Merge pull request #109 from guzba/master

a bit faster

examples/blur.nim

@@ -1,4 +1,4 @@
-import pixie, chroma, vmath
+import chroma, pixie, vmath
 
 let
   trees = readImage("examples/data/trees.png")
@@ -8,7 +8,7 @@ let
 image.fill(rgba(255, 255, 255, 255))
 
 var p: Path
-p.polygon(100, 100, 70, sides=6)
+p.polygon(100, 100, 70, sides = 6)
 p.closePath()
 
 let mask = newMask(200, 200)

examples/heart.nim

@@ -1,4 +1,4 @@
-import pixie, chroma, vmath
+import chroma, pixie, vmath
 
 let
   image = newImage(200, 200)

examples/rounded_rectangle.nim

@@ -1,4 +1,4 @@
-import pixie, chroma, vmath
+import chroma, pixie, vmath
 
 let image = newImage(200, 200)
 image.fill(rgba(255, 255, 255, 255))

examples/shadow.nim

@@ -1,4 +1,4 @@
-import pixie, chroma, vmath
+import chroma, pixie, vmath
 
 let
   trees = readImage("examples/data/trees.png")
@@ -7,7 +7,7 @@ let
 image.fill(rgba(255, 255, 255, 255))
 
 var p: Path
-p.polygon(100, 100, 70, sides=8)
+p.polygon(100, 100, 70, sides = 8)
 p.closePath()
 
 var polyImage = newImage(200, 200)

examples/square.nim

@@ -1,4 +1,4 @@
-import pixie, chroma
+import chroma, pixie
 
 var image = newImage(200, 200)
 image.fill(rgba(255, 255, 255, 255))

examples/tiger.nim

@@ -1,4 +1,4 @@
-import pixie, chroma, vmath
+import chroma, pixie, vmath
 
 let image = newImage(200, 200)
 image.fill(rgba(255, 255, 255, 255))

experiments/benchmark_cairo.nim

@@ -1,4 +1,4 @@
-import cairo, math, benchy, pixie, chroma
+import benchy, cairo, chroma, math, pixie
 
 var
   surface = imageSurfaceCreate(FORMAT_ARGB32, 1000, 1000)

experiments/benchmark_svg_cairo.nim

@@ -1,4 +1,4 @@
-import svg_cairo, benchy
+import benchy, svg_cairo
 
 let data = readFile("tests/images/svg/Ghostscript_Tiger.svg")
 

experiments/svg_cairo.nim

@@ -1,7 +1,7 @@
 ## Load and Save SVG files.
 
-import chroma, pixie/images, pixie/common, pixie/paths, vmath, xmlparser,
-  xmltree, strutils, strutils, cairo
+import cairo, chroma, pixie/common, pixie/images, pixie/paths, strutils, vmath,
+    xmlparser, xmltree
 
 type Path = paths.Path
 

experiments/test_svg_cairo.nim

@@ -1,4 +1,4 @@
-import svg_cairo, pixie, strformat
+import pixie, strformat, svg_cairo
 
 const files = [
   "line01",

experiments/trapezoid.nim

@@ -1,10 +1,9 @@
 
-
-import pixie, pixie/paths, pixie/images, chroma, print, vmath, algorithm, sequtils, bumpy
+import algorithm, bumpy, chroma, pixie, pixie/images, pixie/paths, print,
+    sequtils, vmath
 
 printColors = false
 
-
 proc intersectsInner*(a, b: Segment, at: var Vec2): bool {.inline.} =
   ## Checks if the a segment intersects b segment.
   ## If it returns true, at will have point of intersection
@@ -28,7 +27,7 @@ proc roundBy*(v: Vec2, n: float32): Vec2 {.inline.} =
 
 proc pathToTrapezoids(p: Path): seq[Trapezoid] =
 
-  var polygons = p.commands.commandsToPolygons()
+  var polygons = p.commandsToShapes()
 
   const q = 1/256.0
 
@@ -73,7 +72,7 @@ proc pathToTrapezoids(p: Path): seq[Trapezoid] =
     var collision = false
     for y in yScanLines:
       var at: Vec2
-      if intersects(line(vec2(0,y), vec2(1,y)), s, at):
+      if intersects(line(vec2(0, y), vec2(1, y)), s, at):
         at = at.roundBy(q)
         at.y = y
         if s.at.y != at.y and s.to.y != at.y:
@@ -136,9 +135,9 @@ proc pathToTrapezoids(p: Path): seq[Trapezoid] =
         Trapezoid(
           nw: a.at,
           ne: b.at,
-          se: b.to,# + vec2(0,0.7),
-          sw: a.to# + vec2(0,0.7)
-        )
+          se: b.to, # + vec2(0,0.7),
+        sw: a.to # + vec2(0,0.7)
+      )
       )
 
 proc trapFill(image: Image, t: Trapezoid, color: ColorRGBA) =
@@ -180,7 +179,7 @@ block:
   var image = newImage(200, 200)
   image.fill(rgba(255, 255, 255, 255))
 
