Create GrDraw and start fast pathing src over rects

src/gpu/batches/GrBatch.h

Index: src/gpu/batches/GrBatch.h
diff --git a/src/gpu/batches/GrBatch.h b/src/gpu/batches/GrBatch.h
index c5fc80c5d50dfbd622dfb69e3eb9adfbcd3c31f9..90b2fefb7b10983180d2003fddd674457c8090a1 100644
--- a/src/gpu/batches/GrBatch.h
+++ b/src/gpu/batches/GrBatch.h
@@ -14,6 +14,10 @@
 #include "SkRect.h"
 #include "SkString.h"
 
+#include "draws/GrDraw.h"
+
+#include "../private/SkTemplates.h"
+
 class GrCaps;
 class GrBatchFlushState;
 
@@ -109,6 +113,12 @@ public:
     /** Used for spewing information about batches when debugging. */
     virtual SkString dumpInfo() const = 0;
 
+    template <typename T, typename... Args> void initSnapStorage(Args&&... args) {
+        SkASSERT(!fSnap && sizeof(T) < fSnapStorage.size());
+        fSnap = new (SkTCast<T*>(fSnapStorage.get())) T(skstd::forward<Args>(args)...);
+    }
+    const GrDrawSnap* getDrawSnap() const { return fSnap; }
+
 protected:
     // NOTE, compute some bounds, even if extremely conservative.  Do *NOT* setLargest on the bounds
     // rect because we outset it for dst copy textures
@@ -145,11 +155,14 @@ private:
 
     SkDEBUGCODE(bool                    fUsed;)
     const uint32_t                      fClassID;
+    GrDrawSnap*                         fSnap;
+    SkAlignedSStorage<128>              fSnapStorage; // 128 bytes should be more than enough
 #if GR_BATCH_SPEW
     static uint32_t GenBatchID() { return GenID(&gCurrBatchUniqueID); }
     const uint32_t                      fUniqueID;
     static int32_t                      gCurrBatchUniqueID;
 #endif
+
     static int32_t                      gCurrBatchClassID;
     typedef GrNonAtomicRef INHERITED;
 };