SWDEV-422207 - Capture AQL Packets for graph Kernel nodes during graph Inst. And enqueue AQL packet during launch

Change-Id: I1e5f7f9e2a70bd500d190193cb6ba0867f5a63e7

rocclr/platform/command.cpp

@@ -317,7 +317,6 @@ Command::Command(HostQueue& queue, cl_command_type type, const EventWaitList& ev
       type_(type),
       data_(nullptr),
       waitingEvent_(waitingEvent),
-      buffering_(false),
       eventWaitList_(eventWaitList),
       commandWaitBits_(commandWaitBits) {
   // Retain the commands from the event wait list.
@@ -354,7 +353,7 @@ void Command::enqueue() {
 
     // Notify all commands about the waiter. Barrier will be sent in order to obtain
     // HSA signal for a wait on the current queue
-    for (const auto &event: eventWaitList()) {
+    for (const auto& event : eventWaitList()) {
       event->notifyCmdQueue(!kCpuWait);
     }
 

rocclr/platform/command.hpp

@@ -251,13 +251,12 @@ union CopyMetadata {
  */
 class Command : public Event {
  private:
-  HostQueue* queue_;              //!< The command queue this command is enqueue into
-  Command* next_;                 //!< Next GPU command in the queue list
-  Command* batch_head_ = nullptr; //!< The head of the batch commands
-  cl_command_type     type_;      //!< This command's OpenCL type.
+  HostQueue* queue_;               //!< The command queue this command is enqueue into
+  Command* next_;                  //!< Next GPU command in the queue list
+  Command* batch_head_ = nullptr;  //!< The head of the batch commands
+  cl_command_type type_;           //!< This command's OpenCL type.
   void* data_;
-  const Event* waitingEvent_;     //!< Waiting event associated with the marker
-  bool buffering_;                //!< Flag to enable/disable AQL buffering
+  const Event* waitingEvent_;  //!< Waiting event associated with the marker
 
  protected:
   bool cpu_wait_ = false;         //!< If true, then the command was issued for CPU/GPU sync
@@ -281,7 +280,6 @@ class Command : public Event {
         type_(type),
         data_(nullptr),
         waitingEvent_(nullptr),
-        buffering_(false),
         eventWaitList_(nullWaitList),
         commandWaitBits_(0) {}
 
@@ -296,12 +294,6 @@ class Command : public Event {
   }
 
  public:
-  //! Returns AQL buffer state
-  bool getBufferingState() const { return buffering_; }
-
-  //! Sets AQL buffer state
-  void setBufferingState(bool state) { buffering_ = state; }
-
   //! Return the queue this command is enqueued into.
   HostQueue* queue() const { return queue_; }
 
@@ -1083,6 +1075,10 @@ class NDRangeKernelCommand : public Command {
   uint32_t firstDevice_;    //!< Device index of the first device in the gridc
   uint32_t numWorkgroups_;  //!< Total number of workgroups in the current launch
 
+  bool capturing_ = false;           //!< Flag to enable/disable graph gpu packet capture
+  uint8_t* gpuPacket_ = nullptr;     //!< GPU packet to capture, when graph capturing is enabled
+  address kernArgOffset_ = nullptr;  //!< KernelArg buffer to used when graph capturing is enabled
+
  public:
   enum {
     CooperativeGroups = 0x01,
@@ -1090,6 +1086,22 @@ class NDRangeKernelCommand : public Command {
     AnyOrderLaunch = 0x04,
   };
 
+  //! Returns AQL buffer state
+  bool getCapturingState() const { return capturing_; }
+
+  //! Sets AQL capture state, aql packet to capture and where to copy kernArgs
+  void setCapturingState(bool state, uint8_t* packet, address kernArgOffset) {
+    capturing_ = state;
+    gpuPacket_ = packet;
+    kernArgOffset_ = kernArgOffset;
+  }
+
+  //! returns the graph executable object command belongs to.
+  const uint8_t* getAqlPacket() const { return gpuPacket_; }
+
+  //! returns the graph executable object command belongs to.
+  const address getKernArgOffset() const { return kernArgOffset_; }
+
   //! Construct an ExecuteKernel command
   NDRangeKernelCommand(HostQueue& queue, const EventWaitList& eventWaitList, Kernel& kernel,
                        const NDRangeContainer& sizes, uint32_t sharedMemBytes = 0,