Simplify GrDrawBatch uploads and token uage.

src/gpu/batches/GrVertexBatch.h

Index: src/gpu/batches/GrVertexBatch.h
diff --git a/src/gpu/batches/GrVertexBatch.h b/src/gpu/batches/GrVertexBatch.h
index 2af4dd1caddeb96abbf9d0d172471d1d954df72d..14e380933777ed775abf4b3b96533c4bfc4b5cb2 100644
--- a/src/gpu/batches/GrVertexBatch.h
+++ b/src/gpu/batches/GrVertexBatch.h
@@ -9,8 +9,8 @@
 #define GrVertexBatch_DEFINED
 
 #include "GrDrawBatch.h"
+#include "GrGeometryProcessor.h"
 #include "GrMesh.h"
-#include "GrPrimitiveProcessor.h"
 #include "GrPendingProgramElement.h"
 
 #include "SkTLList.h"
@@ -39,7 +39,7 @@ protected:
                    int instancesToDraw);
 
         /** Call after init() to issue draws to the batch target.*/
-        void recordDraw(Target* target);
+        void recordDraw(Target* target, const GrGeometryProcessor*);
     private:
         GrMesh  fMesh;
     };
@@ -54,7 +54,7 @@ protected:
         /** Finds the cached quad index buffer and reserves vertex space. Returns nullptr on failure
             and on sucess a pointer to the vertex data that the caller should populate before
             calling issueDraws(). */
-        void* init(Target* batchTarget, size_t vertexStride, int quadsToDraw);
+        void* init(Target*, size_t vertexStride, int quadsToDraw);
 
         using InstancedHelper::recordDraw;
     private:
@@ -67,18 +67,21 @@ private:
 
     virtual void onPrepareDraws(Target*) const = 0;
 
-    // A set of contiguous draws with no inline uploads between them that all use the same
-    // primitive processor. All the draws in a DrawArray share a primitive processor and use the
-    // the batch's GrPipeline.
-    struct DrawArray {
-        SkSTArray<1, GrMesh, true>                          fDraws;
-        GrPendingProgramElement<const GrPrimitiveProcessor> fPrimitiveProcessor;
+    // All draws in all the vertex batches have implicit tokens based on the order they are
+    // enqueued globally across all batches. This is the offset of the first entry in fQueuedDraws.
+    GrBatchDrawToken fBaseDrawToken;
+
+    // A set of contiguous draws that share a draw token and primitive processor. The draws all use
+    // the batch's pipeline. The reason for coallescing meshes that share a primitive processor
+    // is that it allows the Gpu object to setup the shared state once and then issue draws for
+    // each mesh.
+    struct QueuedDraw {
+        int fMeshCnt = 0;
+        GrPendingProgramElement<const GrGeometryProcessor> fGeometryProcessor;
     };
 
-    // Array of DrawArray. There may be inline uploads between each DrawArray and each DrawArray
-    // may use a different primitive processor.
-    typedef SkTLList<DrawArray, 4> DrawArrayList;
-    DrawArrayList fDrawArrays;
+    SkSTArray<4, GrMesh>           fMeshes;
+    SkSTArray<4, QueuedDraw, true> fQueuedDraws;
 
     typedef GrDrawBatch INHERITED;
 };