All batches do their own pipeline compare

src/gpu/GrAADistanceFieldPathRenderer.cpp

Index: src/gpu/GrAADistanceFieldPathRenderer.cpp
diff --git a/src/gpu/GrAADistanceFieldPathRenderer.cpp b/src/gpu/GrAADistanceFieldPathRenderer.cpp
index ca0f4e3d651a9eda7939c756eed95a55b0eb2390..830232b4c4048b5280df294a3689bf83a0971fe3 100755
--- a/src/gpu/GrAADistanceFieldPathRenderer.cpp
+++ b/src/gpu/GrAADistanceFieldPathRenderer.cpp
@@ -501,6 +501,10 @@ private:
     bool usesLocalCoords() const { return fBatch.fUsesLocalCoords; }
 
     bool onCombineIfPossible(GrBatch* t) override {
+        if (!this->pipeline()->isEqual(*t->pipeline())) {
+            return false;
+        }
+
         AADistanceFieldPathBatch* that = t->cast<AADistanceFieldPathBatch>();
 
         // TODO we could actually probably do a bunch of this work on the CPU, ie map viewMatrix,