16 files changed, 3119 insertions, 622 deletions
diff --git a/build.bat b/build.bat
index c2735c1..17eb14e 100755
--- a/build.bat
+++ b/build.bat
@@ -8,7 +8,6 @@ REM /Zi /O2
 set OPTIMIZATION=/Zi
 set DEBUG=1
 set IMGUI=0
-set THREADED=1
 set WINDOWS=1
 set ARM=0
 set PACKEDRGB=0
@@ -21,15 +20,14 @@ set IMGUI_FILES=imgui\backends\imgui_impl_sdl.cpp imgui\backends\imgui_impl_open
 set WARNING_FLAGS=/W2 /wd4805 /wd4477 /wd4244 /wd4305
 
 if %DEBUG%==1 ( set PREPROCESSORS="/DDEBUG=1" )
-if %THREADED%==1 ( set PREPROCESSORS=%PREPROCESSORS% "/DTHREADED=1" )
 if %WINDOWS%==1 ( set PREPROCESSORS=%PREPROCESSORS% "/DWINDOWS=1" )
 if %ARM%==1 ( set PREPROCESSORS=%PREPROCESSORS% "/DARM=1" )
 if %PACKEDRGB%==1 ( set PREPROCESSORS=%PREPROCESSORS% "/DPACKEDRGB=1" )
 if %PERF%==1 ( set PREPROCESSORS=%PREPROCESSORS% "/DPERF=1" )
 
-cl /nologo %OPTIMIZATION% /MD /I.. /Iimgui /I%SDL2_DIR%\include my_imgui_internal_widgets.cpp /Fobin/ /c 
+cl /nologo %OPTIMIZATION% /MD /I.. /Iimgui /I%SDL2_DIR%\include my_imgui_internal_widgets.cpp /Fobin/ /c
 if %IMGUI%==1 ( cl /nologo %OPTIMIZATION% /MD /I.. /Iimgui /I%SDL2_DIR%\include %IMGUI_FILES% /Fobin/ /c )
 
-cl /nologo %OPTIMIZATION% /MD /I.. /Ilib/glad/include lib/glad.c /Fobin/ /c 
+cl /nologo %OPTIMIZATION% /MD /I.. /Ilib/glad/include lib/glad.c /Fobin/ /c
 
 cl /nologo /Zi /MD %WARNING_FLAGS% %PREPROCESSORS% %INCLUDES% main.cpp bin/*.obj /Febin/real2d.exe /Fobin/ /link %SDL_LIBS% %FFMPEG_LIBS% /subsystem:console
diff --git a/build.sh b/build.sh
index e0df0d4..31a9eed 100755
--- a/build.sh
+++ b/build.sh
@@ -1,13 +1,10 @@
 #!/bin/bash
 
-WINDOWS=0           # Compile for Windows with Mingw.
 ARM=0               # Compile on ARM machines.
 
 OPTIMIZATION="-O2"  # Enable optimization.
 DEBUG=0             # Compile with debug UI.
 IMGUI=1             # Compile ImGui libs. Our custom ImGui functions still compile on zero.
-THREADED=1          # Compile with threading. Useful to disable when stepping through the renderer.
-PACKEDRGB=0         # Use 4x4 chunks for the software rasterizer. Isn't actually faster at the moment.
 PERF=0              # Print cycle stats.
 
 # Add flag overrides to this file (it's untracked on the tree)
@@ -53,9 +50,6 @@ fi
 if [[ "$THREADED" == 1 ]]; then
 WARNING_FLAGS="$WARNING_FLAGS -DTHREADED=1"
 fi
-if [[ "$WINDOWS" == 1 ]]; then
-WARNING_FLAGS="$WARNING_FLAGS -DWINDOWS=1"
-fi
 if [[ "$ARM" == 1 ]]; then
 WARNING_FLAGS="$WARNING_FLAGS -DARM=1"
 ADDITIONAL_FLAGS="
@@ -66,9 +60,6 @@ ADDITIONAL_FLAGS="
     -march=native
 "
 fi
-if [[ "$PACKEDRGB" == 1 ]]; then
-WARNING_FLAGS="$WARNING_FLAGS -DPACKEDRGB=1"
-fi
 if [[ "$PERF" == 1 ]]; then
 WARNING_FLAGS="$WARNING_FLAGS -DPERF=1"
 fi
@@ -107,19 +98,12 @@ if [[ "$IMGUI" == 1 ]]; then
     done
 fi
 
-if [[ "$WINDOWS" == 1 ]]; then
-clang++ $OPTIMIZATION $WARNING_FLAGS -target x86_64-pc-windows-gnu -march=x86-64-v3 -I .. -Iimgui -Iimgui/backends \
-    main.cpp imgui/imgui*.cpp imgui/backends/imgui_impl_sdl.cpp imgui/backends/imgui_impl_opengl3.cpp \
-    -I/usr/x86_64-w64-mingw32/include/SDL2 -I/usr/x86_64-w64-mingw32/include/GL  \
-    -lmingw32 -lopengl32 -lSDL2main -lSDL2 -llibavcodec -llibswscale -llibavformat -llibavutil \
-    -o bin/real2d
-else
 clang lib/glad.c $GLAD_FLAGS  -I/usr/local/include -I/opt/local/include -c  \
     $WARNING_FLAGS $OPTIMIZATION $ADDITIONAL_FLAGS -o bin/glad.o
 clang main.cpp  $WARNING_FLAGS $OPTIMIZATION $ADDITIONAL_FLAGS -o bin/real2d bin/*.o  \
    $GLAD_FLAGS \
-   -std=c++11 -lstdc++ -lcurl -Iimgui -Iimgui/backends  \
+   -std=c++11 -lstdc++ -Iimgui -Iimgui/backends  \
    $SDL_ARGS \
    -I . \
    -lm -I /usr/local/include $(pkg-config --cflags --libs $FFMPEG_LIBS)
-fi
+# -lcurl
diff --git a/createcalls.cpp b/createcalls.cpp
index e88e39f..4c72c4e 100644
--- a/createcalls.cpp
+++ b/createcalls.cpp
@@ -38,6 +38,22 @@ Source_Delete(project_data *File, memory *Memory, uint32 Index)
     File->Source_Count--;
 }
 
+// These thumbnail textures aren't needed for anything else, so I'm just gonna
+// count on GL to retain them in GPU memory.
+static void
+Source_DumpThumbnail(memory *Memory, block_source *Source, uint32 T_Width, uint32 T_Height)
+{
+    Assert(Source->Type == source_type_principal_temp);
+    void *BitmapAddress = Memory_Block_AddressAtIndex(Memory, F_PrincipalBitmaps, Source->Bitmap_Index);
+    uint32 Size = T_Height*T_Width*4;
+    uint8 *Output = (uint8 *)Memory_PushScratch(Memory, Size);
+    stbir_resize_uint8((uint8 *)BitmapAddress, Source->Width, Source->Height, 0, Output, T_Height, T_Width, 0, 4);
+
+    GL_GenAndBindTexture(&Source->ThumbnailTex, T_Height, T_Width, 4, Output);
+
+    Memory_PopScratch(Memory, Size);
+}
+
 static int16
 Source_Generate(project_data *File, project_state *State, memory *Memory, void *TempString)
 {
@@ -67,10 +83,10 @@ Source_Generate(project_data *File, project_state *State, memory *Memory, void *
 }
 
 static bezier_point *
-Bezier_LookupAddress(memory *Memory, property_channel *Property, uint16 Index)
+Bezier_LookupAddress(memory *Memory, property_channel *Property, uint16 Index, bool32 AssertExists)
 {
     Assert(Index < MAX_KEYFRAMES_PER_BLOCK);
-    block_bezier *Bezier = (block_bezier *)Memory_Block_AddressAtIndex(Memory, F_Bezier, Property->Block_Bezier_Index[0]);
+    block_bezier *Bezier = (block_bezier *)Memory_Block_AddressAtIndex(Memory, F_Bezier, Property->Block_Bezier_Index[0], AssertExists);
     return &Bezier->Point[Index];
 }
 
@@ -100,7 +116,7 @@ Bezier_Add(memory *Memory, property_channel *Property, bezier_point PointData)
 {
     if (!Property->Block_Bezier_Count) {
         Property->Block_Bezier_Index[0] = Memory_Block_AllocateNew(Memory, F_Bezier);
-        block_bezier *Bezier = (block_bezier *)Memory_Block_AddressAtIndex(Memory, F_Bezier, Property->Block_Bezier_Index[0]);
+        block_bezier *Bezier = (block_bezier *)Memory_Block_AddressAtIndex(Memory, F_Bezier, Property->Block_Bezier_Index[0], 0);
         Bezier->Occupied = true;
         // NOTE(fox): Effects will change this!
         History_Action_Swap(Memory, F_Layers, sizeof(Property->Block_Bezier_Count), &Property->Block_Bezier_Count);
@@ -108,7 +124,7 @@ Bezier_Add(memory *Memory, property_channel *Property, bezier_point PointData)
     }
     int k = 0;
     for (;;) {
-        bezier_point *Point = Bezier_LookupAddress(Memory, Property, k);
+        bezier_point *Point = Bezier_LookupAddress(Memory, Property, k, 0);
         if (!Point->Occupied) {
             History_Action_Swap(Memory, F_Bezier, sizeof(*Point), Point);
             *Point = PointData;
@@ -182,6 +198,18 @@ Layer_Interact_Evaluate(memory *Memory, project_state *State, uint16 Layer_Index
 }
 
 static void
+Layer_ToggleChannel(project_data *File, memory *Memory, int32 a)
+{
+    int h = 0, c = 0, i = 0;
+    while (Block_Loop(Memory, F_Layers, File->Layer_Count, &h, &c, &i))
+    {
+        block_layer *Layer = (block_layer *)Memory_Block_AddressAtIndex(Memory, F_Layers, i);
+        if (Layer->IsSelected)
+            Layer->Property[a].IsToggled ^= 1;
+    }
+}
+
+static void
 Layer_Select(memory *Memory, project_state *State, int32 i)
 {
     block_layer *Layer = (block_layer *)Memory_Block_AddressAtIndex(Memory, F_Layers, i);
@@ -208,23 +236,47 @@ void Source_DeselectAll(project_data *File, memory *Memory)
     }
 }
 
-void Clipboard_Paste(project_data *File, project_state *State, memory *Memory, sorted_layer *SortedLayerArray)
+// NOTE(fox): This won't work with precomps!
+
+void Clipboard_Paste(project_data *File, project_state *State, memory *Memory, sorted_comp_info *SortedCompInfo, sorted_layer *SortedLayerInfo)
 {
     clipboard_contents *Contents = (clipboard_contents *)State->ClipboardBuffer;
+    if (Contents->Type == selection_none)
+        return;
     uint64 ClipboardPos = sizeof(clipboard_contents);
     ClipboardPos = sizeof(clipboard_contents);
-    int h = 0, c = 0, i = 0;
-    block_layer *Layer;
+    int i = SortedCompInfo->LayerCount - 1;
+    block_layer *Layer = NULL;
     clipboard_channel *Channel;
     int b = 0;
-    while (b < Contents->ChannelCount) {
+    int LayerCount = 0;
+
+    int NumberOfLayersFromClipboard = 1;
+    int LastOffset = 0;
+    for (int a = 0; a < Contents->ChannelCount; a++) {
+        Channel = &Contents->Channel[a];
+        if (a != 0) {
+            if (Channel->LayerOffset != LastOffset)
+                NumberOfLayersFromClipboard++;
+        }
+        LastOffset = Channel->LayerOffset;
+    }
+
+    for (;;) {
         Channel = &Contents->Channel[b];
-        while (Block_Loop(Memory, F_Layers, File->Layer_Count, &h, &c, &i))
+        while (i >= 0)
         {
-            Layer = (block_layer *)Memory_Block_AddressAtIndex(Memory, F_Layers, i);
-            if (Layer->IsSelected)
+            sorted_layer SortEntry = SortedLayerInfo[i];
+            uint32 Index_Physical = SortEntry.Block_Layer_Index;
+            block_layer *TestLayer = (block_layer *)Memory_Block_AddressAtIndex(Memory, F_Layers, i);
+            if (TestLayer->IsSelected) {
+                Layer = TestLayer;
                 break;
+            }
+            i--;
         }
+        if (Layer == NULL)
+            break;
         // NOTE(fox): This loop assumes all layers and the clipboard have
         // channels laid out in the same way!
         for (int h = 0; h < AmountOf(Layer->Property); h++) {
@@ -232,6 +284,7 @@ void Clipboard_Paste(project_data *File, project_state *State, memory *Memory, s
             if (Property->Name == Channel->Name) {
                 for (int p = 0; p < Channel->KeyframeCount; p++) {
                     bezier_point PointData = *(bezier_point *)((uint8 *)State->ClipboardBuffer + ClipboardPos);
+                    PointData.Pos[0].x += State->Frame_Current;
                     Bezier_Add(Memory, Property, PointData);
                     ClipboardPos += sizeof(bezier_point);
                 }
@@ -239,22 +292,26 @@ void Clipboard_Paste(project_data *File, project_state *State, memory *Memory, s
                 Channel = &Contents->Channel[b];
             }
         }
+        Layer = NULL;
+        if (b < Contents->ChannelCount) {
+            if (NumberOfLayersFromClipboard != 1)
+                break;
+            else
+                b = 0;
+        }
     }
 }
 
 
-void Clipboard_Store(project_data *File, project_state *State, memory *Memory, sorted_property_info *SortedPropertyInfo, uint16 *SortedPropertyArray)
+void Clipboard_Store(project_data *File, project_state *State, memory *Memory, sorted_comp_info *SortedCompInfo, sorted_layer *SortedLayerInfo, sorted_property_info *SortedPropertyInfo, uint16 *SortedPropertyArray)
 {
-    // TODO(fox): Multi-precomp support!
     int LocalOffset = 0;
     clipboard_contents *Contents = (clipboard_contents *)State->ClipboardBuffer;
     *Contents = {};
     uint64 ClipboardPos = sizeof(clipboard_contents);
-    // for (int i = SortedCompInfo.LayerCount - 1; i >= 0; i--)
-    Assert(0);
-    for (int i = 10; i >= 0; i--)
+    for (int i = SortedCompInfo->LayerCount - 1; i >= 0; i--)
     {
-        sorted_layer SortEntry = {}; // SortedLayerInfo[i];
+        sorted_layer SortEntry = SortedLayerInfo[i];
         uint32 Index_Physical = SortEntry.Block_Layer_Index;
         block_layer *Layer = (block_layer *)Memory_Block_AddressAtIndex(Memory, F_Layers, Index_Physical);
         for (int h = 0; h < AmountOf(Layer->Property); h++) {
@@ -263,10 +320,18 @@ void Clipboard_Store(project_data *File, project_state *State, memory *Memory, s
                 sorted_property_info *InfoLocation = SortedPropertyInfo + (i * 7) + h;
                 uint16 *ArrayLocation = SortedPropertyArray + (i * 7 * MAX_KEYFRAMES_PER_BLOCK) + (h * MAX_KEYFRAMES_PER_BLOCK);
                 clipboard_channel *Channel = &Contents->Channel[Contents->ChannelCount];
+                bezier_point *FirstPoint = NULL;
+                int TimeOffset = 0;
                 for (int p = 0; p < Property->Keyframe_Count; p++) {
                     bezier_point *PointAddress = Bezier_LookupAddress(Memory, Property, ArrayLocation[p]);
                     if (PointAddress->IsSelected) {
-                        Memory_Copy((uint8 *)State->ClipboardBuffer + ClipboardPos, (uint8 *)PointAddress, sizeof(bezier_point));
+                        if (!FirstPoint) {
+                            FirstPoint = PointAddress;
+                            TimeOffset = FirstPoint->Pos[0].x;
+                        }
+                        bezier_point PointToCopy = *PointAddress;
+                        PointToCopy.Pos[0].x -= TimeOffset;
+                        Memory_Copy((uint8 *)State->ClipboardBuffer + ClipboardPos, (uint8 *)&PointToCopy, sizeof(bezier_point));
                         ClipboardPos += sizeof(bezier_point);
                         Channel->KeyframeCount++;
                     }
@@ -486,6 +551,35 @@ void Layer_Evaluate_Display(block_layer *Layer, sorted_layer *LayerArrayStart, s
     */
 }
 
+void TempSource_SortAll(project_data *File, project_state *State, memory *Memory, uint16 *SourceArrayStart, uint16 *TempSourceCount)
+{
+    int count = 0;
+    int h = 0, c = 0, i = 0;
+    while (Block_Loop(Memory, F_Sources, File->Source_Count, &h, &c, &i)) {
+        block_source *Source = (block_source *)Memory_Block_AddressAtIndex(Memory, F_Sources, i);
+        if (Source->Type == source_type_principal_temp) {
+            uint32 Playhead = 0;
+            while (Playhead < count) {
+                block_source *TestSource = (block_source *)Memory_Block_AddressAtIndex(Memory, F_Sources, SourceArrayStart[Playhead]);
+                Assert(TestSource->Type == source_type_principal_temp);
+                if (TestSource->RelativeTimestamp > Source->RelativeTimestamp) {
+                    break;
+                } else {
+                   Playhead++;
+                }
+            }
+            if (Playhead != count) {
+                uint8 *Address_Start = (uint8 *)(SourceArrayStart + Playhead);
+                uint8 *Address_End = (uint8 *)(SourceArrayStart + count) - 1;
+                Arbitrary_ShiftData(Address_Start, Address_End, sizeof(uint16), 1);
+            }
+            SourceArrayStart[Playhead] = i;
+            count++;
+        }
+    }
+    *TempSourceCount = count;
+}
+
 // The first loop is for counting how many layers are in each precomp, the
 // second is for sorting the layers by offset, and the third is for applying
 // interactivity if the user is moving any layers.
@@ -591,12 +685,20 @@ sorted_file File_Sort_Push(project_data *File, project_state *State, memory *Mem
     Sorted.PropertyInfo = (sorted_property_info *)Property_SortedArray;
     Sorted.PropertyArray = (uint16 *)((uint8 *)Property_SortedArray + PropertyInfoSize);
 
+    uint64 SourceArraySize = sizeof(uint16) * File->Source_Count;
+    Sorted.Source_SortSize = SourceArraySize;
+    void *Source_SortedArray = Memory_PushScratch(Memory, Sorted.Source_SortSize);
+    Arbitrary_Zero((uint8 *)Source_SortedArray, Sorted.Source_SortSize);
+    Sorted.SourceArray = (uint16 *)Source_SortedArray;
+
+    TempSource_SortAll(File, State, Memory, Sorted.SourceArray, &Sorted.TempSourceCount);
     Layer_SortAll(File, State, Memory, Sorted.LayerArray, Sorted.CompArray, Sorted.PropertyInfo, Sorted.PropertyArray, File->Layer_Count, File->Comp_Count);
     return Sorted;
 }
 
-void File_Sort_Pop(memory *Memory, uint64 Layer_SortSize, uint64 Property_SortSize)
+void File_Sort_Pop(memory *Memory, uint64 Layer_SortSize, uint64 Property_SortSize, uint64 Source_SortSize)
 {
+    Memory_PopScratch(Memory, Source_SortSize);
     Memory_PopScratch(Memory, Property_SortSize);
     Memory_PopScratch(Memory, Layer_SortSize);
 }
@@ -844,6 +946,7 @@ Brush_Info(brush_info *B, brush_state *Brush, block_source *Source, v2 LayerPos,
     }
 
     B->BrushBuffer = Brush->PaintBuffer;
+    B->EraseMode = Brush->EraseMode;
 
     // ImGui's color picker works in sRGB, so we need to convert to linear.
     // real32 R_Brush = (Color.r >= 0.04045) ? pow((Color.r + 0.055) / (1 + 0.055), 2.4) : Color.r / 12.92;
@@ -890,11 +993,22 @@ PaintTest(brush_info B, void *Buffer, rectangle RenderRegion)
             real32 BrushAlpha = Brush_BitmapAlpha * B.A_Brush;
             real32 LayerAlpha = A_Layer;
 
-            real32 A_Blend = LayerAlpha + BrushAlpha;
+            real32 A_Blend = 0;
+            real32 R_Blend = 0;
+            real32 G_Blend = 0;
+            real32 B_Blend = 0;
 
-            real32 R_Blend = (R_Layer * (1.0f - BrushAlpha)) + (B.R_Brush * BrushAlpha);
-            real32 G_Blend = (G_Layer * (1.0f - BrushAlpha)) + (B.G_Brush * BrushAlpha);
-            real32 B_Blend = (B_Layer * (1.0f - BrushAlpha)) + (B.B_Brush * BrushAlpha);
+            if (!B.EraseMode) {
+                A_Blend = LayerAlpha + BrushAlpha;
+                R_Blend = (R_Layer * (1.0f - BrushAlpha)) + (B.R_Brush * BrushAlpha);
+                G_Blend = (G_Layer * (1.0f - BrushAlpha)) + (B.G_Brush * BrushAlpha);
+                B_Blend = (B_Layer * (1.0f - BrushAlpha)) + (B.B_Brush * BrushAlpha);
+            } else {
+                A_Blend = A_Layer * (1.0f - BrushAlpha);
+                R_Blend = R_Layer;
+                G_Blend = G_Layer;
+                B_Blend = B_Layer;
+            }
 
             /*
             R_Blend = (R_Brush * (1.0f - LayerAlpha)) + (R_Blend * LayerAlpha);
diff --git a/debug.h b/debug.h
index 202eb20..6860bdb 100644
--- a/debug.h
+++ b/debug.h
@@ -28,6 +28,7 @@ struct project_debug
 {
     debug_temp Temp;
     bool32 ToggleWindow = 1;
+    bool32 NoThreading = 0;
     uint64 PixelCountTransparent;
     uint64 PixelCountRendered;
     uint64 PixelCountChecked;
diff --git a/functions.h b/functions.h
index 69ce817..9ac14d4 100644
--- a/functions.h
+++ b/functions.h
@@ -16,7 +16,7 @@ static ImVec2 Layer_LocalToScreenSpace(project_state *State, memory *Memory, blo
 
 static v2 TransformPoint(layer_transforms T, real32 Width, real32 Height, v2 Point);
 
-static bezier_point * Bezier_LookupAddress(memory *Memory, property_channel *Property, uint16 Index);
+static bezier_point * Bezier_LookupAddress(memory *Memory, property_channel *Property, uint16 Index, bool32 AssertExists = 1);
 static void           Bezier_EvaluateValue(project_state *State, bezier_point *PointAddress, v2 *Pos, real32 Y_Increment);
 
