errpuc("no clear code", "Corrupt GIF");
+            }
 
-            if (oldcode >= 0) {
+            if (oldcode >= 0)
+            {
                p = &g->codes[avail++];
-               if (avail > 4096)        return stbi__errpuc("too many codes", "Corrupt GIF");
-               p->prefix = (stbi__int16) oldcode;
+               if (avail > 8192)
+               {
+                  return stbi__errpuc("too many codes", "Corrupt GIF");
+               }
+
+               p->prefix = (stbi__int16)oldcode;
                p->first = g->codes[oldcode].first;
                p->suffix = (code == avail) ? p->first : g->codes[code].first;
-            } else if (code == avail)
+            }
+            else if (code == avail)
                return stbi__errpuc("illegal code in raster", "Corrupt GIF");
 
-            stbi__out_gif_code(g, (stbi__uint16) code);
+            stbi__out_gif_code(g, (stbi__uint16)code);
 
-            if ((avail & codemask) == 0 && avail <= 0x0FFF) {
+            if ((avail & codemask) == 0 && avail <= 0x0FFF)
+            {
                codesize++;
                codemask = (1 << codesize) - 1;
             }
 
             oldcode = code;
-         } else {
+         }
+         else
+         {
             return stbi__errpuc("illegal code in raster", "Corrupt GIF");
          }
       }
    }
 }
 
-static void stbi__fill_gif_background(stbi__gif *g, int x0, int y0, int x1, int y1)
-{
-   int x, y;
-   stbi_uc *c = g->pal[g->bgindex];
-   for (y = y0; y < y1; y += 4 * g->w) {
-      for (x = x0; x < x1; x += 4) {
-         stbi_uc *p  = &g->out[y + x];
-         p[0] = c[2];
-         p[1] = c[1];
-         p[2] = c[0];
-         p[3] = 0;
-      }
-   }
-}
-
 // this function is designed to support animated gifs, although stb_image doesn't support it
-static stbi_uc *stbi__gif_load_next(stbi__context *s, stbi__gif *g, int *comp, int req_comp)
+// two back is the image from two frames ago, used for a very specific disposal format
+static stbi_uc* stbi__gif_load_next(stbi__context* s, stbi__gif* g, int* comp, int req_comp, stbi_uc* two_back)
 {
-   int i;
-   stbi_uc *prev_out = 0;
+   int dispose;
+   int first_frame;
+   int pi;
+   int pcount;
+   STBI_NOTUSED(req_comp);
 
-   if (g->out == 0 && !stbi__gif_header(s, g, comp,0))
-      return 0; // stbi__g_failure_reason set by stbi__gif_header
+   // on first frame, any non-written pixels get the background colour (non-transparent)
+   first_frame = 0;
+   if (g->out == 0)
+   {
+      if (!stbi__gif_header(s, g, comp, 0))
+         return 0; // stbi__g_failure_reason set by stbi__gif_header
+      if (!stbi__mad3sizes_valid(4, g->w, g->h, 0))
+         return stbi__errpuc("too large", "GIF image is too large");
+      pcount = g->w * g->h;
+      g->out = (stbi_uc*)stbi__malloc(4 * pcount);
+      g->background = (stbi_uc*)stbi__malloc(4 * pcount);
+      g->history = (stbi_uc*)stbi__malloc(pcount);
+      if (!g->out || !g->background || !g->history)
+         return stbi__errpuc("outofmem", "Out of memory");
+
+      // image is treated as "transparent" at the start - ie, nothing overwrites the current background;
+      // background colour is only used for pixels that are not rendered first frame, after that "background"
+      // color refers to the color that was there the previous frame.
+      memset(g->out, 0x00, 4 * pcount);
+      memset(g->background, 0x00, 4 * pcount); // state of the background (starts transparent)
+      memset(g->history, 0x00, pcount); // pixels that were affected previous frame
+      first_frame = 1;
+   }
+   else
+   {
+      // second frame - how do we dispose of the previous one?
+      dispose = (g->eflags & 0x1C) >> 2;
+      pcount = g->w * g->h;
 
-   prev_out = g->out;
-   g->out = (stbi_uc *) stbi__malloc(4 * g->w * g->h);
-   if (g->out == 0) return stbi__errpuc("outofmem", "Out of memory");
+      if ((dispose == 3) && (two_back == 0))
+      {
+         dispose = 2; // if I don't have an image to revert back to, default to the old background
+      }
 
-   switch ((g->eflags & 0x1C) >> 2) {
-      case 0: // unspecified (also always used on 1st frame)
-         stbi__fill_gif_background(g, 0, 0, 4 * g->w, 4 * g->w * g->h);
-         break;
-      case 1: // do not dispose
-         if (prev_out) memcpy(g->out, prev_out, 4 * g->w * g->h);
-         g->old_out = prev_out;
-         break;
-      case 2: // dispose to background
-         if (prev_out) memcpy(g->out, prev_out, 4 * g->w * g->h);
-         stbi__fill_gif_background(g, g->start_x, g->start_y, g->max_x, g->max_y);
-         break;
-      case 3: // dispose to previous
-         if (g->old_out) {
-            for (i = g->start_y; i < g->max_y; i += 4 * g->w)
-               memcpy(&g->out[i + g->start_x], &g->old_out[i + g->start_x], g->max_x - g->start_x);
+      if (dispose == 3)
+      { // use previous graphic
+         for (pi = 0; pi < pcount; ++pi)
+         {
+            if (g->history[pi])
+            {
+               memcpy(&g->out[pi * 4], &two_back[pi * 4], 4);
+            }
          }
-         break;
+      }
+      else if (dispose == 2)
+      {
+         // restore what was changed last frame to background before that frame;
+         for (pi = 0; pi < pcount; ++pi)
+         {
+            if (g->history[pi])
+            {
+               memcpy(&g->out[pi * 4], &g->background[pi * 4], 4);
+            }
+         }
+      }
+      else
+      {
+         // This is a non-disposal case eithe way, so just
+         // leave the pixels as is, and they will become the new background
+         // 1: do not dispose
+         // 0:  not specified.
+      }
+
+      // background is what out is after the undoing of the previou frame;
+      memcpy(g->background, g->out, 4 * g->w * g->h);
    }
 
-   for (;;) {
-      switch (stbi__get8(s)) {
+   // clear my history;
+   memset(g->history, 0x00, g->w * g->h); // pixels that were affected previous frame
+
+   for (;;)
+   {
+      int tag = stbi__get8(s);
+      switch (tag)
+      {
          case 0x2C: /* Image Descriptor */
          {
-            int prev_trans = -1;
             stbi__int32 x, y, w, h;
-            stbi_uc *o;
+            stbi_uc* o;
 
             x = stbi__get16le(s);
             y = stbi__get16le(s);
@@ -5839,95 +8045,268 @@ static stbi_uc *stbi__gif_load_next(stbi__context *s, stbi__gif *g, int *comp, i
             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->max_x = g->start_x + w * 4;
+            g->max_y = g->start_y + h * g->line_size;
+            g->cur_x = g->start_x;
+            g->cur_y = g->start_y;
+
+            // if the width of the specified rectangle is 0, that means
+            // we may not see *any* pixels or the image is malformed;
+            // to make sure this is caught, move the current y down to
+            // max_y (which is what out_gif_code checks).
+            if (w == 0)
+               g->cur_y = g->max_y;
 
             g->lflags = stbi__get8(s);
 
-            if (g->lflags & 0x40) {
+            if (g->lflags & 0x40)
+            {
                g->step = 8 * g->line_size; // first interlaced spacing
                g->parse = 3;
-            } else {
+            }
+            else
+            {
                g->step = g->line_size;
                g->parse = 0;
             }
 
-            if (g->lflags & 0x80) {
-               stbi__gif_parse_colortable(s,g->lpal, 2 << (g->lflags & 7), g->eflags & 0x01 ? g->transparent : -1);
-               g->color_table = (stbi_uc *) g->lpal;
-            } else if (g->flags & 0x80) {
-               if (g->transparent >= 0 && (g->eflags & 0x01)) {
-                  prev_trans = g->pal[g->transparent][3];
-                  g->pal[g->transparent][3] = 0;
-               }
-               g->color_table = (stbi_uc *) g->pal;
-            } else
+            if (g->lflags & 0x80)
+            {
+               stbi__gif_parse_colortable(s, g->lpal, 2 << (g->lflags & 7), g->eflags & 0x01 ? g->transparent : -1);
+               g->color_table = (stbi_uc*)g->lpal;
+            }
+            else if (g->flags & 0x80)
+            {
+               g->color_table = (stbi_uc*)g->pal;
+            }
+            else
                return stbi__errpuc("missing color table", "Corrupt GIF");
 
             o = stbi__process_gif_raster(s, g);
-            if (o == NULL) return NULL;
+            if (!o)
+               return NULL;
 
-            if (prev_trans != -1)
-               g->pal[g->transparent][3] = (stbi_uc) prev_trans;
+            // if this was the first frame,
+            pcount = g->w * g->h;
+            if (first_frame && (g->bgindex > 0))
+            {
+               // if first frame, any pixel not drawn to gets the background color
+               for (pi = 0; pi < pcount; ++pi)
+               {
+                  if (g->history[pi] == 0)
+                  {
+                     g->pal[g->bgindex][3] = 255; // just in case it was made transparent, undo that; It will be reset next frame if need be;
+                     memcpy(&g->out[pi * 4], &g->pal[g->bgindex], 4);
+                  }
+               }
+            }
 
             return o;
          }
 
-         case 0x21: // Comment Extension.
-         {
-            int len;
-            if (stbi__get8(s) == 0xF9) { // Graphic Control Extension.
-               len = stbi__get8(s);
-               if (len == 4) {
-                  g->eflags = stbi__get8(s);
-                  g->delay = stbi__get16le(s);
-                  g->transparent = stbi__get8(s);
-               } else {
-                  stbi__skip(s, len);
-                  break;
+         case 0x21: // Comment Extension.
+         {
+            int len;
+            int ext = stbi__get8(s);
+            if (ext == 0xF9)
+            { // Graphic Control Extension.
+               len = stbi__get8(s);
+               if (len == 4)
+               {
+                  g->eflags = stbi__get8(s);
+                  g->delay = 10 * stbi__get16le(s); // delay - 1/100th of a second, saving as 1/1000ths.
+
+                  // unset old transparent
+                  if (g->transparent >= 0)
+                  {
+                     g->pal[g->transparent][3] = 255;
+                  }
+                  if (g->eflags & 0x01)
+                  {
+                     g->transparent = stbi__get8(s);
+                     if (g->transparent >= 0)
+                     {
+                        g->pal[g->transparent][3] = 0;
+                     }
+                  }
+                  else
+                  {
+                     // don't need transparent
+                     stbi__skip(s, 1);
+                     g->transparent = -1;
+                  }
+               }
+               else
+               {
+                  stbi__skip(s, len);
+                  break;
+               }
+            }
+            while ((len = stbi__get8(s)) != 0)
+            {
+               stbi__skip(s, len);
+            }
+            break;
+         }
+
+         case 0x3B: // gif stream termination code
+            return (stbi_uc*)s; // using '1' causes warning on some compilers
+
+         default:
+            return stbi__errpuc("unknown code", "Corrupt GIF");
+      }
+   }
+}
+
+static void* stbi__load_gif_main_outofmem(stbi__gif* g, stbi_uc* out, int** delays)
+{
+   STBI_FREE(g->out);
+   STBI_FREE(g->history);
+   STBI_FREE(g->background);
+
+   if (out)
+      STBI_FREE(out);
+   if (delays && *delays)
+      STBI_FREE(*delays);
+   return stbi__errpuc("outofmem", "Out of memory");
+}
+
+static void* stbi__load_gif_main(stbi__context* s, int** delays, int* x, int* y, int* z, int* comp, int req_comp)
+{
+   if (stbi__gif_test(s))
+   {
+      int layers = 0;
+      stbi_uc* u = 0;
+      stbi_uc* out = 0;
+      stbi_uc* two_back = 0;
+      stbi__gif g;
+      int stride;
+      int out_size = 0;
+      int delays_size = 0;
+
+      STBI_NOTUSED(out_size);
+      STBI_NOTUSED(delays_size);
+
+      memset(&g, 0, sizeof(g));
+      if (delays)
+      {
+         *delays = 0;
+      }
+
+      do
+      {
+         u = stbi__gif_load_next(s, &g, comp, req_comp, two_back);
+         if (u == (stbi_uc*)s)
+            u = 0; // end of animated gif marker
+
+         if (u)
+         {
+            *x = g.w;
+            *y = g.h;
+            ++layers;
+            stride = g.w * g.h * 4;
+
+            if (out)
+            {
+               void* tmp = (stbi_uc*)STBI_REALLOC_SIZED(out, out_size, layers * stride);
+               if (!tmp)
+                  return stbi__load_gif_main_outofmem(&g, out, delays);
+               else
+               {
+                  out = (stbi_uc*)tmp;
+                  out_size = layers * stride;
+               }
+
+               if (delays)
+               {
+                  int* new_delays = (int*)STBI_REALLOC_SIZED(*delays, delays_size, sizeof(int) * layers);
+                  if (!new_delays)
+                     return stbi__load_gif_main_outofmem(&g, out, delays);
+                  *delays = new_delays;
+                  delays_size = layers * sizeof(int);
                }
             }
-            while ((len = stbi__get8(s)) != 0)
-               stbi__skip(s, len);
-            break;
+            else
+            {
+               out = (stbi_uc*)stbi__malloc(layers * stride);
+               if (!out)
+                  return stbi__load_gif_main_outofmem(&g, out, delays);
+               out_size = layers * stride;
+               if (delays)
+               {
+                  *delays = (int*)stbi__malloc(layers * sizeof(int));
+                  if (!*delays)
+                     return stbi__load_gif_main_outofmem(&g, out, delays);
+                  delays_size = layers * sizeof(int);
+               }
+            }
+            memcpy(out + ((layers - 1) * stride), u, stride);
+            if (layers >= 2)
+            {
+               two_back = out - 2 * stride;
+            }
+
+            if (delays)
+            {
+               (*delays)[layers - 1U] = g.delay;
+            }
          }
+      } while (u != 0);
 
-         case 0x3B: // gif stream termination code
-            return (stbi_uc *) s; // using '1' causes warning on some compilers
+      // free temp buffer;
+      STBI_FREE(g.out);
+      STBI_FREE(g.history);
+      STBI_FREE(g.background);
 
-         default:
-            return stbi__errpuc("unknown code", "Corrupt GIF");
-      }
-   }
+      // do the final conversion after loading everything;
+      if (req_comp && req_comp != 4)
+         out = stbi__convert_format(out, 4, req_comp, layers * g.w, g.h);
 
-   STBI_NOTUSED(req_comp);
+      *z = layers;
+      return out;
+   }
+   else
+   {
+      return stbi__errpuc("not GIF", "Image was not as a gif type.");
+   }
 }
 
-static stbi_uc *stbi__gif_load(stbi__context *s, int *x, int *y, int *comp, int req_comp)
+static void* stbi__gif_load(stbi__context* s, int* x, int* y, int* comp, int req_comp, stbi__result_info* ri)
 {
-   stbi_uc *u = 0;
+   stbi_uc* u = 0;
    stbi__gif g;
    memset(&g, 0, sizeof(g));
+   STBI_NOTUSED(ri);
 
-   u = stbi__gif_load_next(s, &g, comp, req_comp);
-   if (u == (stbi_uc *) s) u = 0;  // end of animated gif marker
-   if (u) {
+   u = stbi__gif_load_next(s, &g, comp, req_comp, 0);
+   if (u == (stbi_uc*)s)
+      u = 0; // end of animated gif marker
+   if (u)
+   {
       *x = g.w;
       *y = g.h;
+
+      // moved conversion to after successful load so that the same
+      // can be done for multiple frames.
       if (req_comp && req_comp != 4)
          u = stbi__convert_format(u, 4, req_comp, g.w, g.h);
    }
    else if (g.out)
+   {
+      // if there was an error and we allocated an image buffer, free it!
       STBI_FREE(g.out);
+   }
+
+   // free buffers needed for multiple frame loading;
+   STBI_FREE(g.history);
+   STBI_FREE(g.background);
 
    return u;
 }
 
-static int stbi__gif_info(stbi__context *s, int *x, int *y, int *comp)
+static int stbi__gif_info(stbi__context* s, int* x, int* y, int* comp)
 {
-   return stbi__gif_info_raw(s,x,y,comp);
+   return stbi__gif_info_raw(s, x, y, comp);
 }
 #endif
 
@@ -5935,147 +8314,191 @@ static int stbi__gif_info(stbi__context *s, int *x, int *y, int *comp)
 // Radiance RGBE HDR loader
 // originally by Nicolas Schulz
 #ifndef STBI_NO_HDR
-static int stbi__hdr_test_core(stbi__context *s)
+static int stbi__hdr_test_core(stbi__context* s, const char* signature)
 {
-   const char *signature = "#?RADIANCE\n";
    int i;
-   for (i=0; signature[i]; ++i)
+   for (i = 0; signature[i]; ++i)
       if (stbi__get8(s) != signature[i])
          return 0;
+   stbi__rewind(s);
    return 1;
 }
 
 static int stbi__hdr_test(stbi__context* s)
 {
-   int r = stbi__hdr_test_core(s);
+   int r = stbi__hdr_test_core(s, "#?RADIANCE\n");
    stbi__rewind(s);
+   if (!r)
+   {
+      r = stbi__hdr_test_core(s, "#?RGBE\n");
+      stbi__rewind(s);
+   }
    return r;
 }
 
-#define STBI__HDR_BUFLEN  1024
-static char *stbi__hdr_gettoken(stbi__context *z, char *buffer)
+#define STBI__HDR_BUFLEN 1024
+static char* stbi__hdr_gettoken(stbi__context* z, char* buffer)
 {
-   int len=0;
+   int len = 0;
    char c = '\0';
 
-   c = (char) stbi__get8(z);
+   c = (char)stbi__get8(z);
 
-   while (!stbi__at_eof(z) && c != '\n') {
+   while (!stbi__at_eof(z) && c != '\n')
+   {
       buffer[len++] = c;
-      if (len == STBI__HDR_BUFLEN-1) {
+      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);
+      c = (char)stbi__get8(z);
    }
 
    buffer[len] = 0;
    return buffer;
 }
 
-static void stbi__hdr_convert(float *output, stbi_uc *input, int req_comp)
+static void stbi__hdr_convert(float* output, stbi_uc* input, int req_comp)
 {
-   if ( input[3] != 0 ) {
+   if (input[3] != 0)
+   {
       float f1;
       // Exponent
-      f1 = (float) ldexp(1.0f, input[3] - (int)(128 + 8));
+      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 {
+      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) {
+      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 3:
+            output[0] = output[1] = output[2] = 0;
+            break;
          case 2: output[1] = 1; /* fallthrough */
-         case 1: output[0] = 0;
-                 break;
+         case 1:
+            output[0] = 0;
+            break;
       }
    }
 }
 
-static float *stbi__hdr_load(stbi__context *s, int *x, int *y, int *comp, int req_comp)
+static float* stbi__hdr_load(stbi__context* s, int* x, int* y, int* comp, int req_comp, stbi__result_info* ri)
 {
    char buffer[STBI__HDR_BUFLEN];
-   char *token;
+   char* token;
    int valid = 0;
    int width, height;
-   stbi_uc *scanline;
-   float *hdr_data;
+   stbi_uc* scanline;
+   float* hdr_data;
    int len;
    unsigned char count, value;
-   int i, j, k, c1,c2, z;
-
+   int i, j, k, c1, c2, z;
+   const char* headerToken;
+   STBI_NOTUSED(ri);
 
    // Check identifier
-   if (strcmp(stbi__hdr_gettoken(s,buffer), "#?RADIANCE") != 0)
+   headerToken = stbi__hdr_gettoken(s, buffer);
+   if (strcmp(headerToken, "#?RADIANCE") != 0 && strcmp(headerToken, "#?RGBE") != 0)
       return stbi__errpf("not HDR", "Corrupt HDR image");
 
    // Parse header
-   for(;;) {
-      token = stbi__hdr_gettoken(s,buffer);
-      if (token[0] == 0) break;
-      if (strcmp(token, "FORMAT=32-bit_rle_rgbe") == 0) valid = 1;
+   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");
+   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 = 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");
+   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);
+   width = (int)strtol(token, NULL, 10);
+
+   if (height > STBI_MAX_DIMENSIONS)
+      return stbi__errpf("too large", "Very large image (corrupt?)");
+   if (width > STBI_MAX_DIMENSIONS)
+      return stbi__errpf("too large", "Very large image (corrupt?)");
 
    *x = width;
    *y = height;
 
-   if (comp) *comp = 3;
-   if (req_comp == 0) req_comp = 3;
+   if (comp)
+      *comp = 3;
+   if (req_comp == 0)
+      req_comp = 3;
+
+   if (!stbi__mad4sizes_valid(width, height, req_comp, sizeof(float), 0))
+      return stbi__errpf("too large", "HDR image is too large");
 
    // Read data
-   hdr_data = (float *) stbi__malloc(height * width * req_comp * sizeof(float));
+   hdr_data = (float*)stbi__malloc_mad4(width, height, req_comp, sizeof(float), 0);
+   if (!hdr_data)
+      return stbi__errpf("outofmem", "Out of memory");
 
    // Load image data
    // image data is stored as some number of sca
-   if ( width < 8 || width >= 32768) {
+   if (width < 8 || width >= 32768)
+   {
       // Read flat data
-      for (j=0; j < height; ++j) {
-         for (i=0; i < width; ++i) {
+      for (j = 0; j < height; ++j)
+      {
+         for (i = 0; i < width; ++i)
+         {
             stbi_uc rgbe[4];
-           main_decode_loop:
+         main_decode_loop:
             stbi__getn(s, rgbe, 4);
             stbi__hdr_convert(hdr_data + j * width * req_comp + i * req_comp, rgbe, req_comp);
          }
       }
-   } else {
+   }
+   else
+   {
       // Read RLE-encoded data
       scanline = NULL;
 
-      for (j = 0; j < height; ++j) {
+      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)) {
+         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);
+            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;
@@ -6084,133 +8507,242 @@ static float *stbi__hdr_load(stbi__context *s, int *x, int *y, int *comp, int re
          }
          len <<= 8;
          len |= stbi__get8(s);
-         if (len != width) { STBI_FREE(hdr_data); STBI_FREE(scanline); return stbi__errpf("invalid decoded scanline length", "corrupt HDR"); }
-         if (scanline == NULL) scanline = (stbi_uc *) stbi__malloc(width * 4);
+         if (len != width)
+         {
+            STBI_FREE(hdr_data);
+            STBI_FREE(scanline);
+            return stbi__errpf("invalid decoded scanline length", "corrupt HDR");
+         }
+         if (scanline == NULL)
+         {
+            scanline = (stbi_uc*)stbi__malloc_mad2(width, 4, 0);
+            if (!scanline)
+            {
+               STBI_FREE(hdr_data);
+               return stbi__errpf("outofmem", "Out of memory");
+            }
+         }
 
-         for (k = 0; k < 4; ++k) {
+         for (k = 0; k < 4; ++k)
+         {
+            int nleft;
             i = 0;
-            while (i < width) {
+            while ((nleft = width - i) > 0)
+            {
                count = stbi__get8(s);
-               if (count > 128) {
+               if (count > 128)
+               {
                   // Run
                   value = stbi__get8(s);
                   count -= 128;
+                  if ((count == 0) || (count > nleft))
+                  {
+                     STBI_FREE(hdr_data);
+                     STBI_FREE(scanline);
+                     return stbi__errpf("corrupt", "bad RLE data in HDR");
+                  }
                   for (z = 0; z < count; ++z)
                      scanline[i++ * 4 + k] = value;
-               } else {
+               }
+               else
+               {
                   // Dump
+                  if ((count == 0) || (count > nleft))
+                  {
+                     STBI_FREE(hdr_data);
+                     STBI_FREE(scanline);
+                     return stbi__errpf("corrupt", "bad RLE data in HDR");
+                  }
                   for (z = 0; z < count; ++z)
                      scanline[i++ * 4 + k] = stbi__get8(s);
                }
             }
          }
-         for (i=0; i < width; ++i)
-            stbi__hdr_convert(hdr_data+(j*width + i)*req_comp, scanline + i*4, req_comp);
+         for (i = 0; i < width; ++i)
+            stbi__hdr_convert(hdr_data + (j * width + i) * req_comp, scanline + i * 4, req_comp);
       }
-      STBI_FREE(scanline);
+      if (scanline)
+         STBI_FREE(scanline);
    }
 
    return hdr_data;
 }
 
-static int stbi__hdr_info(stbi__context *s, int *x, int *y, int *comp)
+static int stbi__hdr_info(stbi__context* s, int* x, int* y, int* comp)
 {
    char buffer[STBI__HDR_BUFLEN];
-   char *token;
+   char* token;
    int valid = 0;
+   int dummy;
 
-   if (stbi__hdr_test(s) == 0) {
-       stbi__rewind( s );
-       return 0;
+   if (!x)
+      x = &dummy;
+   if (!y)
+      y = &dummy;
+   if (!comp)
+      comp = &dummy;
+
+   if (stbi__hdr_test(s) == 0)
+   {
+      stbi__rewind(s);
+      return 0;
    }
 
-   for(;;) {
-      token = stbi__hdr_gettoken(s,buffer);
-      if (token[0] == 0) break;
-      if (strcmp(token, "FORMAT=32-bit_rle_rgbe") == 0) valid = 1;
+   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;
+   if (!valid)
+   {
+      stbi__rewind(s);
+      return 0;
    }
-   token = stbi__hdr_gettoken(s,buffer);
-   if (strncmp(token, "-Y ", 3)) {
-       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;
+   *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);
+   *x = (int)strtol(token, NULL, 10);
    *comp = 3;
    return 1;
 }
 #endif // STBI_NO_HDR
 
 #ifndef STBI_NO_BMP
-static int stbi__bmp_info(stbi__context *s, int *x, int *y, int *comp)
+static int stbi__bmp_info(stbi__context* s, int* x, int* y, int* comp)
 {
-   void *p;
+   void* p;
    stbi__bmp_data info;
 
-   info.all_a = 255;   
+   info.all_a = 255;
    p = stbi__bmp_parse_header(s, &info);
-   stbi__rewind( s );
    if (p == NULL)
+   {
+      stbi__rewind(s);
       return 0;
-   *x = s->img_x;
-   *y = s->img_y;
-   *comp = info.ma ? 4 : 3;
+   }
+   if (x)
+      *x = s->img_x;
+   if (y)
+      *y = s->img_y;
+   if (comp)
+   {
+      if (info.bpp == 24 && info.ma == 0xff000000)
+         *comp = 3;
+      else
+         *comp = info.ma ? 4 : 3;
+   }
    return 1;
 }
 #endif
 
 #ifndef STBI_NO_PSD
-static int stbi__psd_info(stbi__context *s, int *x, int *y, int *comp)
-{
-   int channelCount;
-   if (stbi__get32be(s) != 0x38425053) {
-       stbi__rewind( s );
-       return 0;
+static int stbi__psd_info(stbi__context* s, int* x, int* y, int* comp)
+{
+   int channelCount, dummy, depth;
+   if (!x)
+      x = &dummy;
+   if (!y)
+      y = &dummy;
+   if (!comp)
+      comp = &dummy;
+   if (stbi__get32be(s) != 0x38425053)
+   {
+      stbi__rewind(s);
+      return 0;
    }
-   if (stbi__get16be(s) != 1) {
-       stbi__rewind( s );
-       return 0;
+   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;
+   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;
+   depth = stbi__get16be(s);
+   if (depth != 8 && depth != 16)
+   {
+      stbi__rewind(s);
+      return 0;
    }
-   if (stbi__get16be(s) != 3) {
-       stbi__rewind( s );
-       return 0;
+   if (stbi__get16be(s) != 3)
+   {
+      stbi__rewind(s);
+      return 0;
    }
    *comp = 4;
    return 1;
 }
+
+static int stbi__psd_is16(stbi__context* s)
+{
+   int channelCount, depth;
+   if (stbi__get32be(s) != 0x38425053)
+   {
+      stbi__rewind(s);
+      return 0;
+   }
+   if (stbi__get16be(s) != 1)
+   {
+      stbi__rewind(s);
+      return 0;
+   }
+   stbi__skip(s, 6);
+   channelCount = stbi__get16be(s);
+   if (channelCount < 0 || channelCount > 16)
+   {
+      stbi__rewind(s);
+      return 0;
+   }
+   STBI_NOTUSED(stbi__get32be(s));
+   STBI_NOTUSED(stbi__get32be(s));
+   depth = stbi__get16be(s);
+   if (depth != 16)
+   {
+      stbi__rewind(s);
+      return 0;
+   }
+   return 1;
+}
 #endif
 
 #ifndef STBI_NO_PIC
-static int stbi__pic_info(stbi__context *s, int *x, int *y, int *comp)
+static int stbi__pic_info(stbi__context* s, int* x, int* y, int* comp)
 {
-   int act_comp=0,num_packets=0,chained;
+   int act_comp = 0, num_packets = 0, chained, dummy;
    stbi__pic_packet packets[10];
 
-   if (!stbi__pic_is4(s,"\x53\x80\xF6\x34")) {
+   if (!x)
+      x = &dummy;
+   if (!y)
+      y = &dummy;
+   if (!comp)
+      comp = &dummy;
+
+   if (!stbi__pic_is4(s, "\x53\x80\xF6\x34"))
+   {
       stbi__rewind(s);
       return 0;
    }
@@ -6219,37 +8751,42 @@ static int stbi__pic_info(stbi__context *s, int *x, int *y, int *comp)
 
    *x = stbi__get16be(s);
    *y = stbi__get16be(s);
-   if (stbi__at_eof(s)) {
-      stbi__rewind( s);
+   if (stbi__at_eof(s))
+   {
+      stbi__rewind(s);
       return 0;
    }
-   if ( (*x) != 0 && (1 << 28) / (*x) < (*y)) {
-      stbi__rewind( s );
+   if ((*x) != 0 && (1 << 28) / (*x) < (*y))
+   {
+      stbi__rewind(s);
       return 0;
    }
 
    stbi__skip(s, 8);
 
-   do {
-      stbi__pic_packet *packet;
+   do
+   {
+      stbi__pic_packet* packet;
 
-      if (num_packets==sizeof(packets)/sizeof(packets[0]))
+      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->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 (stbi__at_eof(s))
+      {
+         stbi__rewind(s);
+         return 0;
       }
-      if (packet->size != 8) {
-          stbi__rewind( s );
-          return 0;
+      if (packet->size != 8)
+      {
+         stbi__rewind(s);
+         return 0;
       }
    } while (chained);
 
@@ -6269,196 +8806,338 @@ static int stbi__pic_info(stbi__context *s, int *x, int *y, int *comp)
 // Known limitations:
 //    Does not support comments in the header section
 //    Does not support ASCII image data (formats P2 and P3)
-//    Does not support 16-bit-per-channel
 
 #ifndef STBI_NO_PNM
 
-static int      stbi__pnm_test(stbi__context *s)
+static int stbi__pnm_test(stbi__context* s)
 {
    char p, t;
-   p = (char) stbi__get8(s);
-   t = (char) stbi__get8(s);
-   if (p != 'P' || (t != '5' && t != '6')) {
-       stbi__rewind( s );
-       return 0;
+   p = (char)stbi__get8(s);
+   t = (char)stbi__get8(s);
+   if (p != 'P' || (t != '5' && t != '6'))
+   {
+      stbi__rewind(s);
+      return 0;
    }
    return 1;
 }
 
-static stbi_uc *stbi__pnm_load(stbi__context *s, int *x, int *y, int *comp, int req_comp)
+static void* stbi__pnm_load(stbi__context* s, int* x, int* y, int* comp, int req_comp, stbi__result_info* ri)
 {
-   stbi_uc *out;
-   if (!stbi__pnm_info(s, (int *)&s->img_x, (int *)&s->img_y, (int *)&s->img_n))
+   stbi_uc* out;
+   STBI_NOTUSED(ri);
+
+   ri->bits_per_channel = stbi__pnm_info(s, (int*)&s->img_x, (int*)&s->img_y, (int*)&s->img_n);
+   if (ri->bits_per_channel == 0)
       return 0;
+
+   if (s->img_y > STBI_MAX_DIMENSIONS)
+      return stbi__errpuc("too large", "Very large image (corrupt?)");
+   if (s->img_x > STBI_MAX_DIMENSIONS)
+      return stbi__errpuc("too large", "Very large image (corrupt?)");
+
    *x = s->img_x;
    *y = s->img_y;
-   *comp = s->img_n;
+   if (comp)
+      *comp = s->img_n;
 
-   out = (stbi_uc *) stbi__malloc(s->img_n * s->img_x * s->img_y);
-   if (!out) return stbi__errpuc("outofmem", "Out of memory");
-   stbi__getn(s, out, s->img_n * s->img_x * s->img_y);
+   if (!stbi__mad4sizes_valid(s->img_n, s->img_x, s->img_y, ri->bits_per_channel / 8, 0))
+      return stbi__errpuc("too large", "PNM too large");
+
+   out = (stbi_uc*)stbi__malloc_mad4(s->img_n, s->img_x, s->img_y, ri->bits_per_channel / 8, 0);
+   if (!out)
+      return stbi__errpuc("outofmem", "Out of memory");
+   if (!stbi__getn(s, out, s->img_n * s->img_x * s->img_y * (ri->bits_per_channel / 8)))
+   {
+      STBI_FREE(out);
+      return stbi__errpuc("bad PNM", "PNM file truncated");
+   }
 
-   if (req_comp && req_comp != s->img_n) {
-      out = stbi__convert_format(out, s->img_n, req_comp, s->img_x, s->img_y);
-      if (out == NULL) return out; // stbi__convert_format frees input on failure
+   if (req_comp && req_comp != s->img_n)
+   {
+      if (ri->bits_per_channel == 16)
+      {
+         out = (stbi_uc*)stbi__convert_format16((stbi__uint16*)out, s->img_n, req_comp, s->img_x, s->img_y);
+      }
+      else
+      {
+         out = stbi__convert_format(out, s->img_n, req_comp, s->img_x, s->img_y);
+      }
+      if (out == NULL)
+         return out; // stbi__convert_format frees input on failure
    }
    return out;
 }
 
-static int      stbi__pnm_isspace(char c)
+static int stbi__pnm_isspace(char c)
 {
    return c == ' ' || c == '\t' || c == '\n' || c == '\v' || c == '\f' || c == '\r';
 }
 
-static void     stbi__pnm_skip_whitespace(stbi__context *s, char *c)
+static void stbi__pnm_skip_whitespace(stbi__context* s, char* c)
 {
-   for (;;) {
+   for (;;)
+   {
       while (!stbi__at_eof(s) && stbi__pnm_isspace(*c))
-         *c = (char) stbi__get8(s);
+         *c = (char)stbi__get8(s);
 
       if (stbi__at_eof(s) || *c != '#')
          break;
 
-      while (!stbi__at_eof(s) && *c != '\n' && *c != '\r' )
-         *c = (char) stbi__get8(s);
+      while (!stbi__at_eof(s) && *c != '\n' && *c != '\r')
+         *c = (char)stbi__get8(s);
    }
 }
 
-static int      stbi__pnm_isdigit(char c)
+static int stbi__pnm_isdigit(char c)
 {
    return c >= '0' && c <= '9';
 }
 
-static int      stbi__pnm_getinteger(stbi__context *s, char *c)
+static int stbi__pnm_getinteger(stbi__context* s, char* c)
 {
    int value = 0;
 
-   while (!stbi__at_eof(s) && stbi__pnm_isdigit(*c)) {
-      value = value*10 + (*c - '0');
-      *c = (char) stbi__get8(s);
+   while (!stbi__at_eof(s) && stbi__pnm_isdigit(*c))
+   {
+      value = value * 10 + (*c - '0');
+      *c = (char)stbi__get8(s);
+      if ((value > 214748364) || (value == 214748364 && *c > '7'))
+         return stbi__err("integer parse overflow", "Parsing an integer in the PPM header overflowed a 32-bit int");
    }
 
    return value;
 }
 
-static int      stbi__pnm_info(stbi__context *s, int *x, int *y, int *comp)
+static int stbi__pnm_info(stbi__context* s, int* x, int* y, int* comp)
 {
-   int maxv;
+   int maxv, dummy;
    char c, p, t;
 
-   stbi__rewind( s );
+   if (!x)
+      x = &dummy;
+   if (!y)
+      y = &dummy;
+   if (!comp)
+      comp = &dummy;
+
+   stbi__rewind(s);
 
