DM: Push GPU-parent child tasks to the front of the queue.

Like yesterday's change to run CPU-parent child tasks serially in thread, this
reduces peak memory usage by improving the temporaly locality of the bitmaps we
create.

E.g. Let's say we start with tasks A B C and D
    Queue: [ A B C D ]
Running A creates A' and A", which depend on a bitmap created by A.
    Queue: [ B C D A' A" * ]
That bitmap now needs sit around in RAM while B C and D run pointlessly and can
only be destroyed at *.  If instead we do this and push dependent child tasks
to the front of the queue, the queue and bitmap lifetime looks like this:
    Queue: [ A' A" * B C D ]

This is much, much worse in practice because the queue is often several thousand
tasks long.  100s of megs of bitmaps can pile up for 10s of seconds pointlessly.

To make this work we add addNext() to SkThreadPool and its cousin DMTaskRunner.
I also took the opportunity to swap head and tail in the threadpool
implementation so it matches the comments and intuition better: we always pop
the head, add() puts it at the tail, addNext() at the head.


Before
  Debug:   49s, 1403352k peak
  Release: 16s, 2064008k peak

After
  Debug:   49s, 1234788k peak
  Release: 15s, 1903424k peak

BUG=skia:2478
R=bsalomon@google.com, borenet@google.com, mtklein@google.com

Author: mtklein@chromium.org

Review URL: https://codereview.chromium.org/263803003

git-svn-id: http://skia.googlecode.com/svn/trunk@14506 2bbb7eff-a529-9590-31e7-b0007b416f81

4 files changed, 9 insertions, 5 deletions

diff --git a/dm/DMTask.cpp b/dm/DMTask.cpp
index 83df84933f..7725780d4c 100644
--- a/dm/DMTask.cpp
+++ b/dm/DMTask.cpp
@@ -37,8 +37,8 @@ void Task::finish() {
     fReporter->finish(this->name(), SkTime::GetMSecs() - fStart);
 }
 
-void Task::reallySpawnChild(CpuTask* task) {
-    fTaskRunner->add(task);
+void Task::spawnChildNext(CpuTask* task) {
+    fTaskRunner->addNext(task);
 }
 
 CpuTask::CpuTask(Reporter* reporter, TaskRunner* taskRunner) : Task(reporter, taskRunner) {}
@@ -55,6 +55,8 @@ void CpuTask::run() {
 
 void CpuTask::spawnChild(CpuTask* task) {
     // Run children serially on this (CPU) thread.  This tends to save RAM and is usually no slower.
+    // Calling spawnChildNext() is nearly equivalent, but it'd pointlessly contend on the
+    // threadpool; spawnChildNext() is most useful when you want to change threadpools.
     task->run();
 }
 
@@ -71,7 +73,8 @@ void GpuTask::run(GrContextFactory& factory) {
 
 void GpuTask::spawnChild(CpuTask* task) {
     // Really spawn a new task so it runs on the CPU threadpool instead of the GPU one we're on now.
-    this->reallySpawnChild(task);
+    // It goes on the front of the queue to minimize the time we must hold reference bitmaps in RAM.
+    this->spawnChildNext(task);
 }
 
 }  // namespace DM
diff --git a/dm/DMTask.h b/dm/DMTask.h
index 02d8a22011..96cf810b29 100644
--- a/dm/DMTask.h
+++ b/dm/DMTask.h
@@ -36,7 +36,7 @@ protected:
     void fail(const char* msg = NULL);
     void finish();
 
-    void reallySpawnChild(CpuTask* task);  // For now we don't allow GPU child tasks.
+    void spawnChildNext(CpuTask* task);  // For now we don't allow GPU child tasks.
 
 private:
     Reporter* fReporter;      // Unowned.
diff --git a/dm/DMTaskRunner.cpp b/dm/DMTaskRunner.cpp
index e0bd977288..8a0bc838e1 100644
--- a/dm/DMTaskRunner.cpp
+++ b/dm/DMTaskRunner.cpp
@@ -6,7 +6,7 @@ namespace DM {
 TaskRunner::TaskRunner(int cpuThreads, int gpuThreads) : fCpu(cpuThreads), fGpu(gpuThreads) {}
 
 void TaskRunner::add(CpuTask* task) { fCpu.add(task); }
-
+void TaskRunner::addNext(CpuTask* task) { fCpu.addNext(task); }
 void TaskRunner::add(GpuTask* task) { fGpu.add(task); }
 
 void TaskRunner::wait() {
diff --git a/dm/DMTaskRunner.h b/dm/DMTaskRunner.h
index b20113f506..dd1440ed9a 100644
--- a/dm/DMTaskRunner.h
+++ b/dm/DMTaskRunner.h
@@ -18,6 +18,7 @@ public:
     explicit TaskRunner(int cpuThreads, int gpuThreads);
 
     void add(CpuTask* task);
+    void addNext(CpuTask* task);
     void add(GpuTask* task);
     void wait();