 static void Transform_ApplyInteractive(interact_transform Interact, real32 *OutputX, real32 *OutputY, real32 *OutputRotation, real32 *OutputScale);
diff --git a/lib/base64.c b/lib/base64.c
index b6a8e5d..07caa0a 100644
--- a/lib/base64.c
+++ b/lib/base64.c
@@ -58,7 +58,7 @@ size_t base64_encode_size(size_t len)
 {
 	size_t olen;
 	olen = len * 4 / 3 + 4; /* 3-byte blocks to 4-byte */
-	olen += olen / 72; /* line feeds */
+	// olen += olen / 72; /* line feeds */
 	olen++; /* nul termination */
 	Assert(olen >= len);
     return olen;
@@ -85,7 +85,7 @@ void base64_encode(const unsigned char *src, size_t len,
 		in += 3;
 		line_len += 4;
 		if (line_len >= 72) {
-			*pos++ = '\n';
+			// *pos++ = '\n';
 			line_len = 0;
 		}
 	}
diff --git a/lib/stb_image_resize.h b/lib/stb_image_resize.h
new file mode 100644
index 0000000..ef9e6fe
--- /dev/null
+++ b/lib/stb_image_resize.h
@@ -0,0 +1,2634 @@
+/* stb_image_resize - v0.97 - public domain image resizing
+   by Jorge L Rodriguez (@VinoBS) - 2014
+   http://github.com/nothings/stb
+
+   Written with emphasis on usability, portability, and efficiency. (No
+   SIMD or threads, so it be easily outperformed by libs that use those.)
+   Only scaling and translation is supported, no rotations or shears.
+   Easy API downsamples w/Mitchell filter, upsamples w/cubic interpolation.
+
+   COMPILING & LINKING
+      In one C/C++ file that #includes this file, do this:
+         #define STB_IMAGE_RESIZE_IMPLEMENTATION
+      before the #include. That will create the implementation in that file.
+
+   QUICKSTART
+      stbir_resize_uint8(      input_pixels , in_w , in_h , 0,
+                               output_pixels, out_w, out_h, 0, num_channels)
+      stbir_resize_float(...)
+      stbir_resize_uint8_srgb( input_pixels , in_w , in_h , 0,
+                               output_pixels, out_w, out_h, 0,
+                               num_channels , alpha_chan  , 0)
+      stbir_resize_uint8_srgb_edgemode(
+                               input_pixels , in_w , in_h , 0,
+                               output_pixels, out_w, out_h, 0,
+                               num_channels , alpha_chan  , 0, STBIR_EDGE_CLAMP)
+                                                            // WRAP/REFLECT/ZERO
+
+   FULL API
+      See the "header file" section of the source for API documentation.
+
+   ADDITIONAL DOCUMENTATION
+
+      SRGB & FLOATING POINT REPRESENTATION
+         The sRGB functions presume IEEE floating point. If you do not have
+         IEEE floating point, define STBIR_NON_IEEE_FLOAT. This will use
+         a slower implementation.
+
+      MEMORY ALLOCATION
+         The resize functions here perform a single memory allocation using
+         malloc. To control the memory allocation, before the #include that
+         triggers the implementation, do:
+
+            #define STBIR_MALLOC(size,context) ...
+            #define STBIR_FREE(ptr,context)   ...
+
+         Each resize function makes exactly one call to malloc/free, so to use
+         temp memory, store the temp memory in the context and return that.
+
+      ASSERT
+         Define STBIR_ASSERT(boolval) to override assert() and not use assert.h
+
+      OPTIMIZATION
+         Define STBIR_SATURATE_INT to compute clamp values in-range using
+         integer operations instead of float operations. This may be faster
+         on some platforms.
+
+      DEFAULT FILTERS
+         For functions which don't provide explicit control over what filters
+         to use, you can change the compile-time defaults with
+
+            #define STBIR_DEFAULT_FILTER_UPSAMPLE     STBIR_FILTER_something
+            #define STBIR_DEFAULT_FILTER_DOWNSAMPLE   STBIR_FILTER_something
+
+         See stbir_filter in the header-file section for the list of filters.
+
+      NEW FILTERS
+         A number of 1D filter kernels are used. For a list of
+         supported filters see the stbir_filter enum. To add a new filter,
+         write a filter function and add it to stbir__filter_info_table.
+
+      PROGRESS
+         For interactive use with slow resize operations, you can install
+         a progress-report callback:
+
+            #define STBIR_PROGRESS_REPORT(val)   some_func(val)
+
+         The parameter val is a float which goes from 0 to 1 as progress is made.
+
+         For example:
+
+            static void my_progress_report(float progress);
+            #define STBIR_PROGRESS_REPORT(val) my_progress_report(val)
+
+            #define STB_IMAGE_RESIZE_IMPLEMENTATION
+            #include "stb_image_resize.h"
+
+            static void my_progress_report(float progress)
+            {
+               printf("Progress: %f%%\n", progress*100);
+            }
+
+      MAX CHANNELS
+         If your image has more than 64 channels, define STBIR_MAX_CHANNELS
+         to the max you'll have.
+
+      ALPHA CHANNEL
+         Most of the resizing functions provide the ability to control how
+         the alpha channel of an image is processed. The important things
+         to know about this:
+
+         1. The best mathematically-behaved version of alpha to use is
+         called "premultiplied alpha", in which the other color channels
+         have had the alpha value multiplied in. If you use premultiplied
+         alpha, linear filtering (such as image resampling done by this
+         library, or performed in texture units on GPUs) does the "right
+         thing". While premultiplied alpha is standard in the movie CGI
+         industry, it is still uncommon in the videogame/real-time world.
+
+         If you linearly filter non-premultiplied alpha, strange effects
+         occur. (For example, the 50/50 average of 99% transparent bright green
+         and 1% transparent black produces 50% transparent dark green when
+         non-premultiplied, whereas premultiplied it produces 50%
+         transparent near-black. The former introduces green energy
+         that doesn't exist in the source image.)
+
+         2. Artists should not edit premultiplied-alpha images; artists
+         want non-premultiplied alpha images. Thus, art tools generally output
+         non-premultiplied alpha images.
+
+         3. You will get best results in most cases by converting images
+         to premultiplied alpha before processing them mathematically.
+
+         4. If you pass the flag STBIR_FLAG_ALPHA_PREMULTIPLIED, the
+         resizer does not do anything special for the alpha channel;
+         it is resampled identically to other channels. This produces
+         the correct results for premultiplied-alpha images, but produces
+         less-than-ideal results for non-premultiplied-alpha images.
+
+         5. If you do not pass the flag STBIR_FLAG_ALPHA_PREMULTIPLIED,
+         then the resizer weights the contribution of input pixels
+         based on their alpha values, or, equivalently, it multiplies
+         the alpha value into the color channels, resamples, then divides
+         by the resultant alpha value. Input pixels which have alpha=0 do
+         not contribute at all to output pixels unless _all_ of the input
+         pixels affecting that output pixel have alpha=0, in which case
+         the result for that pixel is the same as it would be without
+         STBIR_FLAG_ALPHA_PREMULTIPLIED. However, this is only true for
+         input images in integer formats. For input images in float format,
+         input pixels with alpha=0 have no effect, and output pixels
+         which have alpha=0 will be 0 in all channels. (For float images,
+         you can manually achieve the same result by adding a tiny epsilon
+         value to the alpha channel of every image, and then subtracting
+         or clamping it at the end.)
+
+         6. You can suppress the behavior described in #5 and make
+         all-0-alpha pixels have 0 in all channels by #defining
+         STBIR_NO_ALPHA_EPSILON.
+
+         7. You can separately control whether the alpha channel is
+         interpreted as linear or affected by the colorspace. By default
+         it is linear; you almost never want to apply the colorspace.
+         (For example, graphics hardware does not apply sRGB conversion
+         to the alpha channel.)
+
+   CONTRIBUTORS
+      Jorge L Rodriguez: Implementation
+      Sean Barrett: API design, optimizations
+      Aras Pranckevicius: bugfix
+      Nathan Reed: warning fixes
+
+   REVISIONS
+      0.97 (2020-02-02) fixed warning
+      0.96 (2019-03-04) fixed warnings
+      0.95 (2017-07-23) fixed warnings
+      0.94 (2017-03-18) fixed warnings
+      0.93 (2017-03-03) fixed bug with certain combinations of heights
+      0.92 (2017-01-02) fix integer overflow on large (>2GB) images
+      0.91 (2016-04-02) fix warnings; fix handling of subpixel regions
+      0.90 (2014-09-17) first released version
+
+   LICENSE
+     See end of file for license information.
+
+   TODO
+      Don't decode all of the image data when only processing a partial tile
+      Don't use full-width decode buffers when only processing a partial tile
+      When processing wide images, break processing into tiles so data fits in L1 cache
+      Installable filters?
+      Resize that respects alpha test coverage
+         (Reference code: FloatImage::alphaTestCoverage and FloatImage::scaleAlphaToCoverage:
+         https://code.google.com/p/nvidia-texture-tools/source/browse/trunk/src/nvimage/FloatImage.cpp )
+*/
+
+#ifndef STBIR_INCLUDE_STB_IMAGE_RESIZE_H
+#define STBIR_INCLUDE_STB_IMAGE_RESIZE_H
+
+#ifdef _MSC_VER
+typedef unsigned char  stbir_uint8;
+typedef unsigned short stbir_uint16;
+typedef unsigned int   stbir_uint32;
+#else
+#include <stdint.h>
+typedef uint8_t  stbir_uint8;
+typedef uint16_t stbir_uint16;
+typedef uint32_t stbir_uint32;
+#endif
+
+#ifndef STBIRDEF
+#ifdef STB_IMAGE_RESIZE_STATIC
+#define STBIRDEF static
+#else
+#ifdef __cplusplus
+#define STBIRDEF extern "C"
+#else
+#define STBIRDEF extern
+#endif
+#endif
+#endif
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// Easy-to-use API:
+//
+//     * "input pixels" points to an array of image data with 'num_channels' channels (e.g. RGB=3, RGBA=4)
+//     * input_w is input image width (x-axis), input_h is input image height (y-axis)
+//     * stride is the offset between successive rows of image data in memory, in bytes. you can
+//       specify 0 to mean packed continuously in memory
+//     * alpha channel is treated identically to other channels.
+//     * colorspace is linear or sRGB as specified by function name
+//     * returned result is 1 for success or 0 in case of an error.
+//       #define STBIR_ASSERT() to trigger an assert on parameter validation errors.
+//     * Memory required grows approximately linearly with input and output size, but with
+//       discontinuities at input_w == output_w and input_h == output_h.
+//     * These functions use a "default" resampling filter defined at compile time. To change the filter,
+//       you can change the compile-time defaults by #defining STBIR_DEFAULT_FILTER_UPSAMPLE
+//       and STBIR_DEFAULT_FILTER_DOWNSAMPLE, or you can use the medium-complexity API.
+
+STBIRDEF int stbir_resize_uint8(     const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+                                           unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+                                     int num_channels);
+
+STBIRDEF int stbir_resize_float(     const float *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+                                           float *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+                                     int num_channels);
+
+
+// The following functions interpret image data as gamma-corrected sRGB.
+// Specify STBIR_ALPHA_CHANNEL_NONE if you have no alpha channel,
+// or otherwise provide the index of the alpha channel. Flags value
+// of 0 will probably do the right thing if you're not sure what
+// the flags mean.
+
+#define STBIR_ALPHA_CHANNEL_NONE       -1
+
+// Set this flag if your texture has premultiplied alpha. Otherwise, stbir will
+// use alpha-weighted resampling (effectively premultiplying, resampling,
+// then unpremultiplying).
+#define STBIR_FLAG_ALPHA_PREMULTIPLIED    (1 << 0)
+// The specified alpha channel should be handled as gamma-corrected value even
+// when doing sRGB operations.
+#define STBIR_FLAG_ALPHA_USES_COLORSPACE  (1 << 1)
+
+STBIRDEF int stbir_resize_uint8_srgb(const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+                                           unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+                                     int num_channels, int alpha_channel, int flags);
+
+
+typedef enum
+{
+    STBIR_EDGE_CLAMP   = 1,
+    STBIR_EDGE_REFLECT = 2,
+    STBIR_EDGE_WRAP    = 3,
+    STBIR_EDGE_ZERO    = 4,
+} stbir_edge;
+
+// This function adds the ability to specify how requests to sample off the edge of the image are handled.
+STBIRDEF int stbir_resize_uint8_srgb_edgemode(const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+                                                    unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+                                              int num_channels, int alpha_channel, int flags,
+                                              stbir_edge edge_wrap_mode);
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// Medium-complexity API
+//
+// This extends the easy-to-use API as follows:
+//
+//     * Alpha-channel can be processed separately
+//       * If alpha_channel is not STBIR_ALPHA_CHANNEL_NONE
+//         * Alpha channel will not be gamma corrected (unless flags&STBIR_FLAG_GAMMA_CORRECT)
+//         * Filters will be weighted by alpha channel (unless flags&STBIR_FLAG_ALPHA_PREMULTIPLIED)
+//     * Filter can be selected explicitly
+//     * uint16 image type
+//     * sRGB colorspace available for all types
+//     * context parameter for passing to STBIR_MALLOC
+
+typedef enum
+{
+    STBIR_FILTER_DEFAULT      = 0,  // use same filter type that easy-to-use API chooses
+    STBIR_FILTER_BOX          = 1,  // A trapezoid w/1-pixel wide ramps, same result as box for integer scale ratios
+    STBIR_FILTER_TRIANGLE     = 2,  // On upsampling, produces same results as bilinear texture filtering
+    STBIR_FILTER_CUBICBSPLINE = 3,  // The cubic b-spline (aka Mitchell-Netrevalli with B=1,C=0), gaussian-esque
+    STBIR_FILTER_CATMULLROM   = 4,  // An interpolating cubic spline
+    STBIR_FILTER_MITCHELL     = 5,  // Mitchell-Netrevalli filter with B=1/3, C=1/3
+} stbir_filter;
+
+typedef enum
+{
+    STBIR_COLORSPACE_LINEAR,
+    STBIR_COLORSPACE_SRGB,
+
+    STBIR_MAX_COLORSPACES,
+} stbir_colorspace;
+
+// The following functions are all identical except for the type of the image data
+
+STBIRDEF int stbir_resize_uint8_generic( const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+                                               unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+                                         int num_channels, int alpha_channel, int flags,
+                                         stbir_edge edge_wrap_mode, stbir_filter filter, stbir_colorspace space,
+                                         void *alloc_context);
+
+STBIRDEF int stbir_resize_uint16_generic(const stbir_uint16 *input_pixels  , int input_w , int input_h , int input_stride_in_bytes,
+                                               stbir_uint16 *output_pixels , int output_w, int output_h, int output_stride_in_bytes,
+                                         int num_channels, int alpha_channel, int flags,
+                                         stbir_edge edge_wrap_mode, stbir_filter filter, stbir_colorspace space,
+                                         void *alloc_context);
+
+STBIRDEF int stbir_resize_float_generic( const float *input_pixels         , int input_w , int input_h , int input_stride_in_bytes,
+                                               float *output_pixels        , int output_w, int output_h, int output_stride_in_bytes,
+                                         int num_channels, int alpha_channel, int flags,
+                                         stbir_edge edge_wrap_mode, stbir_filter filter, stbir_colorspace space,
+                                         void *alloc_context);
+
+
+
+//////////////////////////////////////////////////////////////////////////////
+//
+// Full-complexity API
+//
+// This extends the medium API as follows:
+//
+//       * uint32 image type
+//     * not typesafe
+//     * separate filter types for each axis
+//     * separate edge modes for each axis
+//     * can specify scale explicitly for subpixel correctness
+//     * can specify image source tile using texture coordinates
+
+typedef enum
+{
+    STBIR_TYPE_UINT8 ,
+    STBIR_TYPE_UINT16,
+    STBIR_TYPE_UINT32,
+    STBIR_TYPE_FLOAT ,
+
+    STBIR_MAX_TYPES
+} stbir_datatype;
+
+STBIRDEF int stbir_resize(         const void *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+                                         void *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+                                   stbir_datatype datatype,
+                                   int num_channels, int alpha_channel, int flags,
+                                   stbir_edge edge_mode_horizontal, stbir_edge edge_mode_vertical,
+                                   stbir_filter filter_horizontal,  stbir_filter filter_vertical,
+                                   stbir_colorspace space, void *alloc_context);
+
+STBIRDEF int stbir_resize_subpixel(const void *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+                                         void *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+                                   stbir_datatype datatype,
+                                   int num_channels, int alpha_channel, int flags,
+                                   stbir_edge edge_mode_horizontal, stbir_edge edge_mode_vertical,
+                                   stbir_filter filter_horizontal,  stbir_filter filter_vertical,
+                                   stbir_colorspace space, void *alloc_context,
+                                   float x_scale, float y_scale,
+                                   float x_offset, float y_offset);
+
+STBIRDEF int stbir_resize_region(  const void *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+                                         void *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+                                   stbir_datatype datatype,
+                                   int num_channels, int alpha_channel, int flags,
+                                   stbir_edge edge_mode_horizontal, stbir_edge edge_mode_vertical,
+                                   stbir_filter filter_horizontal,  stbir_filter filter_vertical,
+                                   stbir_colorspace space, void *alloc_context,
+                                   float s0, float t0, float s1, float t1);
+// (s0, t0) & (s1, t1) are the top-left and bottom right corner (uv addressing style: [0, 1]x[0, 1]) of a region of the input image to use.
+
+//
+//
+////   end header file   /////////////////////////////////////////////////////
+#endif // STBIR_INCLUDE_STB_IMAGE_RESIZE_H
+
+
+
+
+
+#ifdef STB_IMAGE_RESIZE_IMPLEMENTATION
+
+#ifndef STBIR_ASSERT
+#include <assert.h>
+#define STBIR_ASSERT(x) assert(x)
+#endif
+
+// For memset
+#include <string.h>
+
+#include <math.h>
+
+#ifndef STBIR_MALLOC
+#include <stdlib.h>
+// use comma operator to evaluate c, to avoid "unused parameter" warnings
+#define STBIR_MALLOC(size,c) ((void)(c), malloc(size))
+#define STBIR_FREE(ptr,c)    ((void)(c), free(ptr))
+#endif
+
+#ifndef _MSC_VER
+#ifdef __cplusplus
+#define stbir__inline inline
+#else
+#define stbir__inline
+#endif
+#else
+#define stbir__inline __forceinline
+#endif
+
+
+// should produce compiler error if size is wrong
+typedef unsigned char stbir__validate_uint32[sizeof(stbir_uint32) == 4 ? 1 : -1];
+
+#ifdef _MSC_VER
+#define STBIR__NOTUSED(v)  (void)(v)
+#else
+#define STBIR__NOTUSED(v)  (void)sizeof(v)
+#endif
+
+#define STBIR__ARRAY_SIZE(a) (sizeof((a))/sizeof((a)[0]))
+
+#ifndef STBIR_DEFAULT_FILTER_UPSAMPLE
+#define STBIR_DEFAULT_FILTER_UPSAMPLE    STBIR_FILTER_CATMULLROM
+#endif
+
+#ifndef STBIR_DEFAULT_FILTER_DOWNSAMPLE
+#define STBIR_DEFAULT_FILTER_DOWNSAMPLE  STBIR_FILTER_MITCHELL
+#endif
+
+#ifndef STBIR_PROGRESS_REPORT
+#define STBIR_PROGRESS_REPORT(float_0_to_1)
+#endif
+
+#ifndef STBIR_MAX_CHANNELS
+#define STBIR_MAX_CHANNELS 64
+#endif
+
+#if STBIR_MAX_CHANNELS > 65536
+#error "Too many channels; STBIR_MAX_CHANNELS must be no more than 65536."
+// because we store the indices in 16-bit variables
+#endif
+
+// This value is added to alpha just before premultiplication to avoid
+// zeroing out color values. It is equivalent to 2^-80. If you don't want
+// that behavior (it may interfere if you have floating point images with
+// very small alpha values) then you can define STBIR_NO_ALPHA_EPSILON to
+// disable it.
+#ifndef STBIR_ALPHA_EPSILON
+#define STBIR_ALPHA_EPSILON ((float)1 / (1 << 20) / (1 << 20) / (1 << 20) / (1 << 20))
+#endif
+
+
+
+#ifdef _MSC_VER
+#define STBIR__UNUSED_PARAM(v)  (void)(v)
+#else
+#define STBIR__UNUSED_PARAM(v)  (void)sizeof(v)
+#endif
+
+// must match stbir_datatype
+static unsigned char stbir__type_size[] = {
+    1, // STBIR_TYPE_UINT8
+    2, // STBIR_TYPE_UINT16
+    4, // STBIR_TYPE_UINT32
+    4, // STBIR_TYPE_FLOAT
+};
+
+// Kernel function centered at 0
+typedef float (stbir__kernel_fn)(float x, float scale);
+typedef float (stbir__support_fn)(float scale);
+
+typedef struct
+{
+    stbir__kernel_fn* kernel;
+    stbir__support_fn* support;
+} stbir__filter_info;
+
+// When upsampling, the contributors are which source pixels contribute.
+// When downsampling, the contributors are which destination pixels are contributed to.
+typedef struct
+{
+    int n0; // First contributing pixel
+    int n1; // Last contributing pixel
+} stbir__contributors;
+
+typedef struct
+{
+    const void* input_data;
+    int input_w;
+    int input_h;
+    int input_stride_bytes;
+
+    void* output_data;
+    int output_w;
+    int output_h;
+    int output_stride_bytes;
+
+    float s0, t0, s1, t1;
+
+    float horizontal_shift; // Units: output pixels
+    float vertical_shift;   // Units: output pixels
+    float horizontal_scale;
+    float vertical_scale;
+
+    int channels;
+    int alpha_channel;
+    stbir_uint32 flags;
+    stbir_datatype type;
+    stbir_filter horizontal_filter;
+    stbir_filter vertical_filter;
+    stbir_edge edge_horizontal;
+    stbir_edge edge_vertical;
+    stbir_colorspace colorspace;
+
+    stbir__contributors* horizontal_contributors;
+    float* horizontal_coefficients;
+
+    stbir__contributors* vertical_contributors;
+    float* vertical_coefficients;
+
+    int decode_buffer_pixels;
+    float* decode_buffer;
+
+    float* horizontal_buffer;
+
+    // cache these because ceil/floor are inexplicably showing up in profile
+    int horizontal_coefficient_width;
+    int vertical_coefficient_width;
+    int horizontal_filter_pixel_width;
+    int vertical_filter_pixel_width;
+    int horizontal_filter_pixel_margin;
+    int vertical_filter_pixel_margin;
+    int horizontal_num_contributors;
+    int vertical_num_contributors;
+
+    int ring_buffer_length_bytes;   // The length of an individual entry in the ring buffer. The total number of ring buffers is stbir__get_filter_pixel_width(filter)
+    int ring_buffer_num_entries;    // Total number of entries in the ring buffer.
+    int ring_buffer_first_scanline;
+    int ring_buffer_last_scanline;
+    int ring_buffer_begin_index;    // first_scanline is at this index in the ring buffer
+    float* ring_buffer;
+
+    float* encode_buffer; // A temporary buffer to store floats so we don't lose precision while we do multiply-adds.
+
+    int horizontal_contributors_size;
+    int horizontal_coefficients_size;
+    int vertical_contributors_size;
+    int vertical_coefficients_size;
+    int decode_buffer_size;
+    int horizontal_buffer_size;
+    int ring_buffer_size;
+    int encode_buffer_size;
+} stbir__info;
+
+
+static const float stbir__max_uint8_as_float  = 255.0f;
+static const float stbir__max_uint16_as_float = 65535.0f;
+static const double stbir__max_uint32_as_float = 4294967295.0;
+
+
+static stbir__inline int stbir__min(int a, int b)
+{
+    return a < b ? a : b;
+}
+
+static stbir__inline float stbir__saturate(float x)
+{
+    if (x < 0)
+        return 0;
+
+    if (x > 1)
+        return 1;
+
+    return x;
+}
+
+#ifdef STBIR_SATURATE_INT
+static stbir__inline stbir_uint8 stbir__saturate8(int x)
+{
+    if ((unsigned int) x <= 255)
+        return x;
+
+    if (x < 0)
+        return 0;
+
+    return 255;
+}
+
+static stbir__inline stbir_uint16 stbir__saturate16(int x)
+{
+    if ((unsigned int) x <= 65535)
+        return x;
+
+    if (x < 0)
+        return 0;
+
+    return 65535;
+}
+#endif
+
+static float stbir__srgb_uchar_to_linear_float[256] = {
+    0.000000f, 0.000304f, 0.000607f, 0.000911f, 0.001214f, 0.001518f, 0.001821f, 0.002125f, 0.002428f, 0.002732f, 0.003035f,
+    0.003347f, 0.003677f, 0.004025f, 0.004391f, 0.004777f, 0.005182f, 0.005605f, 0.006049f, 0.006512f, 0.006995f, 0.007499f,
+    0.008023f, 0.008568f, 0.009134f, 0.009721f, 0.010330f, 0.010960f, 0.011612f, 0.012286f, 0.012983f, 0.013702f, 0.014444f,
+    0.015209f, 0.015996f, 0.016807f, 0.017642f, 0.018500f, 0.019382f, 0.020289f, 0.021219f, 0.022174f, 0.023153f, 0.024158f,
+    0.025187f, 0.026241f, 0.027321f, 0.028426f, 0.029557f, 0.030713f, 0.031896f, 0.033105f, 0.034340f, 0.035601f, 0.036889f,
+    0.038204f, 0.039546f, 0.040915f, 0.042311f, 0.043735f, 0.045186f, 0.046665f, 0.048172f, 0.049707f, 0.051269f, 0.052861f,
+    0.054480f, 0.056128f, 0.057805f, 0.059511f, 0.061246f, 0.063010f, 0.064803f, 0.066626f, 0.068478f, 0.070360f, 0.072272f,
+    0.074214f, 0.076185f, 0.078187f, 0.080220f, 0.082283f, 0.084376f, 0.086500f, 0.088656f, 0.090842f, 0.093059f, 0.095307f,
+    0.097587f, 0.099899f, 0.102242f, 0.104616f, 0.107023f, 0.109462f, 0.111932f, 0.114435f, 0.116971f, 0.119538f, 0.122139f,
+    0.124772f, 0.127438f, 0.130136f, 0.132868f, 0.135633f, 0.138432f, 0.141263f, 0.144128f, 0.147027f, 0.149960f, 0.152926f,
+    0.155926f, 0.158961f, 0.162029f, 0.165132f, 0.168269f, 0.171441f, 0.174647f, 0.177888f, 0.181164f, 0.184475f, 0.187821f,
+    0.191202f, 0.194618f, 0.198069f, 0.201556f, 0.205079f, 0.208637f, 0.212231f, 0.215861f, 0.219526f, 0.223228f, 0.226966f,
+    0.230740f, 0.234551f, 0.238398f, 0.242281f, 0.246201f, 0.250158f, 0.254152f, 0.258183f, 0.262251f, 0.266356f, 0.270498f,
+    0.274677f, 0.278894f, 0.283149f, 0.287441f, 0.291771f, 0.296138f, 0.300544f, 0.304987f, 0.309469f, 0.313989f, 0.318547f,
+    0.323143f, 0.327778f, 0.332452f, 0.337164f, 0.341914f, 0.346704f, 0.351533f, 0.356400f, 0.361307f, 0.366253f, 0.371238f,
+    0.376262f, 0.381326f, 0.386430f, 0.391573f, 0.396755f, 0.401978f, 0.407240f, 0.412543f, 0.417885f, 0.423268f, 0.428691f,
+    0.434154f, 0.439657f, 0.445201f, 0.450786f, 0.456411f, 0.462077f, 0.467784f, 0.473532f, 0.479320f, 0.485150f, 0.491021f,
+    0.496933f, 0.502887f, 0.508881f, 0.514918f, 0.520996f, 0.527115f, 0.533276f, 0.539480f, 0.545725f, 0.552011f, 0.558340f,
+    0.564712f, 0.571125f, 0.577581f, 0.584078f, 0.590619f, 0.597202f, 0.603827f, 0.610496f, 0.617207f, 0.623960f, 0.630757f,
+    0.637597f, 0.644480f, 0.651406f, 0.658375f, 0.665387f, 0.672443f, 0.679543f, 0.686685f, 0.693872f, 0.701102f, 0.708376f,
+    0.715694f, 0.723055f, 0.730461f, 0.737911f, 0.745404f, 0.752942f, 0.760525f, 0.768151f, 0.775822f, 0.783538f, 0.791298f,
+    0.799103f, 0.806952f, 0.814847f, 0.822786f, 0.830770f, 0.838799f, 0.846873f, 0.854993f, 0.863157f, 0.871367f, 0.879622f,
+    0.887923f, 0.896269f, 0.904661f, 0.913099f, 0.921582f, 0.930111f, 0.938686f, 0.947307f, 0.955974f, 0.964686f, 0.973445f,
+    0.982251f, 0.991102f, 1.0f
+};
+
+static float stbir__srgb_to_linear(float f)
+{
+    if (f <= 0.04045f)
+        return f / 12.92f;
+    else
+        return (float)pow((f + 0.055f) / 1.055f, 2.4f);
+}
+
+static float stbir__linear_to_srgb(float f)
+{
+    if (f <= 0.0031308f)
+        return f * 12.92f;
+    else
+        return 1.055f * (float)pow(f, 1 / 2.4f) - 0.055f;
+}
+
+#ifndef STBIR_NON_IEEE_FLOAT
+// From https://gist.github.com/rygorous/2203834
+
+typedef union
+{
+    stbir_uint32 u;
+    float f;
+} stbir__FP32;
+
+static const stbir_uint32 fp32_to_srgb8_tab4[104] = {
+    0x0073000d, 0x007a000d, 0x0080000d, 0x0087000d, 0x008d000d, 0x0094000d, 0x009a000d, 0x00a1000d,
+    0x00a7001a, 0x00b4001a, 0x00c1001a, 0x00ce001a, 0x00da001a, 0x00e7001a, 0x00f4001a, 0x0101001a,
+    0x010e0033, 0x01280033, 0x01410033, 0x015b0033, 0x01750033, 0x018f0033, 0x01a80033, 0x01c20033,
+    0x01dc0067, 0x020f0067, 0x02430067, 0x02760067, 0x02aa0067, 0x02dd0067, 0x03110067, 0x03440067,
