Generate Signed Distance Field directly from vector path

src/core/SkDistanceFieldGen.cpp

Index: src/core/SkDistanceFieldGen.cpp
diff --git a/src/core/SkDistanceFieldGen.cpp b/src/core/SkDistanceFieldGen.cpp
index 147aefad79052428453494e33fb98ea4e21e9d43..85ad5b95f7d4e2642916741e3dca1166912aa8bc 100755
--- a/src/core/SkDistanceFieldGen.cpp
+++ b/src/core/SkDistanceFieldGen.cpp
@@ -317,13 +317,16 @@ static void B2(DFData* curr, int width) {
 
 #if !DUMP_EDGE
 static unsigned char pack_distance_field_val(float dist, float distanceMagnitude) {
-    if (dist <= -distanceMagnitude) {
-        return 255;
-    } else if (dist > distanceMagnitude) {
-        return 0;
-    } else {
-        return (unsigned char)((distanceMagnitude-dist)*128.0f/distanceMagnitude);
-    }
+    // The distance field is constructed as unsigned char values, so that the zero value is at 128,

[jvanverth1, 2016/02/19 23:36:46]

I was looking at this to understand it a little better, so I could possibly
submit it separately a little early (we have a Chrome branch coming up next
week). However, it looks like (-31/32, 0] maps to 128 (this was true of the
original code). I was thinking it might be better to have the values surrounding
0 map to 128, e.g. (-31/64, 31/64]. What do you think?

[Joel.Liang, 2016/02/22 07:34:00]

Yes, to do that we could add SkScalarRoundToInt to the return statement:
    return (unsigned char)SkScalarRoundToInt(dist / (2 * distanceMagnitude) *
256.0f);

And it's safe, because we have truncated the input by multiply 127/128. (It will
truncate the result which greater than 255.0)

+    // Beside 128, we have 128 values in range [0, 128), but only 127 values in range (128, 255].
+    // So we multiply distanceMagnitude by 127/128 at the latter range to avoid overflow.
+    dist = SkScalarPin(-dist, -distanceMagnitude, distanceMagnitude * 127.0f / 128.0f);
+
+    // Scale into the positive range for unsigned distance
+    dist += distanceMagnitude;
+
+    // Scale into unsigned char range
+    return (unsigned char)(dist / (2 * distanceMagnitude) * 256.0f);
 }
 #endif