Insert CHECK to look for abnormal flushes.

gpu/command_buffer/service/sync_point_manager.cc

Index: gpu/command_buffer/service/sync_point_manager.cc
diff --git a/gpu/command_buffer/service/sync_point_manager.cc b/gpu/command_buffer/service/sync_point_manager.cc
index 9669bb677ab07688bf3f9cf0738500ac403b0207..4dccf384001b1e8af97e07e27021d674c4345ec3 100644
--- a/gpu/command_buffer/service/sync_point_manager.cc
+++ b/gpu/command_buffer/service/sync_point_manager.cc
@@ -255,6 +255,10 @@ void SyncPointClientState::ReleaseFenceSyncLocked(
   fence_sync_lock_.AssertAcquired();
   DCHECK_GT(release, fence_sync_release_);
 
+  // TODO(dyen): https://www.crbug.com/606112 - Hack in a DCHECK which should be
+  // fine under normal circumstances. Under normal usage, this would only
+  // trigger if flushes are arriving out of order.
+  DCHECK(release == fence_sync_release_ + 1 || release == UINT64_MAX);

[piman, 2016/05/11 01:24:11]

nit: make it a CHECK if you want to catch release bots too. It's ok if it's
temporary.

[David Yen, 2016/05/11 17:27:27]

Done.

   fence_sync_release_ = release;
   while (!release_callback_queue_.empty() &&
          release_callback_queue_.top().release_count <= release) {