+    0x037800ce, 0x03df00ce, 0x044600ce, 0x04ad00ce, 0x051400ce, 0x057b00c5, 0x05dd00bc, 0x063b00b5,
+    0x06970158, 0x07420142, 0x07e30130, 0x087b0120, 0x090b0112, 0x09940106, 0x0a1700fc, 0x0a9500f2,
+    0x0b0f01cb, 0x0bf401ae, 0x0ccb0195, 0x0d950180, 0x0e56016e, 0x0f0d015e, 0x0fbc0150, 0x10630143,
+    0x11070264, 0x1238023e, 0x1357021d, 0x14660201, 0x156601e9, 0x165a01d3, 0x174401c0, 0x182401af,
+    0x18fe0331, 0x1a9602fe, 0x1c1502d2, 0x1d7e02ad, 0x1ed4028d, 0x201a0270, 0x21520256, 0x227d0240,
+    0x239f0443, 0x25c003fe, 0x27bf03c4, 0x29a10392, 0x2b6a0367, 0x2d1d0341, 0x2ebe031f, 0x304d0300,
+    0x31d105b0, 0x34a80555, 0x37520507, 0x39d504c5, 0x3c37048b, 0x3e7c0458, 0x40a8042a, 0x42bd0401,
+    0x44c20798, 0x488e071e, 0x4c1c06b6, 0x4f76065d, 0x52a50610, 0x55ac05cc, 0x5892058f, 0x5b590559,
+    0x5e0c0a23, 0x631c0980, 0x67db08f6, 0x6c55087f, 0x70940818, 0x74a007bd, 0x787d076c, 0x7c330723,
+};
+
+static stbir_uint8 stbir__linear_to_srgb_uchar(float in)
+{
+    static const stbir__FP32 almostone = { 0x3f7fffff }; // 1-eps
+    static const stbir__FP32 minval = { (127-13) << 23 };
+    stbir_uint32 tab,bias,scale,t;
+    stbir__FP32 f;
+
+    // Clamp to [2^(-13), 1-eps]; these two values map to 0 and 1, respectively.
+    // The tests are carefully written so that NaNs map to 0, same as in the reference
+    // implementation.
+    if (!(in > minval.f)) // written this way to catch NaNs
+        in = minval.f;
+    if (in > almostone.f)
+        in = almostone.f;
+
+    // Do the table lookup and unpack bias, scale
+    f.f = in;
+    tab = fp32_to_srgb8_tab4[(f.u - minval.u) >> 20];
+    bias = (tab >> 16) << 9;
+    scale = tab & 0xffff;
+
+    // Grab next-highest mantissa bits and perform linear interpolation
+    t = (f.u >> 12) & 0xff;
+    return (unsigned char) ((bias + scale*t) >> 16);
+}
+
+#else
+// sRGB transition values, scaled by 1<<28
+static int stbir__srgb_offset_to_linear_scaled[256] =
+{
+            0,     40738,    122216,    203693,    285170,    366648,    448125,    529603,
+       611080,    692557,    774035,    855852,    942009,   1033024,   1128971,   1229926,
+      1335959,   1447142,   1563542,   1685229,   1812268,   1944725,   2082664,   2226148,
+      2375238,   2529996,   2690481,   2856753,   3028870,   3206888,   3390865,   3580856,
+      3776916,   3979100,   4187460,   4402049,   4622919,   4850123,   5083710,   5323731,
+      5570236,   5823273,   6082892,   6349140,   6622065,   6901714,   7188133,   7481369,
+      7781466,   8088471,   8402427,   8723380,   9051372,   9386448,   9728650,  10078021,
+     10434603,  10798439,  11169569,  11548036,  11933879,  12327139,  12727857,  13136073,
+     13551826,  13975156,  14406100,  14844697,  15290987,  15745007,  16206795,  16676389,
+     17153826,  17639142,  18132374,  18633560,  19142734,  19659934,  20185196,  20718552,
+     21260042,  21809696,  22367554,  22933648,  23508010,  24090680,  24681686,  25281066,
+     25888850,  26505076,  27129772,  27762974,  28404716,  29055026,  29713942,  30381490,
+     31057708,  31742624,  32436272,  33138682,  33849884,  34569912,  35298800,  36036568,
+     36783260,  37538896,  38303512,  39077136,  39859796,  40651528,  41452360,  42262316,
+     43081432,  43909732,  44747252,  45594016,  46450052,  47315392,  48190064,  49074096,
+     49967516,  50870356,  51782636,  52704392,  53635648,  54576432,  55526772,  56486700,
+     57456236,  58435408,  59424248,  60422780,  61431036,  62449032,  63476804,  64514376,
+     65561776,  66619028,  67686160,  68763192,  69850160,  70947088,  72053992,  73170912,
+     74297864,  75434880,  76581976,  77739184,  78906536,  80084040,  81271736,  82469648,
+     83677792,  84896192,  86124888,  87363888,  88613232,  89872928,  91143016,  92423512,
+     93714432,  95015816,  96327688,  97650056,  98982952, 100326408, 101680440, 103045072,
+    104420320, 105806224, 107202800, 108610064, 110028048, 111456776, 112896264, 114346544,
+    115807632, 117279552, 118762328, 120255976, 121760536, 123276016, 124802440, 126339832,
+    127888216, 129447616, 131018048, 132599544, 134192112, 135795792, 137410592, 139036528,
+    140673648, 142321952, 143981456, 145652208, 147334208, 149027488, 150732064, 152447968,
+    154175200, 155913792, 157663776, 159425168, 161197984, 162982240, 164777968, 166585184,
+    168403904, 170234160, 172075968, 173929344, 175794320, 177670896, 179559120, 181458992,
+    183370528, 185293776, 187228736, 189175424, 191133888, 193104112, 195086128, 197079968,
+    199085648, 201103184, 203132592, 205173888, 207227120, 209292272, 211369392, 213458480,
+    215559568, 217672656, 219797792, 221934976, 224084240, 226245600, 228419056, 230604656,
+    232802400, 235012320, 237234432, 239468736, 241715280, 243974080, 246245120, 248528464,
+    250824112, 253132064, 255452368, 257785040, 260130080, 262487520, 264857376, 267239664,
+};
+
+static stbir_uint8 stbir__linear_to_srgb_uchar(float f)
+{
+    int x = (int) (f * (1 << 28)); // has headroom so you don't need to clamp
+    int v = 0;
+    int i;
+
+    // Refine the guess with a short binary search.
+    i = v + 128; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i;
+    i = v +  64; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i;
+    i = v +  32; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i;
+    i = v +  16; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i;
+    i = v +   8; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i;
+    i = v +   4; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i;
+    i = v +   2; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i;
+    i = v +   1; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i;
+
+    return (stbir_uint8) v;
+}
+#endif
+
+static float stbir__filter_trapezoid(float x, float scale)
+{
+    float halfscale = scale / 2;
+    float t = 0.5f + halfscale;
+    STBIR_ASSERT(scale <= 1);
+
+    x = (float)fabs(x);
+
+    if (x >= t)
+        return 0;
+    else
+    {
+        float r = 0.5f - halfscale;
+        if (x <= r)
+            return 1;
+        else
+            return (t - x) / scale;
+    }
+}
+
+static float stbir__support_trapezoid(float scale)
+{
+    STBIR_ASSERT(scale <= 1);
+    return 0.5f + scale / 2;
+}
+
+static float stbir__filter_triangle(float x, float s)
+{
+    STBIR__UNUSED_PARAM(s);
+
+    x = (float)fabs(x);
+
+    if (x <= 1.0f)
+        return 1 - x;
+    else
+        return 0;
+}
+
+static float stbir__filter_cubic(float x, float s)
+{
+    STBIR__UNUSED_PARAM(s);
+
+    x = (float)fabs(x);
+
+    if (x < 1.0f)
+        return (4 + x*x*(3*x - 6))/6;
+    else if (x < 2.0f)
+        return (8 + x*(-12 + x*(6 - x)))/6;
+
+    return (0.0f);
+}
+
+static float stbir__filter_catmullrom(float x, float s)
+{
+    STBIR__UNUSED_PARAM(s);
+
+    x = (float)fabs(x);
+
+    if (x < 1.0f)
+        return 1 - x*x*(2.5f - 1.5f*x);
+    else if (x < 2.0f)
+        return 2 - x*(4 + x*(0.5f*x - 2.5f));
+
+    return (0.0f);
+}
+
+static float stbir__filter_mitchell(float x, float s)
+{
+    STBIR__UNUSED_PARAM(s);
+
+    x = (float)fabs(x);
+
+    if (x < 1.0f)
+        return (16 + x*x*(21 * x - 36))/18;
+    else if (x < 2.0f)
+        return (32 + x*(-60 + x*(36 - 7*x)))/18;
+
+    return (0.0f);
+}
+
+static float stbir__support_zero(float s)
+{
+    STBIR__UNUSED_PARAM(s);
+    return 0;
+}
+
+static float stbir__support_one(float s)
+{
+    STBIR__UNUSED_PARAM(s);
+    return 1;
+}
+
+static float stbir__support_two(float s)
+{
+    STBIR__UNUSED_PARAM(s);
+    return 2;
+}
+
+static stbir__filter_info stbir__filter_info_table[] = {
+        { NULL,                     stbir__support_zero },
+        { stbir__filter_trapezoid,  stbir__support_trapezoid },
+        { stbir__filter_triangle,   stbir__support_one },
+        { stbir__filter_cubic,      stbir__support_two },
+        { stbir__filter_catmullrom, stbir__support_two },
+        { stbir__filter_mitchell,   stbir__support_two },
+};
+
+stbir__inline static int stbir__use_upsampling(float ratio)
+{
+    return ratio > 1;
+}
+
+stbir__inline static int stbir__use_width_upsampling(stbir__info* stbir_info)
+{
+    return stbir__use_upsampling(stbir_info->horizontal_scale);
+}
+
+stbir__inline static int stbir__use_height_upsampling(stbir__info* stbir_info)
+{
+    return stbir__use_upsampling(stbir_info->vertical_scale);
+}
+
+// This is the maximum number of input samples that can affect an output sample
+// with the given filter
+static int stbir__get_filter_pixel_width(stbir_filter filter, float scale)
+{
+    STBIR_ASSERT(filter != 0);
+    STBIR_ASSERT(filter < STBIR__ARRAY_SIZE(stbir__filter_info_table));
+
+    if (stbir__use_upsampling(scale))
+        return (int)ceil(stbir__filter_info_table[filter].support(1/scale) * 2);
+    else
+        return (int)ceil(stbir__filter_info_table[filter].support(scale) * 2 / scale);
+}
+
+// This is how much to expand buffers to account for filters seeking outside
+// the image boundaries.
+static int stbir__get_filter_pixel_margin(stbir_filter filter, float scale)
+{
+    return stbir__get_filter_pixel_width(filter, scale) / 2;
+}
+
+static int stbir__get_coefficient_width(stbir_filter filter, float scale)
+{
+    if (stbir__use_upsampling(scale))
+        return (int)ceil(stbir__filter_info_table[filter].support(1 / scale) * 2);
+    else
+        return (int)ceil(stbir__filter_info_table[filter].support(scale) * 2);
+}
+
+static int stbir__get_contributors(float scale, stbir_filter filter, int input_size, int output_size)
+{
+    if (stbir__use_upsampling(scale))
+        return output_size;
+    else
+        return (input_size + stbir__get_filter_pixel_margin(filter, scale) * 2);
+}
+
+static int stbir__get_total_horizontal_coefficients(stbir__info* info)
+{
+    return info->horizontal_num_contributors
+         * stbir__get_coefficient_width      (info->horizontal_filter, info->horizontal_scale);
+}
+
+static int stbir__get_total_vertical_coefficients(stbir__info* info)
+{
+    return info->vertical_num_contributors
+         * stbir__get_coefficient_width      (info->vertical_filter, info->vertical_scale);
+}
+
+static stbir__contributors* stbir__get_contributor(stbir__contributors* contributors, int n)
+{
+    return &contributors[n];
+}
+
+// For perf reasons this code is duplicated in stbir__resample_horizontal_upsample/downsample,
+// if you change it here change it there too.
+static float* stbir__get_coefficient(float* coefficients, stbir_filter filter, float scale, int n, int c)
+{
+    int width = stbir__get_coefficient_width(filter, scale);
+    return &coefficients[width*n + c];
+}
+
+static int stbir__edge_wrap_slow(stbir_edge edge, int n, int max)
+{
+    switch (edge)
+    {
+    case STBIR_EDGE_ZERO:
+        return 0; // we'll decode the wrong pixel here, and then overwrite with 0s later
+
+    case STBIR_EDGE_CLAMP:
+        if (n < 0)
+            return 0;
+
+        if (n >= max)
+            return max - 1;
+
+        return n; // NOTREACHED
+
+    case STBIR_EDGE_REFLECT:
+    {
+        if (n < 0)
+        {
+            if (n < max)
+                return -n;
+            else
+                return max - 1;
+        }
+
+        if (n >= max)
+        {
+            int max2 = max * 2;
+            if (n >= max2)
+                return 0;
+            else
+                return max2 - n - 1;
+        }
+
+        return n; // NOTREACHED
+    }
+
+    case STBIR_EDGE_WRAP:
+        if (n >= 0)
+            return (n % max);
+        else
+        {
+            int m = (-n) % max;
+
+            if (m != 0)
+                m = max - m;
+
+            return (m);
+        }
+        // NOTREACHED
+
+    default:
+        STBIR_ASSERT(!"Unimplemented edge type");
+        return 0;
+    }
+}
+
+stbir__inline static int stbir__edge_wrap(stbir_edge edge, int n, int max)
+{
+    // avoid per-pixel switch
+    if (n >= 0 && n < max)
+        return n;
+    return stbir__edge_wrap_slow(edge, n, max);
+}
+
+// What input pixels contribute to this output pixel?
+static void stbir__calculate_sample_range_upsample(int n, float out_filter_radius, float scale_ratio, float out_shift, int* in_first_pixel, int* in_last_pixel, float* in_center_of_out)
+{
+    float out_pixel_center = (float)n + 0.5f;
+    float out_pixel_influence_lowerbound = out_pixel_center - out_filter_radius;
+    float out_pixel_influence_upperbound = out_pixel_center + out_filter_radius;
+
+    float in_pixel_influence_lowerbound = (out_pixel_influence_lowerbound + out_shift) / scale_ratio;
+    float in_pixel_influence_upperbound = (out_pixel_influence_upperbound + out_shift) / scale_ratio;
+
+    *in_center_of_out = (out_pixel_center + out_shift) / scale_ratio;
+    *in_first_pixel = (int)(floor(in_pixel_influence_lowerbound + 0.5));
+    *in_last_pixel = (int)(floor(in_pixel_influence_upperbound - 0.5));
+}
+
+// What output pixels does this input pixel contribute to?
+static void stbir__calculate_sample_range_downsample(int n, float in_pixels_radius, float scale_ratio, float out_shift, int* out_first_pixel, int* out_last_pixel, float* out_center_of_in)
+{
+    float in_pixel_center = (float)n + 0.5f;
+    float in_pixel_influence_lowerbound = in_pixel_center - in_pixels_radius;
+    float in_pixel_influence_upperbound = in_pixel_center + in_pixels_radius;
+
+    float out_pixel_influence_lowerbound = in_pixel_influence_lowerbound * scale_ratio - out_shift;
+    float out_pixel_influence_upperbound = in_pixel_influence_upperbound * scale_ratio - out_shift;
+
+    *out_center_of_in = in_pixel_center * scale_ratio - out_shift;
+    *out_first_pixel = (int)(floor(out_pixel_influence_lowerbound + 0.5));
+    *out_last_pixel = (int)(floor(out_pixel_influence_upperbound - 0.5));
+}
+
+static void stbir__calculate_coefficients_upsample(stbir_filter filter, float scale, int in_first_pixel, int in_last_pixel, float in_center_of_out, stbir__contributors* contributor, float* coefficient_group)
+{
+    int i;
+    float total_filter = 0;
+    float filter_scale;
+
+    STBIR_ASSERT(in_last_pixel - in_first_pixel <= (int)ceil(stbir__filter_info_table[filter].support(1/scale) * 2)); // Taken directly from stbir__get_coefficient_width() which we can't call because we don't know if we're horizontal or vertical.
+
+    contributor->n0 = in_first_pixel;
+    contributor->n1 = in_last_pixel;
+
+    STBIR_ASSERT(contributor->n1 >= contributor->n0);
+
+    for (i = 0; i <= in_last_pixel - in_first_pixel; i++)
+    {
+        float in_pixel_center = (float)(i + in_first_pixel) + 0.5f;
+        coefficient_group[i] = stbir__filter_info_table[filter].kernel(in_center_of_out - in_pixel_center, 1 / scale);
+
+        // If the coefficient is zero, skip it. (Don't do the <0 check here, we want the influence of those outside pixels.)
+        if (i == 0 && !coefficient_group[i])
+        {
+            contributor->n0 = ++in_first_pixel;
+            i--;
+            continue;
+        }
+
+        total_filter += coefficient_group[i];
+    }
+
+    // NOTE(fg): Not actually true in general, nor is there any reason to expect it should be.
+    // It would be true in exact math but is at best approximately true in floating-point math,
+    // and it would not make sense to try and put actual bounds on this here because it depends
+    // on the image aspect ratio which can get pretty extreme.
+    //STBIR_ASSERT(stbir__filter_info_table[filter].kernel((float)(in_last_pixel + 1) + 0.5f - in_center_of_out, 1/scale) == 0);
+
+    STBIR_ASSERT(total_filter > 0.9);
+    STBIR_ASSERT(total_filter < 1.1f); // Make sure it's not way off.
+
+    // Make sure the sum of all coefficients is 1.
+    filter_scale = 1 / total_filter;
+
+    for (i = 0; i <= in_last_pixel - in_first_pixel; i++)
+        coefficient_group[i] *= filter_scale;
+
+    for (i = in_last_pixel - in_first_pixel; i >= 0; i--)
+    {
+        if (coefficient_group[i])
+            break;
+
+        // This line has no weight. We can skip it.
+        contributor->n1 = contributor->n0 + i - 1;
+    }
+}
+
+static void stbir__calculate_coefficients_downsample(stbir_filter filter, float scale_ratio, int out_first_pixel, int out_last_pixel, float out_center_of_in, stbir__contributors* contributor, float* coefficient_group)
+{
+    int i;
+
+    STBIR_ASSERT(out_last_pixel - out_first_pixel <= (int)ceil(stbir__filter_info_table[filter].support(scale_ratio) * 2)); // Taken directly from stbir__get_coefficient_width() which we can't call because we don't know if we're horizontal or vertical.
+
+    contributor->n0 = out_first_pixel;
+    contributor->n1 = out_last_pixel;
+
+    STBIR_ASSERT(contributor->n1 >= contributor->n0);
+
+    for (i = 0; i <= out_last_pixel - out_first_pixel; i++)
+    {
+        float out_pixel_center = (float)(i + out_first_pixel) + 0.5f;
+        float x = out_pixel_center - out_center_of_in;
+        coefficient_group[i] = stbir__filter_info_table[filter].kernel(x, scale_ratio) * scale_ratio;
+    }
+
+    // NOTE(fg): Not actually true in general, nor is there any reason to expect it should be.
+    // It would be true in exact math but is at best approximately true in floating-point math,
+    // and it would not make sense to try and put actual bounds on this here because it depends
+    // on the image aspect ratio which can get pretty extreme.
+    //STBIR_ASSERT(stbir__filter_info_table[filter].kernel((float)(out_last_pixel + 1) + 0.5f - out_center_of_in, scale_ratio) == 0);
+
+    for (i = out_last_pixel - out_first_pixel; i >= 0; i--)
+    {
+        if (coefficient_group[i])
+            break;
+
+        // This line has no weight. We can skip it.
+        contributor->n1 = contributor->n0 + i - 1;
+    }
+}
+
+static void stbir__normalize_downsample_coefficients(stbir__contributors* contributors, float* coefficients, stbir_filter filter, float scale_ratio, int input_size, int output_size)
+{
+    int num_contributors = stbir__get_contributors(scale_ratio, filter, input_size, output_size);
+    int num_coefficients = stbir__get_coefficient_width(filter, scale_ratio);
+    int i, j;
+    int skip;
+
+    for (i = 0; i < output_size; i++)
+    {
+        float scale;
+        float total = 0;
+
+        for (j = 0; j < num_contributors; j++)
+        {
+            if (i >= contributors[j].n0 && i <= contributors[j].n1)
+            {
+                float coefficient = *stbir__get_coefficient(coefficients, filter, scale_ratio, j, i - contributors[j].n0);
+                total += coefficient;
+            }
+            else if (i < contributors[j].n0)
+                break;
+        }
+
+        STBIR_ASSERT(total > 0.9f);
+        STBIR_ASSERT(total < 1.1f);
+
+        scale = 1 / total;
+
+        for (j = 0; j < num_contributors; j++)
+        {
+            if (i >= contributors[j].n0 && i <= contributors[j].n1)
+                *stbir__get_coefficient(coefficients, filter, scale_ratio, j, i - contributors[j].n0) *= scale;
+            else if (i < contributors[j].n0)
+                break;
+        }
+    }
+
+    // Optimize: Skip zero coefficients and contributions outside of image bounds.
+    // Do this after normalizing because normalization depends on the n0/n1 values.
+    for (j = 0; j < num_contributors; j++)
+    {
+        int range, max, width;
+
+        skip = 0;
+        while (*stbir__get_coefficient(coefficients, filter, scale_ratio, j, skip) == 0)
+            skip++;
+
+        contributors[j].n0 += skip;
+
+        while (contributors[j].n0 < 0)
+        {
+            contributors[j].n0++;
+            skip++;
+        }
+
+        range = contributors[j].n1 - contributors[j].n0 + 1;
+        max = stbir__min(num_coefficients, range);
+
+        width = stbir__get_coefficient_width(filter, scale_ratio);
+        for (i = 0; i < max; i++)
+        {
+            if (i + skip >= width)
+                break;
+
+            *stbir__get_coefficient(coefficients, filter, scale_ratio, j, i) = *stbir__get_coefficient(coefficients, filter, scale_ratio, j, i + skip);
+        }
+
+        continue;
+    }
+
+    // Using min to avoid writing into invalid pixels.
+    for (i = 0; i < num_contributors; i++)
+        contributors[i].n1 = stbir__min(contributors[i].n1, output_size - 1);
+}
+
+// Each scan line uses the same kernel values so we should calculate the kernel
+// values once and then we can use them for every scan line.
+static void stbir__calculate_filters(stbir__contributors* contributors, float* coefficients, stbir_filter filter, float scale_ratio, float shift, int input_size, int output_size)
+{
+    int n;
+    int total_contributors = stbir__get_contributors(scale_ratio, filter, input_size, output_size);
+
+    if (stbir__use_upsampling(scale_ratio))
+    {
+        float out_pixels_radius = stbir__filter_info_table[filter].support(1 / scale_ratio) * scale_ratio;
+
+        // Looping through out pixels
+        for (n = 0; n < total_contributors; n++)
+        {
+            float in_center_of_out; // Center of the current out pixel in the in pixel space
+            int in_first_pixel, in_last_pixel;
+
+            stbir__calculate_sample_range_upsample(n, out_pixels_radius, scale_ratio, shift, &in_first_pixel, &in_last_pixel, &in_center_of_out);
+
+            stbir__calculate_coefficients_upsample(filter, scale_ratio, in_first_pixel, in_last_pixel, in_center_of_out, stbir__get_contributor(contributors, n), stbir__get_coefficient(coefficients, filter, scale_ratio, n, 0));
+        }
+    }
+    else
+    {
+        float in_pixels_radius = stbir__filter_info_table[filter].support(scale_ratio) / scale_ratio;
+
+        // Looping through in pixels
+        for (n = 0; n < total_contributors; n++)
+        {
+            float out_center_of_in; // Center of the current out pixel in the in pixel space
+            int out_first_pixel, out_last_pixel;
+            int n_adjusted = n - stbir__get_filter_pixel_margin(filter, scale_ratio);
+
+            stbir__calculate_sample_range_downsample(n_adjusted, in_pixels_radius, scale_ratio, shift, &out_first_pixel, &out_last_pixel, &out_center_of_in);
+
+            stbir__calculate_coefficients_downsample(filter, scale_ratio, out_first_pixel, out_last_pixel, out_center_of_in, stbir__get_contributor(contributors, n), stbir__get_coefficient(coefficients, filter, scale_ratio, n, 0));
+        }
+
+        stbir__normalize_downsample_coefficients(contributors, coefficients, filter, scale_ratio, input_size, output_size);
+    }
+}
+
+static float* stbir__get_decode_buffer(stbir__info* stbir_info)
+{
+    // The 0 index of the decode buffer starts after the margin. This makes
+    // it okay to use negative indexes on the decode buffer.
+    return &stbir_info->decode_buffer[stbir_info->horizontal_filter_pixel_margin * stbir_info->channels];
+}
+
+#define STBIR__DECODE(type, colorspace) ((int)(type) * (STBIR_MAX_COLORSPACES) + (int)(colorspace))
+
+static void stbir__decode_scanline(stbir__info* stbir_info, int n)
+{
+    int c;
+    int channels = stbir_info->channels;
+    int alpha_channel = stbir_info->alpha_channel;
+    int type = stbir_info->type;
+    int colorspace = stbir_info->colorspace;
+    int input_w = stbir_info->input_w;
+    size_t input_stride_bytes = stbir_info->input_stride_bytes;
+    float* decode_buffer = stbir__get_decode_buffer(stbir_info);
+    stbir_edge edge_horizontal = stbir_info->edge_horizontal;
+    stbir_edge edge_vertical = stbir_info->edge_vertical;
+    size_t in_buffer_row_offset = stbir__edge_wrap(edge_vertical, n, stbir_info->input_h) * input_stride_bytes;
+    const void* input_data = (char *) stbir_info->input_data + in_buffer_row_offset;
+    int max_x = input_w + stbir_info->horizontal_filter_pixel_margin;
+    int decode = STBIR__DECODE(type, colorspace);
+
+    int x = -stbir_info->horizontal_filter_pixel_margin;
+
+    // special handling for STBIR_EDGE_ZERO because it needs to return an item that doesn't appear in the input,
+    // and we want to avoid paying overhead on every pixel if not STBIR_EDGE_ZERO
+    if (edge_vertical == STBIR_EDGE_ZERO && (n < 0 || n >= stbir_info->input_h))
+    {
+        for (; x < max_x; x++)
+            for (c = 0; c < channels; c++)
+                decode_buffer[x*channels + c] = 0;
+        return;
+    }
+
+    switch (decode)
+    {
+    case STBIR__DECODE(STBIR_TYPE_UINT8, STBIR_COLORSPACE_LINEAR):
+        for (; x < max_x; x++)
+        {
+            int decode_pixel_index = x * channels;
+            int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels;
+            for (c = 0; c < channels; c++)
+                decode_buffer[decode_pixel_index + c] = ((float)((const unsigned char*)input_data)[input_pixel_index + c]) / stbir__max_uint8_as_float;
+        }
+        break;
+
+    case STBIR__DECODE(STBIR_TYPE_UINT8, STBIR_COLORSPACE_SRGB):
+        for (; x < max_x; x++)
+        {
+            int decode_pixel_index = x * channels;
+            int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels;
+            for (c = 0; c < channels; c++)
+                decode_buffer[decode_pixel_index + c] = stbir__srgb_uchar_to_linear_float[((const unsigned char*)input_data)[input_pixel_index + c]];
+
+            if (!(stbir_info->flags&STBIR_FLAG_ALPHA_USES_COLORSPACE))
+                decode_buffer[decode_pixel_index + alpha_channel] = ((float)((const unsigned char*)input_data)[input_pixel_index + alpha_channel]) / stbir__max_uint8_as_float;
+        }
+        break;
+
+    case STBIR__DECODE(STBIR_TYPE_UINT16, STBIR_COLORSPACE_LINEAR):
+        for (; x < max_x; x++)
+        {
+            int decode_pixel_index = x * channels;
+            int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels;
+            for (c = 0; c < channels; c++)
+                decode_buffer[decode_pixel_index + c] = ((float)((const unsigned short*)input_data)[input_pixel_index + c]) / stbir__max_uint16_as_float;
+        }
+        break;
+
+    case STBIR__DECODE(STBIR_TYPE_UINT16, STBIR_COLORSPACE_SRGB):
+        for (; x < max_x; x++)
+        {
+            int decode_pixel_index = x * channels;
+            int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels;
+            for (c = 0; c < channels; c++)
+                decode_buffer[decode_pixel_index + c] = stbir__srgb_to_linear(((float)((const unsigned short*)input_data)[input_pixel_index + c]) / stbir__max_uint16_as_float);
+
+            if (!(stbir_info->flags&STBIR_FLAG_ALPHA_USES_COLORSPACE))
+                decode_buffer[decode_pixel_index + alpha_channel] = ((float)((const unsigned short*)input_data)[input_pixel_index + alpha_channel]) / stbir__max_uint16_as_float;
+        }
+        break;
+
+    case STBIR__DECODE(STBIR_TYPE_UINT32, STBIR_COLORSPACE_LINEAR):
+        for (; x < max_x; x++)
+        {
+            int decode_pixel_index = x * channels;
+            int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels;
+            for (c = 0; c < channels; c++)
+                decode_buffer[decode_pixel_index + c] = (float)(((double)((const unsigned int*)input_data)[input_pixel_index + c]) / stbir__max_uint32_as_float);
+        }
+        break;
+
+    case STBIR__DECODE(STBIR_TYPE_UINT32, STBIR_COLORSPACE_SRGB):
+        for (; x < max_x; x++)
+        {
+            int decode_pixel_index = x * channels;
+            int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels;
+            for (c = 0; c < channels; c++)
+                decode_buffer[decode_pixel_index + c] = stbir__srgb_to_linear((float)(((double)((const unsigned int*)input_data)[input_pixel_index + c]) / stbir__max_uint32_as_float));
+
+            if (!(stbir_info->flags&STBIR_FLAG_ALPHA_USES_COLORSPACE))
+                decode_buffer[decode_pixel_index + alpha_channel] = (float)(((double)((const unsigned int*)input_data)[input_pixel_index + alpha_channel]) / stbir__max_uint32_as_float);
+        }
+        break;
+
+    case STBIR__DECODE(STBIR_TYPE_FLOAT, STBIR_COLORSPACE_LINEAR):
+        for (; x < max_x; x++)
+        {
+            int decode_pixel_index = x * channels;
+            int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels;
+            for (c = 0; c < channels; c++)
+                decode_buffer[decode_pixel_index + c] = ((const float*)input_data)[input_pixel_index + c];
+        }
+        break;
+
+    case STBIR__DECODE(STBIR_TYPE_FLOAT, STBIR_COLORSPACE_SRGB):
+        for (; x < max_x; x++)
+        {
+            int decode_pixel_index = x * channels;
+            int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels;
+            for (c = 0; c < channels; c++)
+                decode_buffer[decode_pixel_index + c] = stbir__srgb_to_linear(((const float*)input_data)[input_pixel_index + c]);
+
+            if (!(stbir_info->flags&STBIR_FLAG_ALPHA_USES_COLORSPACE))
+                decode_buffer[decode_pixel_index + alpha_channel] = ((const float*)input_data)[input_pixel_index + alpha_channel];
+        }
+
+        break;
+
+    default:
+        STBIR_ASSERT(!"Unknown type/colorspace/channels combination.");
+        break;
+    }
+