-  var p = newPath()
+  var p: Path
   p.moveTo(50, 50)
   p.lineTo(50, 150)
   p.lineTo(150, 150)
@@ -190,7 +189,7 @@ block:
   var trapezoids = p.pathToTrapezoids()
   image.drawTrapezoids(trapezoids)
 
-  image.writeFile("trapezoids/rect.png")
+  image.writeFile("experiments/trapezoids/rect.png")
 
 block:
   # Rhombus
@@ -198,7 +197,7 @@ block:
   var image = newImage(200, 200)
   image.fill(rgba(255, 255, 255, 255))
 
-  var p = newPath()
+  var p: Path
   p.moveTo(100, 50)
   p.lineTo(150, 100)
   p.lineTo(100, 150)
@@ -208,7 +207,7 @@ block:
   var trapezoids = p.pathToTrapezoids()
   image.drawTrapezoids(trapezoids)
 
-  image.writeFile("trapezoids/rhombus.png")
+  image.writeFile("experiments/trapezoids/rhombus.png")
 
 block:
   # heart
@@ -224,7 +223,7 @@ block:
   var trapezoids = p.pathToTrapezoids()
   image.drawTrapezoids(trapezoids)
 
-  image.writeFile("trapezoids/heart.png")
+  image.writeFile("experiments/trapezoids/heart.png")
 
 block:
   # l
@@ -241,7 +240,7 @@ block:
   var trapezoids = p.pathToTrapezoids()
   image.drawTrapezoids(trapezoids)
 
-  image.writeFile("trapezoids/l.png")
+  image.writeFile("experiments/trapezoids/l.png")
 
 block:
   # g
@@ -258,4 +257,4 @@ block:
   var trapezoids = p.pathToTrapezoids()
   image.drawTrapezoids(trapezoids)
 
-  image.writeFile("trapezoids/g.png")
+  image.writeFile("experiments/trapezoids/g.png")

pixie.nimble

@@ -11,4 +11,4 @@ requires "chroma >= 0.2.1"
 requires "zippy >= 0.3.5"
 requires "flatty >= 0.1.3"
 requires "nimsimd >= 1.0.0"
-requires "bumpy >= 1.0.1"
+requires "bumpy >= 1.0.2"

src/pixie.nim

@@ -1,8 +1,8 @@
-import pixie/images, pixie/masks, pixie/paths, pixie/common, pixie/blends,
-  pixie/fileformats/bmp, pixie/fileformats/png, pixie/fileformats/jpg,
-  pixie/fileformats/svg, flatty/binny, os, pixie/gradients
+import flatty/binny, os, pixie/blends, pixie/common, pixie/fileformats/bmp,
+    pixie/fileformats/jpg, pixie/fileformats/png, pixie/fileformats/svg,
+    pixie/images, pixie/masks, pixie/paths
 
-export images, masks, paths, common, blends, gradients
+export blends, common, images, masks, paths
 
 type
   FileFormat* = enum

src/pixie/blends.nim

@@ -1,5 +1,6 @@
 ## Blending modes.
-import chroma, math, common
+
+import chroma, common, math
 
 when defined(amd64) and not defined(pixieNoSimd):
   import nimsimd/sse2
@@ -28,8 +29,8 @@ type
     bmColor
     bmLuminosity
 
-    bmMask  ## Special blend mode that is used for masking
-    bmOverwrite  ## Special blend mode that just copies pixels
+    bmMask         ## Special blend mode that is used for masking
+    bmOverwrite    ## Special blend mode that just copies pixels
     bmSubtractMask ## Inverse mask
     bmIntersectMask
     bmExcludeMask
@@ -141,9 +142,9 @@ proc ClipColor(C: var Color) {.inline.} =
     n = min([C.r, C.g, C.b])
     x = max([C.r, C.g, C.b])
   if n < 0:
-      C = L + (((C - L) * L) / (L - n))
+    C = L + (((C - L) * L) / (L - n))
   if x > 1:
-      C = L + (((C - L) * (1 - L)) / (x - L))
+    C = L + (((C - L) * (1 - L)) / (x - L))
 
 proc SetLum(C: Color, l: float32): Color {.inline.} =
   let d = l - Lum(C)
@@ -297,7 +298,12 @@ proc blendSoftLight(backdrop, source: ColorRGBA): ColorRGBA =
 
