Reland - Refactor signaling in RWP

content/renderer/raster_worker_pool.h

 // Copyright 2015 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #ifndef CONTENT_RENDERER_RASTER_WORKER_POOL_H_
 #define CONTENT_RENDERER_RASTER_WORKER_POOL_H_
 
 #include "base/callback.h"
 #include "base/containers/hash_tables.h"
 #include "base/macros.h"
@@ skipping 30 matching lines @@
 
   // Overridden from cc::TaskGraphRunner:
   cc::NamespaceToken GetNamespaceToken() override;
   void ScheduleTasks(cc::NamespaceToken token, cc::TaskGraph* graph) override;
   void WaitForTasksToFinishRunning(cc::NamespaceToken token) override;
   void CollectCompletedTasks(cc::NamespaceToken token,
                              cc::Task::Vector* completed_tasks) override;
 
   // Runs a task from one of the provided categories. Categories listed first
   // have higher priority.
-  void Run(const std::vector<cc::TaskCategory>& categories);
+  void Run(const std::vector<cc::TaskCategory>& categories,
+           base::ConditionVariable* work_ready_cv);
 
   void FlushForTesting();
 
   // Spawn |num_threads| number of threads and start running work on the
   // worker threads.
   void Start(int num_threads,
              const base::SimpleThread::Options& thread_options);
 
   // Finish running all the posted tasks (and nested task posted by those tasks)
   // of all the associated task runners.
@@ skipping 33 matching lines @@
 
    protected:
     ~ClosureTask() override;
 
    private:
     base::Closure closure_;
 
     DISALLOW_COPY_AND_ASSIGN(ClosureTask);
   };
 
+  // Helper function which signals the CV for the next task to run.
+  void SignalTaskReadyConditionVariables();

[reveman, 2016/02/08 19:10:21]

nit: SignalHasReadyToRunTasks(). I think that's more consistent with the naming
of other existing variables and functions and ConditionVariables in the name
doesn't seem necessary and more of an implementation detail

[ericrk, 2016/02/10 18:44:49]

Done.

+
+  // Determines if we should run a new task for the given category. This factors
+  // in whether a task is available and whether the count of running tasks is
+  // low enough to start a new one.
+  bool ShouldRunTaskForCategory(cc::TaskCategory category);
+
   void ScheduleTasksWithLockAcquired(cc::NamespaceToken token,
                                      cc::TaskGraph* graph);
   void CollectCompletedTasksWithLockAcquired(cc::NamespaceToken token,
                                              cc::Task::Vector* completed_tasks);
 
   // The actual threads where work is done.
   std::vector<scoped_ptr<base::SimpleThread>> threads_;
 
   // Lock to exclusively access all the following members that are used to
   // implement the TaskRunner and TaskGraphRunner interfaces.
   base::Lock lock_;
   // Stores the tasks to be run, sorted by priority.
   cc::TaskGraphWorkQueue work_queue_;
   // Namespace used to schedule tasks in the task graph runner.
   cc::NamespaceToken namespace_token_;
   // List of tasks currently queued up for execution.
   cc::Task::Vector tasks_;
   // Graph object used for scheduling tasks.
   cc::TaskGraph graph_;
   // Cached vector to avoid allocation when getting the list of complete
   // tasks.
   cc::Task::Vector completed_tasks_;
-  // Condition variable that is waited on by Run() until new tasks are ready to
-  // run or shutdown starts.
-  base::ConditionVariable has_ready_to_run_tasks_cv_;
+  // Condition variables for foreground and background tasks.
+  base::ConditionVariable foreground_has_ready_to_run_tasks_cv_;
+  base::ConditionVariable background_has_ready_to_run_tasks_cv_;
+  // The target number of threads that should be active at any given time.
+  int target_running_task_count_;

[reveman, 2016/02/08 19:10:21]

Can we remove target_running_task_count_ now?

[ericrk, 2016/02/10 18:44:49]

I guess we don't strictly need this. It gets us the ability to limit the
foreground + background tasks to a fixed number - if a background task is
already running when new work comes in, we won't spin up as many foreground
tasks until that background task finishes, keeping us in the intended limit.

I think it's OK to temporarily exceed the limits in this case, especially as it
simplifies the logic a bit - removing this for now.

WDYT?

[reveman, 2016/02/10 20:49:31]

I think it's best to limit how background work can effect how efficient we run
foreground as much as possible. Not having this limit improves that.

   // Condition variable that is waited on by origin threads until a namespace
   // has finished running all associated tasks.
   base::ConditionVariable has_namespaces_with_finished_running_tasks_cv_;
   // Set during shutdown. Tells Run() to return when no more tasks are pending.
   bool shutdown_;
 };
 
 }  // namespace content
 
 #endif  // CONTENT_RENDERER_RASTER_WORKER_POOL_H_