+    if (!(stbir_info->flags & STBIR_FLAG_ALPHA_PREMULTIPLIED))
+    {
+        for (x = -stbir_info->horizontal_filter_pixel_margin; x < max_x; x++)
+        {
+            int decode_pixel_index = x * channels;
+
+            // If the alpha value is 0 it will clobber the color values. Make sure it's not.
+            float alpha = decode_buffer[decode_pixel_index + alpha_channel];
+#ifndef STBIR_NO_ALPHA_EPSILON
+            if (stbir_info->type != STBIR_TYPE_FLOAT) {
+                alpha += STBIR_ALPHA_EPSILON;
+                decode_buffer[decode_pixel_index + alpha_channel] = alpha;
+            }
+#endif
+            for (c = 0; c < channels; c++)
+            {
+                if (c == alpha_channel)
+                    continue;
+
+                decode_buffer[decode_pixel_index + c] *= alpha;
+            }
+        }
+    }
+
+    if (edge_horizontal == STBIR_EDGE_ZERO)
+    {
+        for (x = -stbir_info->horizontal_filter_pixel_margin; x < 0; x++)
+        {
+            for (c = 0; c < channels; c++)
+                decode_buffer[x*channels + c] = 0;
+        }
+        for (x = input_w; x < max_x; x++)
+        {
+            for (c = 0; c < channels; c++)
+                decode_buffer[x*channels + c] = 0;
+        }
+    }
+}
+
+static float* stbir__get_ring_buffer_entry(float* ring_buffer, int index, int ring_buffer_length)
+{
+    return &ring_buffer[index * ring_buffer_length];
+}
+
+static float* stbir__add_empty_ring_buffer_entry(stbir__info* stbir_info, int n)
+{
+    int ring_buffer_index;
+    float* ring_buffer;
+
+    stbir_info->ring_buffer_last_scanline = n;
+
+    if (stbir_info->ring_buffer_begin_index < 0)
+    {
+        ring_buffer_index = stbir_info->ring_buffer_begin_index = 0;
+        stbir_info->ring_buffer_first_scanline = n;
+    }
+    else
+    {
+        ring_buffer_index = (stbir_info->ring_buffer_begin_index + (stbir_info->ring_buffer_last_scanline - stbir_info->ring_buffer_first_scanline)) % stbir_info->ring_buffer_num_entries;
+        STBIR_ASSERT(ring_buffer_index != stbir_info->ring_buffer_begin_index);
+    }
+
+    ring_buffer = stbir__get_ring_buffer_entry(stbir_info->ring_buffer, ring_buffer_index, stbir_info->ring_buffer_length_bytes / sizeof(float));
+    memset(ring_buffer, 0, stbir_info->ring_buffer_length_bytes);
+
+    return ring_buffer;
+}
+
+
+static void stbir__resample_horizontal_upsample(stbir__info* stbir_info, float* output_buffer)
+{
+    int x, k;
+    int output_w = stbir_info->output_w;
+    int channels = stbir_info->channels;
+    float* decode_buffer = stbir__get_decode_buffer(stbir_info);
+    stbir__contributors* horizontal_contributors = stbir_info->horizontal_contributors;
+    float* horizontal_coefficients = stbir_info->horizontal_coefficients;
+    int coefficient_width = stbir_info->horizontal_coefficient_width;
+
+    for (x = 0; x < output_w; x++)
+    {
+        int n0 = horizontal_contributors[x].n0;
+        int n1 = horizontal_contributors[x].n1;
+
+        int out_pixel_index = x * channels;
+        int coefficient_group = coefficient_width * x;
+        int coefficient_counter = 0;
+
+        STBIR_ASSERT(n1 >= n0);
+        STBIR_ASSERT(n0 >= -stbir_info->horizontal_filter_pixel_margin);
+        STBIR_ASSERT(n1 >= -stbir_info->horizontal_filter_pixel_margin);
+        STBIR_ASSERT(n0 < stbir_info->input_w + stbir_info->horizontal_filter_pixel_margin);
+        STBIR_ASSERT(n1 < stbir_info->input_w + stbir_info->horizontal_filter_pixel_margin);
+
+        switch (channels) {
+            case 1:
+                for (k = n0; k <= n1; k++)
+                {
+                    int in_pixel_index = k * 1;
+                    float coefficient = horizontal_coefficients[coefficient_group + coefficient_counter++];
+                    STBIR_ASSERT(coefficient != 0);
+                    output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
+                }
+                break;
+            case 2:
+                for (k = n0; k <= n1; k++)
+                {
+                    int in_pixel_index = k * 2;
+                    float coefficient = horizontal_coefficients[coefficient_group + coefficient_counter++];
+                    STBIR_ASSERT(coefficient != 0);
+                    output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
+                    output_buffer[out_pixel_index + 1] += decode_buffer[in_pixel_index + 1] * coefficient;
+                }
+                break;
+            case 3:
+                for (k = n0; k <= n1; k++)
+                {
+                    int in_pixel_index = k * 3;
+                    float coefficient = horizontal_coefficients[coefficient_group + coefficient_counter++];
+                    STBIR_ASSERT(coefficient != 0);
+                    output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
+                    output_buffer[out_pixel_index + 1] += decode_buffer[in_pixel_index + 1] * coefficient;
+                    output_buffer[out_pixel_index + 2] += decode_buffer[in_pixel_index + 2] * coefficient;
+                }
+                break;
+            case 4:
+                for (k = n0; k <= n1; k++)
+                {
+                    int in_pixel_index = k * 4;
+                    float coefficient = horizontal_coefficients[coefficient_group + coefficient_counter++];
+                    STBIR_ASSERT(coefficient != 0);
+                    output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
+                    output_buffer[out_pixel_index + 1] += decode_buffer[in_pixel_index + 1] * coefficient;
+                    output_buffer[out_pixel_index + 2] += decode_buffer[in_pixel_index + 2] * coefficient;
+                    output_buffer[out_pixel_index + 3] += decode_buffer[in_pixel_index + 3] * coefficient;
+                }
+                break;
+            default:
+                for (k = n0; k <= n1; k++)
+                {
+                    int in_pixel_index = k * channels;
+                    float coefficient = horizontal_coefficients[coefficient_group + coefficient_counter++];
+                    int c;
+                    STBIR_ASSERT(coefficient != 0);
+                    for (c = 0; c < channels; c++)
+                        output_buffer[out_pixel_index + c] += decode_buffer[in_pixel_index + c] * coefficient;
+                }
+                break;
+        }
+    }
+}
+
+static void stbir__resample_horizontal_downsample(stbir__info* stbir_info, float* output_buffer)
+{
+    int x, k;
+    int input_w = stbir_info->input_w;
+    int channels = stbir_info->channels;
+    float* decode_buffer = stbir__get_decode_buffer(stbir_info);
+    stbir__contributors* horizontal_contributors = stbir_info->horizontal_contributors;
+    float* horizontal_coefficients = stbir_info->horizontal_coefficients;
+    int coefficient_width = stbir_info->horizontal_coefficient_width;
+    int filter_pixel_margin = stbir_info->horizontal_filter_pixel_margin;
+    int max_x = input_w + filter_pixel_margin * 2;
+
+    STBIR_ASSERT(!stbir__use_width_upsampling(stbir_info));
+
+    switch (channels) {
+        case 1:
+            for (x = 0; x < max_x; x++)
+            {
+                int n0 = horizontal_contributors[x].n0;
+                int n1 = horizontal_contributors[x].n1;
+
+                int in_x = x - filter_pixel_margin;
+                int in_pixel_index = in_x * 1;
+                int max_n = n1;
+                int coefficient_group = coefficient_width * x;
+
+                for (k = n0; k <= max_n; k++)
+                {
+                    int out_pixel_index = k * 1;
+                    float coefficient = horizontal_coefficients[coefficient_group + k - n0];
+                    output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
+                }
+            }
+            break;
+
+        case 2:
+            for (x = 0; x < max_x; x++)
+            {
+                int n0 = horizontal_contributors[x].n0;
+                int n1 = horizontal_contributors[x].n1;
+
+                int in_x = x - filter_pixel_margin;
+                int in_pixel_index = in_x * 2;
+                int max_n = n1;
+                int coefficient_group = coefficient_width * x;
+
+                for (k = n0; k <= max_n; k++)
+                {
+                    int out_pixel_index = k * 2;
+                    float coefficient = horizontal_coefficients[coefficient_group + k - n0];
+                    output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
+                    output_buffer[out_pixel_index + 1] += decode_buffer[in_pixel_index + 1] * coefficient;
+                }
+            }
+            break;
+
+        case 3:
+            for (x = 0; x < max_x; x++)
+            {
+                int n0 = horizontal_contributors[x].n0;
+                int n1 = horizontal_contributors[x].n1;
+
+                int in_x = x - filter_pixel_margin;
+                int in_pixel_index = in_x * 3;
+                int max_n = n1;
+                int coefficient_group = coefficient_width * x;
+
+                for (k = n0; k <= max_n; k++)
+                {
+                    int out_pixel_index = k * 3;
+                    float coefficient = horizontal_coefficients[coefficient_group + k - n0];
+                    output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
+                    output_buffer[out_pixel_index + 1] += decode_buffer[in_pixel_index + 1] * coefficient;
+                    output_buffer[out_pixel_index + 2] += decode_buffer[in_pixel_index + 2] * coefficient;
+                }
+            }
+            break;
+
+        case 4:
+            for (x = 0; x < max_x; x++)
+            {
+                int n0 = horizontal_contributors[x].n0;
+                int n1 = horizontal_contributors[x].n1;
+
+                int in_x = x - filter_pixel_margin;
+                int in_pixel_index = in_x * 4;
+                int max_n = n1;
+                int coefficient_group = coefficient_width * x;
+
+                for (k = n0; k <= max_n; k++)
+                {
+                    int out_pixel_index = k * 4;
+                    float coefficient = horizontal_coefficients[coefficient_group + k - n0];
+                    output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient;
+                    output_buffer[out_pixel_index + 1] += decode_buffer[in_pixel_index + 1] * coefficient;
+                    output_buffer[out_pixel_index + 2] += decode_buffer[in_pixel_index + 2] * coefficient;
+                    output_buffer[out_pixel_index + 3] += decode_buffer[in_pixel_index + 3] * coefficient;
+                }
+            }
+            break;
+
+        default:
+            for (x = 0; x < max_x; x++)
+            {
+                int n0 = horizontal_contributors[x].n0;
+                int n1 = horizontal_contributors[x].n1;
+
+                int in_x = x - filter_pixel_margin;
+                int in_pixel_index = in_x * channels;
+                int max_n = n1;
+                int coefficient_group = coefficient_width * x;
+
+                for (k = n0; k <= max_n; k++)
+                {
+                    int c;
+                    int out_pixel_index = k * channels;
+                    float coefficient = horizontal_coefficients[coefficient_group + k - n0];
+                    for (c = 0; c < channels; c++)
+                        output_buffer[out_pixel_index + c] += decode_buffer[in_pixel_index + c] * coefficient;
+                }
+            }
+            break;
+    }
+}
+
+static void stbir__decode_and_resample_upsample(stbir__info* stbir_info, int n)
+{
+    // Decode the nth scanline from the source image into the decode buffer.
+    stbir__decode_scanline(stbir_info, n);
+
+    // Now resample it into the ring buffer.
+    if (stbir__use_width_upsampling(stbir_info))
+        stbir__resample_horizontal_upsample(stbir_info, stbir__add_empty_ring_buffer_entry(stbir_info, n));
+    else
+        stbir__resample_horizontal_downsample(stbir_info, stbir__add_empty_ring_buffer_entry(stbir_info, n));
+
+    // Now it's sitting in the ring buffer ready to be used as source for the vertical sampling.
+}
+
+static void stbir__decode_and_resample_downsample(stbir__info* stbir_info, int n)
+{
+    // Decode the nth scanline from the source image into the decode buffer.
+    stbir__decode_scanline(stbir_info, n);
+
+    memset(stbir_info->horizontal_buffer, 0, stbir_info->output_w * stbir_info->channels * sizeof(float));
+
+    // Now resample it into the horizontal buffer.
+    if (stbir__use_width_upsampling(stbir_info))
+        stbir__resample_horizontal_upsample(stbir_info, stbir_info->horizontal_buffer);
+    else
+        stbir__resample_horizontal_downsample(stbir_info, stbir_info->horizontal_buffer);
+
+    // Now it's sitting in the horizontal buffer ready to be distributed into the ring buffers.
+}
+
+// Get the specified scan line from the ring buffer.
+static float* stbir__get_ring_buffer_scanline(int get_scanline, float* ring_buffer, int begin_index, int first_scanline, int ring_buffer_num_entries, int ring_buffer_length)
+{
+    int ring_buffer_index = (begin_index + (get_scanline - first_scanline)) % ring_buffer_num_entries;
+    return stbir__get_ring_buffer_entry(ring_buffer, ring_buffer_index, ring_buffer_length);
+}
+
+
+static void stbir__encode_scanline(stbir__info* stbir_info, int num_pixels, void *output_buffer, float *encode_buffer, int channels, int alpha_channel, int decode)
+{
+    int x;
+    int n;
+    int num_nonalpha;
+    stbir_uint16 nonalpha[STBIR_MAX_CHANNELS];
+
+    if (!(stbir_info->flags&STBIR_FLAG_ALPHA_PREMULTIPLIED))
+    {
+        for (x=0; x < num_pixels; ++x)
+        {
+            int pixel_index = x*channels;
+
+            float alpha = encode_buffer[pixel_index + alpha_channel];
+            float reciprocal_alpha = alpha ? 1.0f / alpha : 0;
+
+            // unrolling this produced a 1% slowdown upscaling a large RGBA linear-space image on my machine - stb
+            for (n = 0; n < channels; n++)
+                if (n != alpha_channel)
+                    encode_buffer[pixel_index + n] *= reciprocal_alpha;
+
+            // We added in a small epsilon to prevent the color channel from being deleted with zero alpha.
+            // Because we only add it for integer types, it will automatically be discarded on integer
+            // conversion, so we don't need to subtract it back out (which would be problematic for
+            // numeric precision reasons).
+        }
+    }
+
+    // build a table of all channels that need colorspace correction, so
+    // we don't perform colorspace correction on channels that don't need it.
+    for (x = 0, num_nonalpha = 0; x < channels; ++x)
+    {
+        if (x != alpha_channel || (stbir_info->flags & STBIR_FLAG_ALPHA_USES_COLORSPACE))
+        {
+            nonalpha[num_nonalpha++] = (stbir_uint16)x;
+        }
+    }
+
+    #define STBIR__ROUND_INT(f)    ((int)          ((f)+0.5))
+    #define STBIR__ROUND_UINT(f)   ((stbir_uint32) ((f)+0.5))
+
+    #ifdef STBIR__SATURATE_INT
+    #define STBIR__ENCODE_LINEAR8(f)   stbir__saturate8 (STBIR__ROUND_INT((f) * stbir__max_uint8_as_float ))
+    #define STBIR__ENCODE_LINEAR16(f)  stbir__saturate16(STBIR__ROUND_INT((f) * stbir__max_uint16_as_float))
+    #else
+    #define STBIR__ENCODE_LINEAR8(f)   (unsigned char ) STBIR__ROUND_INT(stbir__saturate(f) * stbir__max_uint8_as_float )
+    #define STBIR__ENCODE_LINEAR16(f)  (unsigned short) STBIR__ROUND_INT(stbir__saturate(f) * stbir__max_uint16_as_float)
+    #endif
+
+    switch (decode)
+    {
+        case STBIR__DECODE(STBIR_TYPE_UINT8, STBIR_COLORSPACE_LINEAR):
+            for (x=0; x < num_pixels; ++x)
+            {
+                int pixel_index = x*channels;
+
+                for (n = 0; n < channels; n++)
+                {
+                    int index = pixel_index + n;
+                    ((unsigned char*)output_buffer)[index] = STBIR__ENCODE_LINEAR8(encode_buffer[index]);
+                }
+            }
+            break;
+
+        case STBIR__DECODE(STBIR_TYPE_UINT8, STBIR_COLORSPACE_SRGB):
+            for (x=0; x < num_pixels; ++x)
+            {
+                int pixel_index = x*channels;
+
+                for (n = 0; n < num_nonalpha; n++)
+                {
+                    int index = pixel_index + nonalpha[n];
+                    ((unsigned char*)output_buffer)[index] = stbir__linear_to_srgb_uchar(encode_buffer[index]);
+                }
+
+                if (!(stbir_info->flags & STBIR_FLAG_ALPHA_USES_COLORSPACE))
+                    ((unsigned char *)output_buffer)[pixel_index + alpha_channel] = STBIR__ENCODE_LINEAR8(encode_buffer[pixel_index+alpha_channel]);
+            }
+            break;
+
+        case STBIR__DECODE(STBIR_TYPE_UINT16, STBIR_COLORSPACE_LINEAR):
+            for (x=0; x < num_pixels; ++x)
+            {
+                int pixel_index = x*channels;
+
+                for (n = 0; n < channels; n++)
+                {
+                    int index = pixel_index + n;
+                    ((unsigned short*)output_buffer)[index] = STBIR__ENCODE_LINEAR16(encode_buffer[index]);
+                }
+            }
+            break;
+
+        case STBIR__DECODE(STBIR_TYPE_UINT16, STBIR_COLORSPACE_SRGB):
+            for (x=0; x < num_pixels; ++x)
+            {
+                int pixel_index = x*channels;
+
+                for (n = 0; n < num_nonalpha; n++)
+                {
+                    int index = pixel_index + nonalpha[n];
+                    ((unsigned short*)output_buffer)[index] = (unsigned short)STBIR__ROUND_INT(stbir__linear_to_srgb(stbir__saturate(encode_buffer[index])) * stbir__max_uint16_as_float);
+                }
+
+                if (!(stbir_info->flags&STBIR_FLAG_ALPHA_USES_COLORSPACE))
+                    ((unsigned short*)output_buffer)[pixel_index + alpha_channel] = STBIR__ENCODE_LINEAR16(encode_buffer[pixel_index + alpha_channel]);
+            }
+
+            break;
+
+        case STBIR__DECODE(STBIR_TYPE_UINT32, STBIR_COLORSPACE_LINEAR):
+            for (x=0; x < num_pixels; ++x)
+            {
+                int pixel_index = x*channels;
+
+                for (n = 0; n < channels; n++)
+                {
+                    int index = pixel_index + n;
+                    ((unsigned int*)output_buffer)[index] = (unsigned int)STBIR__ROUND_UINT(((double)stbir__saturate(encode_buffer[index])) * stbir__max_uint32_as_float);
+                }
+            }
+            break;
+
+        case STBIR__DECODE(STBIR_TYPE_UINT32, STBIR_COLORSPACE_SRGB):
+            for (x=0; x < num_pixels; ++x)
+            {
+                int pixel_index = x*channels;
+
+                for (n = 0; n < num_nonalpha; n++)
+                {
+                    int index = pixel_index + nonalpha[n];
+                    ((unsigned int*)output_buffer)[index] = (unsigned int)STBIR__ROUND_UINT(((double)stbir__linear_to_srgb(stbir__saturate(encode_buffer[index]))) * stbir__max_uint32_as_float);
+                }
+
+                if (!(stbir_info->flags&STBIR_FLAG_ALPHA_USES_COLORSPACE))
+                    ((unsigned int*)output_buffer)[pixel_index + alpha_channel] = (unsigned int)STBIR__ROUND_INT(((double)stbir__saturate(encode_buffer[pixel_index + alpha_channel])) * stbir__max_uint32_as_float);
+            }
+            break;
+
+        case STBIR__DECODE(STBIR_TYPE_FLOAT, STBIR_COLORSPACE_LINEAR):
+            for (x=0; x < num_pixels; ++x)
+            {
+                int pixel_index = x*channels;
+
+                for (n = 0; n < channels; n++)
+                {
+                    int index = pixel_index + n;
+                    ((float*)output_buffer)[index] = encode_buffer[index];
+                }
+            }
+            break;
+
+        case STBIR__DECODE(STBIR_TYPE_FLOAT, STBIR_COLORSPACE_SRGB):
+            for (x=0; x < num_pixels; ++x)
+            {
+                int pixel_index = x*channels;
+
+                for (n = 0; n < num_nonalpha; n++)
+                {
+                    int index = pixel_index + nonalpha[n];
+                    ((float*)output_buffer)[index] = stbir__linear_to_srgb(encode_buffer[index]);
+                }
+
+                if (!(stbir_info->flags&STBIR_FLAG_ALPHA_USES_COLORSPACE))
+                    ((float*)output_buffer)[pixel_index + alpha_channel] = encode_buffer[pixel_index + alpha_channel];
+            }
+            break;
+
+        default:
+            STBIR_ASSERT(!"Unknown type/colorspace/channels combination.");
+            break;
+    }
+}
+
+static void stbir__resample_vertical_upsample(stbir__info* stbir_info, int n)
+{
+    int x, k;
+    int output_w = stbir_info->output_w;
+    stbir__contributors* vertical_contributors = stbir_info->vertical_contributors;
+    float* vertical_coefficients = stbir_info->vertical_coefficients;
+    int channels = stbir_info->channels;
+    int alpha_channel = stbir_info->alpha_channel;
+    int type = stbir_info->type;
+    int colorspace = stbir_info->colorspace;
+    int ring_buffer_entries = stbir_info->ring_buffer_num_entries;
+    void* output_data = stbir_info->output_data;
+    float* encode_buffer = stbir_info->encode_buffer;
+    int decode = STBIR__DECODE(type, colorspace);
+    int coefficient_width = stbir_info->vertical_coefficient_width;
+    int coefficient_counter;
+    int contributor = n;
+
+    float* ring_buffer = stbir_info->ring_buffer;
+    int ring_buffer_begin_index = stbir_info->ring_buffer_begin_index;
+    int ring_buffer_first_scanline = stbir_info->ring_buffer_first_scanline;
+    int ring_buffer_length = stbir_info->ring_buffer_length_bytes/sizeof(float);
+
+    int n0,n1, output_row_start;
+    int coefficient_group = coefficient_width * contributor;
+
+    n0 = vertical_contributors[contributor].n0;
+    n1 = vertical_contributors[contributor].n1;
+
+    output_row_start = n * stbir_info->output_stride_bytes;
+
+    STBIR_ASSERT(stbir__use_height_upsampling(stbir_info));
+
+    memset(encode_buffer, 0, output_w * sizeof(float) * channels);
+
+    // I tried reblocking this for better cache usage of encode_buffer
+    // (using x_outer, k, x_inner), but it lost speed. -- stb
+
+    coefficient_counter = 0;
+    switch (channels) {
+        case 1:
+            for (k = n0; k <= n1; k++)
+            {
+                int coefficient_index = coefficient_counter++;
+                float* ring_buffer_entry = stbir__get_ring_buffer_scanline(k, ring_buffer, ring_buffer_begin_index, ring_buffer_first_scanline, ring_buffer_entries, ring_buffer_length);
+                float coefficient = vertical_coefficients[coefficient_group + coefficient_index];
+                for (x = 0; x < output_w; ++x)
+                {
+                    int in_pixel_index = x * 1;
+                    encode_buffer[in_pixel_index + 0] += ring_buffer_entry[in_pixel_index + 0] * coefficient;
+                }
+            }
+            break;
+        case 2:
+            for (k = n0; k <= n1; k++)
+            {
+                int coefficient_index = coefficient_counter++;
+                float* ring_buffer_entry = stbir__get_ring_buffer_scanline(k, ring_buffer, ring_buffer_begin_index, ring_buffer_first_scanline, ring_buffer_entries, ring_buffer_length);
+                float coefficient = vertical_coefficients[coefficient_group + coefficient_index];
+                for (x = 0; x < output_w; ++x)
+                {
+                    int in_pixel_index = x * 2;
+                    encode_buffer[in_pixel_index + 0] += ring_buffer_entry[in_pixel_index + 0] * coefficient;
+                    encode_buffer[in_pixel_index + 1] += ring_buffer_entry[in_pixel_index + 1] * coefficient;
+                }
+            }
+            break;
+        case 3:
+            for (k = n0; k <= n1; k++)
+            {
+                int coefficient_index = coefficient_counter++;
+                float* ring_buffer_entry = stbir__get_ring_buffer_scanline(k, ring_buffer, ring_buffer_begin_index, ring_buffer_first_scanline, ring_buffer_entries, ring_buffer_length);
+                float coefficient = vertical_coefficients[coefficient_group + coefficient_index];
+                for (x = 0; x < output_w; ++x)
+                {
+                    int in_pixel_index = x * 3;
+                    encode_buffer[in_pixel_index + 0] += ring_buffer_entry[in_pixel_index + 0] * coefficient;
+                    encode_buffer[in_pixel_index + 1] += ring_buffer_entry[in_pixel_index + 1] * coefficient;
+                    encode_buffer[in_pixel_index + 2] += ring_buffer_entry[in_pixel_index + 2] * coefficient;
+                }
+            }
+            break;
+        case 4:
+            for (k = n0; k <= n1; k++)
+            {
+                int coefficient_index = coefficient_counter++;
+                float* ring_buffer_entry = stbir__get_ring_buffer_scanline(k, ring_buffer, ring_buffer_begin_index, ring_buffer_first_scanline, ring_buffer_entries, ring_buffer_length);
+                float coefficient = vertical_coefficients[coefficient_group + coefficient_index];
+                for (x = 0; x < output_w; ++x)
+                {
+                    int in_pixel_index = x * 4;
+                    encode_buffer[in_pixel_index + 0] += ring_buffer_entry[in_pixel_index + 0] * coefficient;
+                    encode_buffer[in_pixel_index + 1] += ring_buffer_entry[in_pixel_index + 1] * coefficient;
+                    encode_buffer[in_pixel_index + 2] += ring_buffer_entry[in_pixel_index + 2] * coefficient;
+                    encode_buffer[in_pixel_index + 3] += ring_buffer_entry[in_pixel_index + 3] * coefficient;
+                }
+            }
+            break;
+        default:
+            for (k = n0; k <= n1; k++)
+            {
+                int coefficient_index = coefficient_counter++;
+                float* ring_buffer_entry = stbir__get_ring_buffer_scanline(k, ring_buffer, ring_buffer_begin_index, ring_buffer_first_scanline, ring_buffer_entries, ring_buffer_length);
+                float coefficient = vertical_coefficients[coefficient_group + coefficient_index];
+                for (x = 0; x < output_w; ++x)
+                {
+                    int in_pixel_index = x * channels;
+                    int c;
+                    for (c = 0; c < channels; c++)
+                        encode_buffer[in_pixel_index + c] += ring_buffer_entry[in_pixel_index + c] * coefficient;
+                }
+            }
+            break;
+    }
+    stbir__encode_scanline(stbir_info, output_w, (char *) output_data + output_row_start, encode_buffer, channels, alpha_channel, decode);
+}
+
+static void stbir__resample_vertical_downsample(stbir__info* stbir_info, int n)
+{
+    int x, k;
+    int output_w = stbir_info->output_w;
+    stbir__contributors* vertical_contributors = stbir_info->vertical_contributors;
+    float* vertical_coefficients = stbir_info->vertical_coefficients;
+    int channels = stbir_info->channels;
+    int ring_buffer_entries = stbir_info->ring_buffer_num_entries;
+    float* horizontal_buffer = stbir_info->horizontal_buffer;
+    int coefficient_width = stbir_info->vertical_coefficient_width;
+    int contributor = n + stbir_info->vertical_filter_pixel_margin;
+
+    float* ring_buffer = stbir_info->ring_buffer;
+    int ring_buffer_begin_index = stbir_info->ring_buffer_begin_index;
+    int ring_buffer_first_scanline = stbir_info->ring_buffer_first_scanline;
+    int ring_buffer_length = stbir_info->ring_buffer_length_bytes/sizeof(float);
+    int n0,n1;
+
+    n0 = vertical_contributors[contributor].n0;
+    n1 = vertical_contributors[contributor].n1;
+
+    STBIR_ASSERT(!stbir__use_height_upsampling(stbir_info));
+
+    for (k = n0; k <= n1; k++)
+    {
+        int coefficient_index = k - n0;
+        int coefficient_group = coefficient_width * contributor;
+        float coefficient = vertical_coefficients[coefficient_group + coefficient_index];
+
+        float* ring_buffer_entry = stbir__get_ring_buffer_scanline(k, ring_buffer, ring_buffer_begin_index, ring_buffer_first_scanline, ring_buffer_entries, ring_buffer_length);
+
+        switch (channels) {
+            case 1:
+                for (x = 0; x < output_w; x++)
+                {
+                    int in_pixel_index = x * 1;
+                    ring_buffer_entry[in_pixel_index + 0] += horizontal_buffer[in_pixel_index + 0] * coefficient;
+                }
+                break;
+            case 2:
+                for (x = 0; x < output_w; x++)
+                {
+                    int in_pixel_index = x * 2;
+                    ring_buffer_entry[in_pixel_index + 0] += horizontal_buffer[in_pixel_index + 0] * coefficient;
+                    ring_buffer_entry[in_pixel_index + 1] += horizontal_buffer[in_pixel_index + 1] * coefficient;
+                }
+                break;
+            case 3:
+                for (x = 0; x < output_w; x++)
+                {
