P4 to Git Change 1620765 by skudchad@skudchad_test2_win_opencl on 2018/10/17 16:58:04

SWDEV-145570 - [HIP] Track last used event and use last enqueued command in a stream rather than creating a new event

	ReviewBoardURL = http://ocltc.amd.com/reviews/r/15996/diff/

Affected files ...

... //depot/stg/opencl/drivers/opencl/api/hip/hip_context.cpp#15 edit
... //depot/stg/opencl/drivers/opencl/api/hip/hip_event.cpp#7 edit
... //depot/stg/opencl/drivers/opencl/api/hip/hip_stream.cpp#14 edit
... //depot/stg/opencl/drivers/opencl/runtime/platform/command.cpp#90 edit
... //depot/stg/opencl/drivers/opencl/runtime/platform/commandqueue.cpp#28 edit
... //depot/stg/opencl/drivers/opencl/runtime/platform/commandqueue.hpp#20 edit


[ROCm/clr commit: 19caea543c]

projects/clr/rocclr/runtime/platform/command.cpp

@@ -213,6 +213,9 @@ void Command::enqueue() {
     Agent::postEventCreate(as_cl(static_cast<Event*>(this)), type_);
   }
 
+  if (IS_HIP) {
+    queue_->setLastQueuedCommand(this);
+  }
   queue_->append(*this);
   queue_->flush();
   if (queue_->device().settings().waitCommand_ && (type_ != 0)) {

projects/clr/rocclr/runtime/platform/commandqueue.cpp

@@ -21,7 +21,8 @@ HostQueue::HostQueue(Context& context, Device& device, cl_command_queue_properti
                      uint queueRTCUs, Priority priority)
     : CommandQueue(context, device, properties,
                    device.info().queueProperties_ | CL_QUEUE_COMMAND_INTERCEPT_ENABLE_AMD,
-                   queueRTCUs, priority) {
+                   queueRTCUs, priority),
+      lastEnqueueCommand_(nullptr) {
   if (thread_.state() >= Thread::INITIALIZED) {
     ScopedLock sl(queueLock_);
     thread_.start(this);
@@ -163,12 +164,32 @@ void HostQueue::append(Command& command) {
   queue_.enqueue(&command);
 }
 
-
 bool HostQueue::isEmpty() {
   // Get a snapshot of queue size
   return queue_.empty();
 }
 
+void HostQueue::setLastQueuedCommand(Command* lastCommand) {
+  // Set last submitted command
+  ScopedLock sl(queueLock_);
+  if (lastEnqueueCommand_ != nullptr) {
+    lastEnqueueCommand_->release();
+  }
+  lastEnqueueCommand_ = lastCommand;
+  if (lastCommand != nullptr) {
+    lastEnqueueCommand_->retain();
+  }
+}
+
+Command* HostQueue::getLastQueuedCommand(bool retain) {
+  // Get last submitted command
+  ScopedLock sl(queueLock_);
+  if (retain) {
+    lastEnqueueCommand_->retain();
+  }
+  return lastEnqueueCommand_;
+}
+
 DeviceQueue::~DeviceQueue() {
   delete virtualDevice_;
   ScopedLock lock(context().lock());

projects/clr/rocclr/runtime/platform/commandqueue.hpp

@@ -163,6 +163,8 @@ class HostQueue : public CommandQueue {
  private:
   ConcurrentLinkedQueue<Command*> queue_;  //!< The queue.
 
+  Command* lastEnqueueCommand_;  //!< The last submitted command
+
   //! Await commands and execute them as they become ready.
   void loop(device::VirtualDevice* virtualDevice);
 
@@ -204,6 +206,12 @@ class HostQueue : public CommandQueue {
 
   //! Return the current queue as the HostQueue
   virtual HostQueue* asHostQueue() { return this; }
+
+  //! Get last enqueued command
+  Command* getLastQueuedCommand(bool retain);
+
+  //! Set last enqueued command
+  void setLastQueuedCommand(Command* lastCommand);
 };