   when defined(amd64) and not defined(pixieNoSimd):
     let
-      vb = mm_setr_ps(backdrop.r.float32, backdrop.g.float32, backdrop.b.float32, 0)
+      vb = mm_setr_ps(
+        backdrop.r.float32,
+        backdrop.g.float32,
+        backdrop.b.float32,
+        0
+      )
       vs = mm_setr_ps(source.r.float32, source.g.float32, source.b.float32, 0)
       v2 = mm_set1_ps(2)
       v255 = mm_set1_ps(255)
@@ -519,8 +525,8 @@ when defined(amd64) and not defined(pixieNoSimd):
   proc unpackAlphaValues*(v: M128i): M128i {.inline.} =
     ## Unpack the first 32 bits into 4 rgba(0, 0, 0, value)
     let
-      first32 = cast[M128i]([uint32.high, 0, 0, 0]) # First 32 bits
-      alphaMask = mm_set1_epi32(cast[int32](0xff000000)) # Only `a`
+      first32 = cast[M128i]([uint32.high, 0, 0, 0])                # First 32 bits
+      alphaMask = mm_set1_epi32(cast[int32](0xff000000))           # Only `a`
 
     result = mm_shuffle_epi32(v, MM_SHUFFLE(0, 0, 0, 0))
 

src/pixie/fileformats/bmp.nim

@@ -1,4 +1,4 @@
-import flatty/binny, chroma, pixie/common, pixie/images
+import chroma, flatty/binny, pixie/common, pixie/images
 
 # See: https://en.wikipedia.org/wiki/BMP_file_format
 

src/pixie/fileformats/jpg.nim

@@ -1,4 +1,4 @@
-import pixie/images, pixie/common
+import pixie/common, pixie/images
 
 when defined(pixieUseStb):
   import pixie/fileformats/stb_image/stb_image

src/pixie/fileformats/png.nim

@@ -1,5 +1,5 @@
-import chroma, pixie/common, math, zippy, zippy/crc, flatty/binny,
-  pixie/images, pixie/masks
+import chroma, flatty/binny, math, pixie/common, pixie/images, pixie/masks,
+    zippy, zippy/crc
 
 # See http://www.libpng.org/pub/png/spec/1.2/PNG-Contents.html
 

src/pixie/fileformats/svg.nim

@@ -1,7 +1,7 @@
 ## Load and Save SVG files.
 
-import chroma, pixie/images, pixie/common, pixie/paths, vmath, xmlparser,
-  xmltree, strutils, strutils
+import chroma, pixie/common, pixie/images, pixie/paths, strutils, vmath,
+    xmlparser, xmltree
 
 const svgSignature* = "<?xml"
 

src/pixie/images.nim

@@ -1,4 +1,4 @@
-import chroma, blends, bumpy, vmath, common, system/memory, masks
+import blends, bumpy, chroma, common, masks, system/memory, vmath
 
 when defined(amd64) and not defined(pixieNoSimd):
   import nimsimd/sse2
@@ -68,7 +68,7 @@ proc `[]=`*(image: Image, x, y: int, rgba: ColorRGBA) {.inline.} =
   if image.inside(x, y):
     image.setRgbaUnsafe(x, y, rgba)
 
-proc fillUnsafe(data: var seq[ColorRGBA], rgba: ColorRGBA, start, len: int) =
+proc fillUnsafe*(data: var seq[ColorRGBA], rgba: ColorRGBA, start, len: int) =
   ## Fills the image data with the parameter color starting at index start and
   ## continuing for len indices.
 
@@ -571,7 +571,7 @@ proc drawCorrect(a, b: Image | Mask, mat = mat3(), blendMode = bmNormal) =
         else: # b is a Mask
           let
             sample = b.getValueSmooth(xFloat, yFloat)
-            blended =  blender(backdrop, rgba(0, 0, 0, sample))
+            blended = blender(backdrop, rgba(0, 0, 0, sample))
         a.setRgbaUnsafe(x, y, blended)
       else: # a is a Mask
         let backdrop = a.getValueUnsafe(x, y)
@@ -678,7 +678,7 @@ proc drawUber(a, b: Image | Mask, mat = mat3(), blendMode = bmNormal) =
           else: # b is a Mask
             let
               sample = b.getValueSmooth(xFloat, yFloat)
-              blended =  blender(backdrop, rgba(0, 0, 0, sample))
+              blended = blender(backdrop, rgba(0, 0, 0, sample))
           a.setRgbaUnsafe(x, y, blended)
         else: # a is a Mask
           let backdrop = a.getValueUnsafe(x, y)
@@ -765,7 +765,7 @@ proc drawUber(a, b: Image | Mask, mat = mat3(), blendMode = bmNormal) =
           else: # b is a Mask
             let
               sample = b.getValueUnsafe(xFloat.int, yFloat.int)
-              blended =  blender(backdrop, rgba(0, 0, 0, sample))
+              blended = blender(backdrop, rgba(0, 0, 0, sample))
           a.setRgbaUnsafe(x, y, blended)
         else: # a is a Mask
           let backdrop = a.getValueUnsafe(x, y)
@@ -827,11 +827,13 @@ proc shift*(target: Image | Mask, offset: Vec2) =
   ## Shifts the target by offset.
   if offset != vec2(0, 0):
     let copy = target.copy() # Copy to read from
+
     # Reset target for being drawn to
     when type(target) is Image:
       target.fill(rgba(0, 0, 0, 0))
     else:
       target.fill(0)
+
     target.draw(copy, offset, bmOverwrite) # Draw copy at offset
 