+                    int in_pixel_index = x * 3;
+                    ring_buffer_entry[in_pixel_index + 0] += horizontal_buffer[in_pixel_index + 0] * coefficient;
+                    ring_buffer_entry[in_pixel_index + 1] += horizontal_buffer[in_pixel_index + 1] * coefficient;
+                    ring_buffer_entry[in_pixel_index + 2] += horizontal_buffer[in_pixel_index + 2] * coefficient;
+                }
+                break;
+            case 4:
+                for (x = 0; x < output_w; x++)
+                {
+                    int in_pixel_index = x * 4;
+                    ring_buffer_entry[in_pixel_index + 0] += horizontal_buffer[in_pixel_index + 0] * coefficient;
+                    ring_buffer_entry[in_pixel_index + 1] += horizontal_buffer[in_pixel_index + 1] * coefficient;
+                    ring_buffer_entry[in_pixel_index + 2] += horizontal_buffer[in_pixel_index + 2] * coefficient;
+                    ring_buffer_entry[in_pixel_index + 3] += horizontal_buffer[in_pixel_index + 3] * coefficient;
+                }
+                break;
+            default:
+                for (x = 0; x < output_w; x++)
+                {
+                    int in_pixel_index = x * channels;
+
+                    int c;
+                    for (c = 0; c < channels; c++)
+                        ring_buffer_entry[in_pixel_index + c] += horizontal_buffer[in_pixel_index + c] * coefficient;
+                }
+                break;
+        }
+    }
+}
+
+static void stbir__buffer_loop_upsample(stbir__info* stbir_info)
+{
+    int y;
+    float scale_ratio = stbir_info->vertical_scale;
+    float out_scanlines_radius = stbir__filter_info_table[stbir_info->vertical_filter].support(1/scale_ratio) * scale_ratio;
+
+    STBIR_ASSERT(stbir__use_height_upsampling(stbir_info));
+
+    for (y = 0; y < stbir_info->output_h; y++)
+    {
+        float in_center_of_out = 0; // Center of the current out scanline in the in scanline space
+        int in_first_scanline = 0, in_last_scanline = 0;
+
+        stbir__calculate_sample_range_upsample(y, out_scanlines_radius, scale_ratio, stbir_info->vertical_shift, &in_first_scanline, &in_last_scanline, &in_center_of_out);
+
+        STBIR_ASSERT(in_last_scanline - in_first_scanline + 1 <= stbir_info->ring_buffer_num_entries);
+
+        if (stbir_info->ring_buffer_begin_index >= 0)
+        {
+            // Get rid of whatever we don't need anymore.
+            while (in_first_scanline > stbir_info->ring_buffer_first_scanline)
+            {
+                if (stbir_info->ring_buffer_first_scanline == stbir_info->ring_buffer_last_scanline)
+                {
+                    // We just popped the last scanline off the ring buffer.
+                    // Reset it to the empty state.
+                    stbir_info->ring_buffer_begin_index = -1;
+                    stbir_info->ring_buffer_first_scanline = 0;
+                    stbir_info->ring_buffer_last_scanline = 0;
+                    break;
+                }
+                else
+                {
+                    stbir_info->ring_buffer_first_scanline++;
+                    stbir_info->ring_buffer_begin_index = (stbir_info->ring_buffer_begin_index + 1) % stbir_info->ring_buffer_num_entries;
+                }
+            }
+        }
+
+        // Load in new ones.
+        if (stbir_info->ring_buffer_begin_index < 0)
+            stbir__decode_and_resample_upsample(stbir_info, in_first_scanline);
+
+        while (in_last_scanline > stbir_info->ring_buffer_last_scanline)
+            stbir__decode_and_resample_upsample(stbir_info, stbir_info->ring_buffer_last_scanline + 1);
+
+        // Now all buffers should be ready to write a row of vertical sampling.
+        stbir__resample_vertical_upsample(stbir_info, y);
+
+        STBIR_PROGRESS_REPORT((float)y / stbir_info->output_h);
+    }
+}
+
+static void stbir__empty_ring_buffer(stbir__info* stbir_info, int first_necessary_scanline)
+{
+    int output_stride_bytes = stbir_info->output_stride_bytes;
+    int channels = stbir_info->channels;
+    int alpha_channel = stbir_info->alpha_channel;
+    int type = stbir_info->type;
+    int colorspace = stbir_info->colorspace;
+    int output_w = stbir_info->output_w;
+    void* output_data = stbir_info->output_data;
+    int decode = STBIR__DECODE(type, colorspace);
+
+    float* ring_buffer = stbir_info->ring_buffer;
+    int ring_buffer_length = stbir_info->ring_buffer_length_bytes/sizeof(float);
+
+    if (stbir_info->ring_buffer_begin_index >= 0)
+    {
+        // Get rid of whatever we don't need anymore.
+        while (first_necessary_scanline > stbir_info->ring_buffer_first_scanline)
+        {
+            if (stbir_info->ring_buffer_first_scanline >= 0 && stbir_info->ring_buffer_first_scanline < stbir_info->output_h)
+            {
+                int output_row_start = stbir_info->ring_buffer_first_scanline * output_stride_bytes;
+                float* ring_buffer_entry = stbir__get_ring_buffer_entry(ring_buffer, stbir_info->ring_buffer_begin_index, ring_buffer_length);
+                stbir__encode_scanline(stbir_info, output_w, (char *) output_data + output_row_start, ring_buffer_entry, channels, alpha_channel, decode);
+                STBIR_PROGRESS_REPORT((float)stbir_info->ring_buffer_first_scanline / stbir_info->output_h);
+            }
+
+            if (stbir_info->ring_buffer_first_scanline == stbir_info->ring_buffer_last_scanline)
+            {
+                // We just popped the last scanline off the ring buffer.
+                // Reset it to the empty state.
+                stbir_info->ring_buffer_begin_index = -1;
+                stbir_info->ring_buffer_first_scanline = 0;
+                stbir_info->ring_buffer_last_scanline = 0;
+                break;
+            }
+            else
+            {
+                stbir_info->ring_buffer_first_scanline++;
+                stbir_info->ring_buffer_begin_index = (stbir_info->ring_buffer_begin_index + 1) % stbir_info->ring_buffer_num_entries;
+            }
+        }
+    }
+}
+
+static void stbir__buffer_loop_downsample(stbir__info* stbir_info)
+{
+    int y;
+    float scale_ratio = stbir_info->vertical_scale;
+    int output_h = stbir_info->output_h;
+    float in_pixels_radius = stbir__filter_info_table[stbir_info->vertical_filter].support(scale_ratio) / scale_ratio;
+    int pixel_margin = stbir_info->vertical_filter_pixel_margin;
+    int max_y = stbir_info->input_h + pixel_margin;
+
+    STBIR_ASSERT(!stbir__use_height_upsampling(stbir_info));
+
+    for (y = -pixel_margin; y < max_y; y++)
+    {
+        float out_center_of_in; // Center of the current out scanline in the in scanline space
+        int out_first_scanline, out_last_scanline;
+
+        stbir__calculate_sample_range_downsample(y, in_pixels_radius, scale_ratio, stbir_info->vertical_shift, &out_first_scanline, &out_last_scanline, &out_center_of_in);
+
+        STBIR_ASSERT(out_last_scanline - out_first_scanline + 1 <= stbir_info->ring_buffer_num_entries);
+
+        if (out_last_scanline < 0 || out_first_scanline >= output_h)
+            continue;
+
+        stbir__empty_ring_buffer(stbir_info, out_first_scanline);
+
+        stbir__decode_and_resample_downsample(stbir_info, y);
+
+        // Load in new ones.
+        if (stbir_info->ring_buffer_begin_index < 0)
+            stbir__add_empty_ring_buffer_entry(stbir_info, out_first_scanline);
+
+        while (out_last_scanline > stbir_info->ring_buffer_last_scanline)
+            stbir__add_empty_ring_buffer_entry(stbir_info, stbir_info->ring_buffer_last_scanline + 1);
+
+        // Now the horizontal buffer is ready to write to all ring buffer rows.
+        stbir__resample_vertical_downsample(stbir_info, y);
+    }
+
+    stbir__empty_ring_buffer(stbir_info, stbir_info->output_h);
+}
+
+static void stbir__setup(stbir__info *info, int input_w, int input_h, int output_w, int output_h, int channels)
+{
+    info->input_w = input_w;
+    info->input_h = input_h;
+    info->output_w = output_w;
+    info->output_h = output_h;
+    info->channels = channels;
+}
+
+static void stbir__calculate_transform(stbir__info *info, float s0, float t0, float s1, float t1, float *transform)
+{
+    info->s0 = s0;
+    info->t0 = t0;
+    info->s1 = s1;
+    info->t1 = t1;
+
+    if (transform)
+    {
+        info->horizontal_scale = transform[0];
+        info->vertical_scale   = transform[1];
+        info->horizontal_shift = transform[2];
+        info->vertical_shift   = transform[3];
+    }
+    else
+    {
+        info->horizontal_scale = ((float)info->output_w / info->input_w) / (s1 - s0);
+        info->vertical_scale = ((float)info->output_h / info->input_h) / (t1 - t0);
+
+        info->horizontal_shift = s0 * info->output_w / (s1 - s0);
+        info->vertical_shift = t0 * info->output_h / (t1 - t0);
+    }
+}
+
+static void stbir__choose_filter(stbir__info *info, stbir_filter h_filter, stbir_filter v_filter)
+{
+    if (h_filter == 0)
+        h_filter = stbir__use_upsampling(info->horizontal_scale) ? STBIR_DEFAULT_FILTER_UPSAMPLE : STBIR_DEFAULT_FILTER_DOWNSAMPLE;
+    if (v_filter == 0)
+        v_filter = stbir__use_upsampling(info->vertical_scale)   ? STBIR_DEFAULT_FILTER_UPSAMPLE : STBIR_DEFAULT_FILTER_DOWNSAMPLE;
+    info->horizontal_filter = h_filter;
+    info->vertical_filter = v_filter;
+}
+
+static stbir_uint32 stbir__calculate_memory(stbir__info *info)
+{
+    int pixel_margin = stbir__get_filter_pixel_margin(info->horizontal_filter, info->horizontal_scale);
+    int filter_height = stbir__get_filter_pixel_width(info->vertical_filter, info->vertical_scale);
+
+    info->horizontal_num_contributors = stbir__get_contributors(info->horizontal_scale, info->horizontal_filter, info->input_w, info->output_w);
+    info->vertical_num_contributors   = stbir__get_contributors(info->vertical_scale  , info->vertical_filter  , info->input_h, info->output_h);
+
+    // One extra entry because floating point precision problems sometimes cause an extra to be necessary.
+    info->ring_buffer_num_entries = filter_height + 1;
+
+    info->horizontal_contributors_size = info->horizontal_num_contributors * sizeof(stbir__contributors);
+    info->horizontal_coefficients_size = stbir__get_total_horizontal_coefficients(info) * sizeof(float);
+    info->vertical_contributors_size = info->vertical_num_contributors * sizeof(stbir__contributors);
+    info->vertical_coefficients_size = stbir__get_total_vertical_coefficients(info) * sizeof(float);
+    info->decode_buffer_size = (info->input_w + pixel_margin * 2) * info->channels * sizeof(float);
+    info->horizontal_buffer_size = info->output_w * info->channels * sizeof(float);
+    info->ring_buffer_size = info->output_w * info->channels * info->ring_buffer_num_entries * sizeof(float);
+    info->encode_buffer_size = info->output_w * info->channels * sizeof(float);
+
+    STBIR_ASSERT(info->horizontal_filter != 0);
+    STBIR_ASSERT(info->horizontal_filter < STBIR__ARRAY_SIZE(stbir__filter_info_table)); // this now happens too late
+    STBIR_ASSERT(info->vertical_filter != 0);
+    STBIR_ASSERT(info->vertical_filter < STBIR__ARRAY_SIZE(stbir__filter_info_table)); // this now happens too late
+
+    if (stbir__use_height_upsampling(info))
+        // The horizontal buffer is for when we're downsampling the height and we
+        // can't output the result of sampling the decode buffer directly into the
+        // ring buffers.
+        info->horizontal_buffer_size = 0;
+    else
+        // The encode buffer is to retain precision in the height upsampling method
+        // and isn't used when height downsampling.
+        info->encode_buffer_size = 0;
+
+    return info->horizontal_contributors_size + info->horizontal_coefficients_size
+        + info->vertical_contributors_size + info->vertical_coefficients_size
+        + info->decode_buffer_size + info->horizontal_buffer_size
+        + info->ring_buffer_size + info->encode_buffer_size;
+}
+
+static int stbir__resize_allocated(stbir__info *info,
+    const void* input_data, int input_stride_in_bytes,
+    void* output_data, int output_stride_in_bytes,
+    int alpha_channel, stbir_uint32 flags, stbir_datatype type,
+    stbir_edge edge_horizontal, stbir_edge edge_vertical, stbir_colorspace colorspace,
+    void* tempmem, size_t tempmem_size_in_bytes)
+{
+    size_t memory_required = stbir__calculate_memory(info);
+
+    int width_stride_input = input_stride_in_bytes ? input_stride_in_bytes : info->channels * info->input_w * stbir__type_size[type];
+    int width_stride_output = output_stride_in_bytes ? output_stride_in_bytes : info->channels * info->output_w * stbir__type_size[type];
+
+#ifdef STBIR_DEBUG_OVERWRITE_TEST
+#define OVERWRITE_ARRAY_SIZE 8
+    unsigned char overwrite_output_before_pre[OVERWRITE_ARRAY_SIZE];
+    unsigned char overwrite_tempmem_before_pre[OVERWRITE_ARRAY_SIZE];
+    unsigned char overwrite_output_after_pre[OVERWRITE_ARRAY_SIZE];
+    unsigned char overwrite_tempmem_after_pre[OVERWRITE_ARRAY_SIZE];
+
+    size_t begin_forbidden = width_stride_output * (info->output_h - 1) + info->output_w * info->channels * stbir__type_size[type];
+    memcpy(overwrite_output_before_pre, &((unsigned char*)output_data)[-OVERWRITE_ARRAY_SIZE], OVERWRITE_ARRAY_SIZE);
+    memcpy(overwrite_output_after_pre, &((unsigned char*)output_data)[begin_forbidden], OVERWRITE_ARRAY_SIZE);
+    memcpy(overwrite_tempmem_before_pre, &((unsigned char*)tempmem)[-OVERWRITE_ARRAY_SIZE], OVERWRITE_ARRAY_SIZE);
+    memcpy(overwrite_tempmem_after_pre, &((unsigned char*)tempmem)[tempmem_size_in_bytes], OVERWRITE_ARRAY_SIZE);
+#endif
+
+    STBIR_ASSERT(info->channels >= 0);
+    STBIR_ASSERT(info->channels <= STBIR_MAX_CHANNELS);
+
+    if (info->channels < 0 || info->channels > STBIR_MAX_CHANNELS)
+        return 0;
+
+    STBIR_ASSERT(info->horizontal_filter < STBIR__ARRAY_SIZE(stbir__filter_info_table));
+    STBIR_ASSERT(info->vertical_filter < STBIR__ARRAY_SIZE(stbir__filter_info_table));
+
+    if (info->horizontal_filter >= STBIR__ARRAY_SIZE(stbir__filter_info_table))
+        return 0;
+    if (info->vertical_filter >= STBIR__ARRAY_SIZE(stbir__filter_info_table))
+        return 0;
+
+    if (alpha_channel < 0)
+        flags |= STBIR_FLAG_ALPHA_USES_COLORSPACE | STBIR_FLAG_ALPHA_PREMULTIPLIED;
+
+    if (!(flags&STBIR_FLAG_ALPHA_USES_COLORSPACE) || !(flags&STBIR_FLAG_ALPHA_PREMULTIPLIED)) {
+        STBIR_ASSERT(alpha_channel >= 0 && alpha_channel < info->channels);
+    }
+
+    if (alpha_channel >= info->channels)
+        return 0;
+
+    STBIR_ASSERT(tempmem);
+
+    if (!tempmem)
+        return 0;
+
+    STBIR_ASSERT(tempmem_size_in_bytes >= memory_required);
+
+    if (tempmem_size_in_bytes < memory_required)
+        return 0;
+
+    memset(tempmem, 0, tempmem_size_in_bytes);
+
+    info->input_data = input_data;
+    info->input_stride_bytes = width_stride_input;
+
+    info->output_data = output_data;
+    info->output_stride_bytes = width_stride_output;
+
+    info->alpha_channel = alpha_channel;
+    info->flags = flags;
+    info->type = type;
+    info->edge_horizontal = edge_horizontal;
+    info->edge_vertical = edge_vertical;
+    info->colorspace = colorspace;
+
+    info->horizontal_coefficient_width   = stbir__get_coefficient_width  (info->horizontal_filter, info->horizontal_scale);
+    info->vertical_coefficient_width     = stbir__get_coefficient_width  (info->vertical_filter  , info->vertical_scale  );
+    info->horizontal_filter_pixel_width  = stbir__get_filter_pixel_width (info->horizontal_filter, info->horizontal_scale);
+    info->vertical_filter_pixel_width    = stbir__get_filter_pixel_width (info->vertical_filter  , info->vertical_scale  );
+    info->horizontal_filter_pixel_margin = stbir__get_filter_pixel_margin(info->horizontal_filter, info->horizontal_scale);
+    info->vertical_filter_pixel_margin   = stbir__get_filter_pixel_margin(info->vertical_filter  , info->vertical_scale  );
+
+    info->ring_buffer_length_bytes = info->output_w * info->channels * sizeof(float);
+    info->decode_buffer_pixels = info->input_w + info->horizontal_filter_pixel_margin * 2;
+
+#define STBIR__NEXT_MEMPTR(current, newtype) (newtype*)(((unsigned char*)current) + current##_size)
+
+    info->horizontal_contributors = (stbir__contributors *) tempmem;
+    info->horizontal_coefficients = STBIR__NEXT_MEMPTR(info->horizontal_contributors, float);
+    info->vertical_contributors = STBIR__NEXT_MEMPTR(info->horizontal_coefficients, stbir__contributors);
+    info->vertical_coefficients = STBIR__NEXT_MEMPTR(info->vertical_contributors, float);
+    info->decode_buffer = STBIR__NEXT_MEMPTR(info->vertical_coefficients, float);
+
+    if (stbir__use_height_upsampling(info))
+    {
+        info->horizontal_buffer = NULL;
+        info->ring_buffer = STBIR__NEXT_MEMPTR(info->decode_buffer, float);
+        info->encode_buffer = STBIR__NEXT_MEMPTR(info->ring_buffer, float);
+
+        STBIR_ASSERT((size_t)STBIR__NEXT_MEMPTR(info->encode_buffer, unsigned char) == (size_t)tempmem + tempmem_size_in_bytes);
+    }
+    else
+    {
+        info->horizontal_buffer = STBIR__NEXT_MEMPTR(info->decode_buffer, float);
+        info->ring_buffer = STBIR__NEXT_MEMPTR(info->horizontal_buffer, float);
+        info->encode_buffer = NULL;
+
+        STBIR_ASSERT((size_t)STBIR__NEXT_MEMPTR(info->ring_buffer, unsigned char) == (size_t)tempmem + tempmem_size_in_bytes);
+    }
+
+#undef STBIR__NEXT_MEMPTR
+
+    // This signals that the ring buffer is empty
+    info->ring_buffer_begin_index = -1;
+
+    stbir__calculate_filters(info->horizontal_contributors, info->horizontal_coefficients, info->horizontal_filter, info->horizontal_scale, info->horizontal_shift, info->input_w, info->output_w);
+    stbir__calculate_filters(info->vertical_contributors, info->vertical_coefficients, info->vertical_filter, info->vertical_scale, info->vertical_shift, info->input_h, info->output_h);
+
+    STBIR_PROGRESS_REPORT(0);
+
+    if (stbir__use_height_upsampling(info))
+        stbir__buffer_loop_upsample(info);
+    else
+        stbir__buffer_loop_downsample(info);
+
+    STBIR_PROGRESS_REPORT(1);
+
+#ifdef STBIR_DEBUG_OVERWRITE_TEST
+    STBIR_ASSERT(memcmp(overwrite_output_before_pre, &((unsigned char*)output_data)[-OVERWRITE_ARRAY_SIZE], OVERWRITE_ARRAY_SIZE) == 0);
+    STBIR_ASSERT(memcmp(overwrite_output_after_pre, &((unsigned char*)output_data)[begin_forbidden], OVERWRITE_ARRAY_SIZE) == 0);
+    STBIR_ASSERT(memcmp(overwrite_tempmem_before_pre, &((unsigned char*)tempmem)[-OVERWRITE_ARRAY_SIZE], OVERWRITE_ARRAY_SIZE) == 0);
+    STBIR_ASSERT(memcmp(overwrite_tempmem_after_pre, &((unsigned char*)tempmem)[tempmem_size_in_bytes], OVERWRITE_ARRAY_SIZE) == 0);
+#endif
+
+    return 1;
+}
+
+
+static int stbir__resize_arbitrary(
+    void *alloc_context,
+    const void* input_data, int input_w, int input_h, int input_stride_in_bytes,
+    void* output_data, int output_w, int output_h, int output_stride_in_bytes,
+    float s0, float t0, float s1, float t1, float *transform,
+    int channels, int alpha_channel, stbir_uint32 flags, stbir_datatype type,
+    stbir_filter h_filter, stbir_filter v_filter,
+    stbir_edge edge_horizontal, stbir_edge edge_vertical, stbir_colorspace colorspace)
+{
+    stbir__info info;
+    int result;
+    size_t memory_required;
+    void* extra_memory;
+
+    stbir__setup(&info, input_w, input_h, output_w, output_h, channels);
+    stbir__calculate_transform(&info, s0,t0,s1,t1,transform);
+    stbir__choose_filter(&info, h_filter, v_filter);
+    memory_required = stbir__calculate_memory(&info);
+    extra_memory = STBIR_MALLOC(memory_required, alloc_context);
+
+    if (!extra_memory)
+        return 0;
+
+    result = stbir__resize_allocated(&info, input_data, input_stride_in_bytes,
+                                            output_data, output_stride_in_bytes,
+                                            alpha_channel, flags, type,
+                                            edge_horizontal, edge_vertical,
+                                            colorspace, extra_memory, memory_required);
+
+    STBIR_FREE(extra_memory, alloc_context);
+
+    return result;
+}
+
+STBIRDEF int stbir_resize_uint8(     const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+                                           unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+                                     int num_channels)
+{
+    return stbir__resize_arbitrary(NULL, input_pixels, input_w, input_h, input_stride_in_bytes,
+        output_pixels, output_w, output_h, output_stride_in_bytes,
+        0,0,1,1,NULL,num_channels,-1,0, STBIR_TYPE_UINT8, STBIR_FILTER_DEFAULT, STBIR_FILTER_DEFAULT,
+        STBIR_EDGE_CLAMP, STBIR_EDGE_CLAMP, STBIR_COLORSPACE_LINEAR);
+}
+
+STBIRDEF int stbir_resize_float(     const float *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+                                           float *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+                                     int num_channels)
+{
+    return stbir__resize_arbitrary(NULL, input_pixels, input_w, input_h, input_stride_in_bytes,
+        output_pixels, output_w, output_h, output_stride_in_bytes,
+        0,0,1,1,NULL,num_channels,-1,0, STBIR_TYPE_FLOAT, STBIR_FILTER_DEFAULT, STBIR_FILTER_DEFAULT,
+        STBIR_EDGE_CLAMP, STBIR_EDGE_CLAMP, STBIR_COLORSPACE_LINEAR);
+}
+
+STBIRDEF int stbir_resize_uint8_srgb(const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+                                           unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+                                     int num_channels, int alpha_channel, int flags)
+{
+    return stbir__resize_arbitrary(NULL, input_pixels, input_w, input_h, input_stride_in_bytes,
+        output_pixels, output_w, output_h, output_stride_in_bytes,
+        0,0,1,1,NULL,num_channels,alpha_channel,flags, STBIR_TYPE_UINT8, STBIR_FILTER_DEFAULT, STBIR_FILTER_DEFAULT,
+        STBIR_EDGE_CLAMP, STBIR_EDGE_CLAMP, STBIR_COLORSPACE_SRGB);
+}
+
+STBIRDEF int stbir_resize_uint8_srgb_edgemode(const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+                                                    unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+                                              int num_channels, int alpha_channel, int flags,
+                                              stbir_edge edge_wrap_mode)
+{
+    return stbir__resize_arbitrary(NULL, input_pixels, input_w, input_h, input_stride_in_bytes,
+        output_pixels, output_w, output_h, output_stride_in_bytes,
+        0,0,1,1,NULL,num_channels,alpha_channel,flags, STBIR_TYPE_UINT8, STBIR_FILTER_DEFAULT, STBIR_FILTER_DEFAULT,
+        edge_wrap_mode, edge_wrap_mode, STBIR_COLORSPACE_SRGB);
+}
+
+STBIRDEF int stbir_resize_uint8_generic( const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+                                               unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+                                         int num_channels, int alpha_channel, int flags,
+                                         stbir_edge edge_wrap_mode, stbir_filter filter, stbir_colorspace space,
+                                         void *alloc_context)
+{
+    return stbir__resize_arbitrary(alloc_context, input_pixels, input_w, input_h, input_stride_in_bytes,
+        output_pixels, output_w, output_h, output_stride_in_bytes,
+        0,0,1,1,NULL,num_channels,alpha_channel,flags, STBIR_TYPE_UINT8, filter, filter,
+        edge_wrap_mode, edge_wrap_mode, space);
+}
+
+STBIRDEF int stbir_resize_uint16_generic(const stbir_uint16 *input_pixels  , int input_w , int input_h , int input_stride_in_bytes,
+                                               stbir_uint16 *output_pixels , int output_w, int output_h, int output_stride_in_bytes,
+                                         int num_channels, int alpha_channel, int flags,
+                                         stbir_edge edge_wrap_mode, stbir_filter filter, stbir_colorspace space,
+                                         void *alloc_context)
+{
+    return stbir__resize_arbitrary(alloc_context, input_pixels, input_w, input_h, input_stride_in_bytes,
+        output_pixels, output_w, output_h, output_stride_in_bytes,
+        0,0,1,1,NULL,num_channels,alpha_channel,flags, STBIR_TYPE_UINT16, filter, filter,
+        edge_wrap_mode, edge_wrap_mode, space);
+}
+
+
+STBIRDEF int stbir_resize_float_generic( const float *input_pixels         , int input_w , int input_h , int input_stride_in_bytes,
+                                               float *output_pixels        , int output_w, int output_h, int output_stride_in_bytes,
+                                         int num_channels, int alpha_channel, int flags,
+                                         stbir_edge edge_wrap_mode, stbir_filter filter, stbir_colorspace space,
+                                         void *alloc_context)
+{
+    return stbir__resize_arbitrary(alloc_context, input_pixels, input_w, input_h, input_stride_in_bytes,
+        output_pixels, output_w, output_h, output_stride_in_bytes,
+        0,0,1,1,NULL,num_channels,alpha_channel,flags, STBIR_TYPE_FLOAT, filter, filter,
+        edge_wrap_mode, edge_wrap_mode, space);
+}
+
+
+STBIRDEF int stbir_resize(         const void *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+                                         void *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+                                   stbir_datatype datatype,
+                                   int num_channels, int alpha_channel, int flags,
+                                   stbir_edge edge_mode_horizontal, stbir_edge edge_mode_vertical,
+                                   stbir_filter filter_horizontal,  stbir_filter filter_vertical,
+                                   stbir_colorspace space, void *alloc_context)
+{
+    return stbir__resize_arbitrary(alloc_context, input_pixels, input_w, input_h, input_stride_in_bytes,
+        output_pixels, output_w, output_h, output_stride_in_bytes,
+        0,0,1,1,NULL,num_channels,alpha_channel,flags, datatype, filter_horizontal, filter_vertical,
+        edge_mode_horizontal, edge_mode_vertical, space);
+}
+
+
+STBIRDEF int stbir_resize_subpixel(const void *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+                                         void *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+                                   stbir_datatype datatype,
+                                   int num_channels, int alpha_channel, int flags,
+                                   stbir_edge edge_mode_horizontal, stbir_edge edge_mode_vertical,
+                                   stbir_filter filter_horizontal,  stbir_filter filter_vertical,
+                                   stbir_colorspace space, void *alloc_context,
+                                   float x_scale, float y_scale,
+                                   float x_offset, float y_offset)
+{
+    float transform[4];
+    transform[0] = x_scale;
+    transform[1] = y_scale;
+    transform[2] = x_offset;
+    transform[3] = y_offset;
+    return stbir__resize_arbitrary(alloc_context, input_pixels, input_w, input_h, input_stride_in_bytes,
+        output_pixels, output_w, output_h, output_stride_in_bytes,
+        0,0,1,1,transform,num_channels,alpha_channel,flags, datatype, filter_horizontal, filter_vertical,
+        edge_mode_horizontal, edge_mode_vertical, space);
+}
+
+STBIRDEF int stbir_resize_region(  const void *input_pixels , int input_w , int input_h , int input_stride_in_bytes,
+                                         void *output_pixels, int output_w, int output_h, int output_stride_in_bytes,
+                                   stbir_datatype datatype,
+                                   int num_channels, int alpha_channel, int flags,
+                                   stbir_edge edge_mode_horizontal, stbir_edge edge_mode_vertical,
+                                   stbir_filter filter_horizontal,  stbir_filter filter_vertical,
+                                   stbir_colorspace space, void *alloc_context,
+                                   float s0, float t0, float s1, float t1)
+{
+    return stbir__resize_arbitrary(alloc_context, input_pixels, input_w, input_h, input_stride_in_bytes,
+        output_pixels, output_w, output_h, output_stride_in_bytes,
+        s0,t0,s1,t1,NULL,num_channels,alpha_channel,flags, datatype, filter_horizontal, filter_vertical,
+        edge_mode_horizontal, edge_mode_vertical, space);
+}
+
+#endif // STB_IMAGE_RESIZE_IMPLEMENTATION
+
+/*
+------------------------------------------------------------------------------
+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.
+------------------------------------------------------------------------------
+*/
diff --git a/main.cpp b/main.cpp
index 3cc976c..2d5b501 100644
--- a/main.cpp
+++ b/main.cpp
@@ -36,6 +36,11 @@
 #define STB_IMAGE_WRITE_IMPLEMENTATION
 #include "lib/stb_image_write.h"
 
