use Sk2s for EvalQuadTangent and ChopQuadAt

src/core/SkGeometry.cpp

Index: src/core/SkGeometry.cpp
diff --git a/src/core/SkGeometry.cpp b/src/core/SkGeometry.cpp
index 6d14e4bb54f595996213a85ecd08d4242f8c450c..aea1fe56654468a1d57430aaa3c21b0d1c8f32be 100644
--- a/src/core/SkGeometry.cpp
+++ b/src/core/SkGeometry.cpp
@@ -7,6 +7,7 @@
 
 #include "SkGeometry.h"
 #include "SkMatrix.h"
+#include "Sk2x.h"
 
 /** If defined, this makes eval_quad and eval_cubic do more setup (sometimes
     involving integer multiplies by 2 or 3, but fewer calls to SkScalarMul.
@@ -127,8 +128,6 @@ void SkEvalQuadAt(const SkPoint src[3], SkScalar t, SkPoint* pt, SkVector* tange
     }
 }
 
-#include "Sk2x.h"
-
 SkPoint SkEvalQuadAt(const SkPoint src[3], SkScalar t) {
     SkASSERT(src);
     SkASSERT(t >= 0 && t <= SK_Scalar1);
@@ -147,6 +146,23 @@ SkPoint SkEvalQuadAt(const SkPoint src[3], SkScalar t) {
     return result;
 }
 
+SkVector SkEvalQuadTangentAt(const SkPoint src[3], SkScalar t) {
+    SkASSERT(src);
+    SkASSERT(t >= 0 && t <= SK_Scalar1);
+
+    Sk2f P0 = Sk2f::Load(&src[0].fX);
+    Sk2f P1 = Sk2f::Load(&src[1].fX);
+    Sk2f P2 = Sk2f::Load(&src[2].fX);
+
+    Sk2f B = P1 - P0;
+    Sk2f A = P2 - P1 - B;
+    Sk2f T = A * Sk2f(t) + B;
+
+    SkVector result;
+    (T + T).store(&result.fX);
+    return result;
+}
+
 static void interp_quad_coords(const SkScalar* src, SkScalar* dst, SkScalar t) {
     SkScalar    ab = SkScalarInterp(src[0], src[2], t);
     SkScalar    bc = SkScalarInterp(src[2], src[4], t);
@@ -165,6 +181,28 @@ void SkChopQuadAt(const SkPoint src[3], SkPoint dst[5], SkScalar t) {
     interp_quad_coords(&src[0].fY, &dst[0].fY, t);
 }
 
+static inline Sk2s interp(const Sk2s& v0, const Sk2s& v1, const Sk2s& t) {
+    return v0 + (v1 - v0) * t;
+}
+
+void SkChopQuadAt2(const SkPoint src[3], SkPoint dst[5], SkScalar t) {
+    SkASSERT(t > 0 && t < SK_Scalar1);
+
+    Sk2s p0 = Sk2f::Load(&src[0].fX);
+    Sk2s p1 = Sk2f::Load(&src[1].fX);
+    Sk2s p2 = Sk2f::Load(&src[2].fX);
+    Sk2s tt = Sk2s(t);
+    
+    Sk2s p01 = interp(p0, p1, tt);
+    Sk2s p12 = interp(p1, p2, tt);
+
+    p0.store(&dst[0].fX);
+    p01.store(&dst[1].fX);
+    interp(p01, p12, tt).store(&dst[2].fX);
+    p12.store(&dst[3].fX);
+    p2.store(&dst[4].fX);
+}
+
 void SkChopQuadAtHalf(const SkPoint src[3], SkPoint dst[5]) {
     SkScalar x01 = SkScalarAve(src[0].fX, src[1].fX);
     SkScalar y01 = SkScalarAve(src[0].fY, src[1].fY);