SKPs-as-benches in nanobench

bench/SKPBench.cpp

Index: bench/SKPBench.cpp
diff --git a/bench/SKPBench.cpp b/bench/SKPBench.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..ff4ed113702ab87a24892db5c46ff3fdd573d8ff
--- /dev/null
+++ b/bench/SKPBench.cpp
@@ -0,0 +1,41 @@
+/*
+ * Copyright 2014 Google Inc.
+ *
+ * Use of this source code is governed by a BSD-style license that can be
+ * found in the LICENSE file.
+ */
+
+#include "SKPBench.h"
+
+SKPBench::SKPBench(const char* name, const SkPicture* pic, const SkIRect& clip, SkScalar scale)
+    : fPic(SkRef(pic))
+    , fClip(clip)
+    , fScale(scale) {
+    fName.printf("%s_%.2g", name, scale);
+}
+
+const char* SKPBench::onGetName() {
+    return fName.c_str();
+}
+
+bool SKPBench::isSuitableFor(Backend backend) {
+    return backend != kNonRendering_Backend;
+}
+
+SkIPoint SKPBench::onGetSize() {
+    return SkIPoint::Make(fClip.fRight - fClip.fLeft, fClip.fBottom - fClip.fTop);

[bsalomon, 2014/08/01 13:23:24]

fClip.width(), fClip.height()?

[mtklein, 2014/08/01 13:55:34]

Done.

+}
+
+void SKPBench::onDraw(const int loops, SkCanvas* canvas) {
+    canvas->save();
+        SkRegion deviceClip;

[bsalomon, 2014/08/01 13:23:24]

No regions, please! The GPU doesn't like them and they don't scale well. It
seems this is also the size of the canvas, so probably it isn't necessary at
all.

One thing that bench supports that'd be nice for nanobench is a mode where it
writes a file so a dev can verify that the right thing is being rendered (while
writing a new benchmark or working with a skp and trying to find the right
viewport, etc). I might add something similar to nanobench in a separate CL. I
was thinking about something like -w <dir> [<loopcnt_default_1>].

[mtklein, 2014/08/01 13:55:34]

Done.  Writing output SGTM.  Let's follow up.

+        deviceClip.setRect(fClip);
+        canvas->clipRegion(deviceClip);
+
+        canvas->scale(fScale, fScale);
+
+        for (int i = 0; i < loops; i++) {
+            fPic->draw(canvas);

[bsalomon, 2014/08/01 13:23:24]

I'm not sure whether we should have a canvas flush or not in here...

[mtklein, 2014/08/01 13:55:34]

Good question.  Looks like bench_pictures does, so added here.

+        }
+    canvas->restore();
+}