Thread through a flag to force SkPicture::playback() when recording subpictures.

This makes nanobench picture recording benchmarks somewhat useful again,
as opposed to all taking about 5us to run no matter the content.

ATTN Sheriff: this will probably trigger perf.skia.org alerts.

BUG=skia:

Review URL: https://codereview.chromium.org/1219873002
diff --git a/bench/RecordingBench.cpp b/bench/RecordingBench.cpp
index cd029dd..eada930 100644
--- a/bench/RecordingBench.cpp
+++ b/bench/RecordingBench.cpp
@@ -33,10 +33,11 @@
     const SkScalar w = fSrc->cullRect().width(),
                    h = fSrc->cullRect().height();
 
+    uint32_t flags = SkPictureRecorder::kComputeSaveLayerInfo_RecordFlag
+                   | SkPictureRecorder::kPlaybackDrawPicture_RecordFlag;
     for (int i = 0; i < loops; i++) {
         SkPictureRecorder recorder;
-        fSrc->playback(recorder.beginRecording(w, h, fUseBBH ? &factory : NULL,
-                                               SkPictureRecorder::kComputeSaveLayerInfo_RecordFlag));
+        fSrc->playback(recorder.beginRecording(w, h, fUseBBH ? &factory : NULL, flags));
         SkSafeUnref(recorder.endRecording());
     }
 }
diff --git a/include/core/SkPictureRecorder.h b/include/core/SkPictureRecorder.h
index a84b721..811d02a 100644
--- a/include/core/SkPictureRecorder.h
+++ b/include/core/SkPictureRecorder.h
@@ -33,7 +33,11 @@
     enum RecordFlags {
         // This flag indicates that, if some BHH is being computed, saveLayer
         // information should also be extracted at the same time.
-        kComputeSaveLayerInfo_RecordFlag = 0x01
+        kComputeSaveLayerInfo_RecordFlag = 0x01,
+
+        // If you call drawPicture() on the recording canvas, this flag forces
+        // that to use SkPicture::playback() immediately rather than (e.g.) reffing the picture.
+        kPlaybackDrawPicture_RecordFlag  = 0x02,
     };
 
     /** Returns the canvas that records the drawing commands.
diff --git a/src/core/SkPictureRecorder.cpp b/src/core/SkPictureRecorder.cpp
index c110e0a..877314c 100644
--- a/src/core/SkPictureRecorder.cpp
+++ b/src/core/SkPictureRecorder.cpp
@@ -38,7 +38,10 @@
     if (!fRecord) {
         fRecord.reset(SkNEW(SkRecord));
     }
-    fRecorder->reset(fRecord.get(), cullRect, &fMiniRecorder);
+    SkRecorder::DrawPictureMode dpm = (recordFlags & kPlaybackDrawPicture_RecordFlag)
+        ? SkRecorder::Playback_DrawPictureMode
+        : SkRecorder::Record_DrawPictureMode;
+    fRecorder->reset(fRecord.get(), cullRect, dpm, &fMiniRecorder);
     fActivelyRecording = true;
     return this->getRecordingCanvas();
 }
diff --git a/src/core/SkRecorder.cpp b/src/core/SkRecorder.cpp
index 6d7e5ee..a9d9ba9 100644
--- a/src/core/SkRecorder.cpp
+++ b/src/core/SkRecorder.cpp
@@ -6,6 +6,7 @@
  */
 
 #include "SkBigPicture.h"
+#include "SkCanvasPriv.h"
 #include "SkPatchUtils.h"
 #include "SkPicture.h"
 #include "SkPictureUtils.h"
@@ -35,18 +36,22 @@
 
 SkRecorder::SkRecorder(SkRecord* record, int width, int height, SkMiniRecorder* mr)
     : SkCanvas(SkIRect::MakeWH(width, height), SkCanvas::kConservativeRasterClip_InitFlag)
+    , fDrawPictureMode(Record_DrawPictureMode)
     , fApproxBytesUsedBySubPictures(0)
     , fRecord(record)
     , fMiniRecorder(mr) {}
 
 SkRecorder::SkRecorder(SkRecord* record, const SkRect& bounds, SkMiniRecorder* mr)
     : SkCanvas(bounds.roundOut(), SkCanvas::kConservativeRasterClip_InitFlag)
+    , fDrawPictureMode(Record_DrawPictureMode)
     , fApproxBytesUsedBySubPictures(0)
     , fRecord(record)
     , fMiniRecorder(mr) {}
 
-void SkRecorder::reset(SkRecord* record, const SkRect& bounds, SkMiniRecorder* mr) {
+void SkRecorder::reset(SkRecord* record, const SkRect& bounds,
+                       DrawPictureMode dpm, SkMiniRecorder* mr) {
     this->forgetRecord();
+    fDrawPictureMode = dpm;
     fRecord = record;
     this->resetForNextPicture(bounds.roundOut());
     fMiniRecorder = mr;
@@ -254,8 +259,14 @@
 }
 
 void SkRecorder::onDrawPicture(const SkPicture* pic, const SkMatrix* matrix, const SkPaint* paint) {
-    fApproxBytesUsedBySubPictures += SkPictureUtils::ApproximateBytesUsed(pic);
-    APPEND(DrawPicture, this->copy(paint), pic, matrix ? *matrix : SkMatrix::I());
+    if (fDrawPictureMode == Record_DrawPictureMode) {
+        fApproxBytesUsedBySubPictures += SkPictureUtils::ApproximateBytesUsed(pic);
+        APPEND(DrawPicture, this->copy(paint), pic, matrix ? *matrix : SkMatrix::I());
+    } else {
+        SkASSERT(fDrawPictureMode == Playback_DrawPictureMode);
+        SkAutoCanvasMatrixPaint acmp(this, matrix, paint, pic->cullRect());
+        pic->playback(this);
+    }
 }
 
 void SkRecorder::onDrawVertices(VertexMode vmode,
diff --git a/src/core/SkRecorder.h b/src/core/SkRecorder.h
index 3809ae2..688069b 100644
--- a/src/core/SkRecorder.h
+++ b/src/core/SkRecorder.h
@@ -41,7 +41,8 @@
     SkRecorder(SkRecord*, int width, int height, SkMiniRecorder* = nullptr);   // legacy version
     SkRecorder(SkRecord*, const SkRect& bounds, SkMiniRecorder* = nullptr);
 
-    void reset(SkRecord*, const SkRect& bounds, SkMiniRecorder* = nullptr);
+    enum DrawPictureMode { Record_DrawPictureMode, Playback_DrawPictureMode };
+    void reset(SkRecord*, const SkRect& bounds, DrawPictureMode, SkMiniRecorder* = nullptr);
 
     size_t approxBytesUsedBySubPictures() const { return fApproxBytesUsedBySubPictures; }
 
@@ -137,6 +138,7 @@
         return devBounds;
     }
 
+    DrawPictureMode fDrawPictureMode;
     size_t fApproxBytesUsedBySubPictures;
     SkRecord* fRecord;
     SkAutoTDelete<SkDrawableList> fDrawableList;