 proc shadow*(

src/pixie/masks.nim

@@ -1,4 +1,4 @@
-import common, vmath, system/memory
+import common, system/memory, vmath
 
 when defined(amd64) and not defined(pixieNoSimd):
   import nimsimd/sse2
@@ -84,7 +84,7 @@ proc minifyBy2*(mask: Mask, power = 1): Mask =
           mask.getValueUnsafe(x * 2 + 0, y * 2 + 1)
         result.setValueUnsafe(x, y, (value div 4).uint8)
 
-proc fillUnsafe(data: var seq[uint8], value: uint8, start, len: int) =
+proc fillUnsafe*(data: var seq[uint8], value: uint8, start, len: int) =
   ## Fills the mask data with the parameter value starting at index start and
   ## continuing for len indices.
   nimSetMem(data[start].addr, value.cint, len)

src/pixie/paths.nim

@@ -1,4 +1,4 @@
-import common, strutils, vmath, images, masks, chroma, bumpy, blends
+import blends, bumpy, chroma, common, images, masks, strutils, vmath
 
 when defined(amd64) and not defined(pixieNoSimd):
   import nimsimd/sse2
@@ -382,7 +382,7 @@ proc roundedRect*(
     h = wh.y
     s = splineCircleK
 
-    maxRadius = min(w/2, h/2)
+    maxRadius = min(w / 2, h / 2)
     nw = min(nw, maxRadius)
     ne = min(ne, maxRadius)
     se = min(se, maxRadius)
@@ -397,14 +397,14 @@ proc roundedRect*(
     l1 = vec2(x, y + h - sw)
     l2 = vec2(x, y + nw)
 
-    t1h = t1 + vec2(-nw*s, 0)
-    t2h = t2 + vec2(+ne*s, 0)
-    r1h = r1 + vec2(0, -ne*s)
-    r2h = r2 + vec2(0, +se*s)
-    b1h = b1 + vec2(+se*s, 0)
-    b2h = b2 + vec2(-sw*s, 0)
-    l1h = l1 + vec2(0, +sw*s)
-    l2h = l2 + vec2(0, -nw*s)
+    t1h = t1 + vec2(-nw * s, 0)
+    t2h = t2 + vec2(+ne * s, 0)
+    r1h = r1 + vec2(0, -ne * s)
+    r2h = r2 + vec2(0, +se * s)
+    b1h = b1 + vec2(+se * s, 0)
+    b2h = b2 + vec2(-sw * s, 0)
+    l1h = l1 + vec2(0, +sw * s)
+    l2h = l2 + vec2(0, -nw * s)
 
   if clockwise:
     path.moveTo(t1)
@@ -571,15 +571,15 @@ proc commandsToShapes*(path: Path): seq[seq[Vec2]] =
 
       var q = sqrt(max(0, pq))
       if large == sweep:
-          q = -q
+        q = -q
 
       proc svgAngle(u, v: Vec2): float32 =
         let
-          dot = dot(u,v)
+          dot = dot(u, v)
           len = length(u) * length(v)
         result = arccos(clamp(dot / len, -1, 1))
         if (u.x * v.y - u.y * v.x) < 0:
-            result = -result
+          result = -result
 
       let
         cp = vec2(q * radii.x * p.y / radii.y, -q * radii.y * p.x / radii.x)
@@ -865,13 +865,11 @@ proc partitionSegments(
     segmentCount += shape.len - 1
 
   let
-    maxPartitions = max(1, height div 10)
-    numPartitions = min(maxPartitions, max(1, segmentCount div 10))
-
-  result.setLen(numPartitions)
-
-  let partitionHeight = height div numPartitions
+    maxPartitions = max(1, height div 10).uint32
+    numPartitions = min(maxPartitions, max(1, segmentCount div 10).uint32)
+    partitionHeight = (height.uint32 div numPartitions)
 
