Report texture upload time in renderingStats.

content/common/gpu/client/gpu_channel_host.cc

Index: content/common/gpu/client/gpu_channel_host.cc
diff --git a/content/common/gpu/client/gpu_channel_host.cc b/content/common/gpu/client/gpu_channel_host.cc
index b225a3ad11c110e2d8b3183d837edb291b425db7..ea3e87f9a62fffe52298faf0f8fbb2c2a004095e 100644
--- a/content/common/gpu/client/gpu_channel_host.cc
+++ b/content/common/gpu/client/gpu_channel_host.cc
@@ -214,6 +214,21 @@ void GpuChannelHost::DestroyCommandBuffer(
 #endif
 }
 
+bool GpuChannelHost::CollectRenderingStats(
+    int surface_id, content::GpuRenderingStats* stats) {

[piman, 2012/08/30 22:44:11]

At a high level, why restricting here to "view" command buffers (which have a
non-null surface_id)?
It seems that this would be useful in general for any command buffer (including
"offscreen" ones), by passing their routing id instead.

+  TRACE_EVENT0("gpu", "GpuChannelHost::CollectRenderingStats");
+
+  AutoLock lock(context_lock_);
+  // An error occurred. Need to get the host again to reinitialize it.
+  if (!channel_.get())

[piman, 2012/08/30 22:44:11]

Why this? I think it's racy if called on a non-main thread. In case of lost
channel, the Send below will fail anyway.
I would very much like to avoid holding a lock while sending a sync message
(that can be very slow since it's effectively a command buffer finish).

I think we can simply get rid of the autolock and this test and we'll have the
same behavior, correct?

+    return false;
+
+  if (!Send(new GpuChannelMsg_CollectRenderingStats(surface_id, stats)))
+    return false;
+
+  return true;
+}
+
 void GpuChannelHost::AddRoute(
     int route_id, base::WeakPtr<IPC::Listener> listener) {
   DCHECK(MessageLoopProxy::current());