    // Get identifier
-   p = (char) stbi__get8(s);
-   t = (char) stbi__get8(s);
-   if (p != 'P' || (t != '5' && t != '6')) {
-       stbi__rewind( s );
-       return 0;
+   p = (char)stbi__get8(s);
+   t = (char)stbi__get8(s);
+   if (p != 'P' || (t != '5' && t != '6'))
+   {
+      stbi__rewind(s);
+      return 0;
    }
 
-   *comp = (t == '6') ? 3 : 1;  // '5' is 1-component .pgm; '6' is 3-component .ppm
+   *comp = (t == '6') ? 3 : 1; // '5' is 1-component .pgm; '6' is 3-component .ppm
 
-   c = (char) stbi__get8(s);
+   c = (char)stbi__get8(s);
    stbi__pnm_skip_whitespace(s, &c);
 
    *x = stbi__pnm_getinteger(s, &c); // read width
+   if (*x == 0)
+      return stbi__err("invalid width", "PPM image header had zero or overflowing width");
    stbi__pnm_skip_whitespace(s, &c);
 
    *y = stbi__pnm_getinteger(s, &c); // read height
+   if (*y == 0)
+      return stbi__err("invalid width", "PPM image header had zero or overflowing width");
    stbi__pnm_skip_whitespace(s, &c);
 
-   maxv = stbi__pnm_getinteger(s, &c);  // read max value
-
-   if (maxv > 255)
-      return stbi__err("max value > 255", "PPM image not 8-bit");
+   maxv = stbi__pnm_getinteger(s, &c); // read max value
+   if (maxv > 65535)
+      return stbi__err("max value > 65535", "PPM image supports only 8-bit and 16-bit images");
+   else if (maxv > 255)
+      return 16;
    else
+      return 8;
+}
+
+static int stbi__pnm_is16(stbi__context* s)
+{
+   if (stbi__pnm_info(s, NULL, NULL, NULL) == 16)
       return 1;
+   return 0;
 }
 #endif
 
-static int stbi__info_main(stbi__context *s, int *x, int *y, int *comp)
+static int stbi__info_main(stbi__context* s, int* x, int* y, int* comp)
 {
-   #ifndef STBI_NO_JPEG
-   if (stbi__jpeg_info(s, x, y, comp)) return 1;
-   #endif
+#ifndef STBI_NO_JPEG
+   if (stbi__jpeg_info(s, x, y, comp))
+      return 1;
+#endif
 
-   #ifndef STBI_NO_PNG
-   if (stbi__png_info(s, x, y, comp))  return 1;
-   #endif
+#ifndef STBI_NO_PNG
+   if (stbi__png_info(s, x, y, comp))
+      return 1;
+#endif
 
-   #ifndef STBI_NO_GIF
-   if (stbi__gif_info(s, x, y, comp))  return 1;
-   #endif
+#ifndef STBI_NO_GIF
+   if (stbi__gif_info(s, x, y, comp))
+      return 1;
+#endif
 
-   #ifndef STBI_NO_BMP
-   if (stbi__bmp_info(s, x, y, comp))  return 1;
-   #endif
+#ifndef STBI_NO_BMP
+   if (stbi__bmp_info(s, x, y, comp))
+      return 1;
+#endif
 
-   #ifndef STBI_NO_PSD
-   if (stbi__psd_info(s, x, y, comp))  return 1;
-   #endif
+#ifndef STBI_NO_PSD
+   if (stbi__psd_info(s, x, y, comp))
+      return 1;
+#endif
 
-   #ifndef STBI_NO_PIC
-   if (stbi__pic_info(s, x, y, comp))  return 1;
-   #endif
+#ifndef STBI_NO_PIC
+   if (stbi__pic_info(s, x, y, comp))
+      return 1;
+#endif
 
-   #ifndef STBI_NO_PNM
-   if (stbi__pnm_info(s, x, y, comp))  return 1;
-   #endif
+#ifndef STBI_NO_PNM
+   if (stbi__pnm_info(s, x, y, comp))
+      return 1;
+#endif
 
-   #ifndef STBI_NO_HDR
-   if (stbi__hdr_info(s, x, y, comp))  return 1;
-   #endif
+#ifndef STBI_NO_HDR
+   if (stbi__hdr_info(s, x, y, comp))
+      return 1;
+#endif
 
-   // test tga last because it's a crappy test!
-   #ifndef STBI_NO_TGA
+// test tga last because it's a crappy test!
+#ifndef STBI_NO_TGA
    if (stbi__tga_info(s, x, y, comp))
-       return 1;
-   #endif
+      return 1;
+#endif
    return stbi__err("unknown image type", "Image not of any known type, or corrupt");
 }
 
+static int stbi__is_16_main(stbi__context* s)
+{
+#ifndef STBI_NO_PNG
+   if (stbi__png_is16(s))
+      return 1;
+#endif
+
+#ifndef STBI_NO_PSD
+   if (stbi__psd_is16(s))
+      return 1;
+#endif
+
+#ifndef STBI_NO_PNM
+   if (stbi__pnm_is16(s))
+      return 1;
+#endif
+   return 0;
+}
+
 #ifndef STBI_NO_STDIO
-STBIDEF int stbi_info(char const *filename, int *x, int *y, int *comp)
+STBIDEF int stbi_info(char const* filename, int* x, int* y, int* comp)
+{
+   FILE* f = stbi__fopen(filename, "rb");
+   int result;
+   if (!f)
+      return stbi__err("can't fopen", "Unable to open file");
+   result = stbi_info_from_file(f, x, y, comp);
+   fclose(f);
+   return result;
+}
+
+STBIDEF int stbi_info_from_file(FILE* f, int* x, int* y, int* comp)
+{
+   int r;
+   stbi__context s;
+   long pos = ftell(f);
+   stbi__start_file(&s, f);
+   r = stbi__info_main(&s, x, y, comp);
+   fseek(f, pos, SEEK_SET);
+   return r;
+}
+
+STBIDEF int stbi_is_16_bit(char const* filename)
 {
-    FILE *f = stbi__fopen(filename, "rb");
-    int result;
-    if (!f) return stbi__err("can't fopen", "Unable to open file");
-    result = stbi_info_from_file(f, x, y, comp);
-    fclose(f);
-    return result;
+   FILE* f = stbi__fopen(filename, "rb");
+   int result;
+   if (!f)
+      return stbi__err("can't fopen", "Unable to open file");
+   result = stbi_is_16_bit_from_file(f);
+   fclose(f);
+   return result;
 }
 
-STBIDEF int stbi_info_from_file(FILE *f, int *x, int *y, int *comp)
+STBIDEF int stbi_is_16_bit_from_file(FILE* f)
 {
    int r;
    stbi__context s;
    long pos = ftell(f);
    stbi__start_file(&s, f);
-   r = stbi__info_main(&s,x,y,comp);
-   fseek(f,pos,SEEK_SET);
+   r = stbi__is_16_main(&s);
+   fseek(f, pos, SEEK_SET);
    return r;
 }
 #endif // !STBI_NO_STDIO
 
-STBIDEF int stbi_info_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp)
+STBIDEF int stbi_info_from_memory(stbi_uc const* buffer, int len, int* x, int* y, int* comp)
+{
+   stbi__context s;
+   stbi__start_mem(&s, buffer, len);
+   return stbi__info_main(&s, x, y, comp);
+}
+
+STBIDEF int stbi_info_from_callbacks(stbi_io_callbacks const* c, void* user, int* x, int* y, int* comp)
+{
+   stbi__context s;
+   stbi__start_callbacks(&s, (stbi_io_callbacks*)c, user);
+   return stbi__info_main(&s, x, y, comp);
+}
+
+STBIDEF int stbi_is_16_bit_from_memory(stbi_uc const* buffer, int len)
 {
    stbi__context s;
-   stbi__start_mem(&s,buffer,len);
-   return stbi__info_main(&s,x,y,comp);
+   stbi__start_mem(&s, buffer, len);
+   return stbi__is_16_main(&s);
 }
 
-STBIDEF int stbi_info_from_callbacks(stbi_io_callbacks const *c, void *user, int *x, int *y, int *comp)
+STBIDEF int stbi_is_16_bit_from_callbacks(stbi_io_callbacks const* c, void* user)
 {
    stbi__context s;
-   stbi__start_callbacks(&s, (stbi_io_callbacks *) c, user);
-   return stbi__info_main(&s,x,y,comp);
+   stbi__start_callbacks(&s, (stbi_io_callbacks*)c, user);
+   return stbi__is_16_main(&s);
 }
 
 #endif // STB_IMAGE_IMPLEMENTATION
 
 /*
    revision history:
+      2.20  (2019-02-07) support utf8 filenames in Windows; fix warnings and platform ifdefs
+      2.19  (2018-02-11) fix warning
+      2.18  (2018-01-30) fix warnings
+      2.17  (2018-01-29) change sbti__shiftsigned to avoid clang -O2 bug
+                         1-bit BMP
+                         *_is_16_bit api
+                         avoid warnings
+      2.16  (2017-07-23) all functions have 16-bit variants;
+                         STBI_NO_STDIO works again;
+                         compilation fixes;
+                         fix rounding in unpremultiply;
+                         optimize vertical flip;
+                         disable raw_len validation;
+                         documentation fixes
+      2.15  (2017-03-18) fix png-1,2,4 bug; now all Imagenet JPGs decode;
+                         warning fixes; disable run-time SSE detection on gcc;
+                         uniform handling of optional "return" values;
+                         thread-safe initialization of zlib tables
+      2.14  (2017-03-03) remove deprecated STBI_JPEG_OLD; fixes for Imagenet JPGs
+      2.13  (2016-11-29) add 16-bit API, only supported for PNG right now
+      2.12  (2016-04-02) fix typo in 2.11 PSD fix that caused crashes
+      2.11  (2016-04-02) allocate large structures on the stack
+                         remove white matting for transparent PSD
+                         fix reported channel count for PNG & BMP
+                         re-enable SSE2 in non-gcc 64-bit
+                         support RGB-formatted JPEG
+                         read 16-bit PNGs (only as 8-bit)
       2.10  (2016-01-22) avoid warning introduced in 2.09 by STBI_REALLOC_SIZED
       2.09  (2016-01-16) allow comments in PNM files
                          16-bit-per-pixel TGA (not bit-per-component)
@@ -6612,3 +9291,46 @@ STBIDEF int stbi_info_from_callbacks(stbi_io_callbacks const *c, void *user, int
       0.50  (2006-11-19)
               first released version
 */
+
+
+/*
+------------------------------------------------------------------------------
+This software is available under 2 licenses -- choose whichever you prefer.
+------------------------------------------------------------------------------
+ALTERNATIVE A - MIT License
+Copyright (c) 2017 Sean Barrett
+Permission is hereby granted, free of charge, to any person obtaining a copy of
+this software and associated documentation files (the "Software"), to deal in
+the Software without restriction, including without limitation the rights to
+use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies
+of the Software, and to permit persons to whom the Software is furnished to do
+so, subject to the following conditions:
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
+------------------------------------------------------------------------------
+ALTERNATIVE B - Public Domain (www.unlicense.org)
+This is free and unencumbered software released into the public domain.
+Anyone is free to copy, modify, publish, use, compile, sell, or distribute this
+software, either in source code form or as a compiled binary, for any purpose,
+commercial or non-commercial, and by any means.
+In jurisdictions that recognize copyright laws, the author or authors of this
+software dedicate any and all copyright interest in the software to the public
+domain. We make this dedication for the benefit of the public at large and to
+the detriment of our heirs and successors. We intend this dedication to be an
+overt act of relinquishment in perpetuity of all present and future rights to
+this software under copyright law.
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+------------------------------------------------------------------------------
+*/
\ No newline at end of file
diff --git a/libs/nanovg/nanovg/stb_truetype.h b/libs/nanovg/nanovg/stb_truetype.h
index d04908c84..445d368fa 100755
--- a/libs/nanovg/nanovg/stb_truetype.h
+++ b/libs/nanovg/nanovg/stb_truetype.h
@@ -1,11 +1,21 @@
-// stb_truetype.h - v1.09 - public domain
-// authored from 2009-2015 by Sean Barrett / RAD Game Tools
+// stb_truetype.h - v1.26 - public domain
+// authored from 2009-2021 by Sean Barrett / RAD Game Tools
+//
+// =======================================================================
+//
+//    NO SECURITY GUARANTEE -- DO NOT USE THIS ON UNTRUSTED FONT FILES
+//
+// This library does no range checking of the offsets found in the file,
+// meaning an attacker can use it to read arbitrary memory.
+//
+// =======================================================================
 //
 //   This library processes TrueType files:
 //        parse files
 //        extract glyph metrics
 //        extract glyph shapes
 //        render glyphs to one-channel bitmaps with antialiasing (box filter)
+//        render glyphs to one-channel SDF bitmaps (signed-distance field/function)
 //
 //   Todo:
 //        non-MS cmaps
@@ -20,36 +30,51 @@
 //
 //   Mikko Mononen: compound shape support, more cmap formats
 //   Tor Andersson: kerning, subpixel rendering
-//
-//   Bug/warning reports/fixes:
-//       "Zer" on mollyrocket (with fix)
-//       Cass Everitt
-//       stoiko (Haemimont Games)
-//       Brian Hook 
-//       Walter van Niftrik
-//       David Gow
-//       David Given
-//       Ivan-Assen Ivanov
-//       Anthony Pesch
-//       Johan Duparc
-//       Hou Qiming
-//       Fabian "ryg" Giesen
-//       Martins Mozeiko
-//       Cap Petschulat
-//       Omar Cornut
-//       github:aloucks
-//       Peter LaValle
-//       Sergey Popov
-//       Giumo X. Clanjor
-//       Higor Euripedes
-//       Thomas Fields
-//       Derek Vinyard
+//   Dougall Johnson: OpenType / Type 2 font handling
+//   Daniel Ribeiro Maciel: basic GPOS-based kerning
 //
 //   Misc other:
 //       Ryan Gordon
+//       Simon Glass
+//       github:IntellectualKitty
+//       Imanol Celaya
+//       Daniel Ribeiro Maciel
+//
+//   Bug/warning reports/fixes:
+//       "Zer" on mollyrocket       Fabian "ryg" Giesen   github:NiLuJe
+//       Cass Everitt               Martins Mozeiko       github:aloucks
+//       stoiko (Haemimont Games)   Cap Petschulat        github:oyvindjam
+//       Brian Hook                 Omar Cornut           github:vassvik
+//       Walter van Niftrik         Ryan Griege
+//       David Gow                  Peter LaValle
+//       David Given                Sergey Popov
+//       Ivan-Assen Ivanov          Giumo X. Clanjor
+//       Anthony Pesch              Higor Euripedes
+//       Johan Duparc               Thomas Fields
+//       Hou Qiming                 Derek Vinyard
+//       Rob Loach                  Cort Stratton
+//       Kenney Phillis Jr.         Brian Costabile
+//       Ken Voskuil (kaesve)
 //
 // VERSION HISTORY
 //
+//   1.26 (2021-08-28) fix broken rasterizer
+//   1.25 (2021-07-11) many fixes
+//   1.24 (2020-02-05) fix warning
+//   1.23 (2020-02-02) query SVG data for glyphs; query whole kerning table (but only kern not GPOS)
+//   1.22 (2019-08-11) minimize missing-glyph duplication; fix kerning if both 'GPOS' and 'kern' are defined
+//   1.21 (2019-02-25) fix warning
+//   1.20 (2019-02-07) PackFontRange skips missing codepoints; GetScaleFontVMetrics()
+//   1.19 (2018-02-11) GPOS kerning, STBTT_fmod
+//   1.18 (2018-01-29) add missing function
+//   1.17 (2017-07-23) make more arguments const; doc fix
+//   1.16 (2017-07-12) SDF support
+//   1.15 (2017-03-03) make more arguments const
+//   1.14 (2017-01-16) num-fonts-in-TTC function
+//   1.13 (2017-01-02) support OpenType fonts, certain Apple fonts
+//   1.12 (2016-10-25) suppress warnings about casting away const with -Wcast-qual
+//   1.11 (2016-04-02) fix unused-variable warning
+//   1.10 (2016-04-02) user-defined fabs(); rare memory leak; remove duplicate typedef
 //   1.09 (2016-01-16) warning fix; avoid crash on outofmem; use allocation userdata properly
 //   1.08 (2015-09-13) document stbtt_Rasterize(); fixes for vertical & horizontal edges
 //   1.07 (2015-08-01) allow PackFontRanges to accept arrays of sparse codepoints;
@@ -57,24 +82,16 @@
 //                     fix stbtt_GetFontOFfsetForIndex (never worked for non-0 input?);
 //                     fixed an assert() bug in the new rasterizer
 //                     replace assert() with STBTT_assert() in new rasterizer
-//   1.06 (2015-07-14) performance improvements (~35% faster on x86 and x64 on test machine)
-//                     also more precise AA rasterizer, except if shapes overlap
-//                     remove need for STBTT_sort
-//   1.05 (2015-04-15) fix misplaced definitions for STBTT_STATIC
-//   1.04 (2015-04-15) typo in example
-//   1.03 (2015-04-12) STBTT_STATIC, fix memory leak in new packing, various fixes
 //
 //   Full history can be found at the end of this file.
 //
 // LICENSE
 //
-//   This software is in the public domain. Where that dedication is not
-//   recognized, you are granted a perpetual, irrevocable license to copy,
-//   distribute, and modify this file as you see fit.
+//   See end of file for license information.
 //
 // USAGE
 //
-//   Include this file in whatever places neeed to refer to it. In ONE C/C++
+//   Include this file in whatever places need 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
@@ -90,14 +107,15 @@
 //   Improved 3D API (more shippable):
 //           #include "stb_rect_pack.h"           -- optional, but you really want it
 //           stbtt_PackBegin()
-//           stbtt_PackSetOversample()            -- for improved quality on small fonts
+//           stbtt_PackSetOversampling()          -- for improved quality on small fonts
 //           stbtt_PackFontRanges()               -- pack and renders
 //           stbtt_PackEnd()
 //           stbtt_GetPackedQuad()
 //
 //   "Load" a font file from a memory buffer (you have to keep the buffer loaded)
 //           stbtt_InitFont()
-//           stbtt_GetFontOffsetForIndex()        -- use for TTC font collections
+//           stbtt_GetFontOffsetForIndex()        -- indexing for TTC font collections
+//           stbtt_GetNumberOfFonts()             -- number of fonts for TTC font collections
 //
 //   Render a unicode codepoint to a bitmap
 //           stbtt_GetCodepointBitmap()           -- allocates and returns a bitmap
@@ -107,6 +125,7 @@
 //   Character advance/positioning
 //           stbtt_GetCodepointHMetrics()
 //           stbtt_GetFontVMetrics()
+//           stbtt_GetFontVMetricsOS2()
 //           stbtt_GetCodepointKernAdvance()
 //
 //   Starting with version 1.06, the rasterizer was replaced with a new,
@@ -162,7 +181,7 @@
 //         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
+//         since 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
@@ -172,6 +191,39 @@
 //         for non-commercial fonts, thus making fonts scaled in points
 //         according to the TrueType spec incoherently sized in practice.
 //
+// DETAILED USAGE:
+//
+//  Scale:
+//    Select how high you want the font to be, in points or pixels.
+//    Call ScaleForPixelHeight or ScaleForMappingEmToPixels to compute
+//    a scale factor SF that will be used by all other functions.
+//
+//  Baseline:
+//    You need to select a y-coordinate that is the baseline of where
+//    your text will appear. Call GetFontBoundingBox to get the baseline-relative
+//    bounding box for all characters. SF*-y0 will be the distance in pixels
+//    that the worst-case character could extend above the baseline, so if
+//    you want the top edge of characters to appear at the top of the
+//    screen where y=0, then you would set the baseline to SF*-y0.
+//
+//  Current point:
+//    Set the current point where the first character will appear. The
+//    first character could extend left of the current point; this is font
+//    dependent. You can either choose a current point that is the leftmost
+//    point and hope, or add some padding, or check the bounding box or
+//    left-side-bearing of the first character to be displayed and set
+//    the current point based on that.
+//
+//  Displaying a character:
+//    Compute the bounding box of the character. It will contain signed values
+//    relative to <current_point, baseline>. I.e. if it returns x0,y0,x1,y1,
+//    then the character should be displayed in the rectangle from
+//    <current_point+SF*x0, baseline+SF*y0> to <current_point+SF*x1,baseline+SF*y1).
+//
+//  Advancing for the next character:
+//    Call GlyphHMetrics, and compute 'current_point += SF * advance'.
+//
+//
 // ADVANCED USAGE
 //
 //   Quality:
@@ -206,19 +258,6 @@
 //   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
-//
-//
 // PERFORMANCE MEASUREMENTS FOR 1.06:
 //
 //                      32-bit     64-bit
@@ -233,10 +272,10 @@
 ////  SAMPLE PROGRAMS
 ////
 //
-//  Incomplete text-in-3d-api example, which draws quads properly aligned to be lossless
-//
+//  Incomplete text-in-3d-api example, which draws quads properly aligned to be lossless.
+//  See "tests/truetype_demo_win32.c" for a complete version.
 #if 0
-#define STB_TRUETYPE_IMPLEMENTATION  // force following include to generate implementation
+#define STB_TRUETYPE_IMPLEMENTATION // force following include to generate implementation
 #include "stb_truetype.h"
 
 unsigned char ttf_buffer[1<<20];
@@ -260,6 +299,8 @@ void my_stbtt_initfont(void)
 void my_stbtt_print(float x, float y, char *text)
 {
    // assume orthographic projection with units = screen pixels, origin at top left
+   glEnable(GL_BLEND);
+   glBlendFunc(GL_SRC_ALPHA, GL_ONE_MINUS_SRC_ALPHA);
    glEnable(GL_TEXTURE_2D);
    glBindTexture(GL_TEXTURE_2D, ftex);
    glBegin(GL_QUADS);
@@ -267,10 +308,10 @@ void my_stbtt_print(float x, float y, char *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);
+         glTexCoord2f(q.s0,q.t0); glVertex2f(q.x0,q.y0);
+         glTexCoord2f(q.s1,q.t0); glVertex2f(q.x1,q.y0);
+         glTexCoord2f(q.s1,q.t1); glVertex2f(q.x1,q.y1);
+         glTexCoord2f(q.s0,q.t1); glVertex2f(q.x0,q.y1);
       }
       ++text;
    }
@@ -285,7 +326,7 @@ void my_stbtt_print(float x, float y, char *text)
 //
 #if 0
 #include <stdio.h>
-#define STB_TRUETYPE_IMPLEMENTATION  // force following include to generate implementation
+#define STB_TRUETYPE_IMPLEMENTATION // force following include to generate implementation
 #include "stb_truetype.h"
 
 char ttf_buffer[1<<25];
@@ -308,7 +349,7 @@ int main(int argc, char **argv)
    }
    return 0;
 }
-#endif 
+#endif
 //
 // Output:
 //
@@ -322,9 +363,9 @@ int main(int argc, char **argv)
 //  :@@.  M@M
 //   @@@o@@@@
 //   :M@@V:@@.
-//  
+//
 //////////////////////////////////////////////////////////////////////////////
-// 
+//
 // Complete program: print "Hello World!" banner, with bugs
 //
 #if 0
@@ -378,56 +419,74 @@ int main(int arg, char **argv)
 ////   INTEGRATION WITH YOUR CODEBASE
 ////
 ////   The following sections allow you to supply alternate definitions
-////   of C library functions used by stb_truetype.
+////   of C library functions used by stb_truetype, e.g. if you don't
+////   link with the C runtime library.
 
 #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_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)    ((void)(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
+// #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];
+
+// e.g. #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)
+#define STBTT_pow(x, y) pow(x, y)
+#endif
+
+#ifndef STBTT_fmod
+#include <math.h>
+#define STBTT_fmod(x, y) fmod(x, y)
+#endif
+
+#ifndef STBTT_cos
+#include <math.h>
+#define STBTT_cos(x) cos(x)
+#define STBTT_acos(x) acos(x)
+#endif
+
+#ifndef STBTT_fabs
+#include <math.h>
+#define STBTT_fabs(x) fabs(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) ((void)(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 <string.h>
+#define STBTT_memcpy memcpy
+#define STBTT_memset memset
+#endif
 #endif
 
 ///////////////////////////////////////////////////////////////////////////////
@@ -450,6 +509,14 @@ int main(int arg, char **argv)
 extern "C" {
 #endif
 
+// private structure
+typedef struct
+{
+   unsigned char* data;
+   int cursor;
+   int size;
+} stbtt__buf;
+
 //////////////////////////////////////////////////////////////////////////////
 //
 // TEXTURE BAKING API
@@ -459,15 +526,15 @@ extern "C" {
 
 typedef struct
 {
-   unsigned short x0,y0,x1,y1; // coordinates of bbox in bitmap
-   float xoff,yoff,xadvance;
+   unsigned short x0, y0, x1, y1; // coordinates of bbox in bitmap
+   float xoff, yoff, xadvance;
 } stbtt_bakedchar;
 
-STBTT_DEF 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
+STBTT_DEF 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
@@ -475,15 +542,15 @@ STBTT_DEF int stbtt_BakeFontBitmap(const unsigned char *data, int offset,  // fo
 
 typedef struct
 {
-   float x0,y0,s0,t0; // top-left
-   float x1,y1,s1,t1; // bottom-right
+   float x0, y0, s0, t0; // top-left
+   float x1, y1, s1, t1; // bottom-right
 } stbtt_aligned_quad;
 
-STBTT_DEF void stbtt_GetBakedQuad(stbtt_bakedchar *chardata, int pw, int ph,  // same data as above
-                               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
+STBTT_DEF void stbtt_GetBakedQuad(const stbtt_bakedchar* chardata, int pw, int ph, // same data as above
+                                  int char_index, // character to display
+                                  float* xpos, float* ypos, // pointers to current position in screen pixel space
+                                  stbtt_aligned_quad* q, // output: quad to draw
+                                  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.
 //
@@ -494,6 +561,8 @@ STBTT_DEF void stbtt_GetBakedQuad(stbtt_bakedchar *chardata, int pw, int ph,  //
 //
 // It's inefficient; you might want to c&p it and optimize it.
 
+STBTT_DEF void stbtt_GetScaledFontVMetrics(const unsigned char* fontdata, int index, float size, float* ascent, float* descent, float* lineGap);
+// Query the font vertical metrics without having to create a font first.
 
 
 //////////////////////////////////////////////////////////////////////////////
@@ -505,9 +574,9 @@ STBTT_DEF void stbtt_GetBakedQuad(stbtt_bakedchar *chardata, int pw, int ph,  //
 
 typedef struct
 {
-   unsigned short x0,y0,x1,y1; // coordinates of bbox in bitmap
-   float xoff,yoff,xadvance;
-   float xoff2,yoff2;
+   unsigned short x0, y0, x1, y1; // coordinates of bbox in bitmap
+   float xoff, yoff, xadvance;
+   float xoff2, yoff2;
 } stbtt_packedchar;
 
 typedef struct stbtt_pack_context stbtt_pack_context;
@@ -516,10 +585,10 @@ typedef struct stbtt_fontinfo stbtt_fontinfo;
 typedef struct stbrp_rect stbrp_rect;
 #endif
 
-STBTT_DEF int  stbtt_PackBegin(stbtt_pack_context *spc, unsigned char *pixels, int width, int height, int stride_in_bytes, int padding, void *alloc_context);
+STBTT_DEF int stbtt_PackBegin(stbtt_pack_context* spc, unsigned char* pixels, int width, int height, int stride_in_bytes, int padding, void* alloc_context);
 // Initializes a packing context stored in the passed-in stbtt_pack_context.
 // Future calls using this context will pack characters into the bitmap passed
-// in here: a 1-channel bitmap that is weight x height. stride_in_bytes is
+// in here: a 1-channel bitmap that is width * height. stride_in_bytes is
 // the distance from one row to the next (or 0 to mean they are packed tightly
 // together). "padding" is the amount of padding to leave between each
 // character (normally you want '1' for bitmaps you'll use as textures with
@@ -527,13 +596,13 @@ STBTT_DEF int  stbtt_PackBegin(stbtt_pack_context *spc, unsigned char *pixels, i
 //
 // Returns 0 on failure, 1 on success.
 
-STBTT_DEF void stbtt_PackEnd  (stbtt_pack_context *spc);
+STBTT_DEF void stbtt_PackEnd(stbtt_pack_context* spc);
 // Cleans up the packing context and frees all memory.
 
-#define STBTT_POINT_SIZE(x)   (-(x))
+#define STBTT_POINT_SIZE(x) (-(x))
 
-STBTT_DEF int  stbtt_PackFontRange(stbtt_pack_context *spc, unsigned char *fontdata, int font_index, float font_size,
-                                int first_unicode_char_in_range, int num_chars_in_range, stbtt_packedchar *chardata_for_range);
+STBTT_DEF int stbtt_PackFontRange(stbtt_pack_context* spc, const unsigned char* fontdata, int font_index, float font_size,
+                                  int first_unicode_char_in_range, int num_chars_in_range, stbtt_packedchar* chardata_for_range);
 // Creates character bitmaps from the font_index'th font found in fontdata (use
 // font_index=0 if you don't know what that is). It creates num_chars_in_range
 // bitmaps for characters with unicode values starting at first_unicode_char_in_range
@@ -550,20 +619,20 @@ STBTT_DEF int  stbtt_PackFontRange(stbtt_pack_context *spc, unsigned char *fontd
 typedef struct
 {
    float font_size;
-   int first_unicode_codepoint_in_range;  // if non-zero, then the chars are continuous, and this is the first codepoint
-   int *array_of_unicode_codepoints;       // if non-zero, then this is an array of unicode codepoints
+   int first_unicode_codepoint_in_range; // if non-zero, then the chars are continuous, and this is the first codepoint
+   int* array_of_unicode_codepoints; // if non-zero, then this is an array of unicode codepoints
    int num_chars;
-   stbtt_packedchar *chardata_for_range; // output
+   stbtt_packedchar* chardata_for_range; // output
    unsigned char h_oversample, v_oversample; // don't set these, they're used internally
 } stbtt_pack_range;
 
-STBTT_DEF int  stbtt_PackFontRanges(stbtt_pack_context *spc, unsigned char *fontdata, int font_index, stbtt_pack_range *ranges, int num_ranges);
+STBTT_DEF int stbtt_PackFontRanges(stbtt_pack_context* spc, const unsigned char* fontdata, int font_index, stbtt_pack_range* ranges, int num_ranges);
 // Creates character bitmaps from multiple ranges of characters stored in
 // ranges. This will usually create a better-packed bitmap than multiple
 // calls to stbtt_PackFontRange. Note that you can call this multiple
 // times within a single PackBegin/PackEnd.
 
-STBTT_DEF void stbtt_PackSetOversampling(stbtt_pack_context *spc, unsigned int h_oversample, unsigned int v_oversample);
+STBTT_DEF void stbtt_PackSetOversampling(stbtt_pack_context* spc, unsigned int h_oversample, unsigned int v_oversample);
 // Oversampling a font increases the quality by allowing higher-quality subpixel
 // positioning, and is especially valuable at smaller text sizes.
 //
@@ -579,19 +648,25 @@ STBTT_DEF void stbtt_PackSetOversampling(stbtt_pack_context *spc, unsigned int h
 // To use with PackFontRangesGather etc., you must set it before calls
 // call to PackFontRangesGatherRects.
 
-STBTT_DEF void stbtt_GetPackedQuad(stbtt_packedchar *chardata, int pw, int ph,  // same data as above
-                               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 align_to_integer);
-
-STBTT_DEF int  stbtt_PackFontRangesGatherRects(stbtt_pack_context *spc, stbtt_fontinfo *info, stbtt_pack_range *ranges, int num_ranges, stbrp_rect *rects);
-STBTT_DEF void stbtt_PackFontRangesPackRects(stbtt_pack_context *spc, stbrp_rect *rects, int num_rects);
-STBTT_DEF int  stbtt_PackFontRangesRenderIntoRects(stbtt_pack_context *spc, stbtt_fontinfo *info, stbtt_pack_range *ranges, int num_ranges, stbrp_rect *rects);
+STBTT_DEF void stbtt_PackSetSkipMissingCodepoints(stbtt_pack_context* spc, int skip);
+// If skip != 0, this tells stb_truetype to skip any codepoints for which
+// there is no corresponding glyph. If skip=0, which is the default, then
+// codepoints without a glyph recived the font's "missing character" glyph,
+// typically an empty box by convention.
+
+STBTT_DEF void stbtt_GetPackedQuad(const stbtt_packedchar* chardata, int pw, int ph, // same data as above
+                                   int char_index, // character to display
+                                   float* xpos, float* ypos, // pointers to current position in screen pixel space
+                                   stbtt_aligned_quad* q, // output: quad to draw
+                                   int align_to_integer);
+
+STBTT_DEF int stbtt_PackFontRangesGatherRects(stbtt_pack_context* spc, const stbtt_fontinfo* info, stbtt_pack_range* ranges, int num_ranges, stbrp_rect* rects);
+STBTT_DEF void stbtt_PackFontRangesPackRects(stbtt_pack_context* spc, stbrp_rect* rects, int num_rects);
+STBTT_DEF int stbtt_PackFontRangesRenderIntoRects(stbtt_pack_context* spc, const stbtt_fontinfo* info, stbtt_pack_range* ranges, int num_ranges, stbrp_rect* rects);
 // Calling these functions in sequence is roughly equivalent to calling
 // stbtt_PackFontRanges(). If you more control over the packing of multiple
 // fonts, or if you want to pack custom data into a font texture, take a look
-// at the source to of stbtt_PackFontRanges() and create a custom version 
+// at the source to of stbtt_PackFontRanges() and create a custom version
 // using these functions, e.g. call GatherRects multiple times,
 // building up a single array of rects, then call PackRects once,
 // then call RenderIntoRects repeatedly. This may result in a
@@ -600,16 +675,18 @@ STBTT_DEF int  stbtt_PackFontRangesRenderIntoRects(stbtt_pack_context *spc, stbt
 
 // this is an opaque structure that you shouldn't mess with which holds
 // all the context needed from PackBegin to PackEnd.
-struct stbtt_pack_context {
-   void *user_allocator_context;
-   void *pack_info;
-   int   width;
-   int   height;
-   int   stride_in_bytes;
-   int   padding;
-   unsigned int   h_oversample, v_oversample;
-   unsigned char *pixels;
-   void  *nodes;
+struct stbtt_pack_context
+{
+   void* user_allocator_context;
+   void* pack_info;
+   int width;
+   int height;
+   int stride_in_bytes;
+   int padding;
+   int skip_missing;
+   unsigned int h_oversample, v_oversample;
+   unsigned char* pixels;
+   void* nodes;
 };
 
 //////////////////////////////////////////////////////////////////////////////
@@ -618,31 +695,43 @@ struct stbtt_pack_context {
 //
 //
 
-STBTT_DEF int stbtt_GetFontOffsetForIndex(const unsigned char *data, int index);
+STBTT_DEF int stbtt_GetNumberOfFonts(const unsigned char* data);
+// This function will determine the number of fonts in a font file.  TrueType
+// collection (.ttc) files may contain multiple fonts, while TrueType font
+// (.ttf) files only contain one font. The number of fonts can be used for
+// indexing with the previous function where the index is between zero and one
+// less than the total fonts. If an error occurs, -1 is returned.
+
+STBTT_DEF 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.
+// return '0' for index 0, and -1 for all other indices.
 
-
-// The following structure is defined publically so you can declare one on
+// The following structure is defined publicly so you can declare one on
 // the stack or as a global or etc, but you should treat it as opaque.
 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
+   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, gpos, svg; // 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__buf cff; // cff font data
+   stbtt__buf charstrings; // the charstring index
+   stbtt__buf gsubrs; // global charstring subroutines index
+   stbtt__buf subrs; // private charstring subroutines index
+   stbtt__buf fontdicts; // array of font dicts
+   stbtt__buf fdselect; // map from glyph to fontdict
 };
 
-STBTT_DEF int stbtt_InitFont(stbtt_fontinfo *info, const unsigned char *data, int offset);
+STBTT_DEF 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
@@ -654,11 +743,12 @@ STBTT_DEF int stbtt_InitFont(stbtt_fontinfo *info, const unsigned char *data, in
 //
 // CHARACTER TO GLYPH-INDEX CONVERSIOn
 
-STBTT_DEF int stbtt_FindGlyphIndex(const stbtt_fontinfo *info, int unicode_codepoint);
+STBTT_DEF 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.
+// Returns 0 if the character codepoint is not defined in the font.
 
 
 //////////////////////////////////////////////////////////////////////////////
@@ -666,7 +756,7 @@ STBTT_DEF int stbtt_FindGlyphIndex(const stbtt_fontinfo *info, int unicode_codep
 // CHARACTER PROPERTIES
 //
 
-STBTT_DEF float stbtt_ScaleForPixelHeight(const stbtt_fontinfo *info, float pixels);
+STBTT_DEF 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
@@ -674,12 +764,12 @@ STBTT_DEF float stbtt_ScaleForPixelHeight(const stbtt_fontinfo *info, float pixe
 //       scale = pixels / (ascent - descent)
 // so if you prefer to measure height by the ascent only, use a similar calculation.
 