+  var partitions = newSeq[seq[(Segment, int16)]](numPartitions)
   for shape in shapes:
     for segment in shape.segments:
       if segment.at.y == segment.to.y: # Skip horizontal
@@ -884,41 +882,46 @@ proc partitionSegments(
         winding = -1
 
       if partitionHeight == 0:
-        result[0].add((segment, winding))
+        partitions[0].add((segment, winding))
       else:
-        let
-          atPartition = max(0, segment.at.y).int div partitionHeight
-          toPartition = max(0, ceil(segment.to.y)).int div partitionHeight
-        for i in min(atPartition, result.high) .. min(toPartition, result.high):
-          result[i].add((segment, winding))
+        var
+          atPartition = max(0, segment.at.y).uint32 div partitionHeight
+          toPartition = max(0, ceil(segment.to.y)).uint32 div partitionHeight
+        atPartition = clamp(atPartition, 0, partitions.high.uint32)
+        toPartition = clamp(toPartition, 0, partitions.high.uint32)
+        for i in atPartition .. toPartition:
+          partitions[i].add((segment, winding))
+
+  partitions
 
 proc computeCoverages(
   coverages: var seq[uint8],
   hits: var seq[(float32, int16)],
+  numHits: var int,
   size: Vec2,
   y: int,
   partitions: seq[seq[(Segment, int16)]],
+  partitionHeight: uint32,
   windingRule: WindingRule
-) =
+) {.inline.} =
   const
     quality = 5 # Must divide 255 cleanly (1, 3, 5, 15, 17, 51, 85)
-    sampleCoverage = 255.uint8 div quality
-    ep = 0.0001 * PI
+    sampleCoverage = (255 div quality).uint8
     offset = 1 / quality.float32
     initialOffset = offset / 2
 
-  var numHits: int
-
   let
-    partitionHeight = size.y.int div partitions.len
     partition =
       if partitionHeight == 0:
-        0
+        0.uint32
       else:
-        min(y div partitionHeight, partitions.high)
+        min(y.uint32 div partitionHeight, partitions.high.uint32)
+
+  zeroMem(coverages[0].addr, coverages.len)
 
   # Do scanlines for this row
   for m in 0 ..< quality:
+    const ep = 0.0001 * PI
     let
       yLine = y.float32 + initialOffset + offset * m.float32 + ep
       scanline = Line(a: vec2(0, yLine), b: vec2(size.x, yLine))
@@ -926,7 +929,7 @@ proc computeCoverages(
     for (segment, winding) in partitions[partition]:
       if segment.at.y <= scanline.a.y and segment.to.y >= scanline.a.y:
         var at: Vec2
-        if scanline.intersects(segment, at):# and segment.to != at:
+        if scanline.intersects(segment, at): # and segment.to != at:
           if numHits == hits.len:
             hits.setLen(hits.len * 2)
           hits[numHits] = (at.x.clamp(0, scanline.b.x), winding)
@@ -940,9 +943,9 @@ proc computeCoverages(
     for i in 0 ..< numHits:
       let (at, winding) = hits[i]
 
-      var
-        fillStart = x.int
-        leftCover = if at.int - x.int > 0: trunc(x) + 1 - x else: at - x
+      var fillStart = x.int
+
+      let leftCover = if at.int - x.int > 0: trunc(x) + 1 - x else: at - x
       if leftCover != 0:
         inc fillStart
         if shouldFill(windingRule, count):
@@ -978,7 +981,9 @@ proc fillShapes(
   windingRule: WindingRule,
   blendMode: BlendMode
 ) =
-  let partitions = partitionSegments(shapes, image.height)
+  let
+    partitions = partitionSegments(shapes, image.height)
+    partitionHeight = image.height.uint32 div partitions.len.uint32
 
   # Figure out the total bounds of all the shapes,
   # rasterize only within the total bounds
@@ -995,17 +1000,17 @@ proc fillShapes(
   var
     coverages = newSeq[uint8](image.width)
     hits = newSeq[(float32, int16)](4)
+    numHits: int
 
   for y in startY ..< stopY:
-    # Reset buffer for this row
-    zeroMem(coverages[0].addr, coverages.len)
-
     computeCoverages(
       coverages,
       hits,
+      numHits,
       image.wh,
       y,
       partitions,
+      partitionHeight,
       windingRule
     )
 
@@ -1023,34 +1028,45 @@ proc fillShapes(
         var coverage = mm_loadu_si128(coverages[x].addr)
         coverage = mm_and_si128(coverage, first32)
 
-        let eqZero = mm_cmpeq_epi16(coverage, mm_setzero_si128())
+        let
+          index = image.dataIndex(x, y)
+          eqZero = mm_cmpeq_epi16(coverage, mm_setzero_si128())
         if mm_movemask_epi8(eqZero) != 0xffff:
           # If the coverages are not all zero
-          var source = vColor
-
-          if mm_movemask_epi8(mm_cmpeq_epi32(coverage, first32)) != 0xffff:
-            # If the coverages are not all 255
+          if mm_movemask_epi8(mm_cmpeq_epi32(coverage, first32)) == 0xffff:
+            # Coverages are all 255
+            if color.a == 255 and blendMode == bmNormal:
+              mm_storeu_si128(image.data[index].addr, vColor)
+            else:
+              let backdrop = mm_loadu_si128(image.data[index].addr)
+              mm_storeu_si128(
+                image.data[index].addr,
+                blenderSimd(backdrop, vColor)
+              )
+          else:
+            # Coverages are not all 255
             coverage = unpackAlphaValues(coverage)
             # Shift the coverages from `a` to `g` and `a` for multiplying
             coverage = mm_or_si128(coverage, mm_srli_epi32(coverage, 16))
 
