diff --git a/src/pixie/masks.nim b/src/pixie/masks.nim
index 763718c..9887e6e 100644
--- a/src/pixie/masks.nim
+++ b/src/pixie/masks.nim
@@ -178,25 +178,35 @@ proc getValueSmooth*(mask: Mask, x, y: float32): uint8 =
 
 proc spread*(mask: Mask, spread: float32) =
   ## Grows the mask by spread.
+  let spread = round(spread).int
   if spread == 0:
     return
   if spread < 0:
     raise newException(PixieError, "Cannot apply negative spread")
 
-  let
-    copy = mask.copy()
-    spread = round(spread).int
+  # Spread in the X direction. Store with dimensions swapped for reading later.
+  let spreadX = newMask(mask.height, mask.width)
   for y in 0 ..< mask.height:
     for x in 0 ..< mask.width:
       var maxValue: uint8
-      block blurBox:
-        for by in max(y - spread, 0) .. min(y + spread, mask.height - 1):
-          for bx in max(x - spread, 0) .. min(x + spread, mask.width - 1):
-            let value = copy.getValueUnsafe(bx, by)
-            if value > maxValue:
-              maxValue = value
-            if maxValue == 255:
-              break blurBox
+      for xx in max(x - spread, 0) .. min(x + spread, mask.width - 1):
+        let value = mask.getValueUnsafe(xx, y)
+        if value > maxValue:
+          maxValue = value
+        if maxValue == 255:
+          break
+      spreadX.setValueUnsafe(y, x, maxValue)
+
+  # Spread in the Y direction and modify mask.
+  for y in 0 ..< mask.height:
+    for x in 0 ..< mask.width:
+      var maxValue: uint8
+      for yy in max(y - spread, 0) .. min(y + spread, mask.height - 1):
+        let value = spreadX.getValueUnsafe(yy, x)
+        if value > maxValue:
+          maxValue = value
+        if maxValue == 255:
+          break
       mask.setValueUnsafe(x, y, maxValue)
 
 proc ceil*(mask: Mask) =
diff --git a/src/pixie/paths.nim b/src/pixie/paths.nim
index 124dd5b..e3f8bd3 100644
--- a/src/pixie/paths.nim
+++ b/src/pixie/paths.nim
@@ -1137,12 +1137,10 @@ iterator walk(
       # between zero and nonzero (or the last hit)
       count += winding
       continue
-    if at <= 0:
-      count += winding
-      continue
-    if shouldFill(windingRule, count):
-      yield (prevAt, at, count)
-    prevAt = at
+    if at > 0:
+      if shouldFill(windingRule, count):
+        yield (prevAt, at, count)
+      prevAt = at
     count += winding
 
   when defined(pixieLeakCheck):
diff --git a/tests/benchmark_masks.nim b/tests/benchmark_masks.nim
index f92e8d9..5c0a550 100644
--- a/tests/benchmark_masks.nim
+++ b/tests/benchmark_masks.nim
@@ -33,5 +33,31 @@ timeIt "ceil":
 
 reset()
 
-timeIt "spread":
-  mask.spread(10)
+block spread_1:
+  var p: Path
+  p.rect(500, 500, 500, 500)
+
+  timeIt "spread_1":
+    mask.fill(0)
+    mask.fillPath(p)
+    mask.spread(10)
+
+block spread_2:
+  var p: Path
+  p.rect(500, 500, 1000, 1000)
+
+  timeIt "spread_2":
+    mask.fill(0)
+    mask.fillPath(p)
+    mask.spread(10)
+
+block spread_3:
+  timeIt "spread_3":
+    mask.fill(255)
+    mask.spread(10)
+
+block spread_4:
+  timeIt "spread_4":
+    mask.fill(0)
+    mask.setValueUnsafe(1000, 1000, 255)
+    mask.spread(10)