-STBTT_DEF float stbtt_ScaleForMappingEmToPixels(const stbtt_fontinfo *info, float pixels);
+STBTT_DEF 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.
 
-STBTT_DEF void stbtt_GetFontVMetrics(const stbtt_fontinfo *info, int *ascent, int *descent, int *lineGap);
+STBTT_DEF 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...
@@ -687,25 +777,43 @@ STBTT_DEF void stbtt_GetFontVMetrics(const stbtt_fontinfo *info, int *ascent, in
 //   these are expressed in unscaled coordinates, so you must multiply by
 //   the scale factor for a given size
 
-STBTT_DEF void stbtt_GetFontBoundingBox(const stbtt_fontinfo *info, int *x0, int *y0, int *x1, int *y1);
+STBTT_DEF int stbtt_GetFontVMetricsOS2(const stbtt_fontinfo* info, int* typoAscent, int* typoDescent, int* typoLineGap);
+// analogous to GetFontVMetrics, but returns the "typographic" values from the OS/2
+// table (specific to MS/Windows TTF files).
+//
+// Returns 1 on success (table present), 0 on failure.
+
+STBTT_DEF void stbtt_GetFontBoundingBox(const stbtt_fontinfo* info, int* x0, int* y0, int* x1, int* y1);
 // the bounding box around all possible characters
 
-STBTT_DEF void stbtt_GetCodepointHMetrics(const stbtt_fontinfo *info, int codepoint, int *advanceWidth, int *leftSideBearing);
+STBTT_DEF 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
 
-STBTT_DEF int  stbtt_GetCodepointKernAdvance(const stbtt_fontinfo *info, int ch1, int ch2);
+STBTT_DEF int stbtt_GetCodepointKernAdvance(const stbtt_fontinfo* info, int ch1, int ch2);
 // an additional amount to add to the 'advance' value between ch1 and ch2
 
-STBTT_DEF int stbtt_GetCodepointBox(const stbtt_fontinfo *info, int codepoint, int *x0, int *y0, int *x1, int *y1);
+STBTT_DEF 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
 
-STBTT_DEF void stbtt_GetGlyphHMetrics(const stbtt_fontinfo *info, int glyph_index, int *advanceWidth, int *leftSideBearing);
-STBTT_DEF int  stbtt_GetGlyphKernAdvance(const stbtt_fontinfo *info, int glyph1, int glyph2);
-STBTT_DEF int  stbtt_GetGlyphBox(const stbtt_fontinfo *info, int glyph_index, int *x0, int *y0, int *x1, int *y1);
+STBTT_DEF void stbtt_GetGlyphHMetrics(const stbtt_fontinfo* info, int glyph_index, int* advanceWidth, int* leftSideBearing);
+STBTT_DEF int stbtt_GetGlyphKernAdvance(const stbtt_fontinfo* info, int glyph1, int glyph2);
+STBTT_DEF 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
 
+typedef struct stbtt_kerningentry
+{
+   int glyph1; // use stbtt_FindGlyphIndex
+   int glyph2;
+   int advance;
+} stbtt_kerningentry;
+
+STBTT_DEF int stbtt_GetKerningTableLength(const stbtt_fontinfo* info);
+STBTT_DEF int stbtt_GetKerningTable(const stbtt_fontinfo* info, stbtt_kerningentry* table, int table_length);
+// Retrieves a complete list of all of the kerning pairs provided by the font
+// stbtt_GetKerningTable never writes more than table_length entries and returns how many entries it did write.
+// The table will be sorted by (a.glyph1 == b.glyph1)?(a.glyph2 < b.glyph2):(a.glyph1 < b.glyph1)
 
 //////////////////////////////////////////////////////////////////////////////
 //
@@ -714,50 +822,58 @@ STBTT_DEF int  stbtt_GetGlyphBox(const stbtt_fontinfo *info, int glyph_index, in
 //
 
 #ifndef STBTT_vmove // you can predefine these to use different values (but why?)
-   enum {
-      STBTT_vmove=1,
-      STBTT_vline,
-      STBTT_vcurve
-   };
+enum
+{
+   STBTT_vmove = 1,
+   STBTT_vline,
+   STBTT_vcurve,
+   STBTT_vcubic
+};
 #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;
+// (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, cx1, cy1;
+   unsigned char type, padding;
+} stbtt_vertex;
 #endif
 
-STBTT_DEF int stbtt_IsGlyphEmpty(const stbtt_fontinfo *info, int glyph_index);
+STBTT_DEF int stbtt_IsGlyphEmpty(const stbtt_fontinfo* info, int glyph_index);
 // returns non-zero if nothing is drawn for this glyph
 
-STBTT_DEF int stbtt_GetCodepointShape(const stbtt_fontinfo *info, int unicode_codepoint, stbtt_vertex **vertices);
-STBTT_DEF int stbtt_GetGlyphShape(const stbtt_fontinfo *info, int glyph_index, stbtt_vertex **vertices);
+STBTT_DEF int stbtt_GetCodepointShape(const stbtt_fontinfo* info, int unicode_codepoint, stbtt_vertex** vertices);
+STBTT_DEF 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
+// The shape is a series of contours. 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.
 
-STBTT_DEF void stbtt_FreeShape(const stbtt_fontinfo *info, stbtt_vertex *vertices);
+STBTT_DEF void stbtt_FreeShape(const stbtt_fontinfo* info, stbtt_vertex* vertices);
 // frees the data allocated above
 
+STBTT_DEF unsigned char* stbtt_FindSVGDoc(const stbtt_fontinfo* info, int gl);
+STBTT_DEF int stbtt_GetCodepointSVG(const stbtt_fontinfo* info, int unicode_codepoint, const char** svg);
+STBTT_DEF int stbtt_GetGlyphSVG(const stbtt_fontinfo* info, int gl, const char** svg);
+// fills svg with the character's SVG data.
+// returns data size or 0 if SVG not found.
+
 //////////////////////////////////////////////////////////////////////////////
 //
 // BITMAP RENDERING
 //
 
-STBTT_DEF void stbtt_FreeBitmap(unsigned char *bitmap, void *userdata);
+STBTT_DEF void stbtt_FreeBitmap(unsigned char* bitmap, void* userdata);
 // frees the bitmap allocated below
 
-STBTT_DEF unsigned char *stbtt_GetCodepointBitmap(const stbtt_fontinfo *info, float scale_x, float scale_y, int codepoint, int *width, int *height, int *xoff, int *yoff);
+STBTT_DEF 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).
@@ -766,58 +882,120 @@ STBTT_DEF unsigned char *stbtt_GetCodepointBitmap(const stbtt_fontinfo *info, fl
 //
 // xoff/yoff are the offset it pixel space from the glyph origin to the top-left of the bitmap
 
-STBTT_DEF 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);
+STBTT_DEF 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
 
-STBTT_DEF 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_DEF 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.
 
-STBTT_DEF 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_DEF 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
 
-STBTT_DEF void stbtt_GetCodepointBitmapBox(const stbtt_fontinfo *font, int codepoint, float scale_x, float scale_y, int *ix0, int *iy0, int *ix1, int *iy1);
+STBTT_DEF void stbtt_MakeCodepointBitmapSubpixelPrefilter(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 oversample_x, int oversample_y, float* sub_x, float* sub_y, int codepoint);
+// same as stbtt_MakeCodepointBitmapSubpixel, but prefiltering
+// is performed (see stbtt_PackSetOversampling)
+
+STBTT_DEF 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.)
 
-STBTT_DEF 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_DEF 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)
-STBTT_DEF unsigned char *stbtt_GetGlyphBitmap(const stbtt_fontinfo *info, float scale_x, float scale_y, int glyph, int *width, int *height, int *xoff, int *yoff);
-STBTT_DEF 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);
-STBTT_DEF 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_DEF 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);
-STBTT_DEF void stbtt_GetGlyphBitmapBox(const stbtt_fontinfo *font, int glyph, float scale_x, float scale_y, int *ix0, int *iy0, int *ix1, int *iy1);
-STBTT_DEF 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);
+STBTT_DEF unsigned char* stbtt_GetGlyphBitmap(const stbtt_fontinfo* info, float scale_x, float scale_y, int glyph, int* width, int* height, int* xoff, int* yoff);
+STBTT_DEF 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);
+STBTT_DEF 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_DEF 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);
+STBTT_DEF void stbtt_MakeGlyphBitmapSubpixelPrefilter(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 oversample_x, int oversample_y, float* sub_x, float* sub_y, int glyph);
+STBTT_DEF void stbtt_GetGlyphBitmapBox(const stbtt_fontinfo* font, int glyph, float scale_x, float scale_y, int* ix0, int* iy0, int* ix1, int* iy1);
+STBTT_DEF 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;
+   int w, h, stride;
+   unsigned char* pixels;
 } stbtt__bitmap;
 
 // rasterize a shape with quadratic beziers into a bitmap
-STBTT_DEF void stbtt_Rasterize(stbtt__bitmap *result,        // 1-channel bitmap to draw into
-                               float flatness_in_pixels,     // allowable error of curve in pixels
-                               stbtt_vertex *vertices,       // array of vertices defining shape
-                               int num_verts,                // number of vertices in above array
+STBTT_DEF void stbtt_Rasterize(stbtt__bitmap* result, // 1-channel bitmap to draw into
+                               float flatness_in_pixels, // allowable error of curve in pixels
+                               stbtt_vertex* vertices, // array of vertices defining shape
+                               int num_verts, // number of vertices in above array
                                float scale_x, float scale_y, // scale applied to input vertices
                                float shift_x, float shift_y, // translation applied to input vertices
-                               int x_off, int y_off,         // another translation applied to input
-                               int invert,                   // if non-zero, vertically flip shape
-                               void *userdata);              // context for to STBTT_MALLOC
+                               int x_off, int y_off, // another translation applied to input
+                               int invert, // if non-zero, vertically flip shape
+                               void* userdata); // context for to STBTT_MALLOC
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// Signed Distance Function (or Field) rendering
+
+STBTT_DEF void stbtt_FreeSDF(unsigned char* bitmap, void* userdata);
+// frees the SDF bitmap allocated below
+
+STBTT_DEF unsigned char* stbtt_GetGlyphSDF(const stbtt_fontinfo* info, float scale, int glyph, int padding, unsigned char onedge_value, float pixel_dist_scale, int* width, int* height, int* xoff, int* yoff);
+STBTT_DEF unsigned char* stbtt_GetCodepointSDF(const stbtt_fontinfo* info, float scale, int codepoint, int padding, unsigned char onedge_value, float pixel_dist_scale, int* width, int* height, int* xoff, int* yoff);
+// These functions compute a discretized SDF field for a single character, suitable for storing
+// in a single-channel texture, sampling with bilinear filtering, and testing against
+// larger than some threshold to produce scalable fonts.
+//        info              --  the font
+//        scale             --  controls the size of the resulting SDF bitmap, same as it would be creating a regular bitmap
+//        glyph/codepoint   --  the character to generate the SDF for
+//        padding           --  extra "pixels" around the character which are filled with the distance to the character (not 0),
+//                                 which allows effects like bit outlines
+//        onedge_value      --  value 0-255 to test the SDF against to reconstruct the character (i.e. the isocontour of the character)
+//        pixel_dist_scale  --  what value the SDF should increase by when moving one SDF "pixel" away from the edge (on the 0..255 scale)
+//                                 if positive, > onedge_value is inside; if negative, < onedge_value is inside
+//        width,height      --  output height & width of the SDF bitmap (including padding)
+//        xoff,yoff         --  output origin of the character
+//        return value      --  a 2D array of bytes 0..255, width*height in size
+//
+// pixel_dist_scale & onedge_value are a scale & bias that allows you to make
+// optimal use of the limited 0..255 for your application, trading off precision
+// and special effects. SDF values outside the range 0..255 are clamped to 0..255.
+//
+// Example:
+//      scale = stbtt_ScaleForPixelHeight(22)
+//      padding = 5
+//      onedge_value = 180
+//      pixel_dist_scale = 180/5.0 = 36.0
+//
+//      This will create an SDF bitmap in which the character is about 22 pixels
+//      high but the whole bitmap is about 22+5+5=32 pixels high. To produce a filled
+//      shape, sample the SDF at each pixel and fill the pixel if the SDF value
+//      is greater than or equal to 180/255. (You'll actually want to antialias,
+//      which is beyond the scope of this example.) Additionally, you can compute
+//      offset outlines (e.g. to stroke the character border inside & outside,
+//      or only outside). For example, to fill outside the character up to 3 SDF
+//      pixels, you would compare against (180-36.0*3)/255 = 72/255. The above
+//      choice of variables maps a range from 5 pixels outside the shape to
+//      2 pixels inside the shape to 0..255; this is intended primarily for apply
+//      outside effects only (the interior range is needed to allow proper
+//      antialiasing of the font at *smaller* sizes)
+//
+// The function computes the SDF analytically at each SDF pixel, not by e.g.
+// building a higher-res bitmap and approximating it. In theory the quality
+// should be as high as possible for an SDF of this size & representation, but
+// unclear if this is true in practice (perhaps building a higher-res bitmap
+// and computing from that can allow drop-out prevention).
+//
+// The algorithm has not been optimized at all, so expect it to be slow
+// if computing lots of characters or very large sizes.
+
 
 //////////////////////////////////////////////////////////////////////////////
 //
@@ -841,22 +1019,22 @@ STBTT_DEF void stbtt_Rasterize(stbtt__bitmap *result,        // 1-channel bitmap
 //             You have to have called stbtt_InitFont() first.
 
 
-STBTT_DEF int stbtt_FindMatchingFont(const unsigned char *fontdata, const char *name, int flags);
+STBTT_DEF 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
+#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
 
-STBTT_DEF int stbtt_CompareUTF8toUTF16_bigendian(const char *s1, int len1, const char *s2, int len2);
+STBTT_DEF 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
 
-STBTT_DEF const char *stbtt_GetFontNameString(const stbtt_fontinfo *font, int *length, int platformID, int encodingID, int languageID, int nameID);
+STBTT_DEF 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.
 //
@@ -864,53 +1042,76 @@ STBTT_DEF const char *stbtt_GetFontNameString(const stbtt_fontinfo *font, int *l
 //     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
+{ // 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_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_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
+{ // 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_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
+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
@@ -929,175 +1130,471 @@ enum { // languageID for STBTT_PLATFORM_ID_MAC
 #ifdef STB_TRUETYPE_IMPLEMENTATION
 
 #ifndef STBTT_MAX_OVERSAMPLE
-#define STBTT_MAX_OVERSAMPLE   8
+#define STBTT_MAX_OVERSAMPLE 8
 #endif
 
 #if STBTT_MAX_OVERSAMPLE > 255
 #error "STBTT_MAX_OVERSAMPLE cannot be > 255"
 #endif
 
-typedef int stbtt__test_oversample_pow2[(STBTT_MAX_OVERSAMPLE & (STBTT_MAX_OVERSAMPLE-1)) == 0 ? 1 : -1];
+typedef int stbtt__test_oversample_pow2[(STBTT_MAX_OVERSAMPLE & (STBTT_MAX_OVERSAMPLE - 1)) == 0 ? 1 : -1];
 
 #ifndef STBTT_RASTERIZER_VERSION
 #define STBTT_RASTERIZER_VERSION 2
 #endif
 
+#ifdef _MSC_VER
+#define STBTT__NOTUSED(v) (void)(v)
+#else
+#define STBTT__NOTUSED(v) (void)sizeof(v)
+#endif
+
 //////////////////////////////////////////////////////////////////////////
 //
-// accessors to parse data from file
+// stbtt__buf helpers 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
+static stbtt_uint8 stbtt__buf_get8(stbtt__buf* b)
+{
+   if (b->cursor >= b->size)
+      return 0;
+   return b->data[b->cursor++];
+}
 
-#define ttBYTE(p)     (* (stbtt_uint8 *) (p))
-#define ttCHAR(p)     (* (stbtt_int8 *) (p))
-#define ttFixed(p)    ttLONG(p)
+static stbtt_uint8 stbtt__buf_peek8(stbtt__buf* b)
+{
+   if (b->cursor >= b->size)
+      return 0;
+   return b->data[b->cursor];
+}
 
-#if defined(STB_TRUETYPE_BIGENDIAN) && !defined(ALLOW_UNALIGNED_TRUETYPE)
+static void stbtt__buf_seek(stbtt__buf* b, int o)
+{
+   STBTT_assert(!(o > b->size || o < 0));
+   b->cursor = (o > b->size || o < 0) ? b->size : o;
+}
 
-   #define ttUSHORT(p)   (* (stbtt_uint16 *) (p))
-   #define ttSHORT(p)    (* (stbtt_int16 *) (p))
-   #define ttULONG(p)    (* (stbtt_uint32 *) (p))
-   #define ttLONG(p)     (* (stbtt_int32 *) (p))
+static void stbtt__buf_skip(stbtt__buf* b, int o)
+{
+   stbtt__buf_seek(b, b->cursor + o);
+}
 
-#else
+static stbtt_uint32 stbtt__buf_get(stbtt__buf* b, int n)
+{
+   stbtt_uint32 v = 0;
+   int i;
+   STBTT_assert(n >= 1 && n <= 4);
+   for (i = 0; i < n; i++)
+      v = (v << 8) | stbtt__buf_get8(b);
+   return v;
+}
+
+static stbtt__buf stbtt__new_buf(const void* p, size_t size)
+{
+   stbtt__buf r;
+   STBTT_assert(size < 0x40000000);
+   r.data = (stbtt_uint8*)p;
+   r.size = (int)size;
+   r.cursor = 0;
+   return r;
+}
 
-   static stbtt_uint16 ttUSHORT(const stbtt_uint8 *p) { return p[0]*256 + p[1]; }
-   static stbtt_int16 ttSHORT(const stbtt_uint8 *p)   { return p[0]*256 + p[1]; }
-   static stbtt_uint32 ttULONG(const stbtt_uint8 *p)  { return (p[0]<<24) + (p[1]<<16) + (p[2]<<8) + p[3]; }
-   static stbtt_int32 ttLONG(const stbtt_uint8 *p)    { return (p[0]<<24) + (p[1]<<16) + (p[2]<<8) + p[3]; }
+#define stbtt__buf_get16(b) stbtt__buf_get((b), 2)
+#define stbtt__buf_get32(b) stbtt__buf_get((b), 4)
 
-#endif
+static stbtt__buf stbtt__buf_range(const stbtt__buf* b, int o, int s)
+{
+   stbtt__buf r = stbtt__new_buf(NULL, 0);
+   if (o < 0 || s < 0 || o > b->size || s > b->size - o)
+      return r;
+   r.data = b->data + o;
+   r.size = s;
+   return r;
+}
+
+static stbtt__buf stbtt__cff_get_index(stbtt__buf* b)
+{
+   int count, start, offsize;
+   start = b->cursor;
+   count = stbtt__buf_get16(b);
+   if (count)
+   {
+      offsize = stbtt__buf_get8(b);
+      STBTT_assert(offsize >= 1 && offsize <= 4);
+      stbtt__buf_skip(b, offsize * count);
+      stbtt__buf_skip(b, stbtt__buf_get(b, offsize) - 1);
+   }
+   return stbtt__buf_range(b, start, b->cursor - start);
+}
+
+static stbtt_uint32 stbtt__cff_int(stbtt__buf* b)
+{
+   int b0 = stbtt__buf_get8(b);
+   if (b0 >= 32 && b0 <= 246)
+      return b0 - 139;
+   else if (b0 >= 247 && b0 <= 250)
+      return (b0 - 247) * 256 + stbtt__buf_get8(b) + 108;
+   else if (b0 >= 251 && b0 <= 254)
+      return -(b0 - 251) * 256 - stbtt__buf_get8(b) - 108;
+   else if (b0 == 28)
+      return stbtt__buf_get16(b);
+   else if (b0 == 29)
+      return stbtt__buf_get32(b);
+   STBTT_assert(0);
+   return 0;
+}
+
+static void stbtt__cff_skip_operand(stbtt__buf* b)
+{
+   int v, b0 = stbtt__buf_peek8(b);
+   STBTT_assert(b0 >= 28);
+   if (b0 == 30)
+   {
+      stbtt__buf_skip(b, 1);
+      while (b->cursor < b->size)
+      {
+         v = stbtt__buf_get8(b);
+         if ((v & 0xF) == 0xF || (v >> 4) == 0xF)
+            break;
+      }
+   }
+   else
+   {
+      stbtt__cff_int(b);
+   }
+}
+
+static stbtt__buf stbtt__dict_get(stbtt__buf* b, int key)
+{
+   stbtt__buf_seek(b, 0);
+   while (b->cursor < b->size)
+   {
+      int start = b->cursor, end, op;
+      while (stbtt__buf_peek8(b) >= 28)
+         stbtt__cff_skip_operand(b);
+      end = b->cursor;
+      op = stbtt__buf_get8(b);
+      if (op == 12)
+         op = stbtt__buf_get8(b) | 0x100;
+      if (op == key)
+         return stbtt__buf_range(b, start, end - start);
+   }
+   return stbtt__buf_range(b, 0, 0);
+}
+
+static void stbtt__dict_get_ints(stbtt__buf* b, int key, int outcount, stbtt_uint32* out)
+{
+   int i;
+   stbtt__buf operands = stbtt__dict_get(b, key);
+   for (i = 0; i < outcount && operands.cursor < operands.size; i++)
+      out[i] = stbtt__cff_int(&operands);
+}
+
+static int stbtt__cff_index_count(stbtt__buf* b)
+{
+   stbtt__buf_seek(b, 0);
+   return stbtt__buf_get16(b);
+}
+
+static stbtt__buf stbtt__cff_index_get(stbtt__buf b, int i)
+{
+   int count, offsize, start, end;
+   stbtt__buf_seek(&b, 0);
+   count = stbtt__buf_get16(&b);
+   offsize = stbtt__buf_get8(&b);
+   STBTT_assert(i >= 0 && i < count);
+   STBTT_assert(offsize >= 1 && offsize <= 4);
+   stbtt__buf_skip(&b, i * offsize);
+   start = stbtt__buf_get(&b, offsize);
+   end = stbtt__buf_get(&b, offsize);
+   return stbtt__buf_range(&b, 2 + (count + 1) * offsize + start, end - start);
+}
+
+//////////////////////////////////////////////////////////////////////////
+//
+// 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)
+
+static stbtt_uint16 ttUSHORT(stbtt_uint8* p)
+{
+   return p[0] * 256 + p[1];
+}
+static stbtt_int16 ttSHORT(stbtt_uint8* p) { return p[0] * 256 + p[1]; }
+static stbtt_uint32 ttULONG(stbtt_uint8* p) { return (p[0] << 24) + (p[1] << 16) + (p[2] << 8) + p[3]; }
+static stbtt_int32 ttLONG(stbtt_uint8* p) { return (p[0] << 24) + (p[1] << 16) + (p[2] << 8) + p[3]; }
 
-#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])
+#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)
+static int stbtt__isfont(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
+   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
+   if (stbtt_tag(font, "true"))
+      return 1; // Apple specification for TrueType fonts
    return 0;
 }
 
 // @OPTIMIZE: binary search
-static stbtt_uint32 stbtt__find_table(stbtt_uint8 *data, stbtt_uint32 fontstart, const char *tag)
+static stbtt_uint32 stbtt__find_table(stbtt_uint8* data, stbtt_uint32 fontstart, const char* tag)
 {
-   stbtt_int32 num_tables = ttUSHORT(data+fontstart+4);
+   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);
+   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;
 }
 
-STBTT_DEF int stbtt_GetFontOffsetForIndex(const unsigned char *font_collection, int index)
+static int stbtt_GetFontOffsetForIndex_internal(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")) {
+   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 (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*4);
+         return ttULONG(font_collection + 12 + index * 4);
       }
    }
    return -1;
 }
 
-STBTT_DEF int stbtt_InitFont(stbtt_fontinfo *info, const unsigned char *data2, int fontstart)
+static int stbtt_GetNumberOfFonts_internal(unsigned char* font_collection)
+{
+   // if it's just a font, there's only one valid font
+   if (stbtt__isfont(font_collection))
+      return 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)
+      {
+         return ttLONG(font_collection + 8);
+      }
+   }
+   return 0;
+}
+
+static stbtt__buf stbtt__get_subrs(stbtt__buf cff, stbtt__buf fontdict)
+{
+   stbtt_uint32 subrsoff = 0, private_loc[2] = { 0, 0 };
+   stbtt__buf pdict;
+   stbtt__dict_get_ints(&fontdict, 18, 2, private_loc);
+   if (!private_loc[1] || !private_loc[0])
+      return stbtt__new_buf(NULL, 0);
+   pdict = stbtt__buf_range(&cff, private_loc[1], private_loc[0]);
+   stbtt__dict_get_ints(&pdict, 19, 1, &subrsoff);
+   if (!subrsoff)
+      return stbtt__new_buf(NULL, 0);
+   stbtt__buf_seek(&cff, private_loc[1] + subrsoff);
+   return stbtt__cff_get_index(&cff);
+}
+
+// since most people won't use this, find this table the first time it's needed
+static int stbtt__get_svg(stbtt_fontinfo* info)
+{
+   stbtt_uint32 t;
+   if (info->svg < 0)
+   {
+      t = stbtt__find_table(info->data, info->fontstart, "SVG ");
+      if (t)
+      {
+         stbtt_uint32 offset = ttULONG(info->data + t + 2);
+         info->svg = t + offset;
+      }
+      else
+      {
+         info->svg = 0;
+      }
+   }
+   return info->svg;
+}
+
+static int stbtt_InitFont_internal(stbtt_fontinfo* info, unsigned char* data, int fontstart)
 {
-   stbtt_uint8 *data = (stbtt_uint8 *) data2;
    stbtt_uint32 cmap, t;
-   stbtt_int32 i,numTables;
+   stbtt_int32 i, numTables;
 
    info->data = data;
    info->fontstart = fontstart;
+   info->cff = stbtt__new_buf(NULL, 0);
 
-   cmap = stbtt__find_table(data, fontstart, "cmap");       // required
+   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)
+   info->gpos = stbtt__find_table(data, fontstart, "GPOS"); // not required
+
+   if (!cmap || !info->head || !info->hhea || !info->hmtx)
       return 0;
+   if (info->glyf)
+   {
+      // required for truetype
+      if (!info->loca)
+         return 0;
+   }
+   else
+   {
+      // initialization for CFF / Type2 fonts (OTF)
+      stbtt__buf b, topdict, topdictidx;
+      stbtt_uint32 cstype = 2, charstrings = 0, fdarrayoff = 0, fdselectoff = 0;
+      stbtt_uint32 cff;
+
+      cff = stbtt__find_table(data, fontstart, "CFF ");
+      if (!cff)
+         return 0;
+
+      info->fontdicts = stbtt__new_buf(NULL, 0);
+      info->fdselect = stbtt__new_buf(NULL, 0);
+
+      // @TODO this should use size from table (not 512MB)
+      info->cff = stbtt__new_buf(data + cff, 512 * 1024 * 1024);
+      b = info->cff;
+
+      // read the header
+      stbtt__buf_skip(&b, 2);
+      stbtt__buf_seek(&b, stbtt__buf_get8(&b)); // hdrsize
+
+      // @TODO the name INDEX could list multiple fonts,
+      // but we just use the first one.
+      stbtt__cff_get_index(&b); // name INDEX
+      topdictidx = stbtt__cff_get_index(&b);
+      topdict = stbtt__cff_index_get(topdictidx, 0);
+      stbtt__cff_get_index(&b); // string INDEX
+      info->gsubrs = stbtt__cff_get_index(&b);
+
+      stbtt__dict_get_ints(&topdict, 17, 1, &charstrings);
+      stbtt__dict_get_ints(&topdict, 0x100 | 6, 1, &cstype);
+      stbtt__dict_get_ints(&topdict, 0x100 | 36, 1, &fdarrayoff);
+      stbtt__dict_get_ints(&topdict, 0x100 | 37, 1, &fdselectoff);
+      info->subrs = stbtt__get_subrs(b, topdict);
+
+      // we only support Type 2 charstrings
+      if (cstype != 2)
+         return 0;
+      if (charstrings == 0)
+         return 0;
+
+      if (fdarrayoff)
+      {
+         // looks like a CID font
+         if (!fdselectoff)
+            return 0;
+         stbtt__buf_seek(&b, fdarrayoff);
+         info->fontdicts = stbtt__cff_get_index(&b);
+         info->fdselect = stbtt__buf_range(&b, fdselectoff, b.size - fdselectoff);
+      }
+
+      stbtt__buf_seek(&b, charstrings);
+      info->charstrings = stbtt__cff_get_index(&b);
+   }
 
    t = stbtt__find_table(data, fontstart, "maxp");
    if (t)
-      info->numGlyphs = ttUSHORT(data+t+4);
+      info->numGlyphs = ttUSHORT(data + t + 4);
    else
       info->numGlyphs = 0xffff;
 
+   info->svg = -1;
+
    // 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) {
+   for (i = 0; i < numTables; ++i)
+   {
       stbtt_uint32 encoding_record = cmap + 4 + 8 * i;
       // find an encoding we understand:
-      switch(ttUSHORT(data+encoding_record)) {
+      switch (ttUSHORT(data + encoding_record))
+      {
          case STBTT_PLATFORM_ID_MICROSOFT:
-            switch (ttUSHORT(data+encoding_record+2)) {
+            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);
+                  info->index_map = cmap + ttULONG(data + encoding_record + 4);
                   break;
             }
             break;
-        case STBTT_PLATFORM_ID_UNICODE:
+         case STBTT_PLATFORM_ID_UNICODE:
             // Mac/iOS has these
             // all the encodingIDs are unicode, so we don't bother to check it
-            info->index_map = cmap + ttULONG(data+encoding_record+4);
+            info->index_map = cmap + ttULONG(data + encoding_record + 4);
             break;
       }
    }
    if (info->index_map == 0)
       return 0;
 
-   info->indexToLocFormat = ttUSHORT(data+info->head + 50);
+   info->indexToLocFormat = ttUSHORT(data + info->head + 50);
    return 1;
 }
 
-STBTT_DEF int stbtt_FindGlyphIndex(const stbtt_fontinfo *info, int unicode_codepoint)
+STBTT_DEF int stbtt_FindGlyphIndex(const stbtt_fontinfo* info, int unicode_codepoint)
 {
-   stbtt_uint8 *data = info->data;
+   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
+   if (format == 0)
+   { // apple byte encoding
       stbtt_int32 bytes = ttUSHORT(data + index_map + 2);
-      if (unicode_codepoint < bytes-6)
+      if (unicode_codepoint < bytes - 6)
          return ttBYTE(data + index_map + 6 + unicode_codepoint);
       return 0;
-   } else if (format == 6) {
+   }
+   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);
+      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) {
+   }
+   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;
+   }
+   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;
 
       // do a binary search of the segments
       stbtt_uint32 endCount = index_map + 14;
@@ -1108,53 +1605,59 @@ STBTT_DEF int stbtt_FindGlyphIndex(const stbtt_fontinfo *info, int unicode_codep
 
       // 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;
+      if (unicode_codepoint >= ttUSHORT(data + search + rangeShift * 2))
+         search += rangeShift * 2;
 
       // now decrement to bias correctly to find smallest
       search -= 2;
-      while (entrySelector) {
+      while (entrySelector)
+      {
          stbtt_uint16 end;
          searchRange >>= 1;
-         end = ttUSHORT(data + search + searchRange*2);
+         end = ttUSHORT(data + search + searchRange * 2);
          if (unicode_codepoint > end)
-            search += searchRange*2;
+            search += searchRange * 2;
          --entrySelector;
       }
       search += 2;
 
       {
-         stbtt_uint16 offset, start;
-         stbtt_uint16 item = (stbtt_uint16) ((search - endCount) >> 1);
+         stbtt_uint16 offset, start, last;
+         stbtt_uint16 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);
-         if (unicode_codepoint < start)
+         start = ttUSHORT(data + index_map + 14 + segcount * 2 + 2 + 2 * item);
+         last = ttUSHORT(data + endCount + 2 * item);
+         if (unicode_codepoint < start || unicode_codepoint > last)
             return 0;
 
-         offset = ttUSHORT(data + index_map + 14 + segcount*6 + 2 + 2*item);
+         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 (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);
+         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;
+   }
+   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)
+      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);
+         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;
+               return start_glyph + unicode_codepoint - start_char;
             else // format == 13
                return start_glyph;
          }
@@ -1166,113 +1669,143 @@ STBTT_DEF int stbtt_FindGlyphIndex(const stbtt_fontinfo *info, int unicode_codep
    return 0;
 }
 
-STBTT_DEF int stbtt_GetCodepointShape(const stbtt_fontinfo *info, int unicode_codepoint, stbtt_vertex **vertices)
+STBTT_DEF 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)
+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;
+   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)
+static int stbtt__GetGlyfOffset(const stbtt_fontinfo* info, int glyph_index)
 {
-   int g1,g2;
+   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
+   STBTT_assert(!info->cff.size);
 
-   if (info->indexToLocFormat == 0) {
+   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);
+   }
+   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
+   return g1 == g2 ? -1 : g1; // if length is 0, return -1
 }
 
-STBTT_DEF int stbtt_GetGlyphBox(const stbtt_fontinfo *info, int glyph_index, int *x0, int *y0, int *x1, int *y1)
+static int stbtt__GetGlyphInfoT2(const stbtt_fontinfo* info, int glyph_index, int* x0, int* y0, int* x1, int* y1);
+
+STBTT_DEF 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 (info->cff.size)
+   {
+      stbtt__GetGlyphInfoT2(info, glyph_index, x0, y0, x1, y1);
+   }
+   else
+   {
+      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);
+      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;
 }
 
-STBTT_DEF int stbtt_GetCodepointBox(const stbtt_fontinfo *info, int codepoint, int *x0, int *y0, int *x1, int *y1)
+STBTT_DEF 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);
+   return stbtt_GetGlyphBox(info, stbtt_FindGlyphIndex(info, codepoint), x0, y0, x1, y1);
 }
 
-STBTT_DEF int stbtt_IsGlyphEmpty(const stbtt_fontinfo *info, int glyph_index)
+STBTT_DEF 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;
+   int g;
+   if (info->cff.size)
+      return stbtt__GetGlyphInfoT2(info, glyph_index, NULL, NULL, NULL, NULL) == 0;
+   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)
+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 (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 {
+         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);
+         stbtt_setvertex(&vertices[num_vertices++], STBTT_vcurve, sx, sy, cx, cy);
       else
-         stbtt_setvertex(&vertices[num_vertices++], STBTT_vline,sx,sy,0,0);
+         stbtt_setvertex(&vertices[num_vertices++], STBTT_vline, sx, sy, 0, 0);
    }
    return num_vertices;
 }
 
-STBTT_DEF int stbtt_GetGlyphShape(const stbtt_fontinfo *info, int glyph_index, stbtt_vertex **pvertices)
+static int stbtt__GetGlyphShapeTT(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;
+   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;
+   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;
+   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);
+      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);
+      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;
+      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
@@ -1282,183 +1815,243 @@ STBTT_DEF int stbtt_GetGlyphShape(const stbtt_fontinfo *info, int glyph_index, s
 
       // first load flags
 
-      for (i=0; i < n; ++i) {
-         if (flagcount == 0) {
+      for (i = 0; i < n; ++i)
+      {
+         if (flagcount == 0)
+         {
             flags = *points++;
             if (flags & 8)
                flagcount = *points++;
-         } else
+         }
+         else
             --flagcount;
-         vertices[off+i].type = flags;
+         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) {
+      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]);
+         }
+         else
+         {
+            if (!(flags & 16))
+            {
+               x = x + (stbtt_int16)(points[0] * 256 + points[1]);
                points += 2;
             }
          }
-         vertices[off+i].x = (stbtt_int16) x;
+         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) {
+      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]);
+         }
+         else
+         {
+            if (!(flags & 32))
+            {
+               y = y + (stbtt_int16)(points[0] * 256 + points[1]);
                points += 2;
             }
          }
-         vertices[off+i].y = (stbtt_int16) y;
+         vertices[off + i].y = (stbtt_int16)y;
       }
 
       // now convert them to our format
-      num_vertices=0;
+      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;
+      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 (next_move == i)
+         {
             if (i != 0)
-               num_vertices = stbtt__close_shape(vertices, num_vertices, was_off, start_off, sx,sy,scx,scy,cx,cy);
+               num_vertices = stbtt__close_shape(vertices, num_vertices, was_off, start_off, sx, sy, scx, scy, cx, cy);
 