             var
-              colorEven = mm_slli_epi16(source, 8)
-              colorOdd = mm_and_si128(source, oddMask)
+              source = vColor
+              sourceEven = mm_slli_epi16(source, 8)
+              sourceOdd = mm_and_si128(source, oddMask)
 
-            colorEven = mm_mulhi_epu16(colorEven, coverage)
-            colorOdd = mm_mulhi_epu16(colorOdd, coverage)
+            sourceEven = mm_mulhi_epu16(sourceEven, coverage)
+            sourceOdd = mm_mulhi_epu16(sourceOdd, coverage)
 
-            colorEven = mm_srli_epi16(mm_mulhi_epu16(colorEven, div255), 7)
-            colorOdd = mm_srli_epi16(mm_mulhi_epu16(colorOdd, div255), 7)
+            sourceEven = mm_srli_epi16(mm_mulhi_epu16(sourceEven, div255), 7)
+            sourceOdd = mm_srli_epi16(mm_mulhi_epu16(sourceOdd, div255), 7)
 
-            source = mm_or_si128(colorEven, mm_slli_epi16(colorOdd, 8))
-
-          let
-            index = image.dataIndex(x, y)
-            backdrop = mm_loadu_si128(image.data[index].addr)
-          mm_storeu_si128(image.data[index].addr, blenderSimd(backdrop, source))
+            source = mm_or_si128(sourceEven, mm_slli_epi16(sourceOdd, 8))
 
+            let backdrop = mm_loadu_si128(image.data[index].addr)
+            mm_storeu_si128(
+              image.data[index].addr,
+              blenderSimd(backdrop, source)
+            )
         x += 4
 
     while x < image.width:
@@ -1069,8 +1085,12 @@ proc fillShapes(
           source.b = ((color.b.uint16 * coverage) div 255).uint8
           source.a = ((color.a.uint16 * coverage) div 255).uint8
 
-        let backdrop = image.getRgbaUnsafe(x, y)
-        image.setRgbaUnsafe(x, y, blender(backdrop, source))
+        if source.a == 255 and blendMode == bmNormal:
+          # Skip blending
+          image.setRgbaUnsafe(x, y, source)
+        else:
+          let backdrop = image.getRgbaUnsafe(x, y)
+          image.setRgbaUnsafe(x, y, blender(backdrop, source))
       inc x
 
 proc fillShapes(
@@ -1078,7 +1098,9 @@ proc fillShapes(
   shapes: seq[seq[Vec2]],
   windingRule: WindingRule
 ) =
-  let partitions = partitionSegments(shapes, mask.height)
+  let
+    partitions = partitionSegments(shapes, mask.height)
+    partitionHeight = mask.height.uint32 div partitions.len.uint32
 
   # Figure out the total bounds of all the shapes,
   # rasterize only within the total bounds
@@ -1088,24 +1110,23 @@ proc fillShapes(
     startY = max(0, bounds.y.int)
     stopY = min(mask.height, (bounds.y + bounds.h).int)
 
-  var
-    coverages = newSeq[uint8](mask.width)
-    hits = newSeq[(float32, int16)](4)
-
-
   when defined(amd64) and not defined(pixieNoSimd):
     let maskerSimd = bmNormal.maskerSimd()
 
-  for y in startY ..< stopY:
-    # Reset buffer for this row
-    zeroMem(coverages[0].addr, coverages.len)
+  var
+    coverages = newSeq[uint8](mask.width)
+    hits = newSeq[(float32, int16)](4)
+    numHits: int
 
+  for y in startY ..< stopY:
     computeCoverages(
       coverages,
       hits,
+      numHits,
       mask.wh,
       y,
       partitions,
+      partitionHeight,
       windingRule
     )
 
@@ -1132,6 +1153,7 @@ proc fillShapes(
         if peeked == 0:
           x += 8
           continue
+
       let coverage = coverages[x]
       if coverage != 0:
         let

tests/benchmark_blends.nim

@@ -1,4 +1,4 @@
-import benchy, chroma, vmath, pixie/images
+import benchy, chroma, pixie/images, vmath
 
 include pixie/blends
 

tests/benchmark_drawCorrect_vs_drawUber.nim

@@ -1,4 +1,5 @@
-import chroma, vmath, benchy
+import benchy, chroma, vmath
+
 include pixie/images
 
 block:
@@ -23,7 +24,6 @@ block:
     a.drawUber(b, translate(vec2(25, 25)), bmNormal)
     keep(b)
 
