Merge Group Markers into Chromium Traces.

cc/resources/scoped_gpu_raster.cc

Index: cc/resources/scoped_gpu_raster.cc
diff --git a/cc/resources/scoped_gpu_raster.cc b/cc/resources/scoped_gpu_raster.cc
index 2f5c5c9c3426a42eca7532cb8919611884a983d1..d9ed8055a69b5a4725dc754a0095037cd7df373f 100644
--- a/cc/resources/scoped_gpu_raster.cc
+++ b/cc/resources/scoped_gpu_raster.cc
@@ -28,6 +28,7 @@ void ScopedGpuRaster::BeginGpuRaster() {
   // Using push/pop functions directly incurs cost to evaluate function
   // arguments even when tracing is disabled.
   gl->PushGroupMarkerEXT(0, "GpuRasterization");

[vmiura, 2015/05/12 21:47:41]

Markers and traces each add some overhead, and I think traces will get more
difficult to read with duplicate entries.

I'd like to suggest that we deprecate use of PushGroupMarkerEXT for Chromium
internal code, and keep only the TraceBeginCHROMIUM calls.

GPU logs can also use the TraceBeginCHROMIUM values instead of
PushGroupMarkerEXT values.

Later on we will be able to extend TraceCHROMIUM calls to pass additional JSON
data for viewing in the trace viewer.

+  gl->TraceBeginCHROMIUM("ScopedGpuRaster", "GpuRasterization");
 
   class GrContext* gr_context = context_provider_->GrContext();
   gr_context->resetContext();
@@ -45,6 +46,7 @@ void ScopedGpuRaster::EndGpuRaster() {
   // TODO(alokp): Use a trace macro to push/pop markers.
   // Using push/pop functions directly incurs cost to evaluate function
   // arguments even when tracing is disabled.
+  gl->TraceEndCHROMIUM();
   gl->PopGroupMarkerEXT();
 }