diff --git a/bench/GeometryBench.cpp b/bench/GeometryBench.cpp
index 1d85757..24d5c34 100644
--- a/bench/GeometryBench.cpp
+++ b/bench/GeometryBench.cpp
@@ -132,46 +132,115 @@
 
 ///////////////////////////////////////////////////////////////////////////////////////////////////
 
-class EvalQuadAt0 : public GeometryBench {
+class QuadBenchBase : public GeometryBench {
+protected:
     SkPoint fPts[3];
 public:
-    EvalQuadAt0() : GeometryBench("evalquadat0") {
+    QuadBenchBase(const char name[]) : GeometryBench(name) {
         SkRandom rand;
         for (int i = 0; i < 3; ++i) {
             fPts[i].set(rand.nextUScalar1(), rand.nextUScalar1());
         }
     }
-    
+};
+
+class EvalQuadAt0 : public QuadBenchBase {
+public:
+    EvalQuadAt0() : QuadBenchBase("evalquadat0") {}
 protected:
     void onDraw(const int loops, SkCanvas* canvas) SK_OVERRIDE {
         SkPoint result;
         for (int outer = 0; outer < loops; ++outer) {
-            for (int i = 0; i < 10000; ++i) {
-                SkEvalQuadAt(fPts, 0.5f, &result);
-            }
+            SkEvalQuadAt(fPts, 0.5f, &result);
+            SkEvalQuadAt(fPts, 0.5f, &result);
+            SkEvalQuadAt(fPts, 0.5f, &result);
+            SkEvalQuadAt(fPts, 0.5f, &result);
         }
     }
 };
 DEF_BENCH( return new EvalQuadAt0; )
 
-class EvalQuadAt1 : public GeometryBench {
-    SkPoint fPts[3];
+class EvalQuadAt1 : public QuadBenchBase {
 public:
-    EvalQuadAt1() : GeometryBench("evalquadat1") {
-        SkRandom rand;
-        for (int i = 0; i < 3; ++i) {
-            fPts[i].set(rand.nextUScalar1(), rand.nextUScalar1());
-        }
-    }
-    
+    EvalQuadAt1() : QuadBenchBase("evalquadat1") {}
 protected:
     void onDraw(const int loops, SkCanvas* canvas) SK_OVERRIDE {
+        SkPoint result;
         for (int outer = 0; outer < loops; ++outer) {
-            for (int i = 0; i < 10000; ++i) {
-                SkEvalQuadAt(fPts, 0.5f);
-            }
+            result = SkEvalQuadAt(fPts, 0.5f);
+            result = SkEvalQuadAt(fPts, 0.5f);
+            result = SkEvalQuadAt(fPts, 0.5f);
+            result = SkEvalQuadAt(fPts, 0.5f);
         }
     }
 };
 DEF_BENCH( return new EvalQuadAt1; )
 
+////////
+
+class EvalQuadTangentAt0 : public QuadBenchBase {
+public:
+    EvalQuadTangentAt0() : QuadBenchBase("evalquadtangentat0") {}
+protected:
+    void onDraw(const int loops, SkCanvas* canvas) SK_OVERRIDE {
+        SkPoint result;
+        for (int outer = 0; outer < loops; ++outer) {
+            SkEvalQuadAt(fPts, 0.5f, NULL, &result);
+            SkEvalQuadAt(fPts, 0.5f, NULL, &result);
+            SkEvalQuadAt(fPts, 0.5f, NULL, &result);
+            SkEvalQuadAt(fPts, 0.5f, NULL, &result);
+        }
+    }
+};
+DEF_BENCH( return new EvalQuadTangentAt0; )
+
+class EvalQuadTangentAt1 : public QuadBenchBase {
+public:
+    EvalQuadTangentAt1() : QuadBenchBase("evalquadtangentat1") {}
+protected:
+    void onDraw(const int loops, SkCanvas* canvas) SK_OVERRIDE {
+        SkPoint result;
+        for (int outer = 0; outer < loops; ++outer) {
+            result = SkEvalQuadTangentAt(fPts, 0.5f);
+            result = SkEvalQuadTangentAt(fPts, 0.5f);
+            result = SkEvalQuadTangentAt(fPts, 0.5f);
+            result = SkEvalQuadTangentAt(fPts, 0.5f);
+        }
+    }
+};
+DEF_BENCH( return new EvalQuadTangentAt1; )
+
+////////
+
+class ChopQuadAt0 : public QuadBenchBase {
+public:
+    ChopQuadAt0() : QuadBenchBase("chopquadat0") {}
+protected:
+    void onDraw(const int loops, SkCanvas* canvas) SK_OVERRIDE {
+        SkPoint dst[5];
+        for (int outer = 0; outer < loops; ++outer) {
+            SkChopQuadAt(fPts, dst, 0.5f);
+            SkChopQuadAt(fPts, dst, 0.5f);
+            SkChopQuadAt(fPts, dst, 0.5f);
+            SkChopQuadAt(fPts, dst, 0.5f);
+        }
+    }
+};
+DEF_BENCH( return new ChopQuadAt0; )
+
+class ChopQuadAt1 : public QuadBenchBase {
+public:
+    ChopQuadAt1() : QuadBenchBase("chopquadat1") {}
+protected:
+    void onDraw(const int loops, SkCanvas* canvas) SK_OVERRIDE {
+        SkPoint dst[5];
+        for (int outer = 0; outer < loops; ++outer) {
+            SkChopQuadAt2(fPts, dst, 0.5f);
+            SkChopQuadAt2(fPts, dst, 0.5f);
+            SkChopQuadAt2(fPts, dst, 0.5f);
+            SkChopQuadAt2(fPts, dst, 0.5f);
+        }
+    }
+};
+DEF_BENCH( return new ChopQuadAt1; )
+
diff --git a/src/core/SkGeometry.cpp b/src/core/SkGeometry.cpp
index 6d14e4b..aea1fe5 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 @@
     }
 }
 
