Merge pull request #327 from guzba/master

float32 hit pos, faster simd unpackAlphaValues

src/pixie/blends.nim

@@ -166,9 +166,7 @@ proc SetSat(C: Color, s: float32): Color {.inline.} =
     result = (C - min([C.r, C.g, C.b])) * s / satC
 
 proc blendNormal(backdrop, source: ColorRGBX): ColorRGBX =
-  if backdrop.a == 0:
+  if backdrop.a == 0 or source.a == 255:
-    return source
-  if source.a == 255:
     return source
   if source.a == 0:
     return backdrop

src/pixie/internal.nim

@@ -105,18 +105,7 @@ when defined(amd64) and not defined(pixieNoSimd):
 
   proc unpackAlphaValues*(v: M128i): M128i {.inline, raises: [].} =
     ## Unpack the first 32 bits into 4 rgba(0, 0, 0, value)
-    let mask = cast[M128i]([uint8.high.uint64, 0])
+    let
-
+      a = mm_unpacklo_epi8(v, mm_setzero_si128())
-    var
+      b = mm_unpacklo_epi8(a, mm_setzero_si128())
-      i = mm_and_si128(v, mask)
+    result = mm_slli_epi32(b, 24) # Shift the values to uint32 `a`
-      j = mm_and_si128(v, mm_slli_si128(mask, 1))
-      k = mm_and_si128(v, mm_slli_si128(mask, 2))
-      l = mm_and_si128(v, mm_slli_si128(mask, 3))
-
-    # Shift the values to uint32 `a`
-    i = mm_slli_si128(i, 3)
-    j = mm_slli_si128(j, 6)
-    k = mm_slli_si128(k, 9)
-    l = mm_slli_si128(l, 12)
-
-    result = mm_or_si128(mm_or_si128(i, j), mm_or_si128(k, l))

src/pixie/paths.nim

@@ -1247,7 +1247,7 @@ proc computeCoverage(
           if entry.m == 0:
             entry.b
           else:
-            (yLine - entry.b) / entry.m
+            (yLine.float32 - entry.b) / entry.m
 
         hits[numHits] = (min(x, width), entry.winding)
         inc numHits