-            // now start the new one               
+            // now start the new one
             start_off = !(flags & 1);
-            if (start_off) {
+            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)) {
+               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 {
+                  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;
+                  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 {
+            }
+            else
+            {
                sx = x;
                sy = y;
             }
-            stbtt_setvertex(&vertices[num_vertices++], STBTT_vmove,sx,sy,0,0);
+            stbtt_setvertex(&vertices[num_vertices++], STBTT_vmove, sx, sy, 0, 0);
             was_off = 0;
-            next_move = 1 + ttUSHORT(endPtsOfContours+j*2);
+            next_move = 1 + ttUSHORT(endPtsOfContours + j * 2);
             ++j;
-         } else {
-            if (!(flags & 1)) { // if it's a curve
+         }
+         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);
+                  stbtt_setvertex(&vertices[num_vertices++], STBTT_vcurve, (cx + x) >> 1, (cy + y) >> 1, cx, cy);
                cx = x;
                cy = y;
                was_off = 1;
-            } else {
+            }
+            else
+            {
                if (was_off)
-                  stbtt_setvertex(&vertices[num_vertices++], STBTT_vcurve, x,y, cx, cy);
+                  stbtt_setvertex(&vertices[num_vertices++], STBTT_vcurve, x, y, cx, cy);
                else
-                  stbtt_setvertex(&vertices[num_vertices++], STBTT_vline, x,y,0,0);
+                  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) {
+      num_vertices = stbtt__close_shape(vertices, num_vertices, was_off, start_off, sx, sy, scx, scy, cx, cy);
+   }
+   else if (numberOfContours < 0)
+   {
       // Compound shapes.
       int more = 1;
-      stbtt_uint8 *comp = data + g + 10;
+      stbtt_uint8* comp = data + g + 10;
       num_vertices = 0;
       vertices = 0;
-      while (more) {
+      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;
+         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 {
+         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;
+         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;
+         }
+         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;
+            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]);
+         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) {
+         if (comp_num_verts > 0)
+         {
             // Transform vertices.
-            for (i = 0; i < comp_num_verts; ++i) {
+            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]));
+               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);
+            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);
+            if (num_vertices > 0 && vertices)
+               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);
+         more = flags & (1 << 5);
       }
-   } else if (numberOfContours < 0) {
-      // @TODO other compound variations?
-      STBTT_assert(0);
-   } else {
+   }
+   else
+   {
       // numberOfCounters == 0, do nothing
    }
 
@@ -1466,68 +2059,871 @@ STBTT_DEF int stbtt_GetGlyphShape(const stbtt_fontinfo *info, int glyph_index, s
    return num_vertices;
 }
 
-STBTT_DEF void stbtt_GetGlyphHMetrics(const stbtt_fontinfo *info, int glyph_index, int *advanceWidth, int *leftSideBearing)
+typedef struct
 {
-   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 bounds;
+   int started;
+   float first_x, first_y;
+   float x, y;
+   stbtt_int32 min_x, max_x, min_y, max_y;
+
+   stbtt_vertex* pvertices;
+   int num_vertices;
+} stbtt__csctx;
+
+#define STBTT__CSCTX_INIT(bounds)                \
+   {                                             \
+      bounds, 0, 0, 0, 0, 0, 0, 0, 0, 0, NULL, 0 \
    }
-}
 
-STBTT_DEF int  stbtt_GetGlyphKernAdvance(const stbtt_fontinfo *info, int glyph1, int glyph2)
+static void stbtt__track_vertex(stbtt__csctx* c, stbtt_int32 x, stbtt_int32 y)
 {
-   stbtt_uint8 *data = info->data + info->kern;
+   if (x > c->max_x || !c->started)
+      c->max_x = x;
+   if (y > c->max_y || !c->started)
+      c->max_y = y;
+   if (x < c->min_x || !c->started)
+      c->min_x = x;
+   if (y < c->min_y || !c->started)
+      c->min_y = y;
+   c->started = 1;
+}
+
+static void stbtt__csctx_v(stbtt__csctx* c, stbtt_uint8 type, stbtt_int32 x, stbtt_int32 y, stbtt_int32 cx, stbtt_int32 cy, stbtt_int32 cx1, stbtt_int32 cy1)
+{
+   if (c->bounds)
+   {
+      stbtt__track_vertex(c, x, y);
+      if (type == STBTT_vcubic)
+      {
+         stbtt__track_vertex(c, cx, cy);
+         stbtt__track_vertex(c, cx1, cy1);
+      }
+   }
+   else
+   {
+      stbtt_setvertex(&c->pvertices[c->num_vertices], type, x, y, cx, cy);
+      c->pvertices[c->num_vertices].cx1 = (stbtt_int16)cx1;
+      c->pvertices[c->num_vertices].cy1 = (stbtt_int16)cy1;
+   }
+   c->num_vertices++;
+}
+
+static void stbtt__csctx_close_shape(stbtt__csctx* ctx)
+{
+   if (ctx->first_x != ctx->x || ctx->first_y != ctx->y)
+      stbtt__csctx_v(ctx, STBTT_vline, (int)ctx->first_x, (int)ctx->first_y, 0, 0, 0, 0);
+}
+
+static void stbtt__csctx_rmove_to(stbtt__csctx* ctx, float dx, float dy)
+{
+   stbtt__csctx_close_shape(ctx);
+   ctx->first_x = ctx->x = ctx->x + dx;
+   ctx->first_y = ctx->y = ctx->y + dy;
+   stbtt__csctx_v(ctx, STBTT_vmove, (int)ctx->x, (int)ctx->y, 0, 0, 0, 0);
+}
+
+static void stbtt__csctx_rline_to(stbtt__csctx* ctx, float dx, float dy)
+{
+   ctx->x += dx;
+   ctx->y += dy;
+   stbtt__csctx_v(ctx, STBTT_vline, (int)ctx->x, (int)ctx->y, 0, 0, 0, 0);
+}
+
+static void stbtt__csctx_rccurve_to(stbtt__csctx* ctx, float dx1, float dy1, float dx2, float dy2, float dx3, float dy3)
+{
+   float cx1 = ctx->x + dx1;
+   float cy1 = ctx->y + dy1;
+   float cx2 = cx1 + dx2;
+   float cy2 = cy1 + dy2;
+   ctx->x = cx2 + dx3;
+   ctx->y = cy2 + dy3;
+   stbtt__csctx_v(ctx, STBTT_vcubic, (int)ctx->x, (int)ctx->y, (int)cx1, (int)cy1, (int)cx2, (int)cy2);
+}
+
+static stbtt__buf stbtt__get_subr(stbtt__buf idx, int n)
+{
+   int count = stbtt__cff_index_count(&idx);
+   int bias = 107;
+   if (count >= 33900)
+      bias = 32768;
+   else if (count >= 1240)
+      bias = 1131;
+   n += bias;
+   if (n < 0 || n >= count)
+      return stbtt__new_buf(NULL, 0);
+   return stbtt__cff_index_get(idx, n);
+}
+
+static stbtt__buf stbtt__cid_get_glyph_subrs(const stbtt_fontinfo* info, int glyph_index)
+{
+   stbtt__buf fdselect = info->fdselect;
+   int nranges, start, end, v, fmt, fdselector = -1, i;
+
+   stbtt__buf_seek(&fdselect, 0);
+   fmt = stbtt__buf_get8(&fdselect);
+   if (fmt == 0)
+   {
+      // untested
+      stbtt__buf_skip(&fdselect, glyph_index);
+      fdselector = stbtt__buf_get8(&fdselect);
+   }
+   else if (fmt == 3)
+   {
+      nranges = stbtt__buf_get16(&fdselect);
+      start = stbtt__buf_get16(&fdselect);
+      for (i = 0; i < nranges; i++)
+      {
+         v = stbtt__buf_get8(&fdselect);
+         end = stbtt__buf_get16(&fdselect);
+         if (glyph_index >= start && glyph_index < end)
+         {
+            fdselector = v;
+            break;
+         }
+         start = end;
+      }
+   }
+   if (fdselector == -1)
+      stbtt__new_buf(NULL, 0);
+   return stbtt__get_subrs(info->cff, stbtt__cff_index_get(info->fontdicts, fdselector));
+}
+
+static int stbtt__run_charstring(const stbtt_fontinfo* info, int glyph_index, stbtt__csctx* c)
+{
+   int in_header = 1, maskbits = 0, subr_stack_height = 0, sp = 0, v, i, b0;
+   int has_subrs = 0, clear_stack;
+   float s[48];
+   stbtt__buf subr_stack[10], subrs = info->subrs, b;
+   float f;
+
+#define STBTT__CSERR(s) (0)
+
+   // this currently ignores the initial width value, which isn't needed if we have hmtx
+   b = stbtt__cff_index_get(info->charstrings, glyph_index);
+   while (b.cursor < b.size)
+   {
+      i = 0;
+      clear_stack = 1;
+      b0 = stbtt__buf_get8(&b);
+      switch (b0)
+      {
+         // @TODO implement hinting
+         case 0x13: // hintmask
+         case 0x14: // cntrmask
+            if (in_header)
+               maskbits += (sp / 2); // implicit "vstem"
+            in_header = 0;
+            stbtt__buf_skip(&b, (maskbits + 7) / 8);
+            break;
+
+         case 0x01: // hstem
+         case 0x03: // vstem
+         case 0x12: // hstemhm
+         case 0x17: // vstemhm
+            maskbits += (sp / 2);
+            break;
+
+         case 0x15: // rmoveto
+            in_header = 0;
+            if (sp < 2)
+               return STBTT__CSERR("rmoveto stack");
+            stbtt__csctx_rmove_to(c, s[sp - 2], s[sp - 1]);
+            break;
+         case 0x04: // vmoveto
+            in_header = 0;
+            if (sp < 1)
+               return STBTT__CSERR("vmoveto stack");
+            stbtt__csctx_rmove_to(c, 0, s[sp - 1]);
+            break;
+         case 0x16: // hmoveto
+            in_header = 0;
+            if (sp < 1)
+               return STBTT__CSERR("hmoveto stack");
+            stbtt__csctx_rmove_to(c, s[sp - 1], 0);
+            break;
+
+         case 0x05: // rlineto
+            if (sp < 2)
+               return STBTT__CSERR("rlineto stack");
+            for (; i + 1 < sp; i += 2)
+               stbtt__csctx_rline_to(c, s[i], s[i + 1]);
+            break;
+
+            // hlineto/vlineto and vhcurveto/hvcurveto alternate horizontal and vertical
+            // starting from a different place.
+
+         case 0x07: // vlineto
+            if (sp < 1)
+               return STBTT__CSERR("vlineto stack");
+            goto vlineto;
+         case 0x06: // hlineto
+            if (sp < 1)
+               return STBTT__CSERR("hlineto stack");
+            for (;;)
+            {
+               if (i >= sp)
+                  break;
+               stbtt__csctx_rline_to(c, s[i], 0);
+               i++;
+            vlineto:
+               if (i >= sp)
+                  break;
+               stbtt__csctx_rline_to(c, 0, s[i]);
+               i++;
+            }
+            break;
+
+         case 0x1F: // hvcurveto
+            if (sp < 4)
+               return STBTT__CSERR("hvcurveto stack");
+            goto hvcurveto;
+         case 0x1E: // vhcurveto
+            if (sp < 4)
+               return STBTT__CSERR("vhcurveto stack");
+            for (;;)
+            {
+               if (i + 3 >= sp)
+                  break;
+               stbtt__csctx_rccurve_to(c, 0, s[i], s[i + 1], s[i + 2], s[i + 3], (sp - i == 5) ? s[i + 4] : 0.0f);
+               i += 4;
+            hvcurveto:
+               if (i + 3 >= sp)
+                  break;
+               stbtt__csctx_rccurve_to(c, s[i], 0, s[i + 1], s[i + 2], (sp - i == 5) ? s[i + 4] : 0.0f, s[i + 3]);
+               i += 4;
+            }
+            break;
+
+         case 0x08: // rrcurveto
+            if (sp < 6)
+               return STBTT__CSERR("rcurveline stack");
+            for (; i + 5 < sp; i += 6)
+               stbtt__csctx_rccurve_to(c, s[i], s[i + 1], s[i + 2], s[i + 3], s[i + 4], s[i + 5]);
+            break;
+
+         case 0x18: // rcurveline
+            if (sp < 8)
+               return STBTT__CSERR("rcurveline stack");
+            for (; i + 5 < sp - 2; i += 6)
+               stbtt__csctx_rccurve_to(c, s[i], s[i + 1], s[i + 2], s[i + 3], s[i + 4], s[i + 5]);
+            if (i + 1 >= sp)
+               return STBTT__CSERR("rcurveline stack");
+            stbtt__csctx_rline_to(c, s[i], s[i + 1]);
+            break;
+
+         case 0x19: // rlinecurve
+            if (sp < 8)
+               return STBTT__CSERR("rlinecurve stack");
+            for (; i + 1 < sp - 6; i += 2)
+               stbtt__csctx_rline_to(c, s[i], s[i + 1]);
+            if (i + 5 >= sp)
+               return STBTT__CSERR("rlinecurve stack");
+            stbtt__csctx_rccurve_to(c, s[i], s[i + 1], s[i + 2], s[i + 3], s[i + 4], s[i + 5]);
+            break;
+
+         case 0x1A: // vvcurveto
+         case 0x1B: // hhcurveto
+            if (sp < 4)
+               return STBTT__CSERR("(vv|hh)curveto stack");
+            f = 0.0;
+            if (sp & 1)
+            {
+               f = s[i];
+               i++;
+            }
+            for (; i + 3 < sp; i += 4)
+            {
+               if (b0 == 0x1B)
+                  stbtt__csctx_rccurve_to(c, s[i], f, s[i + 1], s[i + 2], s[i + 3], 0.0);
+               else
+                  stbtt__csctx_rccurve_to(c, f, s[i], s[i + 1], s[i + 2], 0.0, s[i + 3]);
+               f = 0.0;
+            }
+            break;
+
+         case 0x0A: // callsubr
+            if (!has_subrs)
+            {
+               if (info->fdselect.size)
+                  subrs = stbtt__cid_get_glyph_subrs(info, glyph_index);
+               has_subrs = 1;
+            }
+            // FALLTHROUGH
+         case 0x1D: // callgsubr
+            if (sp < 1)
+               return STBTT__CSERR("call(g|)subr stack");
+            v = (int)s[--sp];
+            if (subr_stack_height >= 10)
+               return STBTT__CSERR("recursion limit");
+            subr_stack[subr_stack_height++] = b;
+            b = stbtt__get_subr(b0 == 0x0A ? subrs : info->gsubrs, v);
+            if (b.size == 0)
+               return STBTT__CSERR("subr not found");
+            b.cursor = 0;
+            clear_stack = 0;
+            break;
+
+         case 0x0B: // return
+            if (subr_stack_height <= 0)
+               return STBTT__CSERR("return outside subr");
+            b = subr_stack[--subr_stack_height];
+            clear_stack = 0;
+            break;
+
+         case 0x0E: // endchar
+            stbtt__csctx_close_shape(c);
+            return 1;
+
+         case 0x0C:
+         { // two-byte escape
+            float dx1, dx2, dx3, dx4, dx5, dx6, dy1, dy2, dy3, dy4, dy5, dy6;
+            float dx, dy;
+            int b1 = stbtt__buf_get8(&b);
+            switch (b1)
+            {
+               // @TODO These "flex" implementations ignore the flex-depth and resolution,
+               // and always draw beziers.
+               case 0x22: // hflex
+                  if (sp < 7)
+                     return STBTT__CSERR("hflex stack");
+                  dx1 = s[0];
+                  dx2 = s[1];
+                  dy2 = s[2];
+                  dx3 = s[3];
+                  dx4 = s[4];
+                  dx5 = s[5];
+                  dx6 = s[6];
+                  stbtt__csctx_rccurve_to(c, dx1, 0, dx2, dy2, dx3, 0);
+                  stbtt__csctx_rccurve_to(c, dx4, 0, dx5, -dy2, dx6, 0);
+                  break;
+
+               case 0x23: // flex
+                  if (sp < 13)
+                     return STBTT__CSERR("flex stack");
+                  dx1 = s[0];
+                  dy1 = s[1];
+                  dx2 = s[2];
+                  dy2 = s[3];
+                  dx3 = s[4];
+                  dy3 = s[5];
+                  dx4 = s[6];
+                  dy4 = s[7];
+                  dx5 = s[8];
+                  dy5 = s[9];
+                  dx6 = s[10];
+                  dy6 = s[11];
+                  //fd is s[12]
+                  stbtt__csctx_rccurve_to(c, dx1, dy1, dx2, dy2, dx3, dy3);
+                  stbtt__csctx_rccurve_to(c, dx4, dy4, dx5, dy5, dx6, dy6);
+                  break;
+
+               case 0x24: // hflex1
+                  if (sp < 9)
+                     return STBTT__CSERR("hflex1 stack");
+                  dx1 = s[0];
+                  dy1 = s[1];
+                  dx2 = s[2];
+                  dy2 = s[3];
+                  dx3 = s[4];
+                  dx4 = s[5];
+                  dx5 = s[6];
+                  dy5 = s[7];
+                  dx6 = s[8];
+                  stbtt__csctx_rccurve_to(c, dx1, dy1, dx2, dy2, dx3, 0);
+                  stbtt__csctx_rccurve_to(c, dx4, 0, dx5, dy5, dx6, -(dy1 + dy2 + dy5));
+                  break;
+
+               case 0x25: // flex1
+                  if (sp < 11)
+                     return STBTT__CSERR("flex1 stack");
+                  dx1 = s[0];
+                  dy1 = s[1];
+                  dx2 = s[2];
+                  dy2 = s[3];
+                  dx3 = s[4];
+                  dy3 = s[5];
+                  dx4 = s[6];
+                  dy4 = s[7];
+                  dx5 = s[8];
+                  dy5 = s[9];
+                  dx6 = dy6 = s[10];
+                  dx = dx1 + dx2 + dx3 + dx4 + dx5;
+                  dy = dy1 + dy2 + dy3 + dy4 + dy5;
+                  if (STBTT_fabs(dx) > STBTT_fabs(dy))
+                     dy6 = -dy;
+                  else
+                     dx6 = -dx;
+                  stbtt__csctx_rccurve_to(c, dx1, dy1, dx2, dy2, dx3, dy3);
+                  stbtt__csctx_rccurve_to(c, dx4, dy4, dx5, dy5, dx6, dy6);
+                  break;
+
+               default:
+                  return STBTT__CSERR("unimplemented");
+            }
+         }
+         break;
+
+         default:
+            if (b0 != 255 && b0 != 28 && b0 < 32)
+               return STBTT__CSERR("reserved operator");
+
+            // push immediate
+            if (b0 == 255)
+            {
+               f = (float)(stbtt_int32)stbtt__buf_get32(&b) / 0x10000;
+            }
+            else
+            {
+               stbtt__buf_skip(&b, -1);
+               f = (float)(stbtt_int16)stbtt__cff_int(&b);
+            }
+            if (sp >= 48)
+               return STBTT__CSERR("push stack overflow");
+            s[sp++] = f;
+            clear_stack = 0;
+            break;
+      }
+      if (clear_stack)
+         sp = 0;
+   }
+   return STBTT__CSERR("no endchar");
+
+#undef STBTT__CSERR
+}
+
+static int stbtt__GetGlyphShapeT2(const stbtt_fontinfo* info, int glyph_index, stbtt_vertex** pvertices)
+{
+   // runs the charstring twice, once to count and once to output (to avoid realloc)
+   stbtt__csctx count_ctx = STBTT__CSCTX_INIT(1);
+   stbtt__csctx output_ctx = STBTT__CSCTX_INIT(0);
+   if (stbtt__run_charstring(info, glyph_index, &count_ctx))
+   {
+      *pvertices = (stbtt_vertex*)STBTT_malloc(count_ctx.num_vertices * sizeof(stbtt_vertex), info->userdata);
+      output_ctx.pvertices = *pvertices;
+      if (stbtt__run_charstring(info, glyph_index, &output_ctx))
+      {
+         STBTT_assert(output_ctx.num_vertices == count_ctx.num_vertices);
+         return output_ctx.num_vertices;
+      }
+   }
+   *pvertices = NULL;
+   return 0;
+}
+
+static int stbtt__GetGlyphInfoT2(const stbtt_fontinfo* info, int glyph_index, int* x0, int* y0, int* x1, int* y1)
+{
+   stbtt__csctx c = STBTT__CSCTX_INIT(1);
+   int r = stbtt__run_charstring(info, glyph_index, &c);
+   if (x0)
+      *x0 = r ? c.min_x : 0;
+   if (y0)
+      *y0 = r ? c.min_y : 0;
+   if (x1)
+      *x1 = r ? c.max_x : 0;
+   if (y1)
+      *y1 = r ? c.max_y : 0;
+   return r ? c.num_vertices : 0;
+}
+
+STBTT_DEF int stbtt_GetGlyphShape(const stbtt_fontinfo* info, int glyph_index, stbtt_vertex** pvertices)
+{
+   if (!info->cff.size)
+      return stbtt__GetGlyphShapeTT(info, glyph_index, pvertices);
+   else
+      return stbtt__GetGlyphShapeT2(info, glyph_index, pvertices);
+}
+
+STBTT_DEF 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));
+   }
+}
+
+STBTT_DEF int stbtt_GetKerningTableLength(const stbtt_fontinfo* info)
+{
+   stbtt_uint8* data = info->data + info->kern;
+
+   // 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;
+
+   return ttUSHORT(data + 10);
+}
+
+STBTT_DEF int stbtt_GetKerningTable(const stbtt_fontinfo* info, stbtt_kerningentry* table, int table_length)
+{
+   stbtt_uint8* data = info->data + info->kern;
+   int k, length;
+
+   // 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;
+
+   length = ttUSHORT(data + 10);
+   if (table_length < length)
+      length = table_length;
+
+   for (k = 0; k < length; k++)
+   {
+      table[k].glyph1 = ttUSHORT(data + 18 + (k * 6));
+      table[k].glyph2 = ttUSHORT(data + 20 + (k * 6));
+      table[k].advance = ttSHORT(data + 22 + (k * 6));
+   }
+
+   return length;
+}
+
+static int stbtt__GetGlyphKernInfoAdvance(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
+   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
+   if (ttUSHORT(data + 8) != 1) // horizontal flag must be set in format
       return 0;
 
    l = 0;
-   r = ttUSHORT(data+10) - 1;
+   r = ttUSHORT(data + 10) - 1;
    needle = glyph1 << 16 | glyph2;
-   while (l <= r) {
+   while (l <= r)
+   {
       m = (l + r) >> 1;
-      straw = ttULONG(data+18+(m*6)); // note: unaligned read
+      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 ttSHORT(data + 22 + (m * 6));
+   }
+   return 0;
+}
+
+static stbtt_int32 stbtt__GetCoverageIndex(stbtt_uint8* coverageTable, int glyph)
+{
+   stbtt_uint16 coverageFormat = ttUSHORT(coverageTable);
+   switch (coverageFormat)
+   {
+      case 1:
+      {
+         stbtt_uint16 glyphCount = ttUSHORT(coverageTable + 2);
+
+         // Binary search.
+         stbtt_int32 l = 0, r = glyphCount - 1, m;
+         int straw, needle = glyph;
+         while (l <= r)
+         {
+            stbtt_uint8* glyphArray = coverageTable + 4;
+            stbtt_uint16 glyphID;
+            m = (l + r) >> 1;
+            glyphID = ttUSHORT(glyphArray + 2 * m);
+            straw = glyphID;
+            if (needle < straw)
+               r = m - 1;
+            else if (needle > straw)
+               l = m + 1;
+            else
+            {
+               return m;
+            }
+         }
+         break;
+      }
+
+      case 2:
+      {
+         stbtt_uint16 rangeCount = ttUSHORT(coverageTable + 2);
+         stbtt_uint8* rangeArray = coverageTable + 4;
+
+         // Binary search.
+         stbtt_int32 l = 0, r = rangeCount - 1, m;
+         int strawStart, strawEnd, needle = glyph;
+         while (l <= r)
+         {
+            stbtt_uint8* rangeRecord;
+            m = (l + r) >> 1;
+            rangeRecord = rangeArray + 6 * m;
+            strawStart = ttUSHORT(rangeRecord);
+            strawEnd = ttUSHORT(rangeRecord + 2);
+            if (needle < strawStart)
+               r = m - 1;
+            else if (needle > strawEnd)
+               l = m + 1;
+            else
+            {
+               stbtt_uint16 startCoverageIndex = ttUSHORT(rangeRecord + 4);
+               return startCoverageIndex + glyph - strawStart;
+            }
+         }
+         break;
+      }
+
+      default: return -1; // unsupported
+   }
+
+   return -1;
+}
+
+static stbtt_int32 stbtt__GetGlyphClass(stbtt_uint8* classDefTable, int glyph)
+{
+   stbtt_uint16 classDefFormat = ttUSHORT(classDefTable);
+   switch (classDefFormat)
+   {
+      case 1:
+      {
+         stbtt_uint16 startGlyphID = ttUSHORT(classDefTable + 2);
+         stbtt_uint16 glyphCount = ttUSHORT(classDefTable + 4);
+         stbtt_uint8* classDef1ValueArray = classDefTable + 6;
+
+         if (glyph >= startGlyphID && glyph < startGlyphID + glyphCount)
+            return (stbtt_int32)ttUSHORT(classDef1ValueArray + 2 * (glyph - startGlyphID));
+         break;
+      }
+
+      case 2:
+      {
+         stbtt_uint16 classRangeCount = ttUSHORT(classDefTable + 2);
+         stbtt_uint8* classRangeRecords = classDefTable + 4;
+
+         // Binary search.
+         stbtt_int32 l = 0, r = classRangeCount - 1, m;
+         int strawStart, strawEnd, needle = glyph;
+         while (l <= r)
+         {
+            stbtt_uint8* classRangeRecord;
+            m = (l + r) >> 1;
+            classRangeRecord = classRangeRecords + 6 * m;
+            strawStart = ttUSHORT(classRangeRecord);
+            strawEnd = ttUSHORT(classRangeRecord + 2);
+            if (needle < strawStart)
+               r = m - 1;
+            else if (needle > strawEnd)
+               l = m + 1;
+            else
+               return (stbtt_int32)ttUSHORT(classRangeRecord + 4);
+         }
+         break;
+      }
+
+      default:
+         return -1; // Unsupported definition type, return an error.
    }
+
+   // "All glyphs not assigned to a class fall into class 0". (OpenType spec)
    return 0;
 }
 
-STBTT_DEF int  stbtt_GetCodepointKernAdvance(const stbtt_fontinfo *info, int ch1, int ch2)
+// Define to STBTT_assert(x) if you want to break on unimplemented formats.
+#define STBTT_GPOS_TODO_assert(x)
+
+static stbtt_int32 stbtt__GetGlyphGPOSInfoAdvance(const stbtt_fontinfo* info, int glyph1, int glyph2)
 {
-   if (!info->kern) // if no kerning table, don't waste time looking up both codepoint->glyphs
+   stbtt_uint16 lookupListOffset;
+   stbtt_uint8* lookupList;
+   stbtt_uint16 lookupCount;
+   stbtt_uint8* data;
+   stbtt_int32 i, sti;
+
+   if (!info->gpos)
       return 0;
-   return stbtt_GetGlyphKernAdvance(info, stbtt_FindGlyphIndex(info,ch1), stbtt_FindGlyphIndex(info,ch2));
+
+   data = info->data + info->gpos;
+
+   if (ttUSHORT(data + 0) != 1)
+      return 0; // Major version 1
+   if (ttUSHORT(data + 2) != 0)
+      return 0; // Minor version 0
+
+   lookupListOffset = ttUSHORT(data + 8);
+   lookupList = data + lookupListOffset;
+   lookupCount = ttUSHORT(lookupList);
+
+   for (i = 0; i < lookupCount; ++i)
+   {
+      stbtt_uint16 lookupOffset = ttUSHORT(lookupList + 2 + 2 * i);
+      stbtt_uint8* lookupTable = lookupList + lookupOffset;
+
+      stbtt_uint16 lookupType = ttUSHORT(lookupTable);
+      stbtt_uint16 subTableCount = ttUSHORT(lookupTable + 4);
+      stbtt_uint8* subTableOffsets = lookupTable + 6;
+      if (lookupType != 2) // Pair Adjustment Positioning Subtable
+         continue;
+
+      for (sti = 0; sti < subTableCount; sti++)
+      {
+         stbtt_uint16 subtableOffset = ttUSHORT(subTableOffsets + 2 * sti);
+         stbtt_uint8* table = lookupTable + subtableOffset;
+         stbtt_uint16 posFormat = ttUSHORT(table);
+         stbtt_uint16 coverageOffset = ttUSHORT(table + 2);
+         stbtt_int32 coverageIndex = stbtt__GetCoverageIndex(table + coverageOffset, glyph1);
+         if (coverageIndex == -1)
+            continue;
+
+         switch (posFormat)
+         {
+            case 1:
+            {
+               stbtt_int32 l, r, m;
+               int straw, needle;
+               stbtt_uint16 valueFormat1 = ttUSHORT(table + 4);
+               stbtt_uint16 valueFormat2 = ttUSHORT(table + 6);
+               if (valueFormat1 == 4 && valueFormat2 == 0)
+               { // Support more formats?
+                  stbtt_int32 valueRecordPairSizeInBytes = 2;
+                  stbtt_uint16 pairSetCount = ttUSHORT(table + 8);
+                  stbtt_uint16 pairPosOffset = ttUSHORT(table + 10 + 2 * coverageIndex);
+                  stbtt_uint8* pairValueTable = table + pairPosOffset;
+                  stbtt_uint16 pairValueCount = ttUSHORT(pairValueTable);
+                  stbtt_uint8* pairValueArray = pairValueTable + 2;
+
+                  if (coverageIndex >= pairSetCount)
+                     return 0;
+
+                  needle = glyph2;
+                  r = pairValueCount - 1;
+                  l = 0;
+
+                  // Binary search.
+                  while (l <= r)
+                  {
+                     stbtt_uint16 secondGlyph;
+                     stbtt_uint8* pairValue;
+                     m = (l + r) >> 1;
+                     pairValue = pairValueArray + (2 + valueRecordPairSizeInBytes) * m;
+                     secondGlyph = ttUSHORT(pairValue);
+                     straw = secondGlyph;
+                     if (needle < straw)
+                        r = m - 1;
+                     else if (needle > straw)
+                        l = m + 1;
+                     else
+                     {
+                        stbtt_int16 xAdvance = ttSHORT(pairValue + 2);
+                        return xAdvance;
+                     }
+                  }
+               }
+               else
+                  return 0;
+               break;
+            }
+
+            case 2:
+            {
+               stbtt_uint16 valueFormat1 = ttUSHORT(table + 4);
+               stbtt_uint16 valueFormat2 = ttUSHORT(table + 6);
+               if (valueFormat1 == 4 && valueFormat2 == 0)
+               { // Support more formats?
+                  stbtt_uint16 classDef1Offset = ttUSHORT(table + 8);
+                  stbtt_uint16 classDef2Offset = ttUSHORT(table + 10);
+                  int glyph1class = stbtt__GetGlyphClass(table + classDef1Offset, glyph1);
+                  int glyph2class = stbtt__GetGlyphClass(table + classDef2Offset, glyph2);
+
+                  stbtt_uint16 class1Count = ttUSHORT(table + 12);
+                  stbtt_uint16 class2Count = ttUSHORT(table + 14);
+                  stbtt_uint8 *class1Records, *class2Records;
+                  stbtt_int16 xAdvance;
+
+                  if (glyph1class < 0 || glyph1class >= class1Count)
+                     return 0; // malformed
+                  if (glyph2class < 0 || glyph2class >= class2Count)
+                     return 0; // malformed
+
+                  class1Records = table + 16;
+                  class2Records = class1Records + 2 * (glyph1class * class2Count);
+                  xAdvance = ttSHORT(class2Records + 2 * glyph2class);
+                  return xAdvance;
+               }
+               else
+                  return 0;
+               break;
+            }
+
+            default:
+               return 0; // Unsupported position format
+         }
+      }
+   }
+
+   return 0;
 }
 
-STBTT_DEF void stbtt_GetCodepointHMetrics(const stbtt_fontinfo *info, int codepoint, int *advanceWidth, int *leftSideBearing)
+STBTT_DEF int stbtt_GetGlyphKernAdvance(const stbtt_fontinfo* info, int g1, int g2)
 {
-   stbtt_GetGlyphHMetrics(info, stbtt_FindGlyphIndex(info,codepoint), advanceWidth, leftSideBearing);
+   int xAdvance = 0;
+
+   if (info->gpos)
+      xAdvance += stbtt__GetGlyphGPOSInfoAdvance(info, g1, g2);
+   else if (info->kern)
+      xAdvance += stbtt__GetGlyphKernInfoAdvance(info, g1, g2);
+
+   return xAdvance;
+}
+
+STBTT_DEF int stbtt_GetCodepointKernAdvance(const stbtt_fontinfo* info, int ch1, int ch2)
+{
+   if (!info->kern && !info->gpos) // 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));
 }
 
-STBTT_DEF void stbtt_GetFontVMetrics(const stbtt_fontinfo *info, int *ascent, int *descent, int *lineGap)
+STBTT_DEF void stbtt_GetCodepointHMetrics(const stbtt_fontinfo* info, int codepoint, int* advanceWidth, int* leftSideBearing)
 {
-   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);
+   stbtt_GetGlyphHMetrics(info, stbtt_FindGlyphIndex(info, codepoint), advanceWidth, leftSideBearing);
 }
 
-STBTT_DEF void stbtt_GetFontBoundingBox(const stbtt_fontinfo *info, int *x0, int *y0, int *x1, int *y1)
+STBTT_DEF 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);
+}
+
+STBTT_DEF int stbtt_GetFontVMetricsOS2(const stbtt_fontinfo* info, int* typoAscent, int* typoDescent, int* typoLineGap)
+{
+   int tab = stbtt__find_table(info->data, info->fontstart, "OS/2");
+   if (!tab)
+      return 0;
+   if (typoAscent)
+      *typoAscent = ttSHORT(info->data + tab + 68);
+   if (typoDescent)
+      *typoDescent = ttSHORT(info->data + tab + 70);
+   if (typoLineGap)
+      *typoLineGap = ttSHORT(info->data + tab + 72);
+   return 1;
+}
+
+STBTT_DEF 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);
@@ -1535,59 +2931,113 @@ STBTT_DEF void stbtt_GetFontBoundingBox(const stbtt_fontinfo *info, int *x0, int
    *y1 = ttSHORT(info->data + info->head + 42);
 }
 
-STBTT_DEF float stbtt_ScaleForPixelHeight(const stbtt_fontinfo *info, float height)
+STBTT_DEF 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;
+   return (float)height / fheight;
 }
 
-STBTT_DEF float stbtt_ScaleForMappingEmToPixels(const stbtt_fontinfo *info, float pixels)
+STBTT_DEF float stbtt_ScaleForMappingEmToPixels(const stbtt_fontinfo* info, float pixels)
 {
    int unitsPerEm = ttUSHORT(info->data + info->head + 18);
    return pixels / unitsPerEm;
 }
 
-STBTT_DEF void stbtt_FreeShape(const stbtt_fontinfo *info, stbtt_vertex *v)
+STBTT_DEF void stbtt_FreeShape(const stbtt_fontinfo* info, stbtt_vertex* v)
 {
    STBTT_free(v, info->userdata);
 }
 
+STBTT_DEF stbtt_uint8* stbtt_FindSVGDoc(const stbtt_fontinfo* info, int gl)
+{
+   int i;
+   stbtt_uint8* data = info->data;
+   stbtt_uint8* svg_doc_list = data + stbtt__get_svg((stbtt_fontinfo*)info);
+
+   int numEntries = ttUSHORT(svg_doc_list);
+   stbtt_uint8* svg_docs = svg_doc_list + 2;
+
+   for (i = 0; i < numEntries; i++)
+   {
+      stbtt_uint8* svg_doc = svg_docs + (12 * i);
+      if ((gl >= ttUSHORT(svg_doc)) && (gl <= ttUSHORT(svg_doc + 2)))
+         return svg_doc;
+   }
+   return 0;
+}
+
+STBTT_DEF int stbtt_GetGlyphSVG(const stbtt_fontinfo* info, int gl, const char** svg)
+{
+   stbtt_uint8* data = info->data;
+   stbtt_uint8* svg_doc;
+
+   if (info->svg == 0)
+      return 0;
+
+   svg_doc = stbtt_FindSVGDoc(info, gl);
+   if (svg_doc != NULL)
+   {
+      *svg = (char*)data + info->svg + ttULONG(svg_doc + 4);
+      return ttULONG(svg_doc + 8);
+   }
+   else
+   {
+      return 0;
+   }
+}
+
+STBTT_DEF int stbtt_GetCodepointSVG(const stbtt_fontinfo* info, int unicode_codepoint, const char** svg)
+{
+   return stbtt_GetGlyphSVG(info, stbtt_FindGlyphIndex(info, unicode_codepoint), svg);
+}
+
 //////////////////////////////////////////////////////////////////////////////
 //
 // antialiasing software rasterizer
 //
 
