[Paddle-TRT] Use TRT inspector to show the information inside an engine to verbose log (#38200)

* Inspect the information inside a TRT engine.

* Follow up the google code style.

* Fix code error.

paddle/fluid/inference/tensorrt/engine.cc

@@ -54,6 +54,7 @@ void TensorRTEngine::Execute(int batch_size, std::vector<void *> *buffers,
   } else {
 #if IS_TRT_VERSION_GE(6000)
     infer_context->enqueueV2(buffers->data(), stream, nullptr);
+    GetEngineInfo();
 #endif
   }
   SetRuntimeBatch(batch_size);
@@ -237,6 +238,11 @@ void TensorRTEngine::FreezeNetwork() {
 #endif
   }
 
+#if IS_TRT_VERSION_GE(8200)
+  infer_builder_config_->setProfilingVerbosity(
+      nvinfer1::ProfilingVerbosity::kDETAILED);
+#endif
+
 #if IS_TRT_VERSION_LT(8000)
   infer_engine_.reset(infer_builder_->buildEngineWithConfig(
       *network(), *infer_builder_config_));
@@ -253,6 +259,8 @@ void TensorRTEngine::FreezeNetwork() {
       infer_engine_, platform::errors::Fatal(
                          "Build TensorRT cuda engine failed! Please recheck "
                          "you configurations related to paddle-TensorRT."));
+
+  GetEngineInfo();
 }
 
 nvinfer1::ITensor *TensorRTEngine::DeclareInput(const std::string &name,

paddle/fluid/inference/tensorrt/engine.h

@@ -321,6 +321,8 @@ class TensorRTEngine {
             "on the same GPU architecture;\n2. The Paddle Inference version of "
             "generating serialization file and doing inference are "
             "consistent."));
+
+    GetEngineInfo();
   }
 
   void SetRuntimeBatch(size_t batch_size);
@@ -538,6 +540,18 @@ class TensorRTEngine {
     }
   }
 
+  void GetEngineInfo() {
+#if IS_TRT_VERSION_GE(8200)
+    std::unique_ptr<nvinfer1::IEngineInspector> infer_inspector(
+        infer_engine_->createEngineInspector());
+    infer_inspector->setExecutionContext(context());
+    VLOG(3) << infer_inspector->getEngineInformation(
+        nvinfer1::LayerInformationFormat::kJSON);
+#else
+    VLOG(3) << "Inspector needs TensorRT version 8.2 and after.";
+#endif
+  }
+
  private:
   // Each ICudaEngine object is bound to a specific GPU when it is instantiated,
   // ensure that the thread is associated with the correct device by calling