Add Plugin.PpapiSyncIPCTime metric.

ppapi/proxy/plugin_dispatcher.cc

Index: ppapi/proxy/plugin_dispatcher.cc
diff --git a/ppapi/proxy/plugin_dispatcher.cc b/ppapi/proxy/plugin_dispatcher.cc
index 0f15645e6b7b77b9b46c5ba30e5afcdffcc170d7..91ac68268ccd01c026398cd759d62dfe2f846ff0 100644
--- a/ppapi/proxy/plugin_dispatcher.cc
+++ b/ppapi/proxy/plugin_dispatcher.cc
@@ -9,6 +9,7 @@
 #include "base/compiler_specific.h"
 #include "base/logging.h"
 #include "base/message_loop/message_loop.h"
+#include "base/metrics/histogram_macros.h"
 #include "base/trace_event/trace_event.h"
 #include "ipc/ipc_message.h"
 #include "ipc/ipc_sync_channel.h"
@@ -206,6 +207,7 @@ bool PluginDispatcher::Send(IPC::Message* msg) {
   if (msg->is_sync()) {
     // Synchronous messages might be re-entrant, so we need to drop the lock.
     ProxyAutoUnlock unlock;
+    SCOPED_UMA_HISTOGRAM_TIMER("Plugin.PpapiSyncIPCTime");

[bbudge, 2015/03/25 22:28:14]

There are about a dozen usages of this in Chromium right now, and they all seem
like low frequency things. Could this slow down the Pepper proxy?

[gab, 2015/03/25 22:45:57]

Mark will know better than I, but FWIU, this is a new wrapper for
UMA_HISTOGRAM_TIMES (to avoid every user of that in a small scope to have to do
the local variable/base::Time::Now() dance explicitly).

It probably shouldn't be put in a tight loop -- and even then I'm not sure, it
probably is very minimal overhead -- but other than that I doubt it makes any
difference on the path to a synchronous IPC (which I assume is at least an order
of magnitude slower).

[Mark P, 2015/03/25 22:56:56]

Yes, it should be as fast as doing the local variable time dance explicitly.
All the UMA histogram macros are designed to be extremely low overhead.  They
were written and optimize to use as few machine instructions as possible.  I
forget the exact number but I think it was O(10) machine instructions; unless
you're putting this in a really tight loop, I doubt it can slow anything down
meaningfully.

[bbudge, 2015/03/25 23:24:19]

It's hard to tell looking at the macro definition, which defines a class and
instantiates it.
https://code.google.com/p/chromium/codesearch#chromium/src/base/metrics/histo...

Also, I don't have a good feel for the cost of base::TimeTicks on different
platforms. But it does seem that the cost should be very small relative to a
sync IPC.

[gab, 2015/03/26 00:07:58]

But it's a static instantiation so it only happens once.

     return SendMessage(msg);
   }
   return SendMessage(msg);