-STBTT_DEF 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)
+STBTT_DEF 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=0,y0=0,x1,y1; // =0 suppresses compiler warning
-   if (!stbtt_GetGlyphBox(font, glyph, &x0,&y0,&x1,&y1)) {
+   int x0 = 0, y0 = 0, x1, y1; // =0 suppresses compiler warning
+   if (!stbtt_GetGlyphBox(font, glyph, &x0, &y0, &x1, &y1))
+   {
       // e.g. space character
-      if (ix0) *ix0 = 0;
-      if (iy0) *iy0 = 0;
-      if (ix1) *ix1 = 0;
-      if (iy1) *iy1 = 0;
-   } else {
+      if (ix0)
+         *ix0 = 0;
+      if (iy0)
+         *iy0 = 0;
+      if (ix1)
+         *ix1 = 0;
+      if (iy1)
+         *iy1 = 0;
+   }
+   else
+   {
       // 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_ifloor(-y1 * scale_y + shift_y);
-      if (ix1) *ix1 = STBTT_iceil ( x1 * scale_x + shift_x);
-      if (iy1) *iy1 = STBTT_iceil (-y0 * scale_y + shift_y);
+      if (ix0)
+         *ix0 = STBTT_ifloor(x0 * scale_x + shift_x);
+      if (iy0)
+         *iy0 = STBTT_ifloor(-y1 * scale_y + shift_y);
+      if (ix1)
+         *ix1 = STBTT_iceil(x1 * scale_x + shift_x);
+      if (iy1)
+         *iy1 = STBTT_iceil(-y0 * scale_y + shift_y);
    }
 }
 
-STBTT_DEF void stbtt_GetGlyphBitmapBox(const stbtt_fontinfo *font, int glyph, float scale_x, float scale_y, int *ix0, int *iy0, int *ix1, int *iy1)
+STBTT_DEF 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);
+   stbtt_GetGlyphBitmapBoxSubpixel(font, glyph, scale_x, scale_y, 0.0f, 0.0f, ix0, iy0, ix1, iy1);
 }
 
-STBTT_DEF 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_DEF 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);
+   stbtt_GetGlyphBitmapBoxSubpixel(font, stbtt_FindGlyphIndex(font, codepoint), scale_x, scale_y, shift_x, shift_y, ix0, iy0, ix1, iy1);
 }
 
-STBTT_DEF void stbtt_GetCodepointBitmapBox(const stbtt_fontinfo *font, int codepoint, float scale_x, float scale_y, int *ix0, int *iy0, int *ix1, int *iy1)
+STBTT_DEF 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);
+   stbtt_GetCodepointBitmapBoxSubpixel(font, codepoint, scale_x, scale_y, 0.0f, 0.0f, ix0, iy0, ix1, iy1);
 }
 
 //////////////////////////////////////////////////////////////////////////////
@@ -1596,26 +3046,31 @@ STBTT_DEF void stbtt_GetCodepointBitmapBox(const stbtt_fontinfo *font, int codep
 
 typedef struct stbtt__hheap_chunk
 {
-   struct stbtt__hheap_chunk *next;
+   struct stbtt__hheap_chunk* next;
 } stbtt__hheap_chunk;
 
 typedef struct stbtt__hheap
 {
-   struct stbtt__hheap_chunk *head;
-   void   *first_free;
-   int    num_remaining_in_head_chunk;
+   struct stbtt__hheap_chunk* head;
+   void* first_free;
+   int num_remaining_in_head_chunk;
 } stbtt__hheap;
 
-static void *stbtt__hheap_alloc(stbtt__hheap *hh, size_t size, void *userdata)
+static void* stbtt__hheap_alloc(stbtt__hheap* hh, size_t size, void* userdata)
 {
-   if (hh->first_free) {
-      void *p = hh->first_free;
-      hh->first_free = * (void **) p;
+   if (hh->first_free)
+   {
+      void* p = hh->first_free;
+      hh->first_free = *(void**)p;
       return p;
-   } else {
-      if (hh->num_remaining_in_head_chunk == 0) {
-         int count = (size < 32 ? 2000 : size < 128 ? 800 : 100);
-         stbtt__hheap_chunk *c = (stbtt__hheap_chunk *) STBTT_malloc(sizeof(stbtt__hheap_chunk) + size * count, userdata);
+   }
+   else
+   {
+      if (hh->num_remaining_in_head_chunk == 0)
+      {
+         int count = (size < 32 ? 2000 : size < 128 ? 800
+                                                    : 100);
+         stbtt__hheap_chunk* c = (stbtt__hheap_chunk*)STBTT_malloc(sizeof(stbtt__hheap_chunk) + size * count, userdata);
          if (c == NULL)
             return NULL;
          c->next = hh->head;
@@ -1623,61 +3078,64 @@ static void *stbtt__hheap_alloc(stbtt__hheap *hh, size_t size, void *userdata)
          hh->num_remaining_in_head_chunk = count;
       }
       --hh->num_remaining_in_head_chunk;
-      return (char *) (hh->head) + size * hh->num_remaining_in_head_chunk;
+      return (char*)(hh->head) + sizeof(stbtt__hheap_chunk) + size * hh->num_remaining_in_head_chunk;
    }
 }
 
-static void stbtt__hheap_free(stbtt__hheap *hh, void *p)
+static void stbtt__hheap_free(stbtt__hheap* hh, void* p)
 {
-   *(void **) p = hh->first_free;
+   *(void**)p = hh->first_free;
    hh->first_free = p;
 }
 
-static void stbtt__hheap_cleanup(stbtt__hheap *hh, void *userdata)
+static void stbtt__hheap_cleanup(stbtt__hheap* hh, void* userdata)
 {
-   stbtt__hheap_chunk *c = hh->head;
-   while (c) {
-      stbtt__hheap_chunk *n = c->next;
+   stbtt__hheap_chunk* c = hh->head;
+   while (c)
+   {
+      stbtt__hheap_chunk* n = c->next;
       STBTT_free(c, userdata);
       c = n;
    }
 }
 
-typedef struct stbtt__edge {
-   float x0,y0, x1,y1;
+typedef struct stbtt__edge
+{
+   float x0, y0, x1, y1;
    int invert;
 } stbtt__edge;
 
 
 typedef struct stbtt__active_edge
 {
-   struct stbtt__active_edge *next;
-   #if STBTT_RASTERIZER_VERSION==1
-   int x,dx;
+   struct stbtt__active_edge* next;
+#if STBTT_RASTERIZER_VERSION == 1
+   int x, dx;
    float ey;
    int direction;
-   #elif STBTT_RASTERIZER_VERSION==2
-   float fx,fdx,fdy;
+#elif STBTT_RASTERIZER_VERSION == 2
+   float fx, fdx, fdy;
    float direction;
    float sy;
    float ey;
-   #else
-   #error "Unrecognized value of STBTT_RASTERIZER_VERSION"
-   #endif
+#else
+#error "Unrecognized value of STBTT_RASTERIZER_VERSION"
+#endif
 } stbtt__active_edge;
 
 #if STBTT_RASTERIZER_VERSION == 1
-#define STBTT_FIXSHIFT   10
-#define STBTT_FIX        (1 << STBTT_FIXSHIFT)
-#define STBTT_FIXMASK    (STBTT_FIX-1)
+#define STBTT_FIXSHIFT 10
+#define STBTT_FIX (1 << STBTT_FIXSHIFT)
+#define STBTT_FIXMASK (STBTT_FIX - 1)
 
-static stbtt__active_edge *stbtt__new_active(stbtt__hheap *hh, stbtt__edge *e, int off_x, float start_point, void *userdata)
+static stbtt__active_edge* stbtt__new_active(stbtt__hheap* hh, stbtt__edge* e, int off_x, float start_point, void* userdata)
 {
-   stbtt__active_edge *z = (stbtt__active_edge *) stbtt__hheap_alloc(hh, sizeof(*z), userdata);
+   stbtt__active_edge* z = (stbtt__active_edge*)stbtt__hheap_alloc(hh, sizeof(*z), userdata);
    float dxdy = (e->x1 - e->x0) / (e->y1 - e->y0);
    STBTT_assert(z != NULL);
-   if (!z) return z;
-   
+   if (!z)
+      return z;
+
    // round dx down to avoid overshooting
    if (dxdy < 0)
       z->dx = -STBTT_ifloor(STBTT_FIX * -dxdy);
@@ -1693,15 +3151,16 @@ static stbtt__active_edge *stbtt__new_active(stbtt__hheap *hh, stbtt__edge *e, i
    return z;
 }
 #elif STBTT_RASTERIZER_VERSION == 2
-static stbtt__active_edge *stbtt__new_active(stbtt__hheap *hh, stbtt__edge *e, int off_x, float start_point, void *userdata)
+static stbtt__active_edge* stbtt__new_active(stbtt__hheap* hh, stbtt__edge* e, int off_x, float start_point, void* userdata)
 {
-   stbtt__active_edge *z = (stbtt__active_edge *) stbtt__hheap_alloc(hh, sizeof(*z), userdata);
+   stbtt__active_edge* z = (stbtt__active_edge*)stbtt__hheap_alloc(hh, sizeof(*z), userdata);
    float dxdy = (e->x1 - e->x0) / (e->y1 - e->y0);
    STBTT_assert(z != NULL);
    //STBTT_assert(e->y0 <= start_point);
-   if (!z) return z;
+   if (!z)
+      return z;
    z->fdx = dxdy;
-   z->fdy = dxdy != 0.0f ? (1.0f/dxdy) : 0.0f;
+   z->fdy = dxdy != 0.0f ? (1.0f / dxdy) : 0.0f;
    z->fx = e->x0 + dxdy * (start_point - e->y0);
    z->fx -= off_x;
    z->direction = e->invert ? 1.0f : -1.0f;
@@ -1718,95 +3177,115 @@ static stbtt__active_edge *stbtt__new_active(stbtt__hheap *hh, stbtt__edge *e, i
 // 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)
+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;
+   int x0 = 0, w = 0;
 
-   while (e) {
-      if (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->direction;
-      } else {
-         int x1 = e->x; w += e->direction;
+         x0 = e->x;
+         w += e->direction;
+      }
+      else
+      {
+         int x1 = e->x;
+         w += e->direction;
          // if we went to zero, we need to draw
-         if (w == 0) {
+         if (w == 0)
+         {
             int i = x0 >> STBTT_FIXSHIFT;
             int j = x1 >> STBTT_FIXSHIFT;
 
-            if (i < len && j >= 0) {
-               if (i == j) {
+            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 >> STBTT_FIXSHIFT);
-               } else {
+                  scanline[i] = scanline[i] + (stbtt_uint8)((x1 - x0) * max_weight >> STBTT_FIXSHIFT);
+               }
+               else
+               {
                   if (i >= 0) // add antialiasing for x0
-                     scanline[i] = scanline[i] + (stbtt_uint8) (((STBTT_FIX - (x0 & STBTT_FIXMASK)) * max_weight) >> STBTT_FIXSHIFT);
+                     scanline[i] = scanline[i] + (stbtt_uint8)(((STBTT_FIX - (x0 & STBTT_FIXMASK)) * max_weight) >> STBTT_FIXSHIFT);
                   else
                      i = -1; // clip
 
                   if (j < len) // add antialiasing for x1
-                     scanline[j] = scanline[j] + (stbtt_uint8) (((x1 & STBTT_FIXMASK) * max_weight) >> STBTT_FIXSHIFT);
+                     scanline[j] = scanline[j] + (stbtt_uint8)(((x1 & STBTT_FIXMASK) * max_weight) >> STBTT_FIXSHIFT);
                   else
                      j = len; // clip
 
                   for (++i; i < j; ++i) // fill pixels between x0 and x1
-                     scanline[i] = scanline[i] + (stbtt_uint8) max_weight;
+                     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)
+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__hheap hh = { 0, 0, 0 };
-   stbtt__active_edge *active = NULL;
-   int y,j=0;
-   int max_weight = (255 / vsubsample);  // weight per vertical scanline
+   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);
+      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;
+   e[n].y0 = (off_y + result->h) * (float)vsubsample + 1;
 
-   while (j < result->h) {
+   while (j < result->h)
+   {
       STBTT_memset(scanline, 0, result->w);
-      for (s=0; s < vsubsample; ++s) {
+      for (s = 0; s < vsubsample; ++s)
+      {
          // find center of pixel for this scanline
          float scan_y = y + 0.5f;
-         stbtt__active_edge **step = &active;
+         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) {
+         while (*step)
+         {
+            stbtt__active_edge* z = *step;
+            if (z->ey <= scan_y)
+            {
                *step = z->next; // delete from list
                STBTT_assert(z->direction);
                z->direction = 0;
                stbtt__hheap_free(&hh, z);
-            } else {
+            }
+            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;
+         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;
+            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;
@@ -1815,24 +3294,31 @@ static void stbtt__rasterize_sorted_edges(stbtt__bitmap *result, stbtt__edge *e,
                }
                step = &(*step)->next;
             }
-            if (!changed) break;
+            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 = stbtt__new_active(&hh, e, off_x, scan_y, userdata);
-               if (z != NULL) {
+         while (e->y0 <= scan_y)
+         {
+            if (e->y1 > scan_y)
+            {
+               stbtt__active_edge* z = stbtt__new_active(&hh, e, off_x, scan_y, userdata);
+               if (z != NULL)
+               {
                   // find insertion point
                   if (active == NULL)
                      active = z;
-                  else if (z->x < active->x) {
+                  else if (z->x < active->x)
+                  {
                      // insert at front
                      z->next = active;
                      active = z;
-                  } else {
+                  }
+                  else
+                  {
                      // find thing to insert AFTER
-                     stbtt__active_edge *p = active;
+                     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
@@ -1864,148 +3350,248 @@ static void stbtt__rasterize_sorted_edges(stbtt__bitmap *result, stbtt__edge *e,
 
 // the edge passed in here does not cross the vertical line at x or the vertical line at x+1
 // (i.e. it has already been clipped to those)
-static void stbtt__handle_clipped_edge(float *scanline, int x, stbtt__active_edge *e, float x0, float y0, float x1, float y1)
+static void stbtt__handle_clipped_edge(float* scanline, int x, stbtt__active_edge* e, float x0, float y0, float x1, float y1)
 {
-   if (y0 == y1) return;
+   if (y0 == y1)
+      return;
    STBTT_assert(y0 < y1);
    STBTT_assert(e->sy <= e->ey);
-   if (y0 > e->ey) return;
-   if (y1 < e->sy) return;
-   if (y0 < e->sy) {
-      x0 += (x1-x0) * (e->sy - y0) / (y1-y0);
+   if (y0 > e->ey)
+      return;
+   if (y1 < e->sy)
+      return;
+   if (y0 < e->sy)
+   {
+      x0 += (x1 - x0) * (e->sy - y0) / (y1 - y0);
       y0 = e->sy;
    }
-   if (y1 > e->ey) {
-      x1 += (x1-x0) * (e->ey - y1) / (y1-y0);
+   if (y1 > e->ey)
+   {
+      x1 += (x1 - x0) * (e->ey - y1) / (y1 - y0);
       y1 = e->ey;
    }
 
    if (x0 == x)
-      STBTT_assert(x1 <= x+1);
-   else if (x0 == x+1)
+      STBTT_assert(x1 <= x + 1);
+   else if (x0 == x + 1)
       STBTT_assert(x1 >= x);
    else if (x0 <= x)
       STBTT_assert(x1 <= x);
-   else if (x0 >= x+1)
-      STBTT_assert(x1 >= x+1);
+   else if (x0 >= x + 1)
+      STBTT_assert(x1 >= x + 1);
    else
-      STBTT_assert(x1 >= x && x1 <= x+1);
+      STBTT_assert(x1 >= x && x1 <= x + 1);
 
    if (x0 <= x && x1 <= x)
-      scanline[x] += e->direction * (y1-y0);
-   else if (x0 >= x+1 && x1 >= x+1)
+      scanline[x] += e->direction * (y1 - y0);
+   else if (x0 >= x + 1 && x1 >= x + 1)
       ;
-   else {
-      STBTT_assert(x0 >= x && x0 <= x+1 && x1 >= x && x1 <= x+1);
-      scanline[x] += e->direction * (y1-y0) * (1-((x0-x)+(x1-x))/2); // coverage = 1 - average x position
+   else
+   {
+      STBTT_assert(x0 >= x && x0 <= x + 1 && x1 >= x && x1 <= x + 1);
+      scanline[x] += e->direction * (y1 - y0) * (1 - ((x0 - x) + (x1 - x)) / 2); // coverage = 1 - average x position
    }
 }
 
-static void stbtt__fill_active_edges_new(float *scanline, float *scanline_fill, int len, stbtt__active_edge *e, float y_top)
+static float stbtt__sized_trapezoid_area(float height, float top_width, float bottom_width)
+{
+   STBTT_assert(top_width >= 0);
+   STBTT_assert(bottom_width >= 0);
+   return (top_width + bottom_width) / 2.0f * height;
+}
+
+static float stbtt__position_trapezoid_area(float height, float tx0, float tx1, float bx0, float bx1)
 {
-   float y_bottom = y_top+1;
+   return stbtt__sized_trapezoid_area(height, tx1 - tx0, bx1 - bx0);
+}
 
-   while (e) {
+static float stbtt__sized_triangle_area(float height, float width)
+{
+   return height * width / 2;
+}
+
+static void stbtt__fill_active_edges_new(float* scanline, float* scanline_fill, int len, stbtt__active_edge* e, float y_top)
+{
+   float y_bottom = y_top + 1;
+
+   while (e)
+   {
       // brute force every pixel
 
       // compute intersection points with top & bottom
       STBTT_assert(e->ey >= y_top);
 
-      if (e->fdx == 0) {
+      if (e->fdx == 0)
+      {
          float x0 = e->fx;
-         if (x0 < len) {
-            if (x0 >= 0) {
-               stbtt__handle_clipped_edge(scanline,(int) x0,e, x0,y_top, x0,y_bottom);
-               stbtt__handle_clipped_edge(scanline_fill-1,(int) x0+1,e, x0,y_top, x0,y_bottom);
-            } else {
-               stbtt__handle_clipped_edge(scanline_fill-1,0,e, x0,y_top, x0,y_bottom);
+         if (x0 < len)
+         {
+            if (x0 >= 0)
+            {
+               stbtt__handle_clipped_edge(scanline, (int)x0, e, x0, y_top, x0, y_bottom);
+               stbtt__handle_clipped_edge(scanline_fill - 1, (int)x0 + 1, e, x0, y_top, x0, y_bottom);
+            }
+            else
+            {
+               stbtt__handle_clipped_edge(scanline_fill - 1, 0, e, x0, y_top, x0, y_bottom);
             }
          }
-      } else {
+      }
+      else
+      {
          float x0 = e->fx;
          float dx = e->fdx;
          float xb = x0 + dx;
          float x_top, x_bottom;
-         float sy0,sy1;
+         float sy0, sy1;
          float dy = e->fdy;
          STBTT_assert(e->sy <= y_bottom && e->ey >= y_top);
 
          // compute endpoints of line segment clipped to this scanline (if the
          // line segment starts on this scanline. x0 is the intersection of the
          // line with y_top, but that may be off the line segment.
-         if (e->sy > y_top) {
+         if (e->sy > y_top)
+         {
             x_top = x0 + dx * (e->sy - y_top);
             sy0 = e->sy;
-         } else {
+         }
+         else
+         {
             x_top = x0;
             sy0 = y_top;
          }
-         if (e->ey < y_bottom) {
+         if (e->ey < y_bottom)
+         {
             x_bottom = x0 + dx * (e->ey - y_top);
             sy1 = e->ey;
-         } else {
+         }
+         else
+         {
             x_bottom = xb;
             sy1 = y_bottom;
          }
 
-         if (x_top >= 0 && x_bottom >= 0 && x_top < len && x_bottom < len) {
+         if (x_top >= 0 && x_bottom >= 0 && x_top < len && x_bottom < len)
+         {
             // from here on, we don't have to range check x values
 
-            if ((int) x_top == (int) x_bottom) {
+            if ((int)x_top == (int)x_bottom)
+            {
                float height;
                // simple case, only spans one pixel
-               int x = (int) x_top;
-               height = sy1 - sy0;
+               int x = (int)x_top;
+               height = (sy1 - sy0) * e->direction;
                STBTT_assert(x >= 0 && x < len);
-               scanline[x] += e->direction * (1-((x_top - x) + (x_bottom-x))/2)  * height;
-               scanline_fill[x] += e->direction * height; // everything right of this pixel is filled
-            } else {
-               int x,x1,x2;
-               float y_crossing, step, sign, area;
+               scanline[x] += stbtt__position_trapezoid_area(height, x_top, x + 1.0f, x_bottom, x + 1.0f);
+               scanline_fill[x] += height; // everything right of this pixel is filled
+            }
+            else
+            {
+               int x, x1, x2;
+               float y_crossing, y_final, step, sign, area;
                // covers 2+ pixels
-               if (x_top > x_bottom) {
+               if (x_top > x_bottom)
+               {
                   // flip scanline vertically; signed area is the same
                   float t;
                   sy0 = y_bottom - (sy0 - y_top);
                   sy1 = y_bottom - (sy1 - y_top);
-                  t = sy0; sy0 = sy1; sy1 = t;
-                  t = x_bottom; x_bottom = x_top; x_top = t;
+                  t = sy0, sy0 = sy1, sy1 = t;
+                  t = x_bottom, x_bottom = x_top, x_top = t;
                   dx = -dx;
                   dy = -dy;
-                  t = x0; x0 = xb; xb = t;
+                  t = x0, x0 = xb, xb = t;
                }
+               STBTT_assert(dy >= 0);
+               STBTT_assert(dx >= 0);
 
-               x1 = (int) x_top;
-               x2 = (int) x_bottom;
+               x1 = (int)x_top;
+               x2 = (int)x_bottom;
                // compute intersection with y axis at x1+1
-               y_crossing = (x1+1 - x0) * dy + y_top;
+               y_crossing = y_top + dy * (x1 + 1 - x0);
+
+               // compute intersection with y axis at x2
+               y_final = y_top + dy * (x2 - x0);
+
+               //           x1    x_top                            x2    x_bottom
+               //     y_top  +------|-----+------------+------------+--------|---+------------+
+               //            |            |            |            |            |            |
+               //            |            |            |            |            |            |
+               //       sy0  |      Txxxxx|............|............|............|............|
+               // y_crossing |            *xxxxx.......|............|............|............|
+               //            |            |     xxxxx..|............|............|............|
+               //            |            |     /-   xx*xxxx........|............|............|
+               //            |            | dy <       |    xxxxxx..|............|............|
+               //   y_final  |            |     \-     |          xx*xxx.........|............|
+               //       sy1  |            |            |            |   xxxxxB...|............|
+               //            |            |            |            |            |            |
+               //            |            |            |            |            |            |
+               //  y_bottom  +------------+------------+------------+------------+------------+
+               //
+               // goal is to measure the area covered by '.' in each pixel
+
+               // if x2 is right at the right edge of x1, y_crossing can blow up, github #1057
+               // @TODO: maybe test against sy1 rather than y_bottom?
+               if (y_crossing > y_bottom)
+                  y_crossing = y_bottom;
 
                sign = e->direction;
-               // area of the rectangle covered from y0..y_crossing
-               area = sign * (y_crossing-sy0);
-               // area of the triangle (x_top,y0), (x+1,y0), (x+1,y_crossing)
-               scanline[x1] += area * (1-((x_top - x1)+(x1+1-x1))/2);
-
-               step = sign * dy;
-               for (x = x1+1; x < x2; ++x) {
-                  scanline[x] += area + step/2;
-                  area += step;
+
+               // area of the rectangle covered from sy0..y_crossing
+               area = sign * (y_crossing - sy0);
+
+               // area of the triangle (x_top,sy0), (x1+1,sy0), (x1+1,y_crossing)
+               scanline[x1] += stbtt__sized_triangle_area(area, x1 + 1 - x_top);
+
+               // check if final y_crossing is blown up; no test case for this
+               if (y_final > y_bottom)
+               {
+                  y_final = y_bottom;
+                  dy = (y_final - y_crossing) / (x2 - (x1 + 1)); // if denom=0, y_final = y_crossing, so y_final <= y_bottom
                }
-               y_crossing += dy * (x2 - (x1+1));
 
-               STBTT_assert(fabs(area) <= 1.01f);
+               // in second pixel, area covered by line segment found in first pixel
+               // is always a rectangle 1 wide * the height of that line segment; this
+               // is exactly what the variable 'area' stores. it also gets a contribution
+               // from the line segment within it. the THIRD pixel will get the first
+               // pixel's rectangle contribution, the second pixel's rectangle contribution,
+               // and its own contribution. the 'own contribution' is the same in every pixel except
+               // the leftmost and rightmost, a trapezoid that slides down in each pixel.
+               // the second pixel's contribution to the third pixel will be the
+               // rectangle 1 wide times the height change in the second pixel, which is dy.
+
+               step = sign * dy * 1; // dy is dy/dx, change in y for every 1 change in x,
+               // which multiplied by 1-pixel-width is how much pixel area changes for each step in x
+               // so the area advances by 'step' every time
+
+               for (x = x1 + 1; x < x2; ++x)
+               {
+                  scanline[x] += area + step / 2; // area of trapezoid is 1*step/2
+                  area += step;
+               }
+               STBTT_assert(STBTT_fabs(area) <= 1.01f); // accumulated error from area += step unless we round step down
+               STBTT_assert(sy1 > y_final - 0.01f);
 
-               scanline[x2] += area + sign * (1-((x2-x2)+(x_bottom-x2))/2) * (sy1-y_crossing);
+               // area covered in the last pixel is the rectangle from all the pixels to the left,
+               // plus the trapezoid filled by the line segment in this pixel all the way to the right edge
+               scanline[x2] += area + sign * stbtt__position_trapezoid_area(sy1 - y_final, (float)x2, x2 + 1.0f, x_bottom, x2 + 1.0f);
 
-               scanline_fill[x2] += sign * (sy1-sy0);
+               // the rest of the line is filled based on the total height of the line segment in this pixel
+               scanline_fill[x2] += sign * (sy1 - sy0);
             }
-         } else {
+         }
+         else
+         {
             // if edge goes outside of box we're drawing, we require
             // clipping logic. since this does not match the intended use
             // of this library, we use a different, very slow brute
             // force implementation
+            // note though that this does happen some of the time because
+            // x_top and x_bottom can be extrapolated at the top & bottom of
+            // the shape and actually lie outside the bounding box
             int x;
-            for (x=0; x < len; ++x) {
+            for (x = 0; x < len; ++x)
+            {
                // cases:
                //
                // there can be up to two intersections with the pixel. any intersection
@@ -2019,42 +3605,54 @@ static void stbtt__fill_active_edges_new(float *scanline, float *scanline_fill,
                // from the other y segment, and it might ignored as an empty segment. to avoid
                // that, we need to explicitly produce segments based on x positions.
 
-               // rename variables to clear pairs
+               // rename variables to clearly-defined pairs
                float y0 = y_top;
-               float x1 = (float) (x);
-               float x2 = (float) (x+1);
+               float x1 = (float)(x);
+               float x2 = (float)(x + 1);
                float x3 = xb;
                float y3 = y_bottom;
-               float y1,y2;
 
                // x = e->x + e->dx * (y-y_top)
                // (y-y_top) = (x - e->x) / e->dx
                // y = (x - e->x) / e->dx + y_top
-               y1 = (x - x0) / dx + y_top;
-               y2 = (x+1 - x0) / dx + y_top;
-
-               if (x0 < x1 && x3 > x2) {         // three segments descending down-right
-                  stbtt__handle_clipped_edge(scanline,x,e, x0,y0, x1,y1);
-                  stbtt__handle_clipped_edge(scanline,x,e, x1,y1, x2,y2);
-                  stbtt__handle_clipped_edge(scanline,x,e, x2,y2, x3,y3);
-               } else if (x3 < x1 && x0 > x2) {  // three segments descending down-left
-                  stbtt__handle_clipped_edge(scanline,x,e, x0,y0, x2,y2);
-                  stbtt__handle_clipped_edge(scanline,x,e, x2,y2, x1,y1);
-                  stbtt__handle_clipped_edge(scanline,x,e, x1,y1, x3,y3);
-               } else if (x0 < x1 && x3 > x1) {  // two segments across x, down-right
-                  stbtt__handle_clipped_edge(scanline,x,e, x0,y0, x1,y1);
-                  stbtt__handle_clipped_edge(scanline,x,e, x1,y1, x3,y3);
-               } else if (x3 < x1 && x0 > x1) {  // two segments across x, down-left
-                  stbtt__handle_clipped_edge(scanline,x,e, x0,y0, x1,y1);
-                  stbtt__handle_clipped_edge(scanline,x,e, x1,y1, x3,y3);
-               } else if (x0 < x2 && x3 > x2) {  // two segments across x+1, down-right
-                  stbtt__handle_clipped_edge(scanline,x,e, x0,y0, x2,y2);
-                  stbtt__handle_clipped_edge(scanline,x,e, x2,y2, x3,y3);
-               } else if (x3 < x2 && x0 > x2) {  // two segments across x+1, down-left
-                  stbtt__handle_clipped_edge(scanline,x,e, x0,y0, x2,y2);
-                  stbtt__handle_clipped_edge(scanline,x,e, x2,y2, x3,y3);
-               } else {  // one segment
-                  stbtt__handle_clipped_edge(scanline,x,e, x0,y0, x3,y3);
+               float y1 = (x - x0) / dx + y_top;
+               float y2 = (x + 1 - x0) / dx + y_top;
+
+               if (x0 < x1 && x3 > x2)
+               { // three segments descending down-right
+                  stbtt__handle_clipped_edge(scanline, x, e, x0, y0, x1, y1);
+                  stbtt__handle_clipped_edge(scanline, x, e, x1, y1, x2, y2);
+                  stbtt__handle_clipped_edge(scanline, x, e, x2, y2, x3, y3);
+               }
+               else if (x3 < x1 && x0 > x2)
+               { // three segments descending down-left
+                  stbtt__handle_clipped_edge(scanline, x, e, x0, y0, x2, y2);
+                  stbtt__handle_clipped_edge(scanline, x, e, x2, y2, x1, y1);
+                  stbtt__handle_clipped_edge(scanline, x, e, x1, y1, x3, y3);
+               }
+               else if (x0 < x1 && x3 > x1)
+               { // two segments across x, down-right
+                  stbtt__handle_clipped_edge(scanline, x, e, x0, y0, x1, y1);
+                  stbtt__handle_clipped_edge(scanline, x, e, x1, y1, x3, y3);
+               }
+               else if (x3 < x1 && x0 > x1)
+               { // two segments across x, down-left
+                  stbtt__handle_clipped_edge(scanline, x, e, x0, y0, x1, y1);
+                  stbtt__handle_clipped_edge(scanline, x, e, x1, y1, x3, y3);
+               }
+               else if (x0 < x2 && x3 > x2)
+               { // two segments across x+1, down-right
+                  stbtt__handle_clipped_edge(scanline, x, e, x0, y0, x2, y2);
+                  stbtt__handle_clipped_edge(scanline, x, e, x2, y2, x3, y3);
+               }
+               else if (x3 < x2 && x0 > x2)
+               { // two segments across x+1, down-left
+                  stbtt__handle_clipped_edge(scanline, x, e, x0, y0, x2, y2);
+                  stbtt__handle_clipped_edge(scanline, x, e, x2, y2, x3, y3);
+               }
+               else
+               { // one segment
+                  stbtt__handle_clipped_edge(scanline, x, e, x0, y0, x3, y3);
                }
             }
          }
@@ -2064,52 +3662,70 @@ static void stbtt__fill_active_edges_new(float *scanline, float *scanline_fill,
 }
 
 // directly AA rasterize edges w/o supersampling
-static void stbtt__rasterize_sorted_edges(stbtt__bitmap *result, stbtt__edge *e, int n, int vsubsample, int off_x, int off_y, void *userdata)
+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__hheap hh = { 0, 0, 0 };
-   stbtt__active_edge *active = NULL;
-   int y,j=0, i;
+   stbtt__active_edge* active = NULL;
+   int y, j = 0, i;
    float scanline_data[129], *scanline, *scanline2;
 
+   STBTT__NOTUSED(vsubsample);
+
    if (result->w > 64)
-      scanline = (float *) STBTT_malloc((result->w*2+1) * sizeof(float), userdata);
+      scanline = (float*)STBTT_malloc((result->w * 2 + 1) * sizeof(float), userdata);
    else
       scanline = scanline_data;
 
    scanline2 = scanline + result->w;
 
    y = off_y;
-   e[n].y0 = (float) (off_y + result->h) + 1;
+   e[n].y0 = (float)(off_y + result->h) + 1;
 
-   while (j < result->h) {
+   while (j < result->h)
+   {
       // find center of pixel for this scanline
-      float scan_y_top    = y + 0.0f;
+      float scan_y_top = y + 0.0f;
       float scan_y_bottom = y + 1.0f;
-      stbtt__active_edge **step = &active;
+      stbtt__active_edge** step = &active;
 
-      STBTT_memset(scanline , 0, result->w*sizeof(scanline[0]));
-      STBTT_memset(scanline2, 0, (result->w+1)*sizeof(scanline[0]));
+      STBTT_memset(scanline, 0, result->w * sizeof(scanline[0]));
+      STBTT_memset(scanline2, 0, (result->w + 1) * sizeof(scanline[0]));
 
       // update all active edges;
       // remove all active edges that terminate before the top of this scanline
-      while (*step) {
-         stbtt__active_edge * z = *step;
-         if (z->ey <= scan_y_top) {
+      while (*step)
+      {
+         stbtt__active_edge* z = *step;
+         if (z->ey <= scan_y_top)
+         {
             *step = z->next; // delete from list
             STBTT_assert(z->direction);
             z->direction = 0;
             stbtt__hheap_free(&hh, z);
-         } else {
+         }
+         else
+         {
             step = &((*step)->next); // advance through list
          }
       }
 
       // insert all edges that start before the bottom of this scanline
-      while (e->y0 <= scan_y_bottom) {
-         if (e->y0 != e->y1) {
-            stbtt__active_edge *z = stbtt__new_active(&hh, e, off_x, scan_y_top, userdata);
-            if (z != NULL) {
-               STBTT_assert(z->ey >= scan_y_top);
+      while (e->y0 <= scan_y_bottom)
+      {
+         if (e->y0 != e->y1)
+         {
+            stbtt__active_edge* z = stbtt__new_active(&hh, e, off_x, scan_y_top, userdata);
+            if (z != NULL)
+            {
+               if (j == 0 && off_y != 0)
+               {
+                  if (z->ey < scan_y_top)
+                  {
+                     // this can happen due to subpixel positioning and some kind of fp rounding error i think
+                     z->ey = scan_y_top;
+                  }
+               }
+               STBTT_assert(z->ey >= scan_y_top); // if we get really unlucky a tiny bit of an edge can be out of bounds
                // insert at front
                z->next = active;
                active = z;
@@ -2120,25 +3736,28 @@ static void stbtt__rasterize_sorted_edges(stbtt__bitmap *result, stbtt__edge *e,
 
       // now process all active edges
       if (active)
-         stbtt__fill_active_edges_new(scanline, scanline2+1, result->w, active, scan_y_top);
+         stbtt__fill_active_edges_new(scanline, scanline2 + 1, result->w, active, scan_y_top);
 
       {
          float sum = 0;
-         for (i=0; i < result->w; ++i) {
+         for (i = 0; i < result->w; ++i)
+         {
             float k;
             int m;
             sum += scanline2[i];
             k = scanline[i] + sum;
-            k = (float) fabs(k)*255 + 0.5f;
-            m = (int) k;
-            if (m > 255) m = 255;
-            result->pixels[j*result->stride + i] = (unsigned char) m;
+            k = (float)STBTT_fabs(k) * 255 + 0.5f;
+            m = (int)k;
+            if (m > 255)
+               m = 255;
+            result->pixels[j * result->stride + i] = (unsigned char)m;
          }
       }
       // advance all the edges
       step = &active;
-      while (*step) {
-         stbtt__active_edge *z = *step;
+      while (*step)
+      {
+         stbtt__active_edge* z = *step;
          z->fx += z->fdx; // advance to position for current scanline
          step = &((*step)->next); // advance through list
       }
@@ -2156,19 +3775,22 @@ static void stbtt__rasterize_sorted_edges(stbtt__bitmap *result, stbtt__edge *e,
 #error "Unrecognized value of STBTT_RASTERIZER_VERSION"
 #endif
 
