Proposed fix for small bug in gpu conical gradients

src/effects/gradients/SkTwoPointConicalGradient_gpu.cpp

Index: src/effects/gradients/SkTwoPointConicalGradient_gpu.cpp
diff --git a/src/effects/gradients/SkTwoPointConicalGradient_gpu.cpp b/src/effects/gradients/SkTwoPointConicalGradient_gpu.cpp
index 720dc6345d5014cf3da0a4c9d66bfd386fbd14d0..3b6d5f28a55b363c3859a9200968c2ae35564bd5 100644
--- a/src/effects/gradients/SkTwoPointConicalGradient_gpu.cpp
+++ b/src/effects/gradients/SkTwoPointConicalGradient_gpu.cpp
@@ -785,7 +785,8 @@ static ConicalType set_matrix_circle_conical(const SkTwoPointConicalGradient& sh
     // to the edge shader. kEdgeErrorTol = 5 * kErrorTol was picked after manual testing
     // so that C = 1 / A is stable, and the linear approximation used in the Edge shader is
     // still accurate.
-    if (SkScalarAbs(A) < kEdgeErrorTol) {
+    if (SkScalarAbs(SkScalarAbs(shader.getDiffRadius()) - shader.getCenterX1()) <

[egdaniel, 2014/09/22 13:59:35]

can we do "SkScalarAbs(||centerEndTrans|| - (radiusEnd -1)) < kErrorTol"?

It involves the exact same math as your version and we only have to worry about
the radius scaling of error in the shader instead of both places.

+            kEdgeErrorTol * shader.getStartRadius()) {
         return kEdge_ConicalType;
     }