-
 block:
   let
     a = newImage(1000, 1000)

tests/benchmark_images.nim

@@ -1,4 +1,4 @@
-import chroma, pixie, benchy
+import benchy, chroma, pixie
 
 let image = newImage(2560, 1440)
 

tests/benchmark_images_draw.nim

@@ -1,4 +1,4 @@
-import pixie, chroma, vmath, benchy
+import benchy, chroma, pixie, vmath
 
 block:
   let

tests/benchmark_jpg.nim

@@ -1,4 +1,4 @@
-import pixie/fileformats/jpg, benchy
+import benchy, pixie/fileformats/jpg
 
 let data = readFile("tests/images/jpg/jpeg420exif.jpg")
 

tests/benchmark_masks.nim

@@ -1,4 +1,4 @@
-import chroma, pixie, benchy
+import benchy, chroma, pixie
 
 let mask = newMask(2560, 1440)
 

tests/benchmark_paths.nim

@@ -1,4 +1,4 @@
-import pixie/paths, benchy
+import benchy, pixie/paths
 
 let pathStr = "m57.611-8.591c-1.487,1.851-4.899,4.42-1.982,6.348,0.194,0.129,0.564,0.133,0.737-0.001,2.021-1.565,4.024-2.468,6.46-3.05,0.124-0.029,0.398,0.438,0.767,0.277,1.613-0.703,3.623-0.645,4.807-1.983,3.767,0.224,7.332-0.892,10.723-2.2,1.161-0.448,2.431-1.007,3.632-1.509,1.376-0.576,2.58-1.504,3.692-2.645,0.133-0.136,0.487-0.046,0.754-0.046-0.04-0.863,0.922-0.99,1.169-1.612,0.092-0.232-0.058-0.628,0.075-0.73,2.138-1.63,3.058-3.648,1.889-6.025-0.285-0.578-0.534-1.196-1.1-1.672-1.085-0.911-2.187-0.057-3.234-0.361-0.159,0.628-0.888,0.456-1.274,0.654-0.859,0.439-2.192-0.146-3.051,0.292-1.362,0.695-2.603,0.864-4.025,1.241-0.312,0.082-1.09-0.014-1.25,0.613-0.134-0.134-0.282-0.368-0.388-0.346-1.908,0.396-3.168,0.61-4.469,2.302-0.103,0.133-0.545-0.046-0.704,0.089-0.957,0.808-1.362,2.042-2.463,2.714-0.201,0.123-0.553-0.045-0.747,0.084-0.646,0.431-1.013,1.072-1.655,1.519-0.329,0.229-0.729-0.096-0.697-0.352,0.245-1.947,0.898-3.734,0.323-5.61,2.077-2.52,4.594-4.469,6.4-7.2,0.015-2.166,0.707-4.312,0.594-6.389-0.01-0.193-0.298-0.926-0.424-1.273-0.312-0.854,0.594-1.92-0.25-2.644-1.404-1.203-2.696-0.327-3.52,1.106-1.838,0.39-3.904,1.083-5.482-0.151-1.007-0.787-1.585-1.693-2.384-2.749-0.985-1.302-0.65-2.738-0.58-4.302,0.006-0.128-0.309-0.264-0.309-0.398,0.001-0.135,0.221-0.266,0.355-0.4-0.706-0.626-0.981-1.684-2-2,0.305-1.092-0.371-1.976-1.242-2.278-1.995-0.691-3.672,1.221-5.564,1.294-0.514,0.019-0.981-1.019-1.63-1.344-0.432-0.216-1.136-0.249-1.498,0.017-0.688,0.504-1.277,0.618-2.035,0.823-1.617,0.436-2.895,1.53-4.375,2.385-1.485,0.857-2.44,2.294-3.52,3.614-0.941,1.152-1.077,3.566,0.343,4.066,1.843,0.65,3.147-2.053,5.113-1.727,0.312,0.051,0.518,0.362,0.408,0.75,0.389,0.109,0.607-0.12,0.8-0.4,0.858,1.019,2.022,1.356,2.96,2.229,0.97,0.904,2.716,0.486,3.731,1.483,1.529,1.502,0.97,4.183,2.909,5.488-0.586,1.313-1.193,2.59-1.528,4.017-0.282,1.206,0.712,2.403,1.923,2.312,1.258-0.094,1.52-0.853,2.005-1.929,0.267,0.267,0.736,0.564,0.695,0.78-0.457,2.387-1.484,4.38-1.942,6.811-0.059,0.317-0.364,0.519-0.753,0.409-0.468,4.149-4.52,6.543-7.065,9.708-0.403,0.502-0.407,1.751,0.002,2.154,1.403,1.387,3.363-0.159,5.063-0.662,0.213-1.206,1.072-2.148,2.404-2.092,0.256,0.01,0.491-0.532,0.815-0.662,0.348-0.138,0.85,0.086,1.136-0.112,1.729-1.195,3.137-2.301,4.875-3.49,0.192-0.131,0.536,0.028,0.752-0.08,0.325-0.162,0.512-0.549,0.835-0.734,0.348-0.2,0.59,0.09,0.783,0.37-0.646,0.349-0.65,1.306-1.232,1.508-0.775,0.268-1.336,0.781-2.01,1.228-0.292,0.193-0.951-0.055-1.055,0.124-0.598,1.028-1.782,1.466-2.492,2.349z"
 