-#define STBTT__COMPARE(a,b)  ((a)->y0 < (b)->y0)
+#define STBTT__COMPARE(a, b) ((a)->y0 < (b)->y0)
 
-static void stbtt__sort_edges_ins_sort(stbtt__edge *p, int n)
+static void stbtt__sort_edges_ins_sort(stbtt__edge* p, int n)
 {
-   int i,j;
-   for (i=1; i < n; ++i) {
+   int i, j;
+   for (i = 1; i < n; ++i)
+   {
       stbtt__edge t = p[i], *a = &t;
       j = i;
-      while (j > 0) {
-         stbtt__edge *b = &p[j-1];
-         int c = STBTT__COMPARE(a,b);
-         if (!c) break;
-         p[j] = p[j-1];
+      while (j > 0)
+      {
+         stbtt__edge* b = &p[j - 1];
+         int c = STBTT__COMPARE(a, b);
+         if (!c)
+            break;
+         p[j] = p[j - 1];
          --j;
       }
       if (i != j)
@@ -2176,25 +3798,27 @@ static void stbtt__sort_edges_ins_sort(stbtt__edge *p, int n)
    }
 }
 
-static void stbtt__sort_edges_quicksort(stbtt__edge *p, int n)
+static void stbtt__sort_edges_quicksort(stbtt__edge* p, int n)
 {
-   /* threshhold for transitioning to insertion sort */
-   while (n > 12) {
+   /* threshold for transitioning to insertion sort */
+   while (n > 12)
+   {
       stbtt__edge t;
-      int c01,c12,c,m,i,j;
+      int c01, c12, c, m, i, j;
 
       /* compute median of three */
       m = n >> 1;
-      c01 = STBTT__COMPARE(&p[0],&p[m]);
-      c12 = STBTT__COMPARE(&p[m],&p[n-1]);
+      c01 = STBTT__COMPARE(&p[0], &p[m]);
+      c12 = STBTT__COMPARE(&p[m], &p[n - 1]);
       /* if 0 >= mid >= end, or 0 < mid < end, then use mid */
-      if (c01 != c12) {
+      if (c01 != c12)
+      {
          /* otherwise, we'll need to swap something else to middle */
          int z;
-         c = STBTT__COMPARE(&p[0],&p[n-1]);
+         c = STBTT__COMPARE(&p[0], &p[n - 1]);
          /* 0>mid && mid<n:  0>n => n; 0<n => 0 */
          /* 0<mid && mid>n:  0>n => 0; 0<n => n */
-         z = (c == c12) ? 0 : n-1;
+         z = (c == c12) ? 0 : n - 1;
          t = p[z];
          p[z] = p[m];
          p[m] = t;
@@ -2206,19 +3830,25 @@ static void stbtt__sort_edges_quicksort(stbtt__edge *p, int n)
       p[m] = t;
 
       /* partition loop */
-      i=1;
-      j=n-1;
-      for(;;) {
+      i = 1;
+      j = n - 1;
+      for (;;)
+      {
          /* handling of equality is crucial here */
          /* for sentinels & efficiency with duplicates */
-         for (;;++i) {
-            if (!STBTT__COMPARE(&p[i], &p[0])) break;
+         for (;; ++i)
+         {
+            if (!STBTT__COMPARE(&p[i], &p[0]))
+               break;
          }
-         for (;;--j) {
-            if (!STBTT__COMPARE(&p[0], &p[j])) break;
+         for (;; --j)
+         {
+            if (!STBTT__COMPARE(&p[0], &p[j]))
+               break;
          }
          /* make sure we haven't crossed */
-         if (i >= j) break;
+         if (i >= j)
+            break;
          t = p[i];
          p[i] = p[j];
          p[j] = t;
@@ -2227,18 +3857,21 @@ static void stbtt__sort_edges_quicksort(stbtt__edge *p, int n)
          --j;
       }
       /* recurse on smaller side, iterate on larger */
-      if (j < (n-i)) {
-         stbtt__sort_edges_quicksort(p,j);
-         p = p+i;
-         n = n-i;
-      } else {
-         stbtt__sort_edges_quicksort(p+i, n-i);
+      if (j < (n - i))
+      {
+         stbtt__sort_edges_quicksort(p, j);
+         p = p + i;
+         n = n - i;
+      }
+      else
+      {
+         stbtt__sort_edges_quicksort(p + i, n - i);
          n = j;
       }
    }
 }
 
-static void stbtt__sort_edges(stbtt__edge *p, int n)
+static void stbtt__sort_edges(stbtt__edge* p, int n)
 {
    stbtt__sort_edges_quicksort(p, n);
    stbtt__sort_edges_ins_sort(p, n);
@@ -2246,47 +3879,51 @@ static void stbtt__sort_edges(stbtt__edge *p, int n)
 
 typedef struct
 {
-   float x,y;
+   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)
+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;
+   stbtt__edge* e;
+   int n, i, j, k, m;
 #if STBTT_RASTERIZER_VERSION == 1
    int vsubsample = result->h < 8 ? 15 : 5;
 #elif STBTT_RASTERIZER_VERSION == 2
    int vsubsample = 1;
 #else
-   #error "Unrecognized value of STBTT_RASTERIZER_VERSION"
+#error "Unrecognized value of STBTT_RASTERIZER_VERSION"
 #endif
    // 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)
+   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;
+   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 = 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;
+      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) {
+         if (invert ? p[j].y > p[k].y : p[j].y < p[k].y)
+         {
             e[n].invert = 1;
-            a=j;b=k;
+            a = j, b = k;
          }
          e[n].x0 = p[a].x * scale_x + shift_x;
          e[n].y0 = (p[a].y * y_scale_inv + shift_y) * vsubsample;
@@ -2306,69 +3943,125 @@ static void stbtt__rasterize(stbtt__bitmap *result, stbtt__point *pts, int *wcou
    STBTT_free(e, userdata);
 }
 
-static void stbtt__add_point(stbtt__point *points, int n, float x, float y)
+static void stbtt__add_point(stbtt__point* points, int n, float x, float y)
 {
-   if (!points) return; // during first pass, it's unallocated
+   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)
+// tessellate until threshold 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;
+   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;
+   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;
+   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;
 }
 
+static void stbtt__tesselate_cubic(stbtt__point* points, int* num_points, float x0, float y0, float x1, float y1, float x2, float y2, float x3, float y3, float objspace_flatness_squared, int n)
+{
+   // @TODO this "flatness" calculation is just made-up nonsense that seems to work well enough
+   float dx0 = x1 - x0;
+   float dy0 = y1 - y0;
+   float dx1 = x2 - x1;
+   float dy1 = y2 - y1;
+   float dx2 = x3 - x2;
+   float dy2 = y3 - y2;
+   float dx = x3 - x0;
+   float dy = y3 - y0;
+   float longlen = (float)(STBTT_sqrt(dx0 * dx0 + dy0 * dy0) + STBTT_sqrt(dx1 * dx1 + dy1 * dy1) + STBTT_sqrt(dx2 * dx2 + dy2 * dy2));
+   float shortlen = (float)STBTT_sqrt(dx * dx + dy * dy);
+   float flatness_squared = longlen * longlen - shortlen * shortlen;
+
+   if (n > 16) // 65536 segments on one curve better be enough!
+      return;
+
+   if (flatness_squared > objspace_flatness_squared)
+   {
+      float x01 = (x0 + x1) / 2;
+      float y01 = (y0 + y1) / 2;
+      float x12 = (x1 + x2) / 2;
+      float y12 = (y1 + y2) / 2;
+      float x23 = (x2 + x3) / 2;
+      float y23 = (y2 + y3) / 2;
+
+      float xa = (x01 + x12) / 2;
+      float ya = (y01 + y12) / 2;
+      float xb = (x12 + x23) / 2;
+      float yb = (y12 + y23) / 2;
+
+      float mx = (xa + xb) / 2;
+      float my = (ya + yb) / 2;
+
+      stbtt__tesselate_cubic(points, num_points, x0, y0, x01, y01, xa, ya, mx, my, objspace_flatness_squared, n + 1);
+      stbtt__tesselate_cubic(points, num_points, mx, my, xb, yb, x23, y23, x3, y3, objspace_flatness_squared, n + 1);
+   }
+   else
+   {
+      stbtt__add_point(points, *num_points, x3, y3);
+      *num_points = *num_points + 1;
+   }
+}
+
 // returns number of contours
-static stbtt__point *stbtt_FlattenCurves(stbtt_vertex *vertices, int num_verts, float objspace_flatness, int **contour_lengths, int *num_contours, void *userdata)
+static 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;
+   stbtt__point* points = 0;
+   int num_points = 0;
 
    float objspace_flatness_squared = objspace_flatness * objspace_flatness;
-   int i,n=0,start=0, pass;
+   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)
+   for (i = 0; i < num_verts; ++i)
       if (vertices[i].type == STBTT_vmove)
          ++n;
 
    *num_contours = n;
-   if (n == 0) return 0;
+   if (n == 0)
+      return 0;
 
-   *contour_lengths = (int *) STBTT_malloc(sizeof(**contour_lengths) * n, userdata);
+   *contour_lengths = (int*)STBTT_malloc(sizeof(**contour_lengths) * n, userdata);
 
-   if (*contour_lengths == 0) {
+   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;
+   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) {
+      n = -1;
+      for (i = 0; i < num_verts; ++i)
+      {
+         switch (vertices[i].type)
+         {
             case STBTT_vmove:
                // start the next contour
                if (n >= 0)
@@ -2376,19 +4069,27 @@ static stbtt__point *stbtt_FlattenCurves(stbtt_vertex *vertices, int num_verts,
                ++n;
                start = num_points;
 
-               x = vertices[i].x; y = vertices[i].y;
-               stbtt__add_point(points, num_points++, x,y);
+               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;
+               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;
+               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;
+            case STBTT_vcubic:
+               stbtt__tesselate_cubic(points, &num_points, x, y,
+                                      vertices[i].cx, vertices[i].cy,
+                                      vertices[i].cx1, vertices[i].cy1,
+                                      vertices[i].x, vertices[i].y,
+                                      objspace_flatness_squared, 0);
+               x = vertices[i].x, y = vertices[i].y;
                break;
          }
       }
@@ -2404,51 +4105,65 @@ static stbtt__point *stbtt_FlattenCurves(stbtt_vertex *vertices, int num_verts,
    return NULL;
 }
 
-STBTT_DEF 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)
+STBTT_DEF 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) {
+   int winding_count = 0;
+   int* winding_lengths = NULL;
+   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);
    }
 }
 
-STBTT_DEF void stbtt_FreeBitmap(unsigned char *bitmap, void *userdata)
+STBTT_DEF void stbtt_FreeBitmap(unsigned char* bitmap, void* userdata)
 {
    STBTT_free(bitmap, userdata);
 }
 
-STBTT_DEF 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)
+STBTT_DEF 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;
+   int ix0, iy0, ix1, iy1;
    stbtt__bitmap gbm;
-   stbtt_vertex *vertices;   
+   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;
+   if (scale_x == 0)
+      scale_x = scale_y;
+   if (scale_y == 0)
+   {
+      if (scale_x == 0)
+      {
+         STBTT_free(vertices, info->userdata);
+         return NULL;
+      }
       scale_y = scale_x;
    }
 
-   stbtt_GetGlyphBitmapBoxSubpixel(info, glyph, scale_x, scale_y, shift_x, shift_y, &ix0,&iy0,&ix1,&iy1);
+   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) {
+   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);
@@ -2456,55 +4171,60 @@ STBTT_DEF unsigned char *stbtt_GetGlyphBitmapSubpixel(const stbtt_fontinfo *info
    }
    STBTT_free(vertices, info->userdata);
    return gbm.pixels;
-}   
+}
 
-STBTT_DEF unsigned char *stbtt_GetGlyphBitmap(const stbtt_fontinfo *info, float scale_x, float scale_y, int glyph, int *width, int *height, int *xoff, int *yoff)
+STBTT_DEF 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);
 }
 
-STBTT_DEF 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)
+STBTT_DEF 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 ix0, iy0;
+   stbtt_vertex* vertices;
    int num_verts = stbtt_GetGlyphShape(info, glyph, &vertices);
-   stbtt__bitmap gbm;   
+   stbtt__bitmap gbm;
 
-   stbtt_GetGlyphBitmapBoxSubpixel(info, glyph, scale_x, scale_y, shift_x, shift_y, &ix0,&iy0,0,0);
+   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_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);
 }
 
-STBTT_DEF 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_DEF 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);
+}
+
+STBTT_DEF 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)
 {
-   stbtt_MakeGlyphBitmapSubpixel(info, output, out_w, out_h, out_stride, scale_x, scale_y, 0.0f,0.0f, glyph);
+   return stbtt_GetGlyphBitmapSubpixel(info, scale_x, scale_y, shift_x, shift_y, stbtt_FindGlyphIndex(info, codepoint), width, height, xoff, yoff);
 }
 
-STBTT_DEF 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)
+STBTT_DEF void stbtt_MakeCodepointBitmapSubpixelPrefilter(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 oversample_x, int oversample_y, float* sub_x, float* sub_y, int codepoint)
 {
-   return stbtt_GetGlyphBitmapSubpixel(info, scale_x, scale_y,shift_x,shift_y, stbtt_FindGlyphIndex(info,codepoint), width,height,xoff,yoff);
-}   
+   stbtt_MakeGlyphBitmapSubpixelPrefilter(info, output, out_w, out_h, out_stride, scale_x, scale_y, shift_x, shift_y, oversample_x, oversample_y, sub_x, sub_y, stbtt_FindGlyphIndex(info, codepoint));
+}
 
-STBTT_DEF 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_DEF 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));
+   stbtt_MakeGlyphBitmapSubpixel(info, output, out_w, out_h, out_stride, scale_x, scale_y, shift_x, shift_y, stbtt_FindGlyphIndex(info, codepoint));
 }
 
-STBTT_DEF unsigned char *stbtt_GetCodepointBitmap(const stbtt_fontinfo *info, float scale_x, float scale_y, int codepoint, int *width, int *height, int *xoff, int *yoff)
+STBTT_DEF 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);
-}   
+   return stbtt_GetCodepointBitmapSubpixel(info, scale_x, scale_y, 0.0f, 0.0f, codepoint, width, height, xoff, yoff);
+}
 
-STBTT_DEF 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_DEF 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);
+   stbtt_MakeCodepointBitmapSubpixel(info, output, out_w, out_h, out_stride, scale_x, scale_y, 0.0f, 0.0f, codepoint);
 }
 
 //////////////////////////////////////////////////////////////////////////////
@@ -2513,57 +4233,58 @@ STBTT_DEF void stbtt_MakeCodepointBitmap(const stbtt_fontinfo *info, unsigned ch
 //
 // This is SUPER-CRAPPY packing to keep source code small
 
-STBTT_DEF 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)
+static int stbtt_BakeFontBitmap_internal(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;
+   int x, y, bottom_y, i;
    stbtt_fontinfo f;
    f.userdata = NULL;
    if (!stbtt_InitFont(&f, data, offset))
       return -1;
-   STBTT_memset(pixels, 0, pw*ph); // background of 0 around pixels
-   x=y=1;
+   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;
+   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;
+      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
+         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);
+      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;
+      chardata[i].xoff = (float)x0;
+      chardata[i].yoff = (float)y0;
       x = x + gw + 1;
-      if (y+gh+1 > bottom_y)
-         bottom_y = y+gh+1;
+      if (y + gh + 1 > bottom_y)
+         bottom_y = y + gh + 1;
    }
    return bottom_y;
 }
 
-STBTT_DEF void stbtt_GetBakedQuad(stbtt_bakedchar *chardata, int pw, int ph, int char_index, float *xpos, float *ypos, stbtt_aligned_quad *q, int opengl_fillrule)
+STBTT_DEF void stbtt_GetBakedQuad(const stbtt_bakedchar* chardata, int pw, int ph, int char_index, float* xpos, float* ypos, stbtt_aligned_quad* q, int opengl_fillrule)
 {
    float d3d_bias = opengl_fillrule ? 0 : -0.5f;
    float ipw = 1.0f / pw, iph = 1.0f / ph;
-   stbtt_bakedchar *b = chardata + char_index;
+   const stbtt_bakedchar* b = chardata + char_index;
    int round_x = STBTT_ifloor((*xpos + b->xoff) + 0.5f);
    int round_y = STBTT_ifloor((*ypos + b->yoff) + 0.5f);
 
@@ -2586,11 +4307,6 @@ STBTT_DEF void stbtt_GetBakedQuad(stbtt_bakedchar *chardata, int pw, int ph, int
 //
 
 #ifndef STB_RECT_PACK_VERSION
-#ifdef _MSC_VER
-#define STBTT__NOTUSED(v)  (void)(v)
-#else
-#define STBTT__NOTUSED(v)  (void)sizeof(v)
-#endif
 
 typedef int stbrp_coord;
 
@@ -2607,8 +4323,8 @@ typedef int stbrp_coord;
 
 typedef struct
 {
-   int width,height;
-   int x,y,bottom_y;
+   int width, height;
+   int x, y, bottom_y;
 } stbrp_context;
 
 typedef struct
@@ -2618,26 +4334,28 @@ typedef struct
 
 struct stbrp_rect
 {
-   stbrp_coord x,y;
-   int id,w,h,was_packed;
+   stbrp_coord x, y;
+   int id, w, h, was_packed;
 };
 
-static void stbrp_init_target(stbrp_context *con, int pw, int ph, stbrp_node *nodes, int num_nodes)
+static void stbrp_init_target(stbrp_context* con, int pw, int ph, stbrp_node* nodes, int num_nodes)
 {
-   con->width  = pw;
+   con->width = pw;
    con->height = ph;
    con->x = 0;
    con->y = 0;
    con->bottom_y = 0;
    STBTT__NOTUSED(nodes);
-   STBTT__NOTUSED(num_nodes);   
+   STBTT__NOTUSED(num_nodes);
 }
 
-static void stbrp_pack_rects(stbrp_context *con, stbrp_rect *rects, int num_rects)
+static void stbrp_pack_rects(stbrp_context* con, stbrp_rect* rects, int num_rects)
 {
    int i;
-   for (i=0; i < num_rects; ++i) {
-      if (con->x + rects[i].w > con->width) {
+   for (i = 0; i < num_rects; ++i)
+   {
+      if (con->x + rects[i].w > con->width)
+      {
          con->x = 0;
          con->y = con->bottom_y;
       }
@@ -2650,7 +4368,7 @@ static void stbrp_pack_rects(stbrp_context *con, stbrp_rect *rects, int num_rect
       if (con->y + rects[i].h > con->bottom_y)
          con->bottom_y = con->y + rects[i].h;
    }
-   for (   ; i < num_rects; ++i)
+   for (; i < num_rects; ++i)
       rects[i].was_packed = 0;
 }
 #endif
@@ -2662,15 +4380,18 @@ static void stbrp_pack_rects(stbrp_context *con, stbrp_rect *rects, int num_rect
 // This is SUPER-AWESOME (tm Ryan Gordon) packing using stb_rect_pack.h. If
 // stb_rect_pack.h isn't available, it uses the BakeFontBitmap strategy.
 
-STBTT_DEF int stbtt_PackBegin(stbtt_pack_context *spc, unsigned char *pixels, int pw, int ph, int stride_in_bytes, int padding, void *alloc_context)
+STBTT_DEF int stbtt_PackBegin(stbtt_pack_context* spc, unsigned char* pixels, int pw, int ph, int stride_in_bytes, int padding, void* alloc_context)
 {
-   stbrp_context *context = (stbrp_context *) STBTT_malloc(sizeof(*context)            ,alloc_context);
-   int            num_nodes = pw - padding;
-   stbrp_node    *nodes   = (stbrp_node    *) STBTT_malloc(sizeof(*nodes  ) * num_nodes,alloc_context);
+   stbrp_context* context = (stbrp_context*)STBTT_malloc(sizeof(*context), alloc_context);
+   int num_nodes = pw - padding;
+   stbrp_node* nodes = (stbrp_node*)STBTT_malloc(sizeof(*nodes) * num_nodes, alloc_context);
 
-   if (context == NULL || nodes == NULL) {
-      if (context != NULL) STBTT_free(context, alloc_context);
-      if (nodes   != NULL) STBTT_free(nodes  , alloc_context);
+   if (context == NULL || nodes == NULL)
+   {
+      if (context != NULL)
+         STBTT_free(context, alloc_context);
+      if (nodes != NULL)
+         STBTT_free(nodes, alloc_context);
       return 0;
    }
 
@@ -2684,22 +4405,23 @@ STBTT_DEF int stbtt_PackBegin(stbtt_pack_context *spc, unsigned char *pixels, in
    spc->stride_in_bytes = stride_in_bytes != 0 ? stride_in_bytes : pw;
    spc->h_oversample = 1;
    spc->v_oversample = 1;
+   spc->skip_missing = 0;
 
-   stbrp_init_target(context, pw-padding, ph-padding, nodes, num_nodes);
+   stbrp_init_target(context, pw - padding, ph - padding, nodes, num_nodes);
 
    if (pixels)
-      STBTT_memset(pixels, 0, pw*ph); // background of 0 around pixels
+      STBTT_memset(pixels, 0, pw * ph); // background of 0 around pixels
 
    return 1;
 }
 
-STBTT_DEF void stbtt_PackEnd  (stbtt_pack_context *spc)
+STBTT_DEF void stbtt_PackEnd(stbtt_pack_context* spc)
 {
-   STBTT_free(spc->nodes    , spc->user_allocator_context);
+   STBTT_free(spc->nodes, spc->user_allocator_context);
    STBTT_free(spc->pack_info, spc->user_allocator_context);
 }
 
-STBTT_DEF void stbtt_PackSetOversampling(stbtt_pack_context *spc, unsigned int h_oversample, unsigned int v_oversample)
+STBTT_DEF void stbtt_PackSetOversampling(stbtt_pack_context* spc, unsigned int h_oversample, unsigned int v_oversample)
 {
    STBTT_assert(h_oversample <= STBTT_MAX_OVERSAMPLE);
    STBTT_assert(v_oversample <= STBTT_MAX_OVERSAMPLE);
@@ -2709,15 +4431,21 @@ STBTT_DEF void stbtt_PackSetOversampling(stbtt_pack_context *spc, unsigned int h
       spc->v_oversample = v_oversample;
 }
 
-#define STBTT__OVER_MASK  (STBTT_MAX_OVERSAMPLE-1)
+STBTT_DEF void stbtt_PackSetSkipMissingCodepoints(stbtt_pack_context* spc, int skip)
+{
+   spc->skip_missing = skip;
+}
+
+#define STBTT__OVER_MASK (STBTT_MAX_OVERSAMPLE - 1)
 
-static void stbtt__h_prefilter(unsigned char *pixels, int w, int h, int stride_in_bytes, unsigned int kernel_width)
+static void stbtt__h_prefilter(unsigned char* pixels, int w, int h, int stride_in_bytes, unsigned int kernel_width)
 {
    unsigned char buffer[STBTT_MAX_OVERSAMPLE];
    int safe_w = w - kernel_width;
    int j;
    STBTT_memset(buffer, 0, STBTT_MAX_OVERSAMPLE); // suppress bogus warning from VS2013 -analyze
-   for (j=0; j < h; ++j) {
+   for (j = 0; j < h; ++j)
+   {
       int i;
       unsigned int total;
       STBTT_memset(buffer, 0, kernel_width);
@@ -2725,61 +4453,69 @@ static void stbtt__h_prefilter(unsigned char *pixels, int w, int h, int stride_i
       total = 0;
 
       // make kernel_width a constant in common cases so compiler can optimize out the divide
-      switch (kernel_width) {
+      switch (kernel_width)
+      {
          case 2:
-            for (i=0; i <= safe_w; ++i) {
+            for (i = 0; i <= safe_w; ++i)
+            {
                total += pixels[i] - buffer[i & STBTT__OVER_MASK];
-               buffer[(i+kernel_width) & STBTT__OVER_MASK] = pixels[i];
-               pixels[i] = (unsigned char) (total / 2);
+               buffer[(i + kernel_width) & STBTT__OVER_MASK] = pixels[i];
+               pixels[i] = (unsigned char)(total / 2);
             }
             break;
          case 3:
-            for (i=0; i <= safe_w; ++i) {
+            for (i = 0; i <= safe_w; ++i)
+            {
                total += pixels[i] - buffer[i & STBTT__OVER_MASK];
-               buffer[(i+kernel_width) & STBTT__OVER_MASK] = pixels[i];
-               pixels[i] = (unsigned char) (total / 3);
+               buffer[(i + kernel_width) & STBTT__OVER_MASK] = pixels[i];
+               pixels[i] = (unsigned char)(total / 3);
             }
             break;
          case 4:
-            for (i=0; i <= safe_w; ++i) {
+            for (i = 0; i <= safe_w; ++i)
+            {
                total += pixels[i] - buffer[i & STBTT__OVER_MASK];
-               buffer[(i+kernel_width) & STBTT__OVER_MASK] = pixels[i];
-               pixels[i] = (unsigned char) (total / 4);
+               buffer[(i + kernel_width) & STBTT__OVER_MASK] = pixels[i];
+               pixels[i] = (unsigned char)(total / 4);
             }
             break;
          case 5:
-            for (i=0; i <= safe_w; ++i) {
+            for (i = 0; i <= safe_w; ++i)
+            {
                total += pixels[i] - buffer[i & STBTT__OVER_MASK];
-               buffer[(i+kernel_width) & STBTT__OVER_MASK] = pixels[i];
-               pixels[i] = (unsigned char) (total / 5);
+               buffer[(i + kernel_width) & STBTT__OVER_MASK] = pixels[i];
+               pixels[i] = (unsigned char)(total / 5);
             }
             break;
          default:
-            for (i=0; i <= safe_w; ++i) {
+            for (i = 0; i <= safe_w; ++i)
+            {
                total += pixels[i] - buffer[i & STBTT__OVER_MASK];
-               buffer[(i+kernel_width) & STBTT__OVER_MASK] = pixels[i];
-               pixels[i] = (unsigned char) (total / kernel_width);
+               buffer[(i + kernel_width) & STBTT__OVER_MASK] = pixels[i];
+               pixels[i] = (unsigned char)(total / kernel_width);
             }
             break;
       }
 
-      for (; i < w; ++i) {
+      for (; i < w; ++i)
+      {
          STBTT_assert(pixels[i] == 0);
          total -= buffer[i & STBTT__OVER_MASK];
-         pixels[i] = (unsigned char) (total / kernel_width);
+         pixels[i] = (unsigned char)(total / kernel_width);
       }
 
       pixels += stride_in_bytes;
    }
 }
 
-static void stbtt__v_prefilter(unsigned char *pixels, int w, int h, int stride_in_bytes, unsigned int kernel_width)
+static void stbtt__v_prefilter(unsigned char* pixels, int w, int h, int stride_in_bytes, unsigned int kernel_width)
 {
    unsigned char buffer[STBTT_MAX_OVERSAMPLE];
    int safe_h = h - kernel_width;
    int j;
    STBTT_memset(buffer, 0, STBTT_MAX_OVERSAMPLE); // suppress bogus warning from VS2013 -analyze
-   for (j=0; j < w; ++j) {
+   for (j = 0; j < w; ++j)
+   {
       int i;
       unsigned int total;
       STBTT_memset(buffer, 0, kernel_width);
@@ -2787,48 +4523,55 @@ static void stbtt__v_prefilter(unsigned char *pixels, int w, int h, int stride_i
       total = 0;
 
       // make kernel_width a constant in common cases so compiler can optimize out the divide
-      switch (kernel_width) {
+      switch (kernel_width)
+      {
          case 2:
-            for (i=0; i <= safe_h; ++i) {
-               total += pixels[i*stride_in_bytes] - buffer[i & STBTT__OVER_MASK];
-               buffer[(i+kernel_width) & STBTT__OVER_MASK] = pixels[i*stride_in_bytes];
-               pixels[i*stride_in_bytes] = (unsigned char) (total / 2);
+            for (i = 0; i <= safe_h; ++i)
+            {
+               total += pixels[i * stride_in_bytes] - buffer[i & STBTT__OVER_MASK];
+               buffer[(i + kernel_width) & STBTT__OVER_MASK] = pixels[i * stride_in_bytes];
+               pixels[i * stride_in_bytes] = (unsigned char)(total / 2);
             }
             break;
          case 3:
-            for (i=0; i <= safe_h; ++i) {
-               total += pixels[i*stride_in_bytes] - buffer[i & STBTT__OVER_MASK];
-               buffer[(i+kernel_width) & STBTT__OVER_MASK] = pixels[i*stride_in_bytes];
-               pixels[i*stride_in_bytes] = (unsigned char) (total / 3);
+            for (i = 0; i <= safe_h; ++i)
+            {
+               total += pixels[i * stride_in_bytes] - buffer[i & STBTT__OVER_MASK];
+               buffer[(i + kernel_width) & STBTT__OVER_MASK] = pixels[i * stride_in_bytes];
+               pixels[i * stride_in_bytes] = (unsigned char)(total / 3);
             }
             break;
          case 4:
-            for (i=0; i <= safe_h; ++i) {
-               total += pixels[i*stride_in_bytes] - buffer[i & STBTT__OVER_MASK];
-               buffer[(i+kernel_width) & STBTT__OVER_MASK] = pixels[i*stride_in_bytes];
-               pixels[i*stride_in_bytes] = (unsigned char) (total / 4);
+            for (i = 0; i <= safe_h; ++i)
+            {
+               total += pixels[i * stride_in_bytes] - buffer[i & STBTT__OVER_MASK];
+               buffer[(i + kernel_width) & STBTT__OVER_MASK] = pixels[i * stride_in_bytes];
+               pixels[i * stride_in_bytes] = (unsigned char)(total / 4);
             }
             break;
          case 5:
-            for (i=0; i <= safe_h; ++i) {
-               total += pixels[i*stride_in_bytes] - buffer[i & STBTT__OVER_MASK];
-               buffer[(i+kernel_width) & STBTT__OVER_MASK] = pixels[i*stride_in_bytes];
-               pixels[i*stride_in_bytes] = (unsigned char) (total / 5);
+            for (i = 0; i <= safe_h; ++i)
+            {
+               total += pixels[i * stride_in_bytes] - buffer[i & STBTT__OVER_MASK];
+               buffer[(i + kernel_width) & STBTT__OVER_MASK] = pixels[i * stride_in_bytes];
+               pixels[i * stride_in_bytes] = (unsigned char)(total / 5);
             }
             break;
          default:
-            for (i=0; i <= safe_h; ++i) {
-               total += pixels[i*stride_in_bytes] - buffer[i & STBTT__OVER_MASK];
-               buffer[(i+kernel_width) & STBTT__OVER_MASK] = pixels[i*stride_in_bytes];
-               pixels[i*stride_in_bytes] = (unsigned char) (total / kernel_width);
+            for (i = 0; i <= safe_h; ++i)
+            {
+               total += pixels[i * stride_in_bytes] - buffer[i & STBTT__OVER_MASK];
+               buffer[(i + kernel_width) & STBTT__OVER_MASK] = pixels[i * stride_in_bytes];
+               pixels[i * stride_in_bytes] = (unsigned char)(total / kernel_width);
             }
             break;
       }
 
-      for (; i < h; ++i) {
-         STBTT_assert(pixels[i*stride_in_bytes] == 0);
+      for (; i < h; ++i)
+      {
+         STBTT_assert(pixels[i * stride_in_bytes] == 0);
          total -= buffer[i & STBTT__OVER_MASK];
-         pixels[i*stride_in_bytes] = (unsigned char) (total / kernel_width);
+         pixels[i * stride_in_bytes] = (unsigned char)(total / kernel_width);
       }
 
       pixels += 1;
@@ -2848,27 +4591,39 @@ static float stbtt__oversample_shift(int oversample)
 }
 
 // rects array must be big enough to accommodate all characters in the given ranges
-STBTT_DEF int stbtt_PackFontRangesGatherRects(stbtt_pack_context *spc, stbtt_fontinfo *info, stbtt_pack_range *ranges, int num_ranges, stbrp_rect *rects)
+STBTT_DEF int stbtt_PackFontRangesGatherRects(stbtt_pack_context* spc, const stbtt_fontinfo* info, stbtt_pack_range* ranges, int num_ranges, stbrp_rect* rects)
 {
-   int i,j,k;
+   int i, j, k;
+   int missing_glyph_added = 0;
 
-   k=0;
-   for (i=0; i < num_ranges; ++i) {
+   k = 0;
+   for (i = 0; i < num_ranges; ++i)
+   {
       float fh = ranges[i].font_size;
       float scale = fh > 0 ? stbtt_ScaleForPixelHeight(info, fh) : stbtt_ScaleForMappingEmToPixels(info, -fh);
-      ranges[i].h_oversample = (unsigned char) spc->h_oversample;
-      ranges[i].v_oversample = (unsigned char) spc->v_oversample;
-      for (j=0; j < ranges[i].num_chars; ++j) {
-         int x0,y0,x1,y1;
+      ranges[i].h_oversample = (unsigned char)spc->h_oversample;
+      ranges[i].v_oversample = (unsigned char)spc->v_oversample;
+      for (j = 0; j < ranges[i].num_chars; ++j)
+      {
+         int x0, y0, x1, y1;
          int codepoint = ranges[i].array_of_unicode_codepoints == NULL ? ranges[i].first_unicode_codepoint_in_range + j : ranges[i].array_of_unicode_codepoints[j];
          int glyph = stbtt_FindGlyphIndex(info, codepoint);
-         stbtt_GetGlyphBitmapBoxSubpixel(info,glyph,
-                                         scale * spc->h_oversample,
-                                         scale * spc->v_oversample,
-                                         0,0,
-                                         &x0,&y0,&x1,&y1);
-         rects[k].w = (stbrp_coord) (x1-x0 + spc->padding + spc->h_oversample-1);
-         rects[k].h = (stbrp_coord) (y1-y0 + spc->padding + spc->v_oversample-1);
+         if (glyph == 0 && (spc->skip_missing || missing_glyph_added))
+         {
+            rects[k].w = rects[k].h = 0;
+         }
+         else
+         {
+            stbtt_GetGlyphBitmapBoxSubpixel(info, glyph,
+                                            scale * spc->h_oversample,
+                                            scale * spc->v_oversample,
+                                            0, 0,
+                                            &x0, &y0, &x1, &y1);
+            rects[k].w = (stbrp_coord)(x1 - x0 + spc->padding + spc->h_oversample - 1);
+            rects[k].h = (stbrp_coord)(y1 - y0 + spc->padding + spc->v_oversample - 1);
+            if (glyph == 0)
+               missing_glyph_added = 1;
+         }
          ++k;
       }
    }
@@ -2876,34 +4631,60 @@ STBTT_DEF int stbtt_PackFontRangesGatherRects(stbtt_pack_context *spc, stbtt_fon
    return k;
 }
 
+STBTT_DEF void stbtt_MakeGlyphBitmapSubpixelPrefilter(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 prefilter_x, int prefilter_y, float* sub_x, float* sub_y, int glyph)
+{
+   stbtt_MakeGlyphBitmapSubpixel(info,
+                                 output,
+                                 out_w - (prefilter_x - 1),
+                                 out_h - (prefilter_y - 1),
+                                 out_stride,
+                                 scale_x,
+                                 scale_y,
+                                 shift_x,
+                                 shift_y,
+                                 glyph);
+
+   if (prefilter_x > 1)
+      stbtt__h_prefilter(output, out_w, out_h, out_stride, prefilter_x);
+
+   if (prefilter_y > 1)
+      stbtt__v_prefilter(output, out_w, out_h, out_stride, prefilter_y);
+
+   *sub_x = stbtt__oversample_shift(prefilter_x);
+   *sub_y = stbtt__oversample_shift(prefilter_y);
+}
+
 // rects array must be big enough to accommodate all characters in the given ranges
-STBTT_DEF int stbtt_PackFontRangesRenderIntoRects(stbtt_pack_context *spc, stbtt_fontinfo *info, stbtt_pack_range *ranges, int num_ranges, stbrp_rect *rects)
+STBTT_DEF int stbtt_PackFontRangesRenderIntoRects(stbtt_pack_context* spc, const stbtt_fontinfo* info, stbtt_pack_range* ranges, int num_ranges, stbrp_rect* rects)
 {
-   int i,j,k, return_value = 1;
+   int i, j, k, missing_glyph = -1, return_value = 1;
 
