Add event tracer for backend init (#7575)

Summary:

Per request, backend would like to track model loading time

Reviewed By: tarun292, dbort

Differential Revision: D67987445

runtime/backend/backend_init_context.h

@@ -20,6 +20,7 @@ class BackendInitContext final {
  public:
   explicit BackendInitContext(
       MemoryAllocator* runtime_allocator,
+      EventTracer* event_tracer = nullptr,
       const char* method_name = nullptr)
       : runtime_allocator_(runtime_allocator), method_name_(method_name) {}
 
@@ -31,6 +32,15 @@ class BackendInitContext final {
     return runtime_allocator_;
   }
 
+  /**
+   * Returns a pointer (null if not installed) to an instance of EventTracer to
+   * do profiling/debugging logging inside the delegate backend. Users will need
+   * access to this pointer to use any of the event tracer APIs.  
+   */
+  EventTracer* event_tracer() {
+    return event_tracer_;
+  }
+
   /** Get the loaded method name from ExecuTorch runtime. Usually it's
    * "forward", however, if there are multiple methods in the .pte file, it can
    * be different. One example is that we may have prefill and decode methods in
@@ -44,6 +54,7 @@ class BackendInitContext final {
 
  private:
   MemoryAllocator* runtime_allocator_ = nullptr;
+  EventTracer* event_tracer_ = nullptr;
   const char* method_name_ = nullptr;
 };
 

runtime/executor/method.cpp

@@ -628,6 +628,7 @@ Error Method::init(executorch_flatbuffer::ExecutionPlan* s_plan) {
       const auto& delegate = *delegates->Get(i);
       BackendInitContext backend_init_context(
           method_allocator,
+          /*event_tracer=*/event_tracer_,
           /*method_name=*/serialization_plan_->name()->c_str());
       Error err = BackendDelegate::Init(
           delegate, program_, backend_init_context, &delegates_[i]);