SWDEV-313306 - Fix Co-operative groups dtests

Add a state indicator to retain ExternalSignals when needed.
Co-operative group launch uses external signals to indicate a dependency
to the next command.

Change-Id: I6d0daa006e2377c3bbf4aeca0fd5b63c7ac8fbbb

rocclr/device/rocm/rocvirtual.cpp

@@ -1008,6 +1008,7 @@ bool VirtualGPU::releaseGpuMemoryFence(bool skip_cpu_wait) {
     // Dispatch barrier packet into the queue
     dispatchBarrierPacket(kBarrierPacketHeader);
     hasPendingDispatch_ = false;
+    retainExternalSignals_ = false;
   }
 
   // Check if runtime could skip CPU wait
@@ -1277,7 +1278,9 @@ void VirtualGPU::profilingBegin(amd::Command& command, bool drmProfiling) {
   }
 
   if (AMD_DIRECT_DISPATCH) {
-    Barriers().ClearExternalSignals();
+    if (!retainExternalSignals_) {
+      Barriers().ClearExternalSignals();
+    }
     for (auto it = command.eventWaitList().begin(); it < command.eventWaitList().end(); ++it) {
       void* hw_event = ((*it)->NotifyEvent() != nullptr) ?
         (*it)->NotifyEvent()->HwEvent() : (*it)->HwEvent();
@@ -1304,10 +1307,12 @@ void VirtualGPU::profilingEnd(amd::Command& command) {
     if (!timestamp_->HwProfiling()) {
       timestamp_->end();
     }
-    assert(Barriers().IsExternalSignalListEmpty());
     command.setData(timestamp_);
     timestamp_ = nullptr;
   }
+  if (AMD_DIRECT_DISPATCH) {
+    assert(retainExternalSignals_ || Barriers().IsExternalSignalListEmpty());
+  }
 }
 
 // ================================================================================================
@@ -2991,6 +2996,7 @@ void VirtualGPU::submitKernel(amd::NDRangeKernelCommand& vcmd) {
     // Add a dependency into the current queue on the coop queue
     Barriers().AddExternalSignal(queue->Barriers().GetLastSignal());
     hasPendingDispatch_ = true;
+    retainExternalSignals_ = true;
 
     queue->profilingEnd(vcmd);
   } else {

rocclr/device/rocm/rocvirtual.hpp

@@ -458,12 +458,13 @@ class VirtualGPU : public device::VirtualDevice {
   //! Queue state flags
   union {
     struct {
-      uint32_t hasPendingDispatch_ : 1; //!< A kernel dispatch is outstanding
-      uint32_t profiling_          : 1; //!< Profiling is enabled
-      uint32_t cooperative_        : 1; //!< Cooperative launch is enabled
-      uint32_t addSystemScope_     : 1; //!< Insert a system scope to the next aql
-      uint32_t tracking_created_   : 1; //!< Enabled if tracking object was properly initialized
-      uint32_t profilerAttached_   : 1; //!< Indicates if profiler is attached
+      uint32_t hasPendingDispatch_    : 1; //!< A kernel dispatch is outstanding
+      uint32_t profiling_             : 1; //!< Profiling is enabled
+      uint32_t cooperative_           : 1; //!< Cooperative launch is enabled
+      uint32_t addSystemScope_        : 1; //!< Insert a system scope to the next aql
+      uint32_t tracking_created_      : 1; //!< Enabled if tracking object was properly initialized
+      uint32_t profilerAttached_      : 1; //!< Indicates if profiler is attached
+      uint32_t retainExternalSignals_ : 1; //!< Indicate to retain external signal array
     };
     uint32_t  state_;
   };