tests/benchmark_png.nim

@@ -1,5 +1,5 @@
-import pixie/fileformats/png, stb_image/read as stbi, stb_image/write as stbr,
-  benchy, nimPNG
+import benchy, nimPNG, pixie/fileformats/png, stb_image/read as stbi,
+    stb_image/write as stbr
 
 let data = readFile("tests/images/png/lenna.png")
 

tests/benchmark_svg.nim

@@ -1,4 +1,4 @@
-import pixie/fileformats/svg, benchy
+import benchy, pixie/fileformats/svg
 
 let data = readFile("tests/images/svg/Ghostscript_Tiger.svg")
 

tests/fuzz_jpg.nim

@@ -1,4 +1,4 @@
-import random, strformat, pixie/fileformats/jpg, pixie/common
+import pixie/common, pixie/fileformats/jpg, random, strformat
 
 randomize()
 

tests/fuzz_paths.nim

@@ -1,4 +1,4 @@
-import random, strformat, pixie/paths, pixie/common, strformat
+import pixie/common, pixie/paths, random, strformat
 
 randomize()
 

tests/fuzz_png.nim

@@ -1,4 +1,4 @@
-import random, strformat, pixie/fileformats/png, pixie/common, pngsuite
+import pixie/common, pixie/fileformats/png, pngsuite, random, strformat
 
 randomize()
 

tests/fuzz_svg.nim

@@ -1,4 +1,4 @@
-import random, strformat, pixie/fileformats/svg, pixie/common
+import pixie/common, pixie/fileformats/svg, random, strformat
 
 randomize()
 

tests/test_bmp.nim

@@ -1,4 +1,4 @@
-import pixie, pixie/fileformats/bmp, chroma
+import chroma, pixie, pixie/fileformats/bmp
 
 block:
   var image = newImage(4, 2)
@@ -32,5 +32,6 @@ block:
 
 block:
   for bits in [32, 24]:
-    var image = decodeBmp(readFile("tests/images/bmp/knight." & $bits & ".master.bmp"))
+    let image =
+      decodeBmp(readFile("tests/images/bmp/knight." & $bits & ".master.bmp"))
     writeFile("tests/images/bmp/knight." & $bits & ".bmp", encodeBmp(image))

tests/test_images.nim

@@ -1,4 +1,4 @@
-import pixie, chroma, vmath
+import chroma, pixie, vmath
 
 block:
   let image = newImage(10, 10)

tests/test_paths.nim

@@ -1,4 +1,4 @@
-import pixie, chroma, pixie/fileformats/png
+import chroma, pixie, pixie/fileformats/png
 
 block:
   let pathStr = """

tests/test_png.nim

@@ -1,4 +1,4 @@
-import pixie/fileformats/png, strformat, pngsuite, pixie/common
+import pixie/common, pixie/fileformats/png, pngsuite, strformat
 
 for file in pngSuiteFiles:
   let

tests/test_svg.nim

@@ -1,4 +1,4 @@
-import pixie/fileformats/svg, pixie, strformat
+import pixie, pixie/fileformats/svg, strformat
 
 const files = [
   "line01",

tests/validate_png.nim

@@ -1,4 +1,4 @@
-import chroma, pixie/fileformats/png, stb_image/read as stbi, strformat, pngsuite
+import chroma, pixie/fileformats/png, pngsuite, stb_image/read as stbi, strformat
 
 for file in pngSuiteFiles:
   let

tools/gen_readme.nim

@@ -1,4 +1,4 @@
-import os, strutils, osproc, sequtils, algorithm
+import algorithm, os, osproc, sequtils, strutils
 
 proc cutBetween(str, a, b: string): string =
   let
@@ -30,7 +30,8 @@ for path in exampleFiles:
     md.add "```nim"
     md.add innerCode.strip()
     md.add "```"
-    md.add "![example output](" & path.replace(".nim", ".png").replace("\\", "/") & ")"
+    md.add "![example output](" & path.replace(".nim", ".png").replace("\\",
+        "/") & ")"
     md.add ""
 
 var readme = readFile("README.md")