Fix bad casts in tools.

Also virtualize queue profiling enable.

Change-Id: I761b41269be3df7eb64a5914ee9951ed6b51bb04


[ROCm/ROCR-Runtime commit: 6455a69b03]

projects/rocr-runtime/runtime/hsa-runtime/core/inc/amd_gpu_agent.h

@@ -297,7 +297,7 @@ class GpuAgent : public GpuAgentInt {
     return regions_;
   }
 
-  // @brief OVerride from core::Agent.
+  // @brief Override from core::Agent.
   const core::Isa* isa() const override { return isa_; }
 
   // @brief Override from amd::GpuAgentInt.

projects/rocr-runtime/runtime/hsa-runtime/core/inc/intercept_queue.h

@@ -116,6 +116,7 @@ class QueueWrapper : public Queue {
   void ExecutePM4(uint32_t* cmd_data, size_t cmd_size_b) override {
     wrapped->ExecutePM4(cmd_data, cmd_size_b);
   }
+  void SetProfiling(bool enabled) override { wrapped->SetProfiling(enabled); }
 
  protected:
   void do_set_public_handle(hsa_queue_t* handle) {

projects/rocr-runtime/runtime/hsa-runtime/core/inc/queue.h

@@ -294,6 +294,11 @@ class Queue : public Checked<0xFA3906A679F9DB49>, private LocalQueue {
   // @brief Submits a block of PM4 and waits until it has been executed.
   virtual void ExecutePM4(uint32_t* cmd_data, size_t cmd_size_b) = 0;
 
+  virtual void SetProfiling(bool enabled) {
+    AMD_HSA_BITS_SET(amd_queue_.queue_properties, AMD_QUEUE_PROPERTIES_ENABLE_PROFILING,
+                     (enabled != 0));
+  }
+
   /// @ brief Reports async queue errors to stderr if no other error handler was registered.
   static void DefaultErrorHandler(hsa_status_t status, hsa_queue_t* source, void* data);
 

projects/rocr-runtime/runtime/hsa-runtime/core/runtime/hsa_ext_amd.cpp

@@ -266,11 +266,9 @@ hsa_status_t hsa_amd_profiling_set_profiler_enabled(hsa_queue_t* queue, int enab
   IS_OPEN();
 
   core::Queue* cmd_queue = core::Queue::Convert(queue);
-
   IS_VALID(cmd_queue);
 
-  AMD_HSA_BITS_SET(cmd_queue->amd_queue_.queue_properties,
-                   AMD_QUEUE_PROPERTIES_ENABLE_PROFILING, (enable != 0));
+  cmd_queue->SetProfiling(enable);
 
   return HSA_STATUS_SUCCESS;
   CATCH;

projects/rocr-runtime/runtime/hsa-runtime/core/runtime/intercept_queue.cpp

@@ -235,6 +235,7 @@ void InterceptQueue::StoreRelaxed(hsa_signal_value_t value) {
 
   next_packet_ = i;
   Cursor.queue = nullptr;
+  atomic::Store(&amd_queue_.read_dispatch_id, next_packet_, std::memory_order_release);
 }
 
 }  // namespace core