    // save current values
    int old_h_over = spc->h_oversample;
    int old_v_over = spc->v_oversample;
 
    k = 0;
-   for (i=0; i < num_ranges; ++i) {
+   for (i = 0; i < num_ranges; ++i)
+   {
       float fh = ranges[i].font_size;
       float scale = fh > 0 ? stbtt_ScaleForPixelHeight(info, fh) : stbtt_ScaleForMappingEmToPixels(info, -fh);
-      float recip_h,recip_v,sub_x,sub_y;
+      float recip_h, recip_v, sub_x, sub_y;
       spc->h_oversample = ranges[i].h_oversample;
       spc->v_oversample = ranges[i].v_oversample;
       recip_h = 1.0f / spc->h_oversample;
       recip_v = 1.0f / spc->v_oversample;
       sub_x = stbtt__oversample_shift(spc->h_oversample);
       sub_y = stbtt__oversample_shift(spc->v_oversample);
-      for (j=0; j < ranges[i].num_chars; ++j) {
-         stbrp_rect *r = &rects[k];
-         if (r->was_packed) {
-            stbtt_packedchar *bc = &ranges[i].chardata_for_range[j];
-            int advance, lsb, x0,y0,x1,y1;
+      for (j = 0; j < ranges[i].num_chars; ++j)
+      {
+         stbrp_rect* r = &rects[k];
+         if (r->was_packed && r->w != 0 && r->h != 0)
+         {
+            stbtt_packedchar* bc = &ranges[i].chardata_for_range[j];
+            int advance, lsb, x0, y0, x1, y1;
             int codepoint = ranges[i].array_of_unicode_codepoints == NULL ? ranges[i].first_unicode_codepoint_in_range + j : ranges[i].array_of_unicode_codepoints[j];
             int glyph = stbtt_FindGlyphIndex(info, codepoint);
-            stbrp_coord pad = (stbrp_coord) spc->padding;
+            stbrp_coord pad = (stbrp_coord)spc->padding;
 
             // pad on left and top
             r->x += pad;
@@ -2914,37 +4695,50 @@ STBTT_DEF int stbtt_PackFontRangesRenderIntoRects(stbtt_pack_context *spc, stbtt
             stbtt_GetGlyphBitmapBox(info, glyph,
                                     scale * spc->h_oversample,
                                     scale * spc->v_oversample,
-                                    &x0,&y0,&x1,&y1);
+                                    &x0, &y0, &x1, &y1);
             stbtt_MakeGlyphBitmapSubpixel(info,
-                                          spc->pixels + r->x + r->y*spc->stride_in_bytes,
-                                          r->w - spc->h_oversample+1,
-                                          r->h - spc->v_oversample+1,
+                                          spc->pixels + r->x + r->y * spc->stride_in_bytes,
+                                          r->w - spc->h_oversample + 1,
+                                          r->h - spc->v_oversample + 1,
                                           spc->stride_in_bytes,
                                           scale * spc->h_oversample,
                                           scale * spc->v_oversample,
-                                          0,0,
+                                          0, 0,
                                           glyph);
 
             if (spc->h_oversample > 1)
-               stbtt__h_prefilter(spc->pixels + r->x + r->y*spc->stride_in_bytes,
+               stbtt__h_prefilter(spc->pixels + r->x + r->y * spc->stride_in_bytes,
                                   r->w, r->h, spc->stride_in_bytes,
                                   spc->h_oversample);
 
             if (spc->v_oversample > 1)
-               stbtt__v_prefilter(spc->pixels + r->x + r->y*spc->stride_in_bytes,
+               stbtt__v_prefilter(spc->pixels + r->x + r->y * spc->stride_in_bytes,
                                   r->w, r->h, spc->stride_in_bytes,
                                   spc->v_oversample);
 
-            bc->x0       = (stbtt_int16)  r->x;
-            bc->y0       = (stbtt_int16)  r->y;
-            bc->x1       = (stbtt_int16) (r->x + r->w);
-            bc->y1       = (stbtt_int16) (r->y + r->h);
-            bc->xadvance =                scale * advance;
-            bc->xoff     =       (float)  x0 * recip_h + sub_x;
-            bc->yoff     =       (float)  y0 * recip_v + sub_y;
-            bc->xoff2    =                (x0 + r->w) * recip_h + sub_x;
-            bc->yoff2    =                (y0 + r->h) * recip_v + sub_y;
-         } else {
+            bc->x0 = (stbtt_int16)r->x;
+            bc->y0 = (stbtt_int16)r->y;
+            bc->x1 = (stbtt_int16)(r->x + r->w);
+            bc->y1 = (stbtt_int16)(r->y + r->h);
+            bc->xadvance = scale * advance;
+            bc->xoff = (float)x0 * recip_h + sub_x;
+            bc->yoff = (float)y0 * recip_v + sub_y;
+            bc->xoff2 = (x0 + r->w) * recip_h + sub_x;
+            bc->yoff2 = (y0 + r->h) * recip_v + sub_y;
+
+            if (glyph == 0)
+               missing_glyph = j;
+         }
+         else if (spc->skip_missing)
+         {
+            return_value = 0;
+         }
+         else if (r->was_packed && r->w == 0 && r->h == 0 && missing_glyph >= 0)
+         {
+            ranges[i].chardata_for_range[j] = ranges[i].chardata_for_range[missing_glyph];
+         }
+         else
+         {
             return_value = 0; // if any fail, report failure
          }
 
@@ -2959,72 +4753,88 @@ STBTT_DEF int stbtt_PackFontRangesRenderIntoRects(stbtt_pack_context *spc, stbtt
    return return_value;
 }
 
-STBTT_DEF void stbtt_PackFontRangesPackRects(stbtt_pack_context *spc, stbrp_rect *rects, int num_rects)
+STBTT_DEF void stbtt_PackFontRangesPackRects(stbtt_pack_context* spc, stbrp_rect* rects, int num_rects)
 {
-   stbrp_pack_rects((stbrp_context *) spc->pack_info, rects, num_rects);
+   stbrp_pack_rects((stbrp_context*)spc->pack_info, rects, num_rects);
 }
 
-STBTT_DEF int stbtt_PackFontRanges(stbtt_pack_context *spc, unsigned char *fontdata, int font_index, stbtt_pack_range *ranges, int num_ranges)
+STBTT_DEF int stbtt_PackFontRanges(stbtt_pack_context* spc, const unsigned char* fontdata, int font_index, stbtt_pack_range* ranges, int num_ranges)
 {
    stbtt_fontinfo info;
-   int i,j,n, return_value = 1;
+   int i, j, n, return_value = 1;
    //stbrp_context *context = (stbrp_context *) spc->pack_info;
-   stbrp_rect    *rects;
+   stbrp_rect* rects;
 
    // flag all characters as NOT packed
-   for (i=0; i < num_ranges; ++i)
-      for (j=0; j < ranges[i].num_chars; ++j)
+   for (i = 0; i < num_ranges; ++i)
+      for (j = 0; j < ranges[i].num_chars; ++j)
          ranges[i].chardata_for_range[j].x0 =
          ranges[i].chardata_for_range[j].y0 =
          ranges[i].chardata_for_range[j].x1 =
          ranges[i].chardata_for_range[j].y1 = 0;
 
    n = 0;
-   for (i=0; i < num_ranges; ++i)
+   for (i = 0; i < num_ranges; ++i)
       n += ranges[i].num_chars;
-         
-   rects = (stbrp_rect *) STBTT_malloc(sizeof(*rects) * n, spc->user_allocator_context);
+
+   rects = (stbrp_rect*)STBTT_malloc(sizeof(*rects) * n, spc->user_allocator_context);
    if (rects == NULL)
       return 0;
 
    info.userdata = spc->user_allocator_context;
-   stbtt_InitFont(&info, fontdata, stbtt_GetFontOffsetForIndex(fontdata,font_index));
+   stbtt_InitFont(&info, fontdata, stbtt_GetFontOffsetForIndex(fontdata, font_index));
 
    n = stbtt_PackFontRangesGatherRects(spc, &info, ranges, num_ranges, rects);
 
    stbtt_PackFontRangesPackRects(spc, rects, n);
-  
+
    return_value = stbtt_PackFontRangesRenderIntoRects(spc, &info, ranges, num_ranges, rects);
 
    STBTT_free(rects, spc->user_allocator_context);
    return return_value;
 }
 
-STBTT_DEF int stbtt_PackFontRange(stbtt_pack_context *spc, unsigned char *fontdata, int font_index, float font_size,
-            int first_unicode_codepoint_in_range, int num_chars_in_range, stbtt_packedchar *chardata_for_range)
+STBTT_DEF int stbtt_PackFontRange(stbtt_pack_context* spc, const unsigned char* fontdata, int font_index, float font_size,
+                                  int first_unicode_codepoint_in_range, int num_chars_in_range, stbtt_packedchar* chardata_for_range)
 {
    stbtt_pack_range range;
    range.first_unicode_codepoint_in_range = first_unicode_codepoint_in_range;
    range.array_of_unicode_codepoints = NULL;
-   range.num_chars                   = num_chars_in_range;
-   range.chardata_for_range          = chardata_for_range;
-   range.font_size                   = font_size;
+   range.num_chars = num_chars_in_range;
+   range.chardata_for_range = chardata_for_range;
+   range.font_size = font_size;
    return stbtt_PackFontRanges(spc, fontdata, font_index, &range, 1);
 }
 
-STBTT_DEF void stbtt_GetPackedQuad(stbtt_packedchar *chardata, int pw, int ph, int char_index, float *xpos, float *ypos, stbtt_aligned_quad *q, int align_to_integer)
+STBTT_DEF void stbtt_GetScaledFontVMetrics(const unsigned char* fontdata, int index, float size, float* ascent, float* descent, float* lineGap)
+{
+   int i_ascent, i_descent, i_lineGap;
+   float scale;
+   stbtt_fontinfo info;
+   stbtt_InitFont(&info, fontdata, stbtt_GetFontOffsetForIndex(fontdata, index));
+   scale = size > 0 ? stbtt_ScaleForPixelHeight(&info, size) : stbtt_ScaleForMappingEmToPixels(&info, -size);
+   stbtt_GetFontVMetrics(&info, &i_ascent, &i_descent, &i_lineGap);
+   *ascent = (float)i_ascent * scale;
+   *descent = (float)i_descent * scale;
+   *lineGap = (float)i_lineGap * scale;
+}
+
+STBTT_DEF void stbtt_GetPackedQuad(const stbtt_packedchar* chardata, int pw, int ph, int char_index, float* xpos, float* ypos, stbtt_aligned_quad* q, int align_to_integer)
 {
    float ipw = 1.0f / pw, iph = 1.0f / ph;
-   stbtt_packedchar *b = chardata + char_index;
+   const stbtt_packedchar* b = chardata + char_index;
 
-   if (align_to_integer) {
-      float x = (float) STBTT_ifloor((*xpos + b->xoff) + 0.5f);
-      float y = (float) STBTT_ifloor((*ypos + b->yoff) + 0.5f);
+   if (align_to_integer)
+   {
+      float x = (float)STBTT_ifloor((*xpos + b->xoff) + 0.5f);
+      float y = (float)STBTT_ifloor((*ypos + b->yoff) + 0.5f);
       q->x0 = x;
       q->y0 = y;
       q->x1 = x + b->xoff2 - b->xoff;
       q->y1 = y + b->yoff2 - b->yoff;
-   } else {
+   }
+   else
+   {
       q->x0 = *xpos + b->xoff;
       q->y0 = *ypos + b->yoff;
       q->x1 = *xpos + b->xoff2;
@@ -3039,6 +4849,436 @@ STBTT_DEF void stbtt_GetPackedQuad(stbtt_packedchar *chardata, int pw, int ph, i
    *xpos += b->xadvance;
 }
 
+//////////////////////////////////////////////////////////////////////////////
+//
+// sdf computation
+//
+
+#define STBTT_min(a, b) ((a) < (b) ? (a) : (b))
+#define STBTT_max(a, b) ((a) < (b) ? (b) : (a))
+
+static int stbtt__ray_intersect_bezier(float orig[2], float ray[2], float q0[2], float q1[2], float q2[2], float hits[2][2])
+{
+   float q0perp = q0[1] * ray[0] - q0[0] * ray[1];
+   float q1perp = q1[1] * ray[0] - q1[0] * ray[1];
+   float q2perp = q2[1] * ray[0] - q2[0] * ray[1];
+   float roperp = orig[1] * ray[0] - orig[0] * ray[1];
+
+   float a = q0perp - 2 * q1perp + q2perp;
+   float b = q1perp - q0perp;
+   float c = q0perp - roperp;
+
+   float s0 = 0., s1 = 0.;
+   int num_s = 0;
+
+   if (a != 0.0)
+   {
+      float discr = b * b - a * c;
+      if (discr > 0.0)
+      {
+         float rcpna = -1 / a;
+         float d = (float)STBTT_sqrt(discr);
+         s0 = (b + d) * rcpna;
+         s1 = (b - d) * rcpna;
+         if (s0 >= 0.0 && s0 <= 1.0)
+            num_s = 1;
+         if (d > 0.0 && s1 >= 0.0 && s1 <= 1.0)
+         {
+            if (num_s == 0)
+               s0 = s1;
+            ++num_s;
+         }
+      }
+   }
+   else
+   {
+      // 2*b*s + c = 0
+      // s = -c / (2*b)
+      s0 = c / (-2 * b);
+      if (s0 >= 0.0 && s0 <= 1.0)
+         num_s = 1;
+   }
+
+   if (num_s == 0)
+      return 0;
+   else
+   {
+      float rcp_len2 = 1 / (ray[0] * ray[0] + ray[1] * ray[1]);
+      float rayn_x = ray[0] * rcp_len2, rayn_y = ray[1] * rcp_len2;
+
+      float q0d = q0[0] * rayn_x + q0[1] * rayn_y;
+      float q1d = q1[0] * rayn_x + q1[1] * rayn_y;
+      float q2d = q2[0] * rayn_x + q2[1] * rayn_y;
+      float rod = orig[0] * rayn_x + orig[1] * rayn_y;
+
+      float q10d = q1d - q0d;
+      float q20d = q2d - q0d;
+      float q0rd = q0d - rod;
+
+      hits[0][0] = q0rd + s0 * (2.0f - 2.0f * s0) * q10d + s0 * s0 * q20d;
+      hits[0][1] = a * s0 + b;
+
+      if (num_s > 1)
+      {
+         hits[1][0] = q0rd + s1 * (2.0f - 2.0f * s1) * q10d + s1 * s1 * q20d;
+         hits[1][1] = a * s1 + b;
+         return 2;
+      }
+      else
+      {
+         return 1;
+      }
+   }
+}
+
+static int equal(float* a, float* b)
+{
+   return (a[0] == b[0] && a[1] == b[1]);
+}
+
+static int stbtt__compute_crossings_x(float x, float y, int nverts, stbtt_vertex* verts)
+{
+   int i;
+   float orig[2], ray[2] = { 1, 0 };
+   float y_frac;
+   int winding = 0;
+
+   // make sure y never passes through a vertex of the shape
+   y_frac = (float)STBTT_fmod(y, 1.0f);
+   if (y_frac < 0.01f)
+      y += 0.01f;
+   else if (y_frac > 0.99f)
+      y -= 0.01f;
+
+   orig[0] = x;
+   orig[1] = y;
+
+   // test a ray from (-infinity,y) to (x,y)
+   for (i = 0; i < nverts; ++i)
+   {
+      if (verts[i].type == STBTT_vline)
+      {
+         int x0 = (int)verts[i - 1].x, y0 = (int)verts[i - 1].y;
+         int x1 = (int)verts[i].x, y1 = (int)verts[i].y;
+         if (y > STBTT_min(y0, y1) && y < STBTT_max(y0, y1) && x > STBTT_min(x0, x1))
+         {
+            float x_inter = (y - y0) / (y1 - y0) * (x1 - x0) + x0;
+            if (x_inter < x)
+               winding += (y0 < y1) ? 1 : -1;
+         }
+      }
+      if (verts[i].type == STBTT_vcurve)
+      {
+         int x0 = (int)verts[i - 1].x, y0 = (int)verts[i - 1].y;
+         int x1 = (int)verts[i].cx, y1 = (int)verts[i].cy;
+         int x2 = (int)verts[i].x, y2 = (int)verts[i].y;
+         int ax = STBTT_min(x0, STBTT_min(x1, x2)), ay = STBTT_min(y0, STBTT_min(y1, y2));
+         int by = STBTT_max(y0, STBTT_max(y1, y2));
+         if (y > ay && y < by && x > ax)
+         {
+            float q0[2], q1[2], q2[2];
+            float hits[2][2];
+            q0[0] = (float)x0;
+            q0[1] = (float)y0;
+            q1[0] = (float)x1;
+            q1[1] = (float)y1;
+            q2[0] = (float)x2;
+            q2[1] = (float)y2;
+            if (equal(q0, q1) || equal(q1, q2))
+            {
+               x0 = (int)verts[i - 1].x;
+               y0 = (int)verts[i - 1].y;
+               x1 = (int)verts[i].x;
+               y1 = (int)verts[i].y;
+               if (y > STBTT_min(y0, y1) && y < STBTT_max(y0, y1) && x > STBTT_min(x0, x1))
+               {
+                  float x_inter = (y - y0) / (y1 - y0) * (x1 - x0) + x0;
+                  if (x_inter < x)
+                     winding += (y0 < y1) ? 1 : -1;
+               }
+            }
+            else
+            {
+               int num_hits = stbtt__ray_intersect_bezier(orig, ray, q0, q1, q2, hits);
+               if (num_hits >= 1)
+                  if (hits[0][0] < 0)
+                     winding += (hits[0][1] < 0 ? -1 : 1);
+               if (num_hits >= 2)
+                  if (hits[1][0] < 0)
+                     winding += (hits[1][1] < 0 ? -1 : 1);
+            }
+         }
+      }
+   }
+   return winding;
+}
+
+static float stbtt__cuberoot(float x)
+{
+   if (x < 0)
+      return -(float)STBTT_pow(-x, 1.0f / 3.0f);
+   else
+      return (float)STBTT_pow(x, 1.0f / 3.0f);
+}
+
+// x^3 + a*x^2 + b*x + c = 0
+static int stbtt__solve_cubic(float a, float b, float c, float* r)
+{
+   float s = -a / 3;
+   float p = b - a * a / 3;
+   float q = a * (2 * a * a - 9 * b) / 27 + c;
+   float p3 = p * p * p;
+   float d = q * q + 4 * p3 / 27;
+   if (d >= 0)
+   {
+      float z = (float)STBTT_sqrt(d);
+      float u = (-q + z) / 2;
+      float v = (-q - z) / 2;
+      u = stbtt__cuberoot(u);
+      v = stbtt__cuberoot(v);
+      r[0] = s + u + v;
+      return 1;
+   }
+   else
+   {
+      float u = (float)STBTT_sqrt(-p / 3);
+      float v = (float)STBTT_acos(-STBTT_sqrt(-27 / p3) * q / 2) / 3; // p3 must be negative, since d is negative
+      float m = (float)STBTT_cos(v);
+      float n = (float)STBTT_cos(v - 3.141592 / 2) * 1.732050808f;
+      r[0] = s + u * 2 * m;
+      r[1] = s - u * (m + n);
+      r[2] = s - u * (m - n);
+
+      //STBTT_assert( STBTT_fabs(((r[0]+a)*r[0]+b)*r[0]+c) < 0.05f);  // these asserts may not be safe at all scales, though they're in bezier t parameter units so maybe?
+      //STBTT_assert( STBTT_fabs(((r[1]+a)*r[1]+b)*r[1]+c) < 0.05f);
+      //STBTT_assert( STBTT_fabs(((r[2]+a)*r[2]+b)*r[2]+c) < 0.05f);
+      return 3;
+   }
+}
+
+STBTT_DEF unsigned char* stbtt_GetGlyphSDF(const stbtt_fontinfo* info, float scale, int glyph, int padding, unsigned char onedge_value, float pixel_dist_scale, int* width, int* height, int* xoff, int* yoff)
+{
+   float scale_x = scale, scale_y = scale;
+   int ix0, iy0, ix1, iy1;
+   int w, h;
+   unsigned char* data;
+
+   if (scale == 0)
+      return NULL;
+
+   stbtt_GetGlyphBitmapBoxSubpixel(info, glyph, scale, scale, 0.0f, 0.0f, &ix0, &iy0, &ix1, &iy1);
+
+   // if empty, return NULL
+   if (ix0 == ix1 || iy0 == iy1)
+      return NULL;
+
+   ix0 -= padding;
+   iy0 -= padding;
+   ix1 += padding;
+   iy1 += padding;
+
+   w = (ix1 - ix0);
+   h = (iy1 - iy0);
+
+   if (width)
+      *width = w;
+   if (height)
+      *height = h;
+   if (xoff)
+      *xoff = ix0;
+   if (yoff)
+      *yoff = iy0;
+
+   // invert for y-downwards bitmaps
+   scale_y = -scale_y;
+
+   {
+      int x, y, i, j;
+      float* precompute;
+      stbtt_vertex* verts;
+      int num_verts = stbtt_GetGlyphShape(info, glyph, &verts);
+      data = (unsigned char*)STBTT_malloc(w * h, info->userdata);
+      precompute = (float*)STBTT_malloc(num_verts * sizeof(float), info->userdata);
+
+      for (i = 0, j = num_verts - 1; i < num_verts; j = i++)
+      {
+         if (verts[i].type == STBTT_vline)
+         {
+            float x0 = verts[i].x * scale_x, y0 = verts[i].y * scale_y;
+            float x1 = verts[j].x * scale_x, y1 = verts[j].y * scale_y;
+            float dist = (float)STBTT_sqrt((x1 - x0) * (x1 - x0) + (y1 - y0) * (y1 - y0));
+            precompute[i] = (dist == 0) ? 0.0f : 1.0f / dist;
+         }
+         else if (verts[i].type == STBTT_vcurve)
+         {
+            float x2 = verts[j].x * scale_x, y2 = verts[j].y * scale_y;
+            float x1 = verts[i].cx * scale_x, y1 = verts[i].cy * scale_y;
+            float x0 = verts[i].x * scale_x, y0 = verts[i].y * scale_y;
+            float bx = x0 - 2 * x1 + x2, by = y0 - 2 * y1 + y2;
+            float len2 = bx * bx + by * by;
+            if (len2 != 0.0f)
+               precompute[i] = 1.0f / (bx * bx + by * by);
+            else
+               precompute[i] = 0.0f;
+         }
+         else
+            precompute[i] = 0.0f;
+      }
+
+      for (y = iy0; y < iy1; ++y)
+      {
+         for (x = ix0; x < ix1; ++x)
+         {
+            float val;
+            float min_dist = 999999.0f;
+            float sx = (float)x + 0.5f;
+            float sy = (float)y + 0.5f;
+            float x_gspace = (sx / scale_x);
+            float y_gspace = (sy / scale_y);
+
+            int winding = stbtt__compute_crossings_x(x_gspace, y_gspace, num_verts, verts); // @OPTIMIZE: this could just be a rasterization, but needs to be line vs. non-tesselated curves so a new path
+
+            for (i = 0; i < num_verts; ++i)
+            {
+               float x0 = verts[i].x * scale_x, y0 = verts[i].y * scale_y;
+
+               if (verts[i].type == STBTT_vline && precompute[i] != 0.0f)
+               {
+                  float x1 = verts[i - 1].x * scale_x, y1 = verts[i - 1].y * scale_y;
+
+                  float dist, dist2 = (x0 - sx) * (x0 - sx) + (y0 - sy) * (y0 - sy);
+                  if (dist2 < min_dist * min_dist)
+                     min_dist = (float)STBTT_sqrt(dist2);
+
+                  // coarse culling against bbox
+                  //if (sx > STBTT_min(x0,x1)-min_dist && sx < STBTT_max(x0,x1)+min_dist &&
+                  //    sy > STBTT_min(y0,y1)-min_dist && sy < STBTT_max(y0,y1)+min_dist)
+                  dist = (float)STBTT_fabs((x1 - x0) * (y0 - sy) - (y1 - y0) * (x0 - sx)) * precompute[i];
+                  STBTT_assert(i != 0);
+                  if (dist < min_dist)
+                  {
+                     // check position along line
+                     // x' = x0 + t*(x1-x0), y' = y0 + t*(y1-y0)
+                     // minimize (x'-sx)*(x'-sx)+(y'-sy)*(y'-sy)
+                     float dx = x1 - x0, dy = y1 - y0;
+                     float px = x0 - sx, py = y0 - sy;
+                     // minimize (px+t*dx)^2 + (py+t*dy)^2 = px*px + 2*px*dx*t + t^2*dx*dx + py*py + 2*py*dy*t + t^2*dy*dy
+                     // derivative: 2*px*dx + 2*py*dy + (2*dx*dx+2*dy*dy)*t, set to 0 and solve
+                     float t = -(px * dx + py * dy) / (dx * dx + dy * dy);
+                     if (t >= 0.0f && t <= 1.0f)
+                        min_dist = dist;
+                  }
+               }
+               else if (verts[i].type == STBTT_vcurve)
+               {
+                  float x2 = verts[i - 1].x * scale_x, y2 = verts[i - 1].y * scale_y;
+                  float x1 = verts[i].cx * scale_x, y1 = verts[i].cy * scale_y;
+                  float box_x0 = STBTT_min(STBTT_min(x0, x1), x2);
+                  float box_y0 = STBTT_min(STBTT_min(y0, y1), y2);
+                  float box_x1 = STBTT_max(STBTT_max(x0, x1), x2);
+                  float box_y1 = STBTT_max(STBTT_max(y0, y1), y2);
+                  // coarse culling against bbox to avoid computing cubic unnecessarily
+                  if (sx > box_x0 - min_dist && sx < box_x1 + min_dist && sy > box_y0 - min_dist && sy < box_y1 + min_dist)
+                  {
+                     int num = 0;
+                     float ax = x1 - x0, ay = y1 - y0;
+                     float bx = x0 - 2 * x1 + x2, by = y0 - 2 * y1 + y2;
+                     float mx = x0 - sx, my = y0 - sy;
+                     float res[3] = { 0.f, 0.f, 0.f };
+                     float px, py, t, it, dist2;
+                     float a_inv = precompute[i];
+                     if (a_inv == 0.0)
+                     { // if a_inv is 0, it's 2nd degree so use quadratic formula
+                        float a = 3 * (ax * bx + ay * by);
+                        float b = 2 * (ax * ax + ay * ay) + (mx * bx + my * by);
+                        float c = mx * ax + my * ay;
+                        if (a == 0.0)
+                        { // if a is 0, it's linear
+                           if (b != 0.0)
+                           {
+                              res[num++] = -c / b;
+                           }
+                        }
+                        else
+                        {
+                           float discriminant = b * b - 4 * a * c;
+                           if (discriminant < 0)
+                              num = 0;
+                           else
+                           {
+                              float root = (float)STBTT_sqrt(discriminant);
+                              res[0] = (-b - root) / (2 * a);
+                              res[1] = (-b + root) / (2 * a);
+                              num = 2; // don't bother distinguishing 1-solution case, as code below will still work
+                           }
+                        }
+                     }
+                     else
+                     {
+                        float b = 3 * (ax * bx + ay * by) * a_inv; // could precompute this as it doesn't depend on sample point
+                        float c = (2 * (ax * ax + ay * ay) + (mx * bx + my * by)) * a_inv;
+                        float d = (mx * ax + my * ay) * a_inv;
+                        num = stbtt__solve_cubic(b, c, d, res);
+                     }
+                     dist2 = (x0 - sx) * (x0 - sx) + (y0 - sy) * (y0 - sy);
+                     if (dist2 < min_dist * min_dist)
+                        min_dist = (float)STBTT_sqrt(dist2);
+
+                     if (num >= 1 && res[0] >= 0.0f && res[0] <= 1.0f)
+                     {
+                        t = res[0], it = 1.0f - t;
+                        px = it * it * x0 + 2 * t * it * x1 + t * t * x2;
+                        py = it * it * y0 + 2 * t * it * y1 + t * t * y2;
+                        dist2 = (px - sx) * (px - sx) + (py - sy) * (py - sy);
+                        if (dist2 < min_dist * min_dist)
+                           min_dist = (float)STBTT_sqrt(dist2);
+                     }
+                     if (num >= 2 && res[1] >= 0.0f && res[1] <= 1.0f)
+                     {
+                        t = res[1], it = 1.0f - t;
+                        px = it * it * x0 + 2 * t * it * x1 + t * t * x2;
+                        py = it * it * y0 + 2 * t * it * y1 + t * t * y2;
+                        dist2 = (px - sx) * (px - sx) + (py - sy) * (py - sy);
+                        if (dist2 < min_dist * min_dist)
+                           min_dist = (float)STBTT_sqrt(dist2);
+                     }
+                     if (num >= 3 && res[2] >= 0.0f && res[2] <= 1.0f)
+                     {
+                        t = res[2], it = 1.0f - t;
+                        px = it * it * x0 + 2 * t * it * x1 + t * t * x2;
+                        py = it * it * y0 + 2 * t * it * y1 + t * t * y2;
+                        dist2 = (px - sx) * (px - sx) + (py - sy) * (py - sy);
+                        if (dist2 < min_dist * min_dist)
+                           min_dist = (float)STBTT_sqrt(dist2);
+                     }
+                  }
+               }
+            }
+            if (winding == 0)
+               min_dist = -min_dist; // if outside the shape, value is negative
+            val = onedge_value + pixel_dist_scale * min_dist;
+            if (val < 0)
+               val = 0;
+            else if (val > 255)
+               val = 255;
+            data[(y - iy0) * w + (x - ix0)] = (unsigned char)val;
+         }
+      }
+      STBTT_free(precompute, info->userdata);
+      STBTT_free(verts, info->userdata);
+   }
+   return data;
+}
+
+STBTT_DEF unsigned char* stbtt_GetCodepointSDF(const stbtt_fontinfo* info, float scale, int codepoint, int padding, unsigned char onedge_value, float pixel_dist_scale, int* width, int* height, int* xoff, int* yoff)
+{
+   return stbtt_GetGlyphSDF(info, scale, stbtt_FindGlyphIndex(info, codepoint), padding, onedge_value, pixel_dist_scale, width, height, xoff, yoff);
+}
+
+STBTT_DEF void stbtt_FreeSDF(unsigned char* bitmap, void* userdata)
+{
+   STBTT_free(bitmap, userdata);
+}
 
 //////////////////////////////////////////////////////////////////////////////
 //
@@ -3046,38 +5286,62 @@ STBTT_DEF void stbtt_GetPackedQuad(stbtt_packedchar *chardata, int pw, int ph, i
 //
 
 // 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) 
+static stbtt_int32 stbtt__CompareUTF8toUTF16_bigendian_prefix(stbtt_uint8* s1, stbtt_int32 len1, stbtt_uint8* s2, stbtt_int32 len2)
 {
-   stbtt_int32 i=0;
+   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) {
+   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;
+         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;
+         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) {
+      }
+      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;
+      }
+      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;
@@ -3085,68 +5349,80 @@ static stbtt_int32 stbtt__CompareUTF8toUTF16_bigendian_prefix(const stbtt_uint8
    return i;
 }
 
-STBTT_DEF int stbtt_CompareUTF8toUTF16_bigendian(const char *s1, int len1, const char *s2, int len2) 
+static int stbtt_CompareUTF8toUTF16_bigendian_internal(char* s1, int len1, char* s2, int len2)
 {
-   return len1 == stbtt__CompareUTF8toUTF16_bigendian_prefix((const stbtt_uint8*) s1, len1, (const stbtt_uint8*) s2, len2);
+   return len1 == stbtt__CompareUTF8toUTF16_bigendian_prefix((stbtt_uint8*)s1, len1, (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
-STBTT_DEF const char *stbtt_GetFontNameString(const stbtt_fontinfo *font, int *length, int platformID, int encodingID, int languageID, int nameID)
+STBTT_DEF 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_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;
+   if (!nm)
+      return NULL;
 
-   count = ttUSHORT(fc+nm+2);
-   stringOffset = nm + ttUSHORT(fc+nm+4);
-   for (i=0; i < count; ++i) {
+   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));
+      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)
+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);
+   stbtt_int32 count = ttUSHORT(fc + nm + 2);
+   stbtt_int32 stringOffset = nm + ttUSHORT(fc + nm + 4);
 
-   for (i=0; i < count; ++i) {
+   for (i = 0; i < count; ++i)
+   {
       stbtt_uint32 loc = nm + 6 + 12 * i;
-      stbtt_int32 id = ttUSHORT(fc+loc+6);
-      if (id == target_id) {
+      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);
+         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);
+         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) {
+            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 (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] == ' ') {
+                  }
+                  else if (matchlen < nlen && name[matchlen] == ' ')
+                  {
                      ++matchlen;
-                     if (stbtt_CompareUTF8toUTF16_bigendian((char*) (name+matchlen), nlen-matchlen, (char*)(fc+stringOffset+off),slen))
+                     if (stbtt_CompareUTF8toUTF16_bigendian_internal((char*)(name + matchlen), nlen - matchlen, (char*)(fc + stringOffset + off), slen))
                         return 1;
                   }
-               } else {
+               }
+               else
+               {
                   // if nothing immediately following
                   if (matchlen == nlen)
                      return 1;
@@ -3160,51 +5436,125 @@ static int stbtt__matchpair(stbtt_uint8 *fc, stbtt_uint32 nm, stbtt_uint8 *name,
    return 0;
 }
 
-static int stbtt__matches(stbtt_uint8 *fc, stbtt_uint32 offset, stbtt_uint8 *name, stbtt_int32 flags)
+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;
+   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) {
+   if (flags)
+   {
       hd = stbtt__find_table(fc, offset, "head");
-      if ((ttUSHORT(fc+hd+44) & 7) != (flags & 7)) return 0;
+      if ((ttUSHORT(fc + hd + 44) & 7) != (flags & 7))
+         return 0;
    }
 
    nm = stbtt__find_table(fc, offset, "name");
-   if (!nm) return 0;
+   if (!nm)
+      return 0;
 
-   if (flags) {
+   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;
+      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;
 }
 
-STBTT_DEF int stbtt_FindMatchingFont(const unsigned char *font_collection, const char *name_utf8, stbtt_int32 flags)
+static int stbtt_FindMatchingFont_internal(unsigned char* font_collection, char* name_utf8, stbtt_int32 flags)
 {
    stbtt_int32 i;
-   for (i=0;;++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))
+      if (off < 0)
+         return off;
+      if (stbtt__matches((stbtt_uint8*)font_collection, off, (stbtt_uint8*)name_utf8, flags))
          return off;
    }
 }
 
+#if defined(__GNUC__) || defined(__clang__)
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wcast-qual"
+#endif
+
+STBTT_DEF int stbtt_BakeFontBitmap(const unsigned char* data, int offset,
+                                   float pixel_height, unsigned char* pixels, int pw, int ph,
+                                   int first_char, int num_chars, stbtt_bakedchar* chardata)
+{
+   return stbtt_BakeFontBitmap_internal((unsigned char*)data, offset, pixel_height, pixels, pw, ph, first_char, num_chars, chardata);
+}
+
+STBTT_DEF int stbtt_GetFontOffsetForIndex(const unsigned char* data, int index)
+{
+   return stbtt_GetFontOffsetForIndex_internal((unsigned char*)data, index);
+}
+
+STBTT_DEF int stbtt_GetNumberOfFonts(const unsigned char* data)
+{
+   return stbtt_GetNumberOfFonts_internal((unsigned char*)data);
+}
+
+STBTT_DEF int stbtt_InitFont(stbtt_fontinfo* info, const unsigned char* data, int offset)
+{
+   return stbtt_InitFont_internal(info, (unsigned char*)data, offset);
+}
+
+STBTT_DEF int stbtt_FindMatchingFont(const unsigned char* fontdata, const char* name, int flags)
+{
+   return stbtt_FindMatchingFont_internal((unsigned char*)fontdata, (char*)name, flags);
+}
+
+STBTT_DEF int stbtt_CompareUTF8toUTF16_bigendian(const char* s1, int len1, const char* s2, int len2)
+{
+   return stbtt_CompareUTF8toUTF16_bigendian_internal((char*)s1, len1, (char*)s2, len2);
+}
+
+#if defined(__GNUC__) || defined(__clang__)
+#pragma GCC diagnostic pop
+#endif
+
 #endif // STB_TRUETYPE_IMPLEMENTATION
 
 
 // FULL VERSION HISTORY
 //
+//   1.25 (2021-07-11) many fixes
+//   1.24 (2020-02-05) fix warning
+//   1.23 (2020-02-02) query SVG data for glyphs; query whole kerning table (but only kern not GPOS)
+//   1.22 (2019-08-11) minimize missing-glyph duplication; fix kerning if both 'GPOS' and 'kern' are defined
+//   1.21 (2019-02-25) fix warning
+//   1.20 (2019-02-07) PackFontRange skips missing codepoints; GetScaleFontVMetrics()
+//   1.19 (2018-02-11) OpenType GPOS kerning (horizontal only), STBTT_fmod
+//   1.18 (2018-01-29) add missing function
+//   1.17 (2017-07-23) make more arguments const; doc fix
+//   1.16 (2017-07-12) SDF support
+//   1.15 (2017-03-03) make more arguments const
+//   1.14 (2017-01-16) num-fonts-in-TTC function
+//   1.13 (2017-01-02) support OpenType fonts, certain Apple fonts
+//   1.12 (2016-10-25) suppress warnings about casting away const with -Wcast-qual
+//   1.11 (2016-04-02) fix unused-variable warning
+//   1.10 (2016-04-02) allow user-defined fabs() replacement
+//                     fix memory leak if fontsize=0.0
+//                     fix warning from duplicate typedef
 //   1.09 (2016-01-16) warning fix; avoid crash on outofmem; use alloc userdata for PackFontRanges
 //   1.08 (2015-09-13) document stbtt_Rasterize(); fixes for vertical & horizontal edges
 //   1.07 (2015-08-01) allow PackFontRanges to accept arrays of sparse codepoints;
@@ -3247,3 +5597,45 @@ STBTT_DEF int stbtt_FindMatchingFont(const unsigned char *font_collection, const
 //   0.2  (2009-03-11) Fix unsigned/signed char warnings
 //   0.1  (2009-03-09) First public release
 //
+
+/*
+------------------------------------------------------------------------------
+This software is available under 2 licenses -- choose whichever you prefer.
+------------------------------------------------------------------------------
+ALTERNATIVE A - MIT License
+Copyright (c) 2017 Sean Barrett
+Permission is hereby granted, free of charge, to any person obtaining a copy of
+this software and associated documentation files (the "Software"), to deal in
+the Software without restriction, including without limitation the rights to
+use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies
+of the Software, and to permit persons to whom the Software is furnished to do
+so, subject to the following conditions:
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
+------------------------------------------------------------------------------
+ALTERNATIVE B - Public Domain (www.unlicense.org)
+This is free and unencumbered software released into the public domain.
+Anyone is free to copy, modify, publish, use, compile, sell, or distribute this
+software, either in source code form or as a compiled binary, for any purpose,
+commercial or non-commercial, and by any means.
+In jurisdictions that recognize copyright laws, the author or authors of this
+software dedicate any and all copyright interest in the software to the public
+domain. We make this dedication for the benefit of the public at large and to
+the detriment of our heirs and successors. We intend this dedication to be an
+overt act of relinquishment in perpetuity of all present and future rights to
+this software under copyright law.
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+------------------------------------------------------------------------------
+*/
\ No newline at end of file
diff --git a/libs/nanovg/src/nanovg.c b/libs/nanovg/src/nanovg.c
index 56eb37b00..23f4bbe86 100755
--- a/libs/nanovg/src/nanovg.c
+++ b/libs/nanovg/src/nanovg.c
@@ -24,8 +24,11 @@
 #include "nanovg.h"
 #define FONTSTASH_IMPLEMENTATION
 #include "fontstash.h"
+
+#ifndef NVG_NO_STB
 #define STB_IMAGE_IMPLEMENTATION
 #include "stb_image.h"
+#endif
 
 #ifdef _MSC_VER
 #pragma warning(disable: 4100)  // unreferenced formal parameter
@@ -42,7 +45,10 @@
 #define NVG_INIT_POINTS_SIZE 128
 #define NVG_INIT_PATHS_SIZE 16
 #define NVG_INIT_VERTS_SIZE 256
+
+#ifndef NVG_MAX_STATES
 #define NVG_MAX_STATES 32
+#endif
 
 #define NVG_KAPPA90 0.5522847493f	// Length proportional to radius of a cubic bezier handle for 90deg arcs.
 