+// TODO(fox): Used for thumbnails. The renderer could be expanded to do this
+// much more efficiently.
+#define STB_IMAGE_RESIZE_IMPLEMENTATION
+#include "lib/stb_image_resize.h"
+
 #include "defines.h"
 #include "my_math.h"
 #include "structs.h"
@@ -61,18 +66,18 @@ static uint64 BitmapBlockSize;
 static instruction_mode InstructionMode = instruction_mode_scalar;
 static uint32 RandomGlobalIncrement = 0;
 
+#if SD
 #include "lib/base64.c"
 #include <curl/curl.h>
+#include "stable_diffusion.cpp"
+#endif
 
 #include "memory.cpp"
 #include "undo.cpp"
 #include "strings.cpp"
-#if THREADED
 #include "threading.cpp"
-#endif
 #include "createcalls.cpp"
 // #include "ffmpeg_backend.cpp"
-#include "stable_diffusion.cpp"
 #include "my_imgui_widgets.cpp"
 #include "prenderer.cpp"
 #include "gl_calls.cpp"
@@ -131,27 +136,35 @@ Main_InputTest(project_data *File, project_state *State, memory *Memory, ui *UI,
         UI->Warp_WantSetPos = false;
     }
 
-    // NOTE(fox): The only requirement for something that needs to happen before the sort is anything that creates/deletes layers.
-    // The requirement for something after the sort is things that depend on the sort. (obvious)
     if (!io.WantCaptureKeyboard)
         ImGui_ProcessInputs(File, State, UI, Memory, io);
 
     sorted_file Sorted = File_Sort_Push(File, State, Memory);
 
-    if (ImGui::IsKeyPressed(ImGuiKey_T)) {
-            Interact_Transform_Begin(File, Memory, State, io.MousePos, Sorted.CompArray, Sorted.LayerArray);
-    }
-
-    bool32 mod_key = io.ConfigMacOSXBehaviors ? io.KeySuper : io.KeyCtrl;
-    if (mod_key) {
-        if (ImGui::IsKeyPressed(ImGuiKey_C)) {
-            Clipboard_Store(File, State, Memory, Sorted.PropertyInfo, Sorted.PropertyArray);
-        }
-        if (ImGui::IsKeyPressed(ImGuiKey_V)) {
-            Clipboard_Paste(File, State, Memory, Sorted.LayerArray);
+    // These depend on sorting, so they can't be done in the ProcessInputs function:
+    if (State->HotkeyInput) {
+        switch (State->HotkeyInput) {
+            case hotkey_none:
+            {
+                Assert(0);
+            } break;
+            case hotkey_transform:
+            {
+                Interact_Transform_Begin(File, Memory, State, io.MousePos, Sorted.CompArray, Sorted.LayerArray);
+            } break;
+            case hotkey_copy:
+            {
+                Clipboard_Store(File, State, Memory, Sorted.CompArray, Sorted.LayerArray, Sorted.PropertyInfo, Sorted.PropertyArray);
+            } break;
+            case hotkey_paste:
+            {
+                Clipboard_Paste(File, State, Memory, Sorted.CompArray, Sorted.LayerArray);
+            } break;
         }
+        State->HotkeyInput = hotkey_none;
     }
 
+
     ImGui::DockSpaceOverViewport();
 
     if (Debug.ToggleWindow) {
@@ -162,7 +175,6 @@ Main_InputTest(project_data *File, project_state *State, memory *Memory, ui *UI,
 
     if (State->Initializing == 3) {
         Source_UICreateButton(File, State, Memory, Sorted.CompArray, Sorted.LayerArray);
-        /*
         block_layer *Layera = (block_layer *)Memory_Block_AddressAtIndex(Memory, F_Layers, 2);
         Layera->x.IsToggled = true;
         Layera->y.IsToggled = true;
@@ -172,20 +184,17 @@ Main_InputTest(project_data *File, project_state *State, memory *Memory, ui *UI,
         block_layer *Layerc = (block_layer *)Memory_Block_AddressAtIndex(Memory, F_Layers, 0);
         Layerc->x.IsToggled = true;
         Layerc->y.IsToggled = true;
-        */
-        /*
-        Layer_Select(Memory, State, 0);
-        Layer_Select(Memory, State, 1);
-        Layer_Select(Memory, State, 2);
+        // Layer_Select(Memory, State, 0);
+        // Layer_Select(Memory, State, 1);
+        // Layer_Select(Memory, State, 2);
     }
     if (State->Initializing == 2) {
-        block_composition *MainComp = (block_composition *)Memory_Block_AddressAtIndex(Memory, F_Precomps, File->PrincipalCompIndex);
-        Precomp_UICreateButton(File, State, Memory, MainComp, *Sorted.CompArray, Sorted.LayerArray);
-        Layer_DeselectAll(Memory, State, File->Layer_Count);
-        */
-        // block_layer *Layer = (block_layer *)Memory_Block_AddressAtIndex(Memory, F_Layers, 0);
+        // block_composition *MainComp = (block_composition *)Memory_Block_AddressAtIndex(Memory, F_Precomps, File->PrincipalCompIndex);
+        // Precomp_UICreateButton(File, State, Memory, MainComp, *Sorted.CompArray, Sorted.LayerArray);
+        // Layer_DeselectAll(Memory, State, File->Layer_Count);
+
+        block_layer *Layer = (block_layer *)Memory_Block_AddressAtIndex(Memory, F_Layers, 0);
         Layer_Select(Memory, State, 0);
-        /*
         History_Entry_Commit(Memory, "Add keyframe");
         property_channel *Property = &Layer->x;
         {
@@ -200,6 +209,8 @@ Main_InputTest(project_data *File, project_state *State, memory *Memory, ui *UI,
             bezier_point Point = { 1, {20, 300, -5, 0, 5, 0}, interpolation_type_bezier, 0, {0, 0, 0}, 0 };
             Bezier_Add(Memory, Property, Point);
         }
+        History_Entry_End(Memory);
+        /*
         Property = &Layer->y;
         {
             bezier_point Point = { 1, {10, 50, -5, 0, 5, 0}, interpolation_type_bezier, 0, {0, 0, 0}, 0 };
@@ -239,17 +250,18 @@ Main_InputTest(project_data *File, project_state *State, memory *Memory, ui *UI,
         */
     }
 
-
     ImGui_Viewport(File, State, UI, Memory, io, textureID, Sorted.CompArray, Sorted.LayerArray);
     ImGui_Timeline(File, State, Memory, UI, io, Sorted.CompArray, Sorted.LayerArray, Sorted.PropertyInfo, Sorted.PropertyArray);
     ImGui_File(File, State, Memory, io, Sorted.CompArray, Sorted.LayerArray);
     ImGui_PropertiesPanel(File, State, UI, Memory, io);
     ImGui_ColorPanel(File, State, UI, Memory, io);
+#if SD
     ImGui_SD_Prompt(File, State, UI, Memory, io);
-    ImGui_SD_Thumbnail(File, State, UI, Memory, io);
+    ImGui_SD_Thumbnail(File, State, UI, Memory, io, Sorted.SourceArray, Sorted.TempSourceCount, textureID);
+#endif
     ImGui_Menu(File, State, UI, Memory, io);
 
-    File_Sort_Pop(Memory, Sorted.Layer_SortSize, Sorted.Property_SortSize);
+    File_Sort_Pop(Memory, Sorted.Layer_SortSize, Sorted.Property_SortSize, Sorted.Source_SortSize);
 
 #if DEBUG
     Debug.Temp = {};
@@ -264,10 +276,9 @@ Render_Main(void *Data, void *OutputBuffer, render_type RenderType, rectangle Re
     bool32 IsRendering = true;
     Renderer_Start(Data, OutputBuffer, RenderType, RenderRegion);
     while (IsRendering) {
-        // SDL_Delay(1);
+
         // Main_InputTest(File, State, &Memory, &UI);
         Renderer_Check(&IsRendering, RenderType);
-        // TODO(fox): Make interruptable if the render time gets too long.
     }
 }
 
@@ -346,16 +357,24 @@ Render_Comp(project_data *File, project_state *State, memory *Memory, ImGuiIO io
 
             layer_bitmap_state *BitmapState = &State->Render.Bitmap[Index_Physical];
             void *BitmapAddress = NULL;
+            uint64 ScratchActive = 0;
             if (!Layer->IsPrecomp) {
                 block_source *Source = (block_source *)Memory_Block_AddressAtIndex(Memory, F_Sources, Layer->Block_Source_Index);
-
                 if (Source->Type == source_type_principal) {
                     if (State->Interact_Active == interact_type_brush && State->Brush.LayerToPaint_Index == Index_Physical) {
-                        // Assert(0);
-                        // Render_Main((void *)T, , RenderRegion);
-                        BitmapAddress = Memory_Block_AddressAtIndex(Memory, F_PrincipalBitmaps, Source->Bitmap_Index);
+                        void *SourceBitmapAddress = Memory_Block_AddressAtIndex(Memory, F_PrincipalBitmaps, Source->Bitmap_Index, 0);
+                        rectangle RenderRegion = { 0, 0, Source->Width, Source->Height };
+                        // TODO(fox): Do all these extra precomputes really make a difference in the renderer?
+                        direct_info Info = { (real32)Source->Width, (real32)Source->Height, (real32)Source->BytesPerPixel, (real32)Source->Width * Source->BytesPerPixel,
+                                             Bitmap_ByteInfo(Source->BytesPerPixel), blend_normal,
+                                             RenderRegion, State->Brush.TransientBitmap};
+                        ScratchActive = Source->Width * Source->Height * Source->BytesPerPixel;
+                        void *SecondSourceBitmap = Memory_PushScratch(Memory, ScratchActive);
+                        Memory_Copy((uint8 *)SecondSourceBitmap, (uint8 *)SourceBitmapAddress, ScratchActive);
+                        Render_Main((void *)&Info, SecondSourceBitmap, render_type_notransform, RenderRegion);
+                        BitmapAddress = SecondSourceBitmap;
                     } else {
-                        BitmapAddress = Memory_Block_AddressAtIndex(Memory, F_PrincipalBitmaps, Source->Bitmap_Index);
+                        BitmapAddress = Memory_Block_AddressAtIndex(Memory, F_PrincipalBitmaps, Source->Bitmap_Index, 0);
                     }
                 } else {
                     cache_entry *Entry = Memory_Cache_Search(State, Memory, cache_entry_type_source, Layer->Block_Source_Index, 0);
@@ -407,6 +426,8 @@ Render_Comp(project_data *File, project_state *State, memory *Memory, ImGuiIO io
             rectangle RenderRegion = {0, 0, Comp->Width, Comp->Height};
 
             Render_Main((void *)&T, CompBuffer, render_type_main, RenderRegion);
+            if (ScratchActive)
+                Memory_PopScratch(Memory, ScratchActive);
         }
     }
     Entry_Main->CycleTime = GetCPUTime() - Comp_TimeStart;
@@ -426,7 +447,7 @@ Main_Renderer(project_data *File, project_state *State, memory *Memory, SDL_Wind
     void *MainCompBuffer = Render_Comp(File, State, Memory, io, Sorted.CompArray, Sorted.LayerArray,
                                       Sorted.PropertyInfo, Sorted.PropertyArray, File->PrincipalCompIndex, State->Frame_Current);
 
-    File_Sort_Pop(Memory, Sorted.Layer_SortSize, Sorted.Property_SortSize);
+    File_Sort_Pop(Memory, Sorted.Layer_SortSize, Sorted.Property_SortSize, Sorted.Source_SortSize);
 
     glBindTexture(GL_TEXTURE_2D, textureID);
 
@@ -476,6 +497,7 @@ int main(int argc, char *argv[]) {
     Memory_InitTable(&GlobalMemory, &Memory, 4 * 1024 * 1024, F_Strings, "Strings", sizeof(block_string));
     Memory_InitTable(&GlobalMemory, &Memory, (uint64)50 * 1024 * 1024, F_PrincipalBitmaps, "Principal bitmap data", BitmapBlockSize);
 
+    Memory_InitTable(&GlobalMemory, &Memory, (uint64)5 * 1024 * 1024, B_Thumbnails, "Thumbnails");
     Memory_InitTable(&GlobalMemory, &Memory, (uint64)64 * 1024 * 1024, B_ScratchSpace, "Scratch");
     Memory_InitTable(&GlobalMemory, &Memory, (uint64)50 * 1024 * 1024, B_CachedBitmaps, "Cached bitmap buffer");
 
@@ -483,9 +505,9 @@ int main(int argc, char *argv[]) {
 #if ARM
     InstructionMode = instruction_mode_neon;
 #else
-    if (SDL_HasSSE2()) {
-        InstructionMode = instruction_mode_sse;
-    }
+    // if (SDL_HasSSE2()) {
+    //     InstructionMode = instruction_mode_sse;
+    // }
     if (SDL_HasAVX2()) {
         InstructionMode = instruction_mode_avx;
     }
@@ -518,12 +540,6 @@ int main(int argc, char *argv[]) {
     ui UI = {};
     UI.Test = ImDrawListSplitter();
 
-    // int ToolCount = (int)tool_count;
-    // for (int i = 0; i < ToolCount; i++) {
-    //     glBindTexture(GL_TEXTURE_2D, textureID);
-    //     glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA, MainComp->Width, MainComp->Height, 0, GL_RGBA, ByteFlag2, MainCompBuffer);
-    // }
-
     block_composition *MainComp = (block_composition *)Memory_Block_AllocateAddress(&Memory, F_Precomps);
 
     MainComp->Width = 1280;
@@ -537,232 +553,35 @@ int main(int argc, char *argv[]) {
 
     File->Comp_Count = 1;
 
-    /*
-    block_composition *Comp2 = (block_composition *)Memory_Block_AllocateAddress(&Memory, F_Precomps);
-    Comp2->Width = 500;
-    Comp2->Height = 500;
-    Comp2->FPS = 24;
-    Comp2->BytesPerPixel = 4;
-    Comp2->Frame_Count = 48;
-    Comp2->Frame_End = 48;
-    Comp2->Occupied = 1;
-    Comp2->Name_String_Index = String_AddToFile(&Memory, "Another comp");
-    */
-
+#if 1
     {
         uint16 SourceIndex = Source_Generate(File, State, &Memory, (void *)"../asset/a.jpg");
         block_source *Source = (block_source *)Memory_Block_AddressAtIndex(&Memory, F_Sources, 0);
         Source->IsSelected = true;
-
-        /*
-        {
-            Layer_CreateFromSource(File, State, &Memory, SourceIndex, MainComp->Frame_End);
-            block_layer *Layer = (block_layer *)Memory_Block_AddressAtIndex(&Memory, F_Layers, File->Layer_Count - 1);
-            Layer->Vertical_Offset = 6;
-            Layer->Col[0] = 1;
-            Layer->Col[1] = 0;
-            Layer->Col[2] = 0;
-            Layer->Block_Composition_Index = 0;
-            Layer->IsSelected = true;
-
-            property_channel *Property = &Layer->x;
-            Property->Block_Bezier_Index[0] = Memory_Block_AllocateNew(&Memory, F_Bezier);
-            Property->Block_Bezier_Count = 1;
-
-            block_bezier *Bezier = (block_bezier *)Memory_Block_AddressAtIndex(&Memory, F_Bezier, Property->Block_Bezier_Index[0]);
-            Bezier->Occupied = 1;
-
-            Bezier->Point[0].Pos[0] = V2(0, 0);
-            Bezier->Point[1].Pos[0] = V2(20, 50);
-            Bezier->Point[2].Pos[0] = V2(40, -50);
-            Bezier->Point[0].Pos[1] = V2(-4, 0);
-            Bezier->Point[1].Pos[1] = V2(-4, 0);
-            Bezier->Point[2].Pos[1] = V2(-4, 0);
-            Bezier->Point[0].Pos[2] = V2(4, 0);
-            Bezier->Point[1].Pos[2] = V2(4, 0);
-            Bezier->Point[2].Pos[2] = V2(4, 0);
-            Bezier->Point[0].Type = interpolation_type_bezier;
-            Bezier->Point[1].Type = interpolation_type_bezier;
-            Bezier->Point[2].Type = interpolation_type_bezier;
-            Bezier->Point[1].IsSelected = false;
-            Bezier->Point[0].Occupied = true;
-            Bezier->Point[1].Occupied = true;
-            Bezier->Point[2].Occupied = true;
-
-            property_channel *Property2 = &Layer->opacity;
-            Property2->Block_Bezier_Index[0] = Memory_Block_AllocateNew(&Memory, F_Bezier);
-            Property2->Block_Bezier_Count = 1;
-
-            block_bezier *Bezier2 = (block_bezier *)Memory_Block_AddressAtIndex(&Memory, F_Bezier, Property2->Block_Bezier_Index[0]);
-            Bezier2->Occupied = 1;
-
-            Bezier2->Point[0].Pos[0] = V2(0, 0);
-            Bezier2->Point[1].Pos[0] = V2(20, 1);
-            Bezier2->Point[0].Occupied = true;
-            Bezier2->Point[1].Occupied = true;
-
-            property_channel *Property3 = &Layer->y;
-            Property3->Block_Bezier_Index[0] = Memory_Block_AllocateNew(&Memory, F_Bezier);
-            Property3->Block_Bezier_Count = 1;
-
-            block_bezier *Bezier3 = (block_bezier *)Memory_Block_AddressAtIndex(&Memory, F_Bezier, Property3->Block_Bezier_Index[0]);
-            Bezier3->Occupied = 1;
-
-            Bezier3->Point[0].Pos[0] = V2(0, -20);
-            Bezier3->Point[1].Pos[0] = V2(20, 300);
-            Bezier3->Point[2].Pos[0] = V2(40, 100);
-            Bezier3->Point[0].Pos[1] = V2(-4, 0);
-            Bezier3->Point[1].Pos[1] = V2(-4, 0);
-            Bezier3->Point[2].Pos[1] = V2(-4, 0);
-            Bezier3->Point[0].Pos[2] = V2(4, 0);
-            Bezier3->Point[1].Pos[2] = V2(4, 0);
-            Bezier3->Point[2].Pos[2] = V2(4, 0);
-            Bezier3->Point[0].Type = interpolation_type_bezier;
-            Bezier3->Point[1].Type = interpolation_type_bezier;
-            Bezier3->Point[2].Type = interpolation_type_bezier;
-            Bezier3->Point[0].Occupied = true;
-            Bezier3->Point[1].Occupied = true;
-            Bezier3->Point[2].Occupied = true;
-        }
-        */
-        /*
-        {
-            Layer_CreateFromSource(File, State, &Memory, SourceIndex, MainComp->Frame_End);
-            block_layer *Layer = (block_layer *)Memory_Block_AddressAtIndex(&Memory, F_Layers, File->Layer_Count - 1);
-            Layer->Vertical_Offset = 2;
-            Layer->IsSelected = true;
-            Layer->ColIndex = 2;
-            Layer->Block_Composition_Index = 0;
-        }
-        {
-            Layer_CreateFromSource(File, State, &Memory, SourceIndex, MainComp->Frame_End);
-            block_layer *Layer = (block_layer *)Memory_Block_AddressAtIndex(&Memory, F_Layers, File->Layer_Count - 1);
-            Layer->Vertical_Offset = 5;
-            Layer->ColIndex = 0;
-            Layer->Block_Composition_Index = 0;
-        }
-        */
     }
 
