SWDEV-382397 - Move VirtualGPU destruction back to the thread exit

OS can terminate unfinished queue thread from default stream at any
time. Potentially leaving the queue lock in a bad state and causing a
deadlock if runtime destroys VirtualGPU later from the host thread.

Change-Id: I247f102ee84e6b4dba947504933395071945c85d


[ROCm/clr commit: 28daf98f1f]

projects/clr/rocclr/platform/commandqueue.cpp

@@ -66,6 +66,7 @@ bool HostQueue::terminate() {
       marker->awaitCompletion();
       marker->release();
     }
+    thread_.Release();
     thread_.acceptingCommands_ = false;
   } else {
     if (Os::isThreadAlive(thread_)) {

projects/clr/rocclr/platform/commandqueue.hpp

@@ -164,19 +164,13 @@ class HostQueue : public CommandQueue {
           acceptingCommands_(false),
           virtualDevice_(nullptr) {}
 
-    virtual ~Thread() {
-      if (virtualDevice_ != nullptr) {
-        delete virtualDevice_;
-        virtualDevice_ = nullptr;
-      }
-    }
-
     //! The command queue thread entry point.
     void run(void* data) {
       HostQueue* queue = static_cast<HostQueue*>(data);
       virtualDevice_ = queue->device().createVirtualDevice(queue);
       if (virtualDevice_ != nullptr) {
         queue->loop(virtualDevice_);
+        Release();
       } else {
         acceptingCommands_ = false;
         queue->flush();
@@ -190,6 +184,8 @@ class HostQueue : public CommandQueue {
       }
     }
 
+    void Release() const { delete virtualDevice_; }
+
     //! Get virtual device for the current thread
     device::VirtualDevice* vdev() const { return virtualDevice_; }