Timestamp Packet ownership

- Tag allocator: reference count tracking
- Obtain tag by command queue and pass to Event if exist during enqueue
- Handle Timestamp Packet lifetime on Event and CmdQueue destruction

Change-Id: I9a5969830ea7a9d729e6f70519d8c28ff70fcf06
Signed-off-by: Dunajski, Bartosz <bartosz.dunajski@intel.com>

runtime/command_queue/command_queue.cpp

@@ -36,6 +36,7 @@
 #include "runtime/helpers/mipmap.h"
 #include "runtime/helpers/options.h"
 #include "runtime/helpers/ptr_math.h"
+#include "runtime/helpers/timestamp_packet.h"
 #include "runtime/mem_obj/buffer.h"
 #include "runtime/mem_obj/image.h"
 #include "runtime/helpers/surface_formats.h"
@@ -43,6 +44,7 @@
 #include "runtime/helpers/string.h"
 #include "CL/cl_ext.h"
 #include "runtime/utilities/api_intercept.h"
+#include "runtime/utilities/tag_allocator.h"
 #include "runtime/helpers/convert_color.h"
 #include "runtime/helpers/queue_helpers.h"
 #include <map>
@@ -102,6 +104,10 @@ CommandQueue::~CommandQueue() {
         auto memoryManager = device->getMemoryManager();
         DEBUG_BREAK_IF(nullptr == memoryManager);
 
+        if (timestampPacketNode) {
+            memoryManager->getTimestampPacketAllocator()->returnTag(timestampPacketNode);
+        }
+
         if (commandStream && commandStream->getGraphicsAllocation()) {
             memoryManager->storeAllocation(std::unique_ptr<GraphicsAllocation>(commandStream->getGraphicsAllocation()), REUSABLE_ALLOCATION);
             commandStream->replaceGraphicsAllocation(nullptr);
@@ -604,4 +610,13 @@ void CommandQueue::dispatchAuxTranslation(MultiDispatchInfo &multiDispatchInfo,
 
     builder.buildDispatchInfos(multiDispatchInfo, dispatchParams);
 }
+
+void CommandQueue::obtainNewTimestampPacketNode() {
+    auto allocator = device->getMemoryManager()->getTimestampPacketAllocator();
+
+    if (timestampPacketNode) {
+        allocator->returnTag(timestampPacketNode);
+    }
+    timestampPacketNode = allocator->getTag();
+}
 } // namespace OCLRT

runtime/command_queue/command_queue.h

@@ -40,6 +40,7 @@ class Image;
 class IndirectHeap;
 class Kernel;
 class MemObj;
+class TimestampPacket;
 struct CompletionStamp;
 
 enum class QueuePriority {
@@ -418,6 +419,8 @@ class CommandQueue : public BaseObject<_cl_command_queue> {
     MOCKABLE_VIRTUAL void dispatchAuxTranslation(MultiDispatchInfo &multiDispatchInfo, BuffersForAuxTranslation &buffersForAuxTranslation,
                                                  AuxTranslationDirection auxTranslationDirection);
 
+    void obtainNewTimestampPacketNode();
+
     Context *context;
     Device *device;
 
@@ -438,6 +441,8 @@ class CommandQueue : public BaseObject<_cl_command_queue> {
     bool mapDcFlushRequired = false;
     bool isSpecialCommandQueue = false;
 
+    TagNode<TimestampPacket> *timestampPacketNode = nullptr;
+
   private:
     void providePerformanceHint(TransferProperties &transferProperties);
 };

runtime/command_queue/enqueue_common.h

@@ -240,13 +240,22 @@ void CommandQueueHw<GfxFamily>::enqueueHandler(Surface **surfacesForResidency,
             }
         }
 
-        if ((this->isProfilingEnabled() && (eventBuilder.getEvent() != nullptr))) {
-            // Get allocation for timestamps
-            hwTimeStamps = eventBuilder.getEvent()->getHwTimeStamp();
-            if (this->isPerfCountersEnabled()) {
-                hwPerfCounter = eventBuilder.getEvent()->getHwPerfCounter();
-                //PERF COUNTER: copy current configuration from queue to event
-                eventBuilder.getEvent()->copyPerfCounters(this->getPerfCountersConfigData());
+        if (DebugManager.flags.EnableTimestampPacket.get()) {
+            obtainNewTimestampPacketNode();
+        }
+
+        if (eventBuilder.getEvent()) {
+            if (timestampPacketNode) {
+                eventBuilder.getEvent()->setTimestampPacketNode(timestampPacketNode);
+            }
+            if (this->isProfilingEnabled()) {
+                // Get allocation for timestamps
+                hwTimeStamps = eventBuilder.getEvent()->getHwTimeStamp();
+                if (this->isPerfCountersEnabled()) {
+                    hwPerfCounter = eventBuilder.getEvent()->getHwPerfCounter();
+                    // PERF COUNTER: copy current configuration from queue to event
+                    eventBuilder.getEvent()->copyPerfCounters(this->getPerfCountersConfigData());
+                }
             }
         }
 

runtime/command_queue/gpgpu_walker.inl

@@ -610,7 +610,7 @@ void GpgpuWalkerHelper<GfxFamily>::dispatchWalker(
         // Implement enabling special WA DisableLSQCROPERFforOCL if needed
         GpgpuWalkerHelper<GfxFamily>::applyWADisableLSQCROPERFforOCL(commandStream, kernel, true);
 
-        bool setupTimestampPacket = (DebugManager.flags.EnableTimestampPacket.get()) && (currentDispatchIndex == numDispatches - 1);
+        bool setupTimestampPacket = (timestampPacket && (currentDispatchIndex == numDispatches - 1));
         if (setupTimestampPacket) {
             GpgpuWalkerHelper<GfxFamily>::setupTimestampPacket(commandStream, nullptr, timestampPacket, TimestampPacket::WriteOperationType::Start);
         }