     {
         uint16 SourceIndex = Source_Generate(File, State, &Memory, (void *)"../asset/b.jpg");
         block_source *Source = (block_source *)Memory_Block_AddressAtIndex(&Memory, F_Sources, 1);
-        // Source->IsSelected = true;
-        /*
-        {
-            Layer_CreateFromSource(File, State, &Memory, SourceIndex, MainComp->Frame_End);
-            block_layer *Layer = (block_layer *)Memory_Block_AddressAtIndex(&Memory, F_Layers, File->Layer_Count - 1);
-            Layer->IsSelected = true;
-            Layer->Vertical_Offset = 3;
-            Layer->ColIndex = 1;
-            Layer->Block_Composition_Index = 0;
-            // Layer->IsSelected = true;
-
-            property_channel *Property = &Layer->x;
-            Property->Block_Bezier_Index[0] = Memory_Block_AllocateNew(&Memory, F_Bezier);
-            Property->Block_Bezier_Count = 1;
-
-            block_bezier *Bezier = (block_bezier *)Memory_Block_AddressAtIndex(&Memory, F_Bezier, Property->Block_Bezier_Index[0]);
-            Bezier->Occupied = 1;
-
-            Bezier->Point[0].Pos[0] = V2(30, 0);
-            Bezier->Point[1].Pos[0] = V2(33, 250);
-            Bezier->Point[2].Pos[0] = V2(35, -50);
-            Bezier->Point[3].Pos[0] = V2(55, 0);
-            Bezier->Point[0].Pos[1] = V2(-1, 0);
-            Bezier->Point[1].Pos[1] = V2(-1, 0);
-            Bezier->Point[2].Pos[1] = V2(-1, 0);
-            Bezier->Point[3].Pos[1] = V2(-1, 0);
-            Bezier->Point[0].Pos[2] = V2(1, 0);
-            Bezier->Point[1].Pos[2] = V2(1, 0);
-            Bezier->Point[2].Pos[2] = V2(1, 0);
-            Bezier->Point[3].Pos[2] = V2(1, 0);
-            Bezier->Point[0].Type = interpolation_type_bezier;
-            Bezier->Point[1].Type = interpolation_type_bezier;
-            Bezier->Point[2].Type = interpolation_type_bezier;
-            Bezier->Point[3].Type = interpolation_type_bezier;
-            Bezier->Point[0].Occupied = true;
-            Bezier->Point[1].Occupied = true;
-            Bezier->Point[2].Occupied = true;
-            Bezier->Point[3].Occupied = true;
-
-            property_channel *Property2 = &Layer->opacity;
-            Property2->Block_Bezier_Index[0] = Memory_Block_AllocateNew(&Memory, F_Bezier);
-            Property2->Block_Bezier_Count = 1;
-
-            block_bezier *Bezier2 = (block_bezier *)Memory_Block_AddressAtIndex(&Memory, F_Bezier, Property2->Block_Bezier_Index[0]);
-            Bezier2->Occupied = 1;
-
-            Bezier2->Point[0].Pos[0] = V2(25, 1);
-            Bezier2->Point[1].Pos[0] = V2(40, 0);
-            Bezier2->Point[0].Occupied = true;
-            Bezier2->Point[1].Occupied = true;
-        }
-        */
+        Source->IsSelected = true;
     }
 
     {
         uint16 SourceIndex = Source_Generate(File, State, &Memory, (void *)"../asset/c.jpg");
         block_source *Source = (block_source *)Memory_Block_AddressAtIndex(&Memory, F_Sources, 2);
-        // Source->IsSelected = true;
-        /*
-        {
-            Layer_CreateFromSource(File, State, &Memory, SourceIndex, MainComp->Frame_End);
-            block_layer *Layer = (block_layer *)Memory_Block_AddressAtIndex(&Memory, F_Layers, File->Layer_Count - 1);
-            Layer->Vertical_Offset = 1;
-            Layer->ColIndex = 3;
-            Layer->Block_Composition_Index = 0;
-        }
-        */
-        // {
-        //     Layer_CreateFromSource(File, State, &Memory, SourceIndex, MainComp->Frame_End);
-        //     block_layer *Layer = (block_layer *)Memory_Block_AddressAtIndex(&Memory, F_Layers, File->Layer_Count - 1);
-        //     Layer->Vertical_Offset = 4;
-        //     Layer->Col[0] = 0;
-        //     Layer->Col[1] = 1;
-        //     Layer->Col[2] = 1;
-        //     Layer->Block_Composition_Index = 1;
-        // }
-    }
-
-    /*
-    {
-        Layer_CreateFromSource(File, State, &Memory, 1, MainComp->Frame_End);
-        block_layer *Layer = (block_layer *)Memory_Block_AddressAtIndex(&Memory, F_Layers, File->Layer_Count - 1);
-        Layer->IsPrecomp = true;
-        Layer->Vertical_Offset = 9;
-        Layer->Col[0] = 1;
-        Layer->Col[1] = 1;
-        Layer->Col[2] = 1;
-        Layer->Block_Composition_Index = 0;
-    }
-
-    {
-        Layer_CreateFromSource(File, State, &Memory, 1, MainComp->Frame_End);
-        block_layer *Layer = (block_layer *)Memory_Block_AddressAtIndex(&Memory, F_Layers, File->Layer_Count - 1);
-        Layer->IsPrecomp = true;
-        Layer->Vertical_Offset = 10;
-        Layer->Col[0] = 1;
-        Layer->Col[1] = 1;
-        Layer->Col[2] = 1;
-        Layer->Block_Composition_Index = 0;
+        Source->IsSelected = true;
     }
-    */
-
-    // History_Undo(&Memory);
-    // History_Redo(&Memory);
+#endif
 
     SDL_Init(SDL_INIT_VIDEO);
 
     Semaphore = SDL_CreateSemaphore(0);
 
-#if THREADED
     int Index[7];
     for (int i = 0; i < 7; i++) {
         Index[i] = i;
         Thread[i] = SDL_CreateThread(TestThread, "thread", (void *)&Index[i]);
     }
-#endif
 
     // Decide GL+GLSL versions
     SDL_GL_SetAttribute(SDL_GL_CONTEXT_FLAGS, SDL_GL_CONTEXT_DEBUG_FLAG);
@@ -859,46 +678,23 @@ int main(int argc, char *argv[]) {
     Brush_CalcBitmapAlphaFromSize(&Memory, &State->Brush, 4);
     State_BindBrushTexture(&Memory, &State->Brush, 4);
 
+#if SD
     curl_global_init(CURL_GLOBAL_ALL);
     curl_state MainHandle = {};
     curl_state ProgHandle = {};
     uint32 Inc = 0;
+#endif
 
     while (State->IsRunning)
     {
         // State->Interact_Active = interact_type_layer_move;
         // State->Interact_Offset[1] = -3.0f;
 
+#if SD
         if (State->CurlActive) {
-            if (State->CurlActive == -1) {
-                Curl_GET_Init(&MainHandle, State->Dump1, State->JSONPayload, State->SD.ServerAddress, State->SD.Mode);
-                Curl_Prog_Init(&ProgHandle, State->Dump2);
-                State->CurlActive = 1;
-            } else {
-                if (Curl_Check(&MainHandle) == 1) {
-                    SD_JSONToSource(File, State, &Memory, State->Dump1, State->SD.Height, State->SD.Width);
-                    Curl_Free(&ProgHandle);
-                    curl_slist_free_all(ProgHandle.list);
-                    Curl_Free(&MainHandle);
-                    State->CurlActive = 0;
-                }
-                uint64 Time = ImGui::GetTime();
-                if (Time - State->SDTimer > 0.3f) {
-                    if (Curl_Check(&ProgHandle) == 1) {
-                        SD_ParseProgress(State, (char *)State->Dump2);
-                        curl_multi_remove_handle(ProgHandle.curlm, ProgHandle.curl);
-                        curl_easy_reset(ProgHandle.curl);
-                        ProgHandle.CurlData.size = 0;
-                        curl_easy_setopt(ProgHandle.curl, CURLOPT_URL, "http://127.0.0.1:7860/sdapi/v1/progress");
-                        curl_easy_setopt(ProgHandle.curl, CURLOPT_WRITEFUNCTION, dumbcurlcallback);
-                        curl_easy_setopt(ProgHandle.curl, CURLOPT_WRITEDATA, (void *)&ProgHandle.CurlData);
-                        curl_multi_add_handle(ProgHandle.curlm, ProgHandle.curl);
-                    }
-                    State->SDTimer = Time;
-                }
-                Inc++;
-            }
+            Curl_Main();
         }
+#endif
 
         Main_InputTest(File, State, &Memory, &UI, window, textureID);
 
@@ -916,7 +712,6 @@ int main(int argc, char *argv[]) {
 
         Main_RenderUI(io, clear_color, window);
 
-        // TODO(fox): Fix things that rely on this.
         if (State->Initializing)
             State->Initializing--;
     }
diff --git a/main.h b/main.h
index 3d7b3e5..9630faf 100644
--- a/main.h
+++ b/main.h
@@ -163,8 +163,11 @@ struct sorted_file
     sorted_comp_info *CompArray;
     sorted_property_info *PropertyInfo;
     uint16 *PropertyArray;
+    uint16 *SourceArray;        // sort by recency, temp only
+    uint16 TempSourceCount;
     uint64 Layer_SortSize;
     uint64 Property_SortSize;
+    uint64 Source_SortSize;
 };
 
 enum timeline_mode
@@ -251,9 +254,10 @@ char *BrushNames[brush_amount] = {
 struct brush_state
 {
     ImVec2 UIPos;               // Initial position when ctrl is held
-    real32 Size = 512;          // Maxes at 1024 for now
+    real32 Size = 64;          // Maxes at 1024 for now
     real32 Hardness = 1.0f;     // From 1 to 100
     real32 Spacing = 1.0f;
+    bool32 EraseMode = 0;
     brush_type Type = brush_normal;
     GLuint GLTexture;
     void *PaintBuffer;
@@ -300,6 +304,16 @@ struct interact_transform
     uint32 TransformMode;
 };
 
+
+
+enum hotkey_input
+{
+    hotkey_none,
+    hotkey_transform,
+    hotkey_copy,
+    hotkey_paste
+};
+
 struct project_state
 {
     bool32 UpdateKeyframes = 1;
@@ -315,12 +329,14 @@ struct project_state
     brush_state Brush;
 
     sd_state SD;
-    char JSONPayload[1024];
+    char JSONPayload[1024*1024*4];
     int32 CurlActive = 0;
     real32 SDPercentDone;
     real32 SDTimeEstimate;
     real64 SDTimer;
 
+    hotkey_input HotkeyInput;
+
     void *Dump1;
     void *Dump2;
 
@@ -407,6 +423,9 @@ struct block_source
     uint16 Height;
     uint16 BytesPerPixel;
 
+    uint32 RelativeTimestamp; //
+    GLuint ThumbnailTex;
+
     source_type Type;
 };
 
@@ -518,6 +537,18 @@ struct transform_info {
     void *SourceBuffer;
 };
 
+struct direct_info {
+    real32 BufferWidth;
+    real32 BufferHeight;
+    real32 BufferBytesPerPixel;
+    real32 BufferPitch;
+    render_byte_info BufferBits;
+
+    blend_mode BlendMode;
+    rectangle ClipRect;
+    void *SourceBuffer;
+};
+
 struct brush_info {
     uint32 BrushLength;
     rectangle LayerBounds;
@@ -535,12 +566,13 @@ struct brush_info {
     real32 G_Brush;
     real32 B_Brush;
     real32 A_Brush;
+    bool32 EraseMode;
     uint8 *BrushRow;
 };
 
 enum render_type {
     render_type_main,
-    render_type_direct,
+    render_type_notransform,
     render_type_brush
 };
 
diff --git a/memory.cpp b/memory.cpp
index 64aadc1..47252fe 100644
--- a/memory.cpp
+++ b/memory.cpp
@@ -51,7 +51,7 @@ Memory_Block_AllocateAddress(memory *Memory, memory_table_list TableName)
     return Memory_Block_AddressAtIndex(Memory, TableName, FileIndex, 0);
 }
 
-// IMPORTANT(fox): All block data has to start with a uint8 Occupied variable!
+// IMPORTANT(fox): All block data structs have to start with a uint8 Occupied variable!
 static bool32
 Block_Loop(memory *Memory, memory_table_list TableName, uint32 TotalCount, int *HasIncremented, int *CurrentCount, int *Index)
 {
diff --git a/memory.h b/memory.h
index 2067e16..5f15bca 100644
--- a/memory.h
+++ b/memory.h
@@ -14,6 +14,7 @@ enum memory_table_list {
     F_Strings,
     F_PrincipalBitmaps,
 
+    B_Thumbnails,
     B_ScratchSpace,
     B_CachedBitmaps,
 };
diff --git a/my_imgui_widgets.cpp b/my_imgui_widgets.cpp
index b3339d2..0b1e017 100644
--- a/my_imgui_widgets.cpp
+++ b/my_imgui_widgets.cpp
@@ -236,10 +236,53 @@ ImGui_File(project_data *File, project_state *State, memory *Memory, ImGuiIO io,
     ImGui::End();
 }
 
+#if SD
 static void
-ImGui_SD_Thumbnail(project_data *File, project_state *State, ui *UI, memory *Memory, ImGuiIO io)
+ImGui_SD_Thumbnail(project_data *File, project_state *State, ui *UI, memory *Memory, ImGuiIO io, uint16 *SourceArray, uint16 SourceCount, GLuint textureID)
 {
     ImGui::Begin("SD gallery");
+    ImDrawList* draw_list = ImGui::GetWindowDrawList();
+    ImVec2 ViewportMin = ImGui::GetCursorScreenPos();
+    ImVec2 ViewportScale = ImGui::GetContentRegionAvail();
+    ImVec2 ViewportMax = ImVec2(ViewportMin.x + ViewportScale.x, ViewportMin.y + ViewportScale.y);
+    if (!SourceCount) {
+        ImGui::End();
+        return;
+    }
+    int test[16];
+    int count = 0;
+    // draw_list->AddImage((void *)(intptr_t)textureID, ViewportMin, ViewportMax);
+    uint32 T_Height = 256;
+    real32 RowPercent = (real32)ViewportScale.x / T_Height;
+    uint32 PerRow = (uint32)RowPercent;
+    uint32 UI_Size = T_Height;
+    for (int i = 0; i < SourceCount; i++) {
+        int32 Y = i / PerRow;
+        int32 X = i % PerRow;
+        ImVec2 ImgMin = ViewportMin + ImVec2(X * UI_Size, Y * UI_Size);
+        ImVec2 ImgMax = ImgMin + ImVec2(UI_Size, UI_Size);
+        block_source *Source = (block_source *)Memory_Block_AddressAtIndex(Memory, F_Sources, SourceArray[i]);
+        // ImGui::Text("Count: %i", Source->RelativeTimestamp);
+        real32 Ratio = (real32)Source->Width / Source->Height;
+        uint32 T_Width = (uint32)(Ratio * T_Height);
+        if (Source->ThumbnailTex == 0) {
+            Source_DumpThumbnail(Memory, Source, T_Width, T_Height);
+        }
+        // draw_list->AddRect(ImgMin, ImgMax, IM_COL32(255, 255, 255, 64));
+        ImGui::SetCursorScreenPos(ImgMin);
+        ImGui::Button("asda", ImVec2(UI_Size, UI_Size));
+        draw_list->AddImage((void *)(intptr_t)Source->ThumbnailTex, ImgMin, ImgMax);
+
+        // block_string *String = (block_string *)Memory_Block_AddressAtIndex(Memory, F_Strings, Source->Path_String_Index);
+        // ImGui::Selectable(String->Char, Source->IsSelected);
+        // if (ImGui::IsItemClicked() || ImGui::IsItemClicked(ImGuiMouseButton_Right)) {
+        //     if (!io.KeyShift && !Source->IsSelected) {
+        //         Source_DeselectAll(File, Memory);
+        //     }
+        //     Source->IsSelected = 1;
+        // }
+        // ImGui::OpenPopupOnItemClick("sourcecontext", ImGuiPopupFlags_MouseButtonRight);
+    }
     ImGui::End();
 }
 
@@ -265,18 +308,18 @@ ImGui_SD_Prompt(project_data *File, project_state *State, ui *UI, memory *Memory
                                       Source->Width, Source->Height, Source->BytesPerPixel, &len);
                 Assert(PNGBitmap);
 
-                uint64 EncodedLength = 0;
-                uint64 EncodedAllocSize = base64_encode_size(Size);
-                uint8 *EncodedOutput = (uint8 *)Memory_PushScratch(Memory, EncodedAllocSize);
-                uint64 EncodedTrueSize;
+                uint64 EncodedEstimateSize = base64_encode_size(Size);
+                uint8 *EncodedOutput = (uint8 *)Memory_PushScratch(Memory, EncodedEstimateSize);
+                uint64 EncodedTrueSize = 0;
 
-                base64_encode((uint8 *)BitmapAddress, Size, EncodedOutput, &EncodedTrueSize);
+                base64_encode((uint8 *)PNGBitmap, len, EncodedOutput, &EncodedTrueSize);
                 Assert(EncodedOutput[EncodedTrueSize] == '\0');
+                // printf("%s", EncodedOutput);
 
                 STBIW_FREE(PNGBitmap);
 
-                SD_AssembleJSON(SD, (char *)State->Dump1, EncodedOutput);
-                Memory_PopScratch(Memory, EncodedAllocSize);
+                SD_AssembleJSON(SD, (char *)State->JSONPayload, EncodedOutput);
+                Memory_PopScratch(Memory, EncodedEstimateSize);
             } else {
                 SD_AssembleJSON(SD, State->JSONPayload);
                 // SD_AssembleJSON(SD, (char *)State->Dump2);
@@ -317,6 +360,7 @@ ImGui_SD_Prompt(project_data *File, project_state *State, ui *UI, memory *Memory
     ImGui::InputInt("Seed", &SD->Seed);
     ImGui::End();
 }
+#endif
 
 static void
 ImGui_ColorPanel(project_data *File, project_state *State, ui *UI, memory *Memory, ImGuiIO io)
@@ -833,7 +877,6 @@ ImGui_Viewport(project_data *File, project_state *State, ui *UI, memory *Memory,
 
     block_composition *MainComp = (block_composition *)Memory_Block_AddressAtIndex(Memory, F_Precomps, File->PrincipalCompIndex);
 
-
     ImVec2 ViewportMin = ImGui::GetCursorScreenPos();
     ImVec2 ViewportScale = ImGui::GetContentRegionAvail();
     ImVec2 ViewportMax = ImVec2(ViewportMin.x + ViewportScale.x, ViewportMin.y + ViewportScale.y);
@@ -874,14 +917,15 @@ ImGui_Viewport(project_data *File, project_state *State, ui *UI, memory *Memory,
 
     ImGui::InvisibleButton("canvas", ViewportScale, ImGuiButtonFlags_MouseButtonLeft | ImGuiButtonFlags_MouseButtonRight);
     bool32 IsHovered = ImGui::IsItemHovered();
-#if 1
+#if 0
     bool32 IsActive = ImGui::IsItemActive();
     bool32 IsActivated = ImGui::IsItemActivated();
+    bool32 IsDeactivated = ImGui::IsItemDeactivated();
 #else
     bool32 IsActive = ImGui::IsKeyDown(ImGuiKey_3);
     bool32 IsActivated = ImGui::IsKeyPressed(ImGuiKey_3);
+    bool32 IsDeactivated = ImGui::IsKeyReleased(ImGuiKey_3);
 #endif
-    bool32 IsDeactivated = ImGui::IsItemDeactivated();
 
     if (IsHovered && IsActivated && !ImGui::IsMouseDown(ImGuiMouseButton_Right))
     {
@@ -891,14 +935,17 @@ ImGui_Viewport(project_data *File, project_state *State, ui *UI, memory *Memory,
         if (State->Tool == tool_brush) {
             sorted_layer *SortedLayerInfo = Layer_GetSortedArray(SortedLayerArray, SortedCompArray, File->PrincipalCompIndex);
             sorted_comp_info SortedCompInfo = SortedCompArray[File->PrincipalCompIndex];
-            for (int i = SortedCompInfo.LayerCount - 1; i >= 0; i--) {
-                sorted_layer SortEntry = SortedLayerInfo[i];
-                uint32 Index_Physical = SortEntry.Block_Layer_Index;
-                block_layer *Layer = (block_layer *)Memory_Block_AddressAtIndex(Memory, F_Layers, Index_Physical);
-                block_source *Source = (block_source *)Memory_Block_AddressAtIndex(Memory, F_Sources, Layer->Block_Source_Index);
-                if (Layer->IsSelected && Source->Type == source_type_principal) {
-                    State->Interact_Active = interact_type_brush;
-                    State->Brush.LayerToPaint_Index = Layer->Block_Source_Index;
+            if (!io.KeyCtrl) {
+                for (int i = SortedCompInfo.LayerCount - 1; i >= 0; i--) {
+                    sorted_layer SortEntry = SortedLayerInfo[i];
+                    uint32 Index_Physical = SortEntry.Block_Layer_Index;
+                    block_layer *Layer = (block_layer *)Memory_Block_AddressAtIndex(Memory, F_Layers, Index_Physical);
+                    block_source *Source = (block_source *)Memory_Block_AddressAtIndex(Memory, F_Sources, Layer->Block_Source_Index);
+                    if (Layer->IsSelected && Source->Type == source_type_principal) {
+                        State->Interact_Active = interact_type_brush;
+                        State->Brush.LayerToPaint_Index = Index_Physical;
+                        break;
+                    }
                 }
             }
             if (State->Brush.LayerToPaint_Index == -1) {
@@ -963,7 +1010,7 @@ ImGui_Viewport(project_data *File, project_state *State, ui *UI, memory *Memory,
             block_layer *Layer = (block_layer *)Memory_Block_AddressAtIndex(Memory, F_Layers, State->Brush.LayerToPaint_Index);
             layer_transforms T_Layer = Layer_GetTransforms(Layer);
             block_source *Source = (block_source *)Memory_Block_AddressAtIndex(Memory, F_Sources, Layer->Block_Source_Index);
-            void *SourceBitmapAddress = Memory_Block_AddressAtIndex(Memory, F_PrincipalBitmaps, Source->Bitmap_Index);
+            void *SourceBitmapAddress = Memory_Block_AddressAtIndex(Memory, F_PrincipalBitmaps, Source->Bitmap_Index, 0);
             ImVec2 MouseDelta = io.MouseDelta;
             real32 Delta = MouseDelta.x + MouseDelta.y;
             if (Delta != 0.0f) {
@@ -971,11 +1018,11 @@ ImGui_Viewport(project_data *File, project_state *State, ui *UI, memory *Memory,
                 real32 DeltaSlope = MouseDelta.y / MouseDelta.x;
                 for (real32 i = 0; i < DeltaDistance; i += State->Brush.Spacing) {
                     ImVec2 MousePos = ImGui_Brush_CalcMousePos(State, io, MouseDelta, i, DeltaDistance, DeltaSlope);
-                    Brush_Render(State, UI, T_Layer, MainComp, Source, SourceBitmapAddress, ViewportMin, MousePos);
+                    Brush_Render(State, UI, T_Layer, MainComp, Source, State->Brush.TransientBitmap, ViewportMin, MousePos);
                 }
             } else if (IsActivated) {
                 ImVec2 MousePos = io.MousePos;
-                Brush_Render(State, UI, T_Layer, MainComp, Source, SourceBitmapAddress, ViewportMin, MousePos);
+                Brush_Render(State, UI, T_Layer, MainComp, Source, State->Brush.TransientBitmap, ViewportMin, MousePos);
             }
             // Memory_Cache_Invalidate(State, Memory, cache_entry_type_comp, Layer->Block_Composition_Index, 0);
             State->UpdateFrame = true;
@@ -1974,16 +2021,19 @@ ImGui_ProcessInputs(project_data *File, project_state *State, ui *UI, memory *Me
         State->UpdateFrame = true;
     }
     if (ImGui::IsKeyPressed(ImGuiKey_E)) {
-        block_composition *MainComp = (block_composition *)Memory_Block_AddressAtIndex(Memory, F_Precomps, File->PrincipalCompIndex);
-        State->Frame_Current = ((State->Frame_Current + 1) >= MainComp->Frame_Count) ? 0 : State->Frame_Current + 1;
-        State->UpdateFrame = true;
+        if (!io.KeyShift) {
+            block_composition *MainComp = (block_composition *)Memory_Block_AddressAtIndex(Memory, F_Precomps, File->PrincipalCompIndex);
+            State->Frame_Current = ((State->Frame_Current + 1) >= MainComp->Frame_Count) ? 0 : State->Frame_Current + 1;
+            State->UpdateFrame = true;
+        } else {
+            State->Brush.EraseMode ^= 1;
+        }
     }
     if (ImGui::IsKeyPressed(ImGuiKey_X)) {
         v4 Temp = UI->Color;
         UI->Color = UI->AltColor;
         UI->AltColor = Temp;
     }
-
     if (ImGui::IsKeyPressed(ImGuiKey_V)) {
         State->Tool = tool_default;
     }
@@ -1995,19 +2045,43 @@ ImGui_ProcessInputs(project_data *File, project_state *State, ui *UI, memory *Me
         UI->GraphZoomSize = ImVec2(0, 0);
         UI->GraphMoveSize = ImVec2(0, 0);
     }
-    if (UI->TimelineMode == timeline_mode_graph) {
-        if (ImGui::IsKeyPressed(ImGuiKey_G)) {
-                State->Interact_Offset[2] = io.MousePos.x;
-                State->Interact_Offset[3] = io.MousePos.y;
-                State->Interact_Active = interact_type_keyframe_move;
-        } else if (ImGui::IsKeyPressed(ImGuiKey_R)) {
-                // State->Interact_Offset[2] = io.MousePos.x;
-                // State->Interact_Offset[3] = io.MousePos.y;
-                // State->Interact_Active = interact_type_keyframe_rotate;
-        } else if (ImGui::IsKeyPressed(ImGuiKey_S)) {
-                State->Interact_Offset[2] = io.MousePos.x;
-                State->Interact_Offset[3] = io.MousePos.y;
-                State->Interact_Active = interact_type_keyframe_scale;
+    if (UI->FocusedWindow == focus_timeline) {
+        if (UI->TimelineMode == timeline_mode_default) {
+            if (ImGui::IsKeyPressed(ImGuiKey_G)) {
+                Layer_ToggleChannel(File, Memory, 0);
+                Layer_ToggleChannel(File, Memory, 1);
+            } else if (ImGui::IsKeyPressed(ImGuiKey_A)) {
+                Layer_ToggleChannel(File, Memory, 2);
+                Layer_ToggleChannel(File, Memory, 3);
+            } else if (ImGui::IsKeyPressed(ImGuiKey_R)) {
+                Layer_ToggleChannel(File, Memory, 4);
+            } else if (ImGui::IsKeyPressed(ImGuiKey_S)) {
+                Layer_ToggleChannel(File, Memory, 5);
+            } else if (ImGui::IsKeyPressed(ImGuiKey_T)) {
+                if (io.KeyShift) {
+                    Layer_ToggleChannel(File, Memory, 6);
+                } else {
+                    Layer_ToggleChannel(File, Memory, 7);
+                }
+            }
+        } else if (UI->TimelineMode == timeline_mode_graph) {
+            if (ImGui::IsKeyPressed(ImGuiKey_G)) {
+                    State->Interact_Offset[2] = io.MousePos.x;
+                    State->Interact_Offset[3] = io.MousePos.y;
+                    State->Interact_Active = interact_type_keyframe_move;
+            } else if (ImGui::IsKeyPressed(ImGuiKey_R)) {
+                    // State->Interact_Offset[2] = io.MousePos.x;
+                    // State->Interact_Offset[3] = io.MousePos.y;
+                    // State->Interact_Active = interact_type_keyframe_rotate;
+            } else if (ImGui::IsKeyPressed(ImGuiKey_S)) {
+                    State->Interact_Offset[2] = io.MousePos.x;
+                    State->Interact_Offset[3] = io.MousePos.y;
+                    State->Interact_Active = interact_type_keyframe_scale;
+            }
+        }
+    } else if (UI->FocusedWindow == focus_viewport) {
+        if (ImGui::IsKeyPressed(ImGuiKey_T)) {
+            State->HotkeyInput = hotkey_transform;
         }
     }
     if (ImGui::IsKeyPressed(ImGuiKey_Escape)) {
@@ -2066,6 +2140,23 @@ ImGui_ProcessInputs(project_data *File, project_state *State, ui *UI, memory *Me
 
     bool32 mod_key = io.ConfigMacOSXBehaviors ? io.KeySuper : io.KeyCtrl;
     if (mod_key) {
+        if (ImGui::IsKeyPressed(ImGuiKey_C)) {
+            State->HotkeyInput = hotkey_copy;
+        }
+        if (ImGui::IsKeyPressed(ImGuiKey_V)) {
+            State->HotkeyInput = hotkey_paste;
+        }
+        if (ImGui::IsKeyPressed(ImGuiKey_Z)) {
+            if (io.KeyShift) {
+                History_Redo(Memory);
+                State->UpdateFrame = true;
+            } else {
+                History_Undo(Memory);
+                State->UpdateFrame = true;
+            }
+        }
+    }
+
         /*
         if (ImGui::IsKeyPressed(ImGuiKey_S)) {
             if (io.KeyShift) {
@@ -2089,16 +2180,6 @@ ImGui_ProcessInputs(project_data *File, project_state *State, ui *UI, memory *Me
                 State->ImGuiPopups = popup_newlayer;
         }
         */
-        if (ImGui::IsKeyPressed(ImGuiKey_Z)) {
-            if (io.KeyShift) {
-                History_Redo(Memory);
-                State->UpdateFrame = true;
-            } else {
-                History_Undo(Memory);
-                State->UpdateFrame = true;
-            }
-        }
-    }
 