@@ -66,6 +72,8 @@ enum NVGpointFlags
 };
 
 struct NVGstate {
+	NVGcompositeOperationState compositeOperation;
+	int shapeAntiAlias;
 	NVGpaint fill;
 	NVGpaint stroke;
 	float strokeWidth;
@@ -203,6 +211,84 @@ static void nvg__setDevicePixelRatio(NVGcontext* ctx, float ratio)
 	ctx->devicePxRatio = ratio;
 }
 
+static NVGcompositeOperationState nvg__compositeOperationState(int op)
+{
+	int sfactor, dfactor;
+
+	if (op == NVG_SOURCE_OVER)
+	{
+		sfactor = NVG_ONE;
+		dfactor = NVG_ONE_MINUS_SRC_ALPHA;
+	}
+	else if (op == NVG_SOURCE_IN)
+	{
+		sfactor = NVG_DST_ALPHA;
+		dfactor = NVG_ZERO;
+	}
+	else if (op == NVG_SOURCE_OUT)
+	{
+		sfactor = NVG_ONE_MINUS_DST_ALPHA;
+		dfactor = NVG_ZERO;
+	}
+	else if (op == NVG_ATOP)
+	{
+		sfactor = NVG_DST_ALPHA;
+		dfactor = NVG_ONE_MINUS_SRC_ALPHA;
+	}
+	else if (op == NVG_DESTINATION_OVER)
+	{
+		sfactor = NVG_ONE_MINUS_DST_ALPHA;
+		dfactor = NVG_ONE;
+	}
+	else if (op == NVG_DESTINATION_IN)
+	{
+		sfactor = NVG_ZERO;
+		dfactor = NVG_SRC_ALPHA;
+	}
+	else if (op == NVG_DESTINATION_OUT)
+	{
+		sfactor = NVG_ZERO;
+		dfactor = NVG_ONE_MINUS_SRC_ALPHA;
+	}
+	else if (op == NVG_DESTINATION_ATOP)
+	{
+		sfactor = NVG_ONE_MINUS_DST_ALPHA;
+		dfactor = NVG_SRC_ALPHA;
+	}
+	else if (op == NVG_LIGHTER)
+	{
+		sfactor = NVG_ONE;
+		dfactor = NVG_ONE;
+	}
+	else if (op == NVG_COPY)
+	{
+		sfactor = NVG_ONE;
+		dfactor = NVG_ZERO;
+	}
+	else if (op == NVG_XOR)
+	{
+		sfactor = NVG_ONE_MINUS_DST_ALPHA;
+		dfactor = NVG_ONE_MINUS_SRC_ALPHA;
+	}
+	else
+	{
+		sfactor = NVG_ONE;
+		dfactor = NVG_ZERO;
+	}
+
+	NVGcompositeOperationState state;
+	state.srcRGB = sfactor;
+	state.dstRGB = dfactor;
+	state.srcAlpha = sfactor;
+	state.dstAlpha = dfactor;
+	return state;
+}
+
+static NVGstate* nvg__getState(NVGcontext* ctx)
+{
+	return &ctx->states[ctx->nstates-1];
+}
+
 NVGcontext* nvgCreateInternal(NVGparams* params)
 {
 	FONSparams fontParams;
@@ -283,7 +369,7 @@ void nvgDeleteInternal(NVGcontext* ctx)
 	free(ctx);
 }
 
-void nvgBeginFrame(NVGcontext* ctx, int windowWidth, int windowHeight, float devicePixelRatio)
+void nvgBeginFrame(NVGcontext* ctx, float windowWidth, float windowHeight, float devicePixelRatio)
 {
 /*	printf("Tris: draws:%d  fill:%d  stroke:%d  text:%d  TOT:%d\n",
 		ctx->drawCallCount, ctx->fillTriCount, ctx->strokeTriCount, ctx->textTriCount,
@@ -295,7 +381,7 @@ void nvgBeginFrame(NVGcontext* ctx, int windowWidth, int windowHeight, float dev
 
 	nvg__setDevicePixelRatio(ctx, devicePixelRatio);
 
-	ctx->params.renderViewport(ctx->params.userPtr, windowWidth, windowHeight);
+	ctx->params.renderViewport(ctx->params.userPtr, windowWidth, windowHeight, devicePixelRatio);
 
 	ctx->drawCallCount = 0;
 	ctx->fillTriCount = 0;
@@ -313,6 +399,7 @@ void nvgEndFrame(NVGcontext* ctx)
 	ctx->params.renderFlush(ctx->params.userPtr);
 	if (ctx->fontImageIdx != 0) {
 		int fontImage = ctx->fontImages[ctx->fontImageIdx];
+		ctx->fontImages[ctx->fontImageIdx] = 0;
 		int i, j, iw, ih;
 		// delete images that smaller than current one
 		if (fontImage == 0)
@@ -321,20 +408,19 @@ void nvgEndFrame(NVGcontext* ctx)
 		for (i = j = 0; i < ctx->fontImageIdx; i++) {
 			if (ctx->fontImages[i] != 0) {
 				int nw, nh;
-				nvgImageSize(ctx, ctx->fontImages[i], &nw, &nh);
+				int image = ctx->fontImages[i];
+				ctx->fontImages[i] = 0;
+				nvgImageSize(ctx, image, &nw, &nh);
 				if (nw < iw || nh < ih)
-					nvgDeleteImage(ctx, ctx->fontImages[i]);
+					nvgDeleteImage(ctx, image);
 				else
-					ctx->fontImages[j++] = ctx->fontImages[i];
+					ctx->fontImages[j++] = image;
 			}
 		}
 		// make current font image to first
-		ctx->fontImages[j++] = ctx->fontImages[0];
+		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;
 	}
 }
 
@@ -386,7 +472,7 @@ NVGcolor nvgLerpRGBA(NVGcolor c0, NVGcolor c1, float u)
 {
 	int i;
 	float oneminu;
-	NVGcolor cint;
+	NVGcolor cint = {{{0}}};
 
 	u = nvg__clampf(u, 0.0f, 1.0f);
 	oneminu = 1.0f - u;
@@ -433,12 +519,6 @@ NVGcolor nvgHSLA(float h, float s, float l, unsigned char a)
 	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;
@@ -571,6 +651,8 @@ void nvgReset(NVGcontext* ctx)
 
 	nvg__setPaintColor(&state->fill, nvgRGBA(255,255,255,255));
 	nvg__setPaintColor(&state->stroke, nvgRGBA(0,0,0,255));
+	state->compositeOperation = nvg__compositeOperationState(NVG_SOURCE_OVER);
+	state->shapeAntiAlias = 1;
 	state->strokeWidth = 1.0f;
 	state->miterLimit = 10.0f;
 	state->lineCap = NVG_BUTT;
@@ -590,6 +672,12 @@ void nvgReset(NVGcontext* ctx)
 }
 
 // State setting
+void nvgShapeAntiAlias(NVGcontext* ctx, int enabled)
+{
+	NVGstate* state = nvg__getState(ctx);
+	state->shapeAntiAlias = enabled;
+}
+
 void nvgStrokeWidth(NVGcontext* ctx, float width)
 {
 	NVGstate* state = nvg__getState(ctx);
@@ -706,6 +794,7 @@ void nvgFillPaint(NVGcontext* ctx, NVGpaint paint)
 	nvgTransformMultiply(state->fill.xform, state->xform);
 }
 
+#ifndef NVG_NO_STB
 int nvgCreateImage(NVGcontext* ctx, const char* filename, int imageFlags)
 {
 	int w, h, n, image;
@@ -734,6 +823,7 @@ int nvgCreateImageMem(NVGcontext* ctx, int imageFlags, unsigned char* data, int
 	stbi_image_free(img);
 	return image;
 }
+#endif
 
 int nvgCreateImageRGBA(NVGcontext* ctx, int w, int h, int imageFlags, const unsigned char* data)
 {
@@ -939,6 +1029,30 @@ void nvgResetScissor(NVGcontext* ctx)
 	state->scissor.extent[1] = -1.0f;
 }
 
+// Global composite operation.
+void nvgGlobalCompositeOperation(NVGcontext* ctx, int op)
+{
+	NVGstate* state = nvg__getState(ctx);
+	state->compositeOperation = nvg__compositeOperationState(op);
+}
+
+void nvgGlobalCompositeBlendFunc(NVGcontext* ctx, int sfactor, int dfactor)
+{
+	nvgGlobalCompositeBlendFuncSeparate(ctx, sfactor, dfactor, sfactor, dfactor);
+}
+
+void nvgGlobalCompositeBlendFuncSeparate(NVGcontext* ctx, int srcRGB, int dstRGB, int srcAlpha, int dstAlpha)
+{
+	NVGcompositeOperationState op;
+	op.srcRGB = srcRGB;
+	op.dstRGB = dstRGB;
+	op.srcAlpha = srcAlpha;
+	op.dstAlpha = dstAlpha;
+
+	NVGstate* state = nvg__getState(ctx);
+	state->compositeOperation = op;
+}
+
 static int nvg__ptEquals(float x1, float y1, float x2, float y2, float tol)
 {
 	float dx = x2 - x1;
@@ -1334,7 +1448,8 @@ static void nvg__chooseBevel(int bevel, NVGpoint* p0, NVGpoint* p1, float w,
 }
 
 static NVGvertex* nvg__roundJoin(NVGvertex* dst, NVGpoint* p0, NVGpoint* p1,
-										float lw, float rw, float lu, float ru, int ncap, float fringe)
+								 float lw, float rw, float lu, float ru, int ncap,
+								 float fringe)
 {
 	int i, n;
 	float dlx0 = p0->dy;
@@ -1467,36 +1582,39 @@ static NVGvertex* nvg__bevelJoin(NVGvertex* dst, NVGpoint* p0, NVGpoint* p1,
 }
 
 static NVGvertex* nvg__buttCapStart(NVGvertex* dst, NVGpoint* p,
-										   float dx, float dy, float w, float d, float aa)
+									float dx, float dy, float w, float d,
+									float aa, float u0, float u1)
 {
 	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++;
+	nvg__vset(dst, px + dlx*w - dx*aa, py + dly*w - dy*aa, u0,0); dst++;
+	nvg__vset(dst, px - dlx*w - dx*aa, py - dly*w - dy*aa, u1,0); dst++;
+	nvg__vset(dst, px + dlx*w, py + dly*w, u0,1); dst++;
+	nvg__vset(dst, px - dlx*w, py - dly*w, u1,1); dst++;
 	return dst;
 }
 
 static NVGvertex* nvg__buttCapEnd(NVGvertex* dst, NVGpoint* p,
-										   float dx, float dy, float w, float d, float aa)
+								  float dx, float dy, float w, float d,
+								  float aa, float u0, float u1)
 {
 	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++;
+	nvg__vset(dst, px + dlx*w, py + dly*w, u0,1); dst++;
+	nvg__vset(dst, px - dlx*w, py - dly*w, u1,1); dst++;
+	nvg__vset(dst, px + dlx*w + dx*aa, py + dly*w + dy*aa, u0,0); dst++;
+	nvg__vset(dst, px - dlx*w + dx*aa, py - dly*w + dy*aa, u1,0); dst++;
 	return dst;
 }
 
 
 static NVGvertex* nvg__roundCapStart(NVGvertex* dst, NVGpoint* p,
-											float dx, float dy, float w, int ncap, float aa)
+									 float dx, float dy, float w, int ncap,
+									 float aa, float u0, float u1)
 {
 	int i;
 	float px = p->x;
@@ -1507,16 +1625,17 @@ static NVGvertex* nvg__roundCapStart(NVGvertex* dst, NVGpoint* p,
 	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 - dlx*ax - dx*ay, py - dly*ax - dy*ay, u0,1); 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++;
+	nvg__vset(dst, px + dlx*w, py + dly*w, u0,1); dst++;
+	nvg__vset(dst, px - dlx*w, py - dly*w, u1,1); dst++;
 	return dst;
 }
 
 static NVGvertex* nvg__roundCapEnd(NVGvertex* dst, NVGpoint* p,
-										  float dx, float dy, float w, int ncap, float aa)
+								   float dx, float dy, float w, int ncap,
+								   float aa, float u0, float u1)
 {
 	int i;
 	float px = p->x;
@@ -1524,13 +1643,13 @@ static NVGvertex* nvg__roundCapEnd(NVGvertex* dst, NVGpoint* p,
 	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++;
+	nvg__vset(dst, px + dlx*w, py + dly*w, u0,1); dst++;
+	nvg__vset(dst, px - dlx*w, py - dly*w, u1,1); 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++;
+		nvg__vset(dst, px - dlx*ax + dx*ay, py - dly*ax + dy*ay, u0,1); dst++;
 	}
 	return dst;
 }
@@ -1606,15 +1725,24 @@ static void nvg__calculateJoins(NVGcontext* ctx, float w, int lineJoin, float mi
 }
 
 
-static int nvg__expandStroke(NVGcontext* ctx, float w, int lineCap, int lineJoin, float miterLimit)
+static int nvg__expandStroke(NVGcontext* ctx, float w, float fringe, int lineCap, int lineJoin, float miterLimit)
 {
 	NVGpathCache* cache = ctx->cache;
 	NVGvertex* verts;
 	NVGvertex* dst;
 	int cverts, i, j;
-	float aa = ctx->fringeWidth;
+	float aa = fringe;//ctx->fringeWidth;
+	float u0 = 0.0f, u1 = 1.0f;
 	int ncap = nvg__curveDivs(w, NVG_PI, ctx->tessTol);	// Calculate divisions per half circle.
 
+	w += aa * 0.5f;
+
+	// Disable the gradient used for antialiasing when antialiasing is not used.
+	if (aa == 0.0f) {
+		u0 = 0.5f;
+		u1 = 0.5f;
+	}
+
 	nvg__calculateJoins(ctx, w, lineJoin, miterLimit);
 
 	// Calculate max vertex usage.
@@ -1675,42 +1803,42 @@ static int nvg__expandStroke(NVGcontext* ctx, float w, int lineCap, int lineJoin
 			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);
+				dst = nvg__buttCapStart(dst, p0, dx, dy, w, -aa*0.5f, aa, u0, u1);
 			else if (lineCap == NVG_BUTT || lineCap == NVG_SQUARE)
-				dst = nvg__buttCapStart(dst, p0, dx, dy, w, w-aa, aa);
+				dst = nvg__buttCapStart(dst, p0, dx, dy, w, w-aa, aa, u0, u1);
 			else if (lineCap == NVG_ROUND)
-				dst = nvg__roundCapStart(dst, p0, dx, dy, w, ncap, aa);
+				dst = nvg__roundCapStart(dst, p0, dx, dy, w, ncap, aa, u0, u1);
 		}
 
 		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);
+					dst = nvg__roundJoin(dst, p0, p1, w, w, u0, u1, ncap, aa);
 				} else {
-					dst = nvg__bevelJoin(dst, p0, p1, w, w, 0, 1, aa);
+					dst = nvg__bevelJoin(dst, p0, p1, w, w, u0, u1, 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++;
+				nvg__vset(dst, p1->x + (p1->dmx * w), p1->y + (p1->dmy * w), u0,1); dst++;
+				nvg__vset(dst, p1->x - (p1->dmx * w), p1->y - (p1->dmy * w), u1,1); 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++;
+			nvg__vset(dst, verts[0].x, verts[0].y, u0,1); dst++;
+			nvg__vset(dst, verts[1].x, verts[1].y, u1,1); 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);
+				dst = nvg__buttCapEnd(dst, p1, dx, dy, w, -aa*0.5f, aa, u0, u1);
 			else if (lineCap == NVG_BUTT || lineCap == NVG_SQUARE)
-				dst = nvg__buttCapEnd(dst, p1, dx, dy, w, w-aa, aa);
+				dst = nvg__buttCapEnd(dst, p1, dx, dy, w, w-aa, aa, u0, u1);
 			else if (lineCap == NVG_ROUND)
-				dst = nvg__roundCapEnd(dst, p1, dx, dy, w, ncap, aa);
+				dst = nvg__roundCapEnd(dst, p1, dx, dy, w, ncap, aa, u0, u1);
 		}
 
 		path->nstroke = (int)(dst - verts);
@@ -2025,22 +2153,31 @@ void nvgRect(NVGcontext* ctx, float x, float y, float w, float h)
 
 void nvgRoundedRect(NVGcontext* ctx, float x, float y, float w, float h, float r)
 {
-	if (r < 0.1f) {
-		nvgRect(ctx, x,y,w,h);
+	nvgRoundedRectVarying(ctx, x, y, w, h, r, r, r, r);
+}
+
+void nvgRoundedRectVarying(NVGcontext* ctx, float x, float y, float w, float h, float radTopLeft, float radTopRight, float radBottomRight, float radBottomLeft)
+{
+	if(radTopLeft < 0.1f && radTopRight < 0.1f && radBottomRight < 0.1f && radBottomLeft < 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);
+	} else {
+		float halfw = nvg__absf(w)*0.5f;
+		float halfh = nvg__absf(h)*0.5f;
+		float rxBL = nvg__minf(radBottomLeft, halfw) * nvg__signf(w), ryBL = nvg__minf(radBottomLeft, halfh) * nvg__signf(h);
+		float rxBR = nvg__minf(radBottomRight, halfw) * nvg__signf(w), ryBR = nvg__minf(radBottomRight, halfh) * nvg__signf(h);
+		float rxTR = nvg__minf(radTopRight, halfw) * nvg__signf(w), ryTR = nvg__minf(radTopRight, halfh) * nvg__signf(h);
+		float rxTL = nvg__minf(radTopLeft, halfw) * nvg__signf(w), ryTL = nvg__minf(radTopLeft, halfh) * 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_MOVETO, x, y + ryTL,
+			NVG_LINETO, x, y + h - ryBL,
+			NVG_BEZIERTO, x, y + h - ryBL*(1 - NVG_KAPPA90), x + rxBL*(1 - NVG_KAPPA90), y + h, x + rxBL, y + h,
+			NVG_LINETO, x + w - rxBR, y + h,
+			NVG_BEZIERTO, x + w - rxBR*(1 - NVG_KAPPA90), y + h, x + w, y + h - ryBR*(1 - NVG_KAPPA90), x + w, y + h - ryBR,
+			NVG_LINETO, x + w, y + ryTR,
+			NVG_BEZIERTO, x + w, y + ryTR*(1 - NVG_KAPPA90), x + w - rxTR*(1 - NVG_KAPPA90), y, x + w - rxTR, y,
+			NVG_LINETO, x + rxTL, y,
+			NVG_BEZIERTO, x + rxTL*(1 - NVG_KAPPA90), y, x, y + ryTL*(1 - NVG_KAPPA90), x, y + ryTL,
 			NVG_CLOSE
 		};
 		nvg__appendCommands(ctx, vals, NVG_COUNTOF(vals));
@@ -2095,7 +2232,7 @@ void nvgFill(NVGcontext* ctx)
 	int i;
 
 	nvg__flattenPaths(ctx);
-	if (ctx->params.edgeAntiAlias)
+	if (ctx->params.edgeAntiAlias && state->shapeAntiAlias)
 		nvg__expandFill(ctx, ctx->fringeWidth, NVG_MITER, 2.4f);
 	else
 		nvg__expandFill(ctx, 0.0f, NVG_MITER, 2.4f);
@@ -2104,7 +2241,7 @@ void nvgFill(NVGcontext* ctx)
 	fillPaint.innerColor.a *= state->alpha;
 	fillPaint.outerColor.a *= state->alpha;
 
-	ctx->params.renderFill(ctx->params.userPtr, &fillPaint, &state->scissor, ctx->fringeWidth,
+	ctx->params.renderFill(ctx->params.userPtr, &fillPaint, state->compositeOperation, &state->scissor, ctx->fringeWidth,
 						   ctx->cache->bounds, ctx->cache->paths, ctx->cache->npaths);
 
 	// Count triangles
@@ -2125,6 +2262,7 @@ void nvgStroke(NVGcontext* ctx)
 	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.
@@ -2140,12 +2278,12 @@ void nvgStroke(NVGcontext* ctx)
 
 	nvg__flattenPaths(ctx);
 
-	if (ctx->params.edgeAntiAlias)
-		nvg__expandStroke(ctx, strokeWidth*0.5f + ctx->fringeWidth*0.5f, state->lineCap, state->lineJoin, state->miterLimit);
+	if (ctx->params.edgeAntiAlias && state->shapeAntiAlias)
+		nvg__expandStroke(ctx, strokeWidth*0.5f, ctx->fringeWidth, state->lineCap, state->lineJoin, state->miterLimit);
 	else
-		nvg__expandStroke(ctx, strokeWidth*0.5f, state->lineCap, state->lineJoin, state->miterLimit);
+		nvg__expandStroke(ctx, strokeWidth*0.5f, 0.0f, state->lineCap, state->lineJoin, state->miterLimit);
 
-	ctx->params.renderStroke(ctx->params.userPtr, &strokePaint, &state->scissor, ctx->fringeWidth,
+	ctx->params.renderStroke(ctx->params.userPtr, &strokePaint, state->compositeOperation, &state->scissor, ctx->fringeWidth,
 							 strokeWidth, ctx->cache->paths, ctx->cache->npaths);
 
 	// Count triangles
@@ -2157,14 +2295,24 @@ void nvgStroke(NVGcontext* ctx)
 }
 
 // Add fonts
-int nvgCreateFont(NVGcontext* ctx, const char* name, const char* path)
+int nvgCreateFont(NVGcontext* ctx, const char* name, const char* filename)
+{
+	return fonsAddFont(ctx->fs, name, filename, 0);
+}
+
+int nvgCreateFontAtIndex(NVGcontext* ctx, const char* name, const char* filename, const int fontIndex)
 {
-	return fonsAddFont(ctx->fs, name, path);
+	return fonsAddFont(ctx->fs, name, filename, fontIndex);
 }
 
 int nvgCreateFontMem(NVGcontext* ctx, const char* name, unsigned char* data, int ndata, int freeData)
 {
-	return fonsAddFontMem(ctx->fs, name, data, ndata, freeData);
+	return fonsAddFontMem(ctx->fs, name, data, ndata, freeData, 0);
+}
+
+int nvgCreateFontMemAtIndex(NVGcontext* ctx, const char* name, unsigned char* data, int ndata, int freeData, const int fontIndex)
+{
+	return fonsAddFontMem(ctx->fs, name, data, ndata, freeData, fontIndex);
 }
 
 int nvgFindFont(NVGcontext* ctx, const char* name)
@@ -2173,6 +2321,28 @@ int nvgFindFont(NVGcontext* ctx, const char* name)
 	return fonsGetFontByName(ctx->fs, name);
 }
 
+
+int nvgAddFallbackFontId(NVGcontext* ctx, int baseFont, int fallbackFont)
+{
+	if(baseFont == -1 || fallbackFont == -1) return 0;
+	return fonsAddFallbackFont(ctx->fs, baseFont, fallbackFont);
+}
+
+int nvgAddFallbackFont(NVGcontext* ctx, const char* baseFont, const char* fallbackFont)
+{
+	return nvgAddFallbackFontId(ctx, nvgFindFont(ctx, baseFont), nvgFindFont(ctx, fallbackFont));
+}
+
+void nvgResetFallbackFontsId(NVGcontext* ctx, int baseFont)
+{
+	fonsResetFallbackFont(ctx->fs, baseFont);
+}
+
+void nvgResetFallbackFonts(NVGcontext* ctx, const char* baseFont)
+{
+	nvgResetFallbackFontsId(ctx, nvgFindFont(ctx, baseFont));
+}
+
 // State setting
 void nvgFontSize(NVGcontext* ctx, float size)
 {
@@ -2281,12 +2451,18 @@ static void nvg__renderText(NVGcontext* ctx, NVGvertex* verts, int nverts)
 	paint.innerColor.a *= state->alpha;
 	paint.outerColor.a *= state->alpha;
 
-	ctx->params.renderTriangles(ctx->params.userPtr, &paint, &state->scissor, verts, nverts);
+	ctx->params.renderTriangles(ctx->params.userPtr, &paint, state->compositeOperation, &state->scissor, verts, nverts, ctx->fringeWidth);
 
 	ctx->drawCallCount++;
 	ctx->textTriCount += nverts/3;
 }
 
+static int nvg__isTransformFlipped(const float *xform)
+{
+	float det = xform[0] * xform[3] - xform[2] * xform[1];
+	return( det < 0);
+}
+
 float nvgText(NVGcontext* ctx, float x, float y, const char* string, const char* end)
 {
 	NVGstate* state = nvg__getState(ctx);
@@ -2297,6 +2473,7 @@ float nvgText(NVGcontext* ctx, float x, float y, const char* string, const char*
 	float invscale = 1.0f / scale;
 	int cverts = 0;
 	int nverts = 0;
+	int isFlipped = nvg__isTransformFlipped(state->xform);
 
 	if (end == NULL)
 		end = string + strlen(string);
@@ -2313,23 +2490,29 @@ float nvgText(NVGcontext* ctx, float x, float y, const char* string, const char*
 	verts = nvg__allocTempVerts(ctx, cverts);
 	if (verts == NULL) return x;
 
-	fonsTextIterInit(ctx->fs, &iter, x*scale, y*scale, string, end);
+	fonsTextIterInit(ctx->fs, &iter, x*scale, y*scale, string, end, FONS_GLYPH_BITMAP_REQUIRED);
 	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;
 			}
+			if (!nvg__allocTextAtlas(ctx))
+				break; // no memory :(
 			iter = prevIter;
 			fonsTextIterNext(ctx->fs, &iter, &q); // try again
 			if (iter.prevGlyphIndex == -1) // still can not find glyph?
 				break;
 		}
 		prevIter = iter;
+		if(isFlipped) {
+			float tmp;
+
+			tmp = q.y0; q.y0 = q.y1; q.y1 = tmp;
+			tmp = q.t0; q.t0 = q.t1; q.t1 = tmp;
+		}
 		// 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);
@@ -2351,7 +2534,7 @@ float nvgText(NVGcontext* ctx, float x, float y, const char* string, const char*
 
 	nvg__renderText(ctx, verts, nverts);
 
-	return iter.x;
+	return iter.nextx / scale;
 }
 
 void nvgTextBox(NVGcontext* ctx, float x, float y, float breakRowWidth, const char* string, const char* end)
@@ -2360,7 +2543,7 @@ void nvgTextBox(NVGcontext* ctx, float x, float y, float breakRowWidth, const ch
 	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 halign = 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;
 
@@ -2373,11 +2556,11 @@ void nvgTextBox(NVGcontext* ctx, float x, float y, float breakRowWidth, const ch
 	while ((nrows = nvgTextBreakLines(ctx, string, end, breakRowWidth, rows, 2))) {
 		for (i = 0; i < nrows; i++) {
 			NVGtextRow* row = &rows[i];
-			if (haling & NVG_ALIGN_LEFT)
+			if (halign & NVG_ALIGN_LEFT)
 				nvgText(ctx, x, y, row->start, row->end);
-			else if (haling & NVG_ALIGN_CENTER)
+			else if (halign & NVG_ALIGN_CENTER)
 				nvgText(ctx, x + breakRowWidth*0.5f - row->width*0.5f, y, row->start, row->end);
-			else if (haling & NVG_ALIGN_RIGHT)
+			else if (halign & NVG_ALIGN_RIGHT)
 				nvgText(ctx, x + breakRowWidth - row->width, y, row->start, row->end);
 			y += lineh * state->lineHeight;
 		}
@@ -2410,7 +2593,7 @@ int nvgTextGlyphPositions(NVGcontext* ctx, float x, float y, const char* string,
 	fonsSetAlign(ctx->fs, state->textAlign);
 	fonsSetFont(ctx->fs, state->fontId);
 
-	fonsTextIterInit(ctx->fs, &iter, x*scale, y*scale, string, end);
+	fonsTextIterInit(ctx->fs, &iter, x*scale, y*scale, string, end, FONS_GLYPH_BITMAP_OPTIONAL);
 	prevIter = iter;
 	while (fonsTextIterNext(ctx->fs, &iter, &q)) {
 		if (iter.prevGlyphIndex < 0 && nvg__allocTextAtlas(ctx)) { // can not retrieve glyph?
@@ -2434,6 +2617,7 @@ enum NVGcodepointType {
 	NVG_SPACE,
 	NVG_NEWLINE,
 	NVG_CHAR,
+	NVG_CJK_CHAR,
 };
 
 int nvgTextBreakLines(NVGcontext* ctx, const char* string, const char* end, float breakRowWidth, NVGtextRow* rows, int maxRows)
@@ -2475,7 +2659,7 @@ int nvgTextBreakLines(NVGcontext* ctx, const char* string, const char* end, floa
 
 	breakRowWidth *= scale;
 
-	fonsTextIterInit(ctx->fs, &iter, 0, 0, string, end);
+	fonsTextIterInit(ctx->fs, &iter, 0, 0, string, end, FONS_GLYPH_BITMAP_OPTIONAL);
 	prevIter = iter;
 	while (fonsTextIterNext(ctx->fs, &iter, &q)) {
 		if (iter.prevGlyphIndex < 0 && nvg__allocTextAtlas(ctx)) { // can not retrieve glyph?
@@ -2501,7 +2685,15 @@ int nvgTextBreakLines(NVGcontext* ctx, const char* string, const char* end, floa
 				type = NVG_NEWLINE;
 				break;
 			default:
-				type = NVG_CHAR;
+				if ((iter.codepoint >= 0x4E00 && iter.codepoint <= 0x9FFF) ||
+					(iter.codepoint >= 0x3000 && iter.codepoint <= 0x30FF) ||
+					(iter.codepoint >= 0xFF00 && iter.codepoint <= 0xFFEF) ||
+					(iter.codepoint >= 0x1100 && iter.codepoint <= 0x11FF) ||
+					(iter.codepoint >= 0x3130 && iter.codepoint <= 0x318F) ||
+					(iter.codepoint >= 0xAC00 && iter.codepoint <= 0xD7AF))
+					type = NVG_CJK_CHAR;
+				else
+					type = NVG_CHAR;
 				break;
 		}
 
@@ -2528,12 +2720,12 @@ int nvgTextBreakLines(NVGcontext* ctx, const char* string, const char* end, floa
 		} else {
 			if (rowStart == NULL) {
 				// Skip white space until the beginning of the line
-				if (type == NVG_CHAR) {
+				if (type == NVG_CHAR || type == NVG_CJK_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;
+					rowWidth = iter.nextx - rowStartX;
 					rowMinX = q.x0 - rowStartX;
 					rowMaxX = q.x1 - rowStartX;
 					wordStart = iter.str;
@@ -2548,26 +2740,26 @@ int nvgTextBreakLines(NVGcontext* ctx, const char* string, const char* end, floa
 				float nextWidth = iter.nextx - rowStartX;
 
 				// track last non-white space character
-				if (type == NVG_CHAR) {
+				if (type == NVG_CHAR || type == NVG_CJK_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) {
+				if (((ptype == NVG_CHAR || ptype == NVG_CJK_CHAR) && type == NVG_SPACE) || type == NVG_CJK_CHAR) {
 					breakEnd = iter.str;
 					breakWidth = rowWidth;
 					breakMaxX = rowMaxX;
 				}
 				// track last beginning of a word
-				if (ptype == NVG_SPACE && type == NVG_CHAR) {
+				if ((ptype == NVG_SPACE && (type == NVG_CHAR || type == NVG_CJK_CHAR)) || type == NVG_CJK_CHAR) {
 					wordStart = iter.str;
 					wordStartX = iter.x;
-					wordMinX = q.x0 - rowStartX;
+					wordMinX = q.x0;
 				}
 
 				// Break to new line when a character is beyond break width.
-				if (type == NVG_CHAR && nextWidth > breakRowWidth) {
+				if ((type == NVG_CHAR || type == NVG_CJK_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.
@@ -2600,13 +2792,13 @@ int nvgTextBreakLines(NVGcontext* ctx, const char* string, const char* end, floa
 						nrows++;
 						if (nrows >= maxRows)
 							return nrows;
+						// Update row
 						rowStartX = wordStartX;
 						rowStart = wordStart;
 						rowEnd = iter.next;
 						rowWidth = iter.nextx - rowStartX;
-						rowMinX = wordMinX;
+						rowMinX = wordMinX - rowStartX;
 						rowMaxX = q.x1 - rowStartX;
-						// No change to the word start
 					}
 					// Set null break point
 					breakEnd = rowStart;
@@ -2669,7 +2861,7 @@ void nvgTextBoxBounds(NVGcontext* ctx, float x, float y, float breakRowWidth, co
 	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 halign = 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;
@@ -2701,11 +2893,11 @@ void nvgTextBoxBounds(NVGcontext* ctx, float x, float y, float breakRowWidth, co
 			NVGtextRow* row = &rows[i];
 			float rminx, rmaxx, dx = 0;
 			// Horizontal bounds
-			if (haling & NVG_ALIGN_LEFT)
+			if (halign & NVG_ALIGN_LEFT)
 				dx = 0;
-			else if (haling & NVG_ALIGN_CENTER)
+			else if (halign & NVG_ALIGN_CENTER)
 				dx = breakRowWidth*0.5f - row->width*0.5f;
-			else if (haling & NVG_ALIGN_RIGHT)
+			else if (halign & NVG_ALIGN_RIGHT)
 				dx = breakRowWidth - row->width;
 			rminx = x + row->minx + dx;
 			rmaxx = x + row->maxx + dx;