SWDEV-480209 - Make internal callbacks non-blocking

Change-Id: Ic918d08f341abfd9a7c167d09f9c723cdc43157f

rocclr/device/rocm/rocvirtual.cpp

@@ -239,7 +239,7 @@ bool HsaAmdSignalHandler(hsa_signal_value_t value, void* arg) {
   gpu->updateCommandsState(ts->command().GetBatchHead());
 
   // Reset API callback signal. It will release AQL queue and start commands processing
-  if (callback_signal.handle != 0) {
+  if (callback_signal.handle != 0 && ts->GetBlocking()) {
     hsa_signal_subtract_relaxed(callback_signal, 1);
   }
 
@@ -515,10 +515,13 @@ hsa_signal_t VirtualGPU::HwQueueTracker::ActiveSignal(
         // If API callback is enabled, then use a blocking signal for AQL queue.
         // HSA signal will be acquired in SW and released after HSA signal callback
         if (ts->command().Callback() != nullptr) {
-          ts->SetCallbackSignal(prof_signal->signal_);
+          bool blocking = ts->command().Callback()->blocking_;
+          ts->SetCallbackSignal(prof_signal->signal_, blocking);
           // Blocks AQL queue from further processing
-          hsa_signal_add_relaxed(prof_signal->signal_, 1);
-          init_value += 1;
+          if (blocking) {
+            hsa_signal_add_relaxed(prof_signal->signal_, 1);
+            init_value += 1;
+          }
         }
         gpu_.QueuedAsyncHandlers()++;
         hsa_status_t result = hsa_amd_signal_async_handler(prof_signal->signal_,

rocclr/device/rocm/rocvirtual.hpp

@@ -112,8 +112,9 @@ class Timestamp : public amd::ReferenceCountedObject {
   std::vector<ProfilingSignal*> signals_; //!< The list of all signals, associated with the TS
   hsa_signal_t callback_signal_;  //!< Signal associated with a callback for possible later update
   amd::Monitor  lock_;            //!< Serialize timestamp update
-  bool        accum_ena_ = false; //!< If TRUE then the accumulation of execution times has started
-  bool        hasHwProfiling_ = false; //!< If TRUE then HwProfiling is enabled for the command
+  bool accum_ena_ = false;        //!< If TRUE then the accumulation of execution times has started
+  bool hasHwProfiling_ = false;   //!< If TRUE then HwProfiling is enabled for the command
+  bool blocking_ = true;          //!< If TRUE callback is blocking
 
   Timestamp(const Timestamp&) = delete;
   Timestamp& operator=(const Timestamp&) = delete;
@@ -177,12 +178,15 @@ class Timestamp : public amd::ReferenceCountedObject {
   VirtualGPU* gpu() const { return gpu_; }
 
   //! Updates the callback signal
-  void SetCallbackSignal(hsa_signal_t callback_signal) {
+  void SetCallbackSignal(hsa_signal_t callback_signal, bool blocking = true) {
     callback_signal_ = callback_signal;
+    blocking_ = blocking;
   }
-
   //! Returns the callback signal
   hsa_signal_t GetCallbackSignal() const { return callback_signal_; }
+
+  //! Return if callback is blocking/non-blocking
+  bool GetBlocking() { return blocking_; }
 };
 
 class VirtualGPU : public device::VirtualDevice {