 }
 
@@ -3562,222 +3643,6 @@ ImGui_EffectsPanel(project_data *File, project_state *State, memory *Memory, ui
 
 
 
-static void
-ImGui_ProcessInputs(project_data *File, project_state *State, comp_buffer *CompBuffer, memory *Memory, ui *UI, ImGuiIO io)
-{
-    if (io.KeysData[ImGuiKey_Q].Down) {
-        State->IsRunning = false;
-    }
-#if DEBUG
-        // ImGui::SaveIniSettingsToDisk("asda");
-#endif
-    if (ImGui::IsKeyPressed(ImGuiKey_R)) {
-        /*
-        UI->Y_TimelinePercentZoomed = UI->Default_Y_TimelinePercentZoomed;
-        UI->Y_TimelinePercentOffset = UI->Default_Y_TimelinePercentOffset;
-        keyframe *Keyframe = KeyframeLookup(&File->Layer[0]->x, 0);
-        Keyframe->Value.f = -10;
-        Keyframe = KeyframeLookup(&File->Layer[0]->x, 1);
-        Keyframe->Value.f = -5;
-        Keyframe = KeyframeLookup(&File->Layer[0]->x, 2);
-        Keyframe->Value.f = 0;
-        Keyframe = KeyframeLookup(&File->Layer[0]->x, 3);
-        Keyframe->Value.f = 5;
-        Keyframe = KeyframeLookup(&File->Layer[0]->x, 4);
-        Keyframe->Value.f = 10;
-        */
-    }
-
-    if (ImGui::IsKeyPressed(ImGuiKey_D)) {
-        IncrementFrame(File, -1);
-        State->UpdateFrame = true;
-        State->UpdateKeyframes = true;
-    }
-
-    if (ImGui::IsKeyPressed(ImGuiKey_F)) {
-        IncrementFrame(File, 1);
-        State->UpdateFrame = true;
-        State->UpdateKeyframes = true;
-    }
-
-    if (ImGui::IsKeyPressed(ImGuiKey_Z)) {
-        if (io.KeyCtrl && io.KeyShift) {
-            History_Redo(Memory);
-        } else if (io.KeyCtrl) {
-            History_Undo(Memory);
-        }
-        State->UpdateFrame = true;
-        State->UpdateKeyframes = true;
-    }
-
-    if (ImGui::IsKeyPressed(ImGuiKey_V)) {
-        if (State->Tool == tool_pen) {
-            State->Tool = tool_default;
-        } else {
-            State->Tool = tool_pen;
-        }
-    }
-
-    if (ImGui::IsKeyPressed(ImGuiKey_Space)) {
-        if (io.KeyShift) {
-            State->RerouteEffects = true;
-        } else {
-            State->IsPlaying ^= 1;
-        }
-    }
-
-
-    if (State->IsPlaying && !IsRendering) {
-        IncrementFrame(File, 1);
-        State->UpdateFrame = true;
-        State->UpdateKeyframes = true;
-    }
-
-    // if (ImGui::IsKeyPressed(ImGuiKey_R) && State->NumberOfSelectedLayers)
-    //     TransformsInteract(File, State, Memory, UI, sliding_rotation);
-    if (ImGui::IsKeyPressed(ImGuiKey_S)  && State->NumberOfSelectedLayers)
-        TransformsInteract(File, State, Memory, UI, sliding_scale);
-    if (ImGui::IsKeyPressed(ImGuiKey_G)  && State->NumberOfSelectedLayers)
-        TransformsInteract(File, State, Memory, UI, sliding_position);
-    if (ImGui::IsKeyPressed(ImGuiKey_A)  && State->NumberOfSelectedLayers)
-        TransformsInteract(File, State, Memory, UI, sliding_anchorpoint);
-
-    if (ImGui::IsKeyPressed(ImGuiKey_1))
-        LoadTestFootage(File, State, Memory);
-
-    if (ImGui::IsKeyPressed(ImGuiKey_D) && io.KeyCtrl)
-    {
-    }
-
-    if (ImGui::IsKeyPressed(ImGuiKey_Delete))
-    {
-        switch (State->RecentSelectionType)
-        {
-            case selection_none:
-            {
-            } break;
-            case selection_layer:
-            {
-            } break;
-            case selection_effect:
-            {
-            } break;
-            case selection_keyframe:
-            {
-                DeleteSelectedKeyframes(File, Memory);
-                State->UpdateKeyframes = true;
-                State->UpdateFrame = true;
-            }
-            case selection_maskpoint:
-            {
-            }
-            case selection_source:
-            {
-            }
-            break;
-        }
-    }
-
-    if (State->IsInteracting) {
-        ImVec2 MouseIncrement = io.MouseDelta * (ImVec2(CompBuffer->Width, CompBuffer->Height) / UI->CompZoom);
-        project_layer *Layer = File->Layer[State->MostRecentlySelectedLayer];
-        switch (State->TransformsHotkeyInteract)
-        {
-            case sliding_position:
-            {
-                Layer->x.CurrentValue.f += MouseIncrement.x;
-                Layer->y.CurrentValue.f += MouseIncrement.y;
-            } break;
-            case sliding_anchorpoint:
-            {
-                source *Source = Layer->Source;
-                Layer->x.CurrentValue.f += MouseIncrement.x;
-                Layer->y.CurrentValue.f += MouseIncrement.y;
-                Layer_CalcRotatedOffset(Layer, V2(MouseIncrement), V2(Source->Info.Width, Source->Info.Height),
-                                        &Layer->ax.CurrentValue.f, &Layer->ay.CurrentValue.f);
-            } break;
-            case sliding_rotation:
-            {
-                Layer->rotation.CurrentValue.f += MouseIncrement.x / 10.0;
-            } break;
-            case sliding_scale:
-            {
-                Layer->scale.CurrentValue.f += MouseIncrement.x / 200.0;
-            } break;
-        }
-        if (ImGui::IsKeyPressed(ImGuiKey_Escape)) {
-            switch (State->TransformsHotkeyInteract)
-            {
-                case sliding_position:
-                {
-                    Layer->x.CurrentValue.f = State->InteractCache[0];
-                    Layer->y.CurrentValue.f = State->InteractCache[1];
-                } break;
-                case sliding_anchorpoint:
-                {
-                    Layer->x.CurrentValue.f = State->InteractCache[0];
-                    Layer->y.CurrentValue.f = State->InteractCache[1];
-                    Layer->ax.CurrentValue.f = State->InteractCache[2];
-                    Layer->ay.CurrentValue.f = State->InteractCache[3];
-                } break;
-                case sliding_rotation:
-                {
-                    Layer->rotation.CurrentValue.f = State->InteractCache[0];
-                } break;
-                case sliding_scale:
-                {
-                    Layer->scale.CurrentValue.f = State->InteractCache[0];
-                } break;
-            }
-            State->IsInteracting = false;
-            State->UpdateFrame = true;
-        }
-        if (ImGui::IsMouseDown(ImGuiMouseButton_Left)) {
-            History_Entry_Commit(Memory, action_entry_default, "Tranforms interact");
-            switch (State->TransformsHotkeyInteract)
-            {
-                case sliding_position:
-                {
-                    History_Action_Change(Memory, &Layer->x.CurrentValue.f, &State->InteractCache[0],
-                                         &Layer->x.CurrentValue.f, action_type_change_r32);
-                    History_Action_Change(Memory, &Layer->y.CurrentValue.f, &State->InteractCache[1],
-                                         &Layer->y.CurrentValue.f, action_type_change_r32);
-                } break;
-                case sliding_anchorpoint:
-                {
-                    History_Action_Change(Memory, &Layer->x.CurrentValue.f, &State->InteractCache[0],
-                                         &Layer->x.CurrentValue.f, action_type_change_r32);
-                    History_Action_Change(Memory, &Layer->y.CurrentValue.f, &State->InteractCache[1],
-                                         &Layer->y.CurrentValue.f, action_type_change_r32);
-                    History_Action_Change(Memory, &Layer->ax.CurrentValue.f, &State->InteractCache[2],
-                                         &Layer->ax.CurrentValue.f, action_type_change_r32);
-                    History_Action_Change(Memory, &Layer->ay.CurrentValue.f, &State->InteractCache[3],
-                                         &Layer->ay.CurrentValue.f, action_type_change_r32);
-                } break;
-                case sliding_rotation:
-                {
-                    History_Action_Change(Memory, &Layer->rotation.CurrentValue.f, &State->InteractCache[0],
-                                         &Layer->rotation.CurrentValue.f, action_type_change_r32);
-                } break;
-                case sliding_scale:
-                {
-                    History_Action_Change(Memory, &Layer->scale.CurrentValue.f, &State->InteractCache[0],
-                                         &Layer->scale.CurrentValue.f, action_type_change_r32);
-                } break;
-            }
-            History_Entry_End(Memory);
-            State->IsInteracting = false;
-        }
-        State->UpdateFrame = true;
-    }
-
-    if (!ImGui::IsMouseDown(ImGuiMouseButton_Left)) {
-        UI->DraggingLayerThreshold = 0;
-        UI->DraggingTimelineThreshold = 0;
-        UI->DraggingKeyframeThreshold = 0;
-    }
-}
-
 static char ImGuiPrefs[] = "[Window][DockSpaceViewport_11111111]"
 "\nPos=0,0"
 "\nSize=3200,1800"
diff --git a/prenderer.cpp b/prenderer.cpp
index aa44a5a..8690f4e 100644
--- a/prenderer.cpp
+++ b/prenderer.cpp
@@ -257,6 +257,8 @@ Layer_LocalToScreenSpace(project_state *State, memory *Memory, block_layer *Laye
 
 static void
 Fallback_RenderLayer(transform_info T, void *OutputBuffer, rectangle RenderRegion);
+static void
+Fallback_RenderDirect(direct_info T, void *OutputBuffer, rectangle RenderRegion);
 
 static void
 RenderLayers(render_entry Entry) {
@@ -266,14 +268,13 @@ RenderLayers(render_entry Entry) {
         {
             Fallback_RenderLayer(*(transform_info *)Entry.RenderData, Entry.OutputBuffer, Entry.RenderRegion);
         } break;
-        case render_type_direct:
+        case render_type_notransform:
         {
-            Assert(0);
-            // Fallback_RenderDirect(*(direct_info *)Entry.RenderData, Entry.OutputBuffer, Entry.RenderRegion);
+            Fallback_RenderDirect(*(direct_info *)Entry.RenderData, Entry.OutputBuffer, Entry.RenderRegion);
         } break;
         case render_type_brush:
         {
-            PaintTest_AVX2(*(brush_info *)Entry.RenderData, Entry.OutputBuffer, Entry.RenderRegion);
+            PaintTest(*(brush_info *)Entry.RenderData, Entry.OutputBuffer, Entry.RenderRegion);
         } break;
         default:
         {
@@ -295,6 +296,13 @@ RenderLayers(render_entry Entry) {
 static void
 Renderer_Start(void *Data, void *OutputBuffer, render_type RenderType, rectangle RenderRegion)
 {
+#if DEBUG
+    if (Debug.NoThreading) {
+        render_entry Entry = { Data, OutputBuffer, RenderType, RenderRegion };
+        RenderLayers(Entry);
+        return;
+    }
+#endif
     // CPU
     Threading_BitmapOp(Data, OutputBuffer, RenderType, RenderRegion);
 }
@@ -302,6 +310,12 @@ Renderer_Start(void *Data, void *OutputBuffer, render_type RenderType, rectangle
 static void
 Renderer_Check(bool32 *Test, render_type RenderType)
 {
+#if DEBUG
+    if (Debug.NoThreading) {
+        *Test = true;
+        return;
+    }
+#endif
     // CPU
     *Test = Threading_IsActive(RenderType);
 }
@@ -521,7 +535,7 @@ Transform_Calculate(project_state *State, memory *Memory, project_data *File, bl
             A_Blend = A_Dest;\
 
 static void
-Fallback_RenderDirect(transform_info T, void *OutputBuffer, rectangle RenderRegion)
+Fallback_RenderDirect(direct_info T, void *OutputBuffer, rectangle RenderRegion)
 {
     rectangle LayerBounds = ClipRectangle( T.ClipRect, RenderRegion);
 
@@ -534,16 +548,16 @@ Fallback_RenderDirect(transform_info T, void *OutputBuffer, rectangle RenderRegi
         {
             uint16 LX = X;
             uint16 LY = Y;
-            uint16 LXPlus = Ceil(X+1, (uint32)T.LayerWidth - 1);
-            uint16 LYPlus = Ceil(Y+1, (uint32)T.LayerHeight - 1);
+            uint16 LXPlus = Ceil(X+1, (uint32)T.BufferWidth - 1);
+            uint16 LYPlus = Ceil(Y+1, (uint32)T.BufferHeight - 1);
 
-            uint8 *TexPTR0 = ((uint8 *)T.SourceBuffer + ((uint16)T.LayerPitch * LY)     + (LX     * (uint16)T.LayerBytesPerPixel));
+            uint8 *TexPTR0 = ((uint8 *)T.SourceBuffer + ((uint16)T.BufferPitch * LY)     + (LX     * (uint16)T.BufferBytesPerPixel));
             uint32 PixelA = *(uint32 *)TexPTR0;
 
-            real32 TexRA = (real32)(*(uint32 *)(TexPTR0 + T.LayerBits.ByteOffset * 0) & T.LayerBits.MaskPixel) * T.LayerBits.Normalized;
-            real32 TexGA = (real32)(*(uint32 *)(TexPTR0 + T.LayerBits.ByteOffset * 1) & T.LayerBits.MaskPixel) * T.LayerBits.Normalized;
-            real32 TexBA = (real32)(*(uint32 *)(TexPTR0 + T.LayerBits.ByteOffset * 2) & T.LayerBits.MaskPixel) * T.LayerBits.Normalized;
-            real32 TexAA = (real32)(*(uint32 *)(TexPTR0 + T.LayerBits.ByteOffset * 3) & T.LayerBits.MaskPixel) * T.LayerBits.Normalized;
+            real32 TexRA = (real32)(*(uint32 *)(TexPTR0 + T.BufferBits.ByteOffset * 0) & T.BufferBits.MaskPixel) * T.BufferBits.Normalized;
+            real32 TexGA = (real32)(*(uint32 *)(TexPTR0 + T.BufferBits.ByteOffset * 1) & T.BufferBits.MaskPixel) * T.BufferBits.Normalized;
+            real32 TexBA = (real32)(*(uint32 *)(TexPTR0 + T.BufferBits.ByteOffset * 2) & T.BufferBits.MaskPixel) * T.BufferBits.Normalized;
+            real32 TexAA = (real32)(*(uint32 *)(TexPTR0 + T.BufferBits.ByteOffset * 3) & T.BufferBits.MaskPixel) * T.BufferBits.Normalized;
 
             real32 LayerAlpha = TexAA * 1; // brush opacity
 
@@ -850,11 +864,9 @@ EndRenderState(project_state *State)
       }
 
       State->NumberOfLayersToRender = 0;
-#if THREADED
       SDL_AtomicSet(&CurrentEntry, 0);
       SDL_AtomicSet(&QueuedEntries, 0);
       SDL_AtomicSet(&CompletedEntries, 0);
-#endif
 
 }
 
@@ -930,7 +942,6 @@ QueueCurrentFrame(project_data *File, comp_buffer *CompBuffer, project_state *St
         }
     }
 
-#if THREADED
 
     uint16 TileWidth = CompBuffer->Width / 4;
     uint16 TileHeight = CompBuffer->Height / 4;
@@ -957,12 +968,10 @@ QueueCurrentFrame(project_data *File, comp_buffer *CompBuffer, project_state *St
         }
     }
 
-#else
 
     rectangle RenderRegion = {0, 0, (int32)CompBuffer->Width, (int32)CompBuffer->Height};
     RenderLayers(&RenderInfo, RenderRegion);
 
-#endif
 }
 
 #if ARM
diff --git a/stable_diffusion.cpp b/stable_diffusion.cpp
index 2eb230b..dc12bbc 100644
--- a/stable_diffusion.cpp
+++ b/stable_diffusion.cpp
@@ -38,9 +38,19 @@ SD_JSONToSource(project_data *File, project_state *State, memory *Memory, void *
     void *RawData = stbi_load_from_memory((stbi_uc *)PNGData, PNGSize, &x, &y, &a, 4);
     Assert(x == Width && y == Height);
     Memory_PopScratch(Memory, MaxSize);
+    int32 Highest = 0;
+    {
+        int h = 0, c = 0, i = 0;
+        while (Block_Loop(Memory, F_Sources, File->Source_Count, &h, &c, &i)) {
+            block_source *Source = (block_source *)Memory_Block_AddressAtIndex(Memory, F_Sources, i);
+            if (Source->Type == source_type_principal_temp && Source->RelativeTimestamp > Highest)
+                Highest = Source->RelativeTimestamp;
+        }
+    }
     int SrcIdx = Source_Generate_Blank(File, State, Memory, Width, Height, 4);
     block_source *Source = (block_source *)Memory_Block_AddressAtIndex(Memory, F_Sources, SrcIdx);
-    // Source->Type = source_type_principal_temp;
+    Source->Type = source_type_principal_temp;
+    Source->RelativeTimestamp = Highest + 1;
     void *BitmapAddress = Memory_Block_AddressAtIndex(Memory, F_PrincipalBitmaps, Source->Bitmap_Index, 0);
     Memory_Copy((uint8 *)BitmapAddress, (uint8 *)RawData, MaxSize);
     stbi_image_free(RawData);
@@ -74,6 +84,8 @@ SD_ParseProgress(project_state *State, char *JSONInfo)
     // Assert(0);
 }
 
+static char *pre = "data:image/png;base64,";
+
 static void
 JSON_AppendParam_String(char *String, uint64 *i, char *P1, char *P2)
 {
@@ -83,17 +95,22 @@ JSON_AppendParam_String(char *String, uint64 *i, char *P1, char *P2)
     while(P1[a] != '\0') {
         String[c++] = P1[a++];
     }
+    String[c++] = '"';
     String[c++] = ':';
     String[c++] = ' ';
+    String[c++] = '[';
     String[c++] = '"';
     a = 0;
+    while(pre[a] != '\0') {
+        String[c++] = pre[a++];
+    }
+    a = 0;
     while(P2[a] != '\0') {
         String[c++] = P2[a++];
-        if (a > 64)
-            break;
     }
     c--;
     String[c++] = '"';
+    String[c++] = ']';
     String[c++] = ',';
     String[c++] = '\n';
     String[c++] = '\0';
@@ -114,8 +131,9 @@ SD_AssembleJSON(sd_state *SD, char *JSONPayload, void *Base64Bitmap = NULL)
     JSONPayload[1] = '\n';
     JSONPayload[2] = '\0';
     uint64 i = 2;
-    if (SD->Mode)
+    if (SD->Mode) {
         JSON_AppendParam_String(JSONPayload, &i, "init_images", (char *)Base64Bitmap);
+    }
     for (int i = 0; i < 6; i++) {
         if (Type[i] == 0) {
             sprintf(JSONPayload, "%s\"%s\": \"%s\",\n",   JSONPayload, Test[i], (char *)Test2[i]);
@@ -131,6 +149,7 @@ SD_AssembleJSON(sd_state *SD, char *JSONPayload, void *Base64Bitmap = NULL)
         sprintf(JSONPayload, "%s\"%s\": %.2f,\n", JSONPayload, "denoising_strength", SD->DenoisingStrength);
     sprintf(JSONPayload, "%s%s\n", JSONPayload, "\"sampler_index\": \"DDIM\"");
     sprintf(JSONPayload, "%s}\n", JSONPayload);
+    printf("%s\n", JSONPayload);
     // sprintf(CurlCommand, "curl -X POST -H 'Content-Type: application/json' -i '%s/sdapi/v1/txt2img' --data '%s'", SD->ServerAddress, JSONPayload);
     // printf("%s\n", CurlCommand);
 };
@@ -144,6 +163,43 @@ struct curl_state
 };
 
 static void
+Curl_Main()
+{
+    if (State->CurlActive == -1) {
+        Curl_GET_Init(&MainHandle, State->Dump1, State->JSONPayload, State->SD.ServerAddress, State->SD.Mode);
+        Curl_Prog_Init(&ProgHandle, State->Dump2);
+        State->CurlActive = 1;
+    } else {
+        if (Curl_Check(&MainHandle) == 1) {
+            SD_JSONToSource(File, State, &Memory, State->Dump1, State->SD.Height, State->SD.Width);
+            Curl_StopAll(State, &ProgHandle, &MainHandle);
+        }
+        uint64 Time = ImGui::GetTime();
+        if (Time - State->SDTimer > 0.3f) {
+            int Test = Curl_Check(&ProgHandle);
+            if (Test == 1) {
+                SD_ParseProgress(State, (char *)State->Dump2);
+                curl_multi_remove_handle(ProgHandle.curlm, ProgHandle.curl);
+                curl_easy_reset(ProgHandle.curl);
+                ProgHandle.CurlData.size = 0;
+                curl_easy_setopt(ProgHandle.curl, CURLOPT_URL, "http://127.0.0.1:7860/sdapi/v1/progress");
+                curl_easy_setopt(ProgHandle.curl, CURLOPT_WRITEFUNCTION, dumbcurlcallback);
+                curl_easy_setopt(ProgHandle.curl, CURLOPT_WRITEDATA, (void *)&ProgHandle.CurlData);
+                curl_multi_add_handle(ProgHandle.curlm, ProgHandle.curl);
+            } else if (Test == -1) {
+                PostMsg(State, "Active stable-diffusion-webui instance not found at URL.");
+                Curl_StopAll(State, &ProgHandle, &MainHandle);
+            } else if (Test == -2) {
+                PostMsg(State, "CURL error; see command line.");
+                Curl_StopAll(State, &ProgHandle, &MainHandle);
+            }
+            State->SDTimer = Time;
+        }
+        Inc++;
+    }
+}
+
+static void
 Curl_Free(curl_state *Handle)
 {
     curl_multi_remove_handle(Handle->curlm, Handle->curl);
@@ -151,7 +207,16 @@ Curl_Free(curl_state *Handle)
     curl_multi_cleanup(Handle->curlm);
 }
 
-static bool32
+static void
+Curl_StopAll(project_state *State, curl_state *ProgHandle, curl_state *MainHandle)
+{
+    Curl_Free(ProgHandle);
+    curl_slist_free_all(ProgHandle->list);
+    Curl_Free(MainHandle);
+    State->CurlActive = 0;
+}
+
+static int
 Curl_Check(curl_state *Handle)
 {
     int IsActive;
@@ -167,11 +232,10 @@ Curl_Check(curl_state *Handle)
             if (!msg->data.result) {
                 return 1;
             } else if (msg->data.result == CURLE_COULDNT_CONNECT) {
-                // printf("Active stable-diffusion-webui instance not found at URL.\n");
                 return -1;
             } else {
-                // printf("curl error: %s!\n", curl_easy_strerror(msg->data.result));
-                return -1;
+                printf("curl error: %s!\n", curl_easy_strerror(msg->data.result));
+                return -2;
             }
         }
     }
diff --git a/stable_diffusion.h b/stable_diffusion.h
index 97ae076..febb8f6 100644
--- a/stable_diffusion.h
+++ b/stable_diffusion.h
@@ -26,7 +26,7 @@ struct sd_state
     int32 Height = 384;
     int32 SamplerIndex = 0;
     real32 CFG = 7;
-    real32 DenoisingStrength = 0.2;
+    real32 DenoisingStrength = 0.8;
     int32 Seed = -1;
 };
 
diff --git a/undo.cpp b/undo.cpp
index e568675..d985cee 100644
--- a/undo.cpp
+++ b/undo.cpp
@@ -173,7 +173,7 @@ void History_Entry_Commit(memory *Memory, char *Name)
     History->EntryPlayhead++;
     Memory->IsFileSaved = false;
 #if DEBUG
-    Assert(Debug.UndoState != 1);
+    Assert(Debug.UndoState != 1);   // You forgot to end a History_Entry_Commit()!
     Debug.UndoState = 1;
 #endif
 }