-#include "Sk2x.h"
-
 SkPoint SkEvalQuadAt(const SkPoint src[3], SkScalar t) {
     SkASSERT(src);
     SkASSERT(t >= 0 && t <= SK_Scalar1);
@@ -147,6 +146,23 @@
     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 @@
     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);
diff --git a/src/core/SkGeometry.h b/src/core/SkGeometry.h
index ff863a2..5f7aa5e 100644
--- a/src/core/SkGeometry.h
+++ b/src/core/SkGeometry.h
@@ -17,11 +17,14 @@
 
 ///////////////////////////////////////////////////////////////////////////////
 
+SkPoint SkEvalQuadAt(const SkPoint src[3], SkScalar t);
+SkPoint SkEvalQuadTangentAt(const SkPoint src[3], SkScalar t);
+void SkChopQuadAt2(const SkPoint src[3], SkPoint dst[5], SkScalar t);
+
 /** Set pt to the point on the src quadratic specified by t. t must be
     0 <= t <= 1.0
 */
 void SkEvalQuadAt(const SkPoint src[3], SkScalar t, SkPoint* pt, SkVector* tangent = NULL);
-SkPoint SkEvalQuadAt(const SkPoint src[3], SkScalar t);
 
 /** Given a src quadratic bezier, chop it at the specified t value,
     where 0 < t < 1, and return the two new quadratics in dst:
diff --git a/tests/GeometryTest.cpp b/tests/GeometryTest.cpp
index 9be6000..aba8dd2 100644
--- a/tests/GeometryTest.cpp
+++ b/tests/GeometryTest.cpp
@@ -34,6 +34,16 @@
     }
 }
 
+static void check_pairs(skiatest::Reporter* reporter, int index, SkScalar t, const char name[],
+                        SkScalar x0, SkScalar y0, SkScalar x1, SkScalar y1) {
+    bool eq = SkScalarNearlyEqual(x0, x1) && SkScalarNearlyEqual(y0, y1);
+    if (!eq) {
+        SkDebugf("%s [%d %g] p0 [%10.8f %10.8f] p1 [%10.8f %10.8f]\n",
+                 name, index, t, x0, y0, x1, y1);
+        REPORTER_ASSERT(reporter, eq);
+    }
+}
+
 static void test_evalquadat(skiatest::Reporter* reporter) {
     SkRandom rand;
     for (int i = 0; i < 1000; ++i) {
@@ -41,17 +51,27 @@
         for (int j = 0; j < 3; ++j) {
             pts[j].set(rand.nextSScalar1() * 100, rand.nextSScalar1() * 100);
         }
-        SkScalar t = 0;
         const SkScalar dt = SK_Scalar1 / 128;
-        for (int j = 0; j < 128; ++j) {
+        SkScalar t = dt;
+        for (int j = 1; j < 128; ++j) {
             SkPoint r0;
             SkEvalQuadAt(pts, t, &r0);
             SkPoint r1 = SkEvalQuadAt(pts, t);
-            bool eq = SkScalarNearlyEqual(r0.fX, r1.fX) && SkScalarNearlyEqual(r0.fY, r1.fY);
-            if (!eq) {
-                SkDebugf("[%d %g] p0 [%10.8f %10.8f] p1 [%10.8f %10.8f]\n", i, t, r0.fX, r0.fY, r1.fX, r1.fY);
-                REPORTER_ASSERT(reporter, eq);
+            check_pairs(reporter, i, t, "quad-pos", r0.fX, r0.fY, r1.fX, r1.fY);
+
+            SkVector v0;
+            SkEvalQuadAt(pts, t, NULL, &v0);
+            SkVector v1 = SkEvalQuadTangentAt(pts, t);
+            check_pairs(reporter, i, t, "quad-tan", v0.fX, v0.fY, v1.fX, v1.fY);
+
+            SkPoint dst0[5], dst1[5];
+            SkChopQuadAt(pts,  dst0, t);
+            SkChopQuadAt2(pts, dst1, t);
+            for (int k = 0; k < 5; ++k) {
+                check_pairs(reporter, i, t, "chop-quad",
+                            dst0[k].fX, dst0[k].fY, dst1[k].fX, dst1[k].fY);
             }
+
             t += dt;
         }
     }