fix emb_eltwise_ln gpu_id bug (#33701)

paddle/fluid/inference/api/analysis_config.cc

@@ -421,7 +421,6 @@ void AnalysisConfig::Update() {
       pass_builder()->AppendPass(pass);
     }
   }
-  LOG(INFO) << "use_dlnne_:" << use_dlnne_ << std::endl;
   if (use_dlnne_) {
     pass_builder()->ClearPasses();
     for (const auto &pass : kDlnneSubgraphPasses) {

paddle/fluid/inference/api/analysis_predictor.cc

@@ -152,8 +152,8 @@ bool AnalysisPredictor::Init(
                                              : platform::ProfilerState::kCPU;
     platform::EnableProfiler(tracking_device);
   } else {
-    LOG(INFO) << "Profiler is deactivated, and no profiling report will be "
-                 "generated.";
+    VLOG(2) << "Profiler is deactivated, and no profiling report will be "
+               "generated.";
   }
 
   // no matter with or without MKLDNN

paddle/fluid/inference/api/paddle_analysis_config.h

@@ -315,7 +315,7 @@ struct PD_INFER_DECL AnalysisConfig {
   /// workspace.
   /// \param max_batch_size The maximum batch size of this prediction task,
   /// better set as small as possible for less performance loss.
-  /// \param min_subgrpah_size The minimum TensorRT subgraph size needed, if a
+  /// \param min_subgraph_size The minimum TensorRT subgraph size needed, if a
   /// subgraph is smaller than this, it will not be transferred to TensorRT
   /// engine.
   /// \param precision The precision used in TensorRT.

paddle/fluid/inference/tensorrt/plugin/emb_eltwise_layernorm_plugin.cu

@@ -134,7 +134,7 @@ int EmbEltwiseLayernormPluginDynamicImpl<T>::enqueue(
   int batch = id_dims.d[0];
   int seq_len = id_dims.d[1];
   int input_num = embs_.size();
-
+  cudaGetDevice(&device_id_);
   auto in_ptr_gpu_d =
       in_ptr_tensor_.mutable_data<int64_t>(platform::CUDAPlace(device_id_));
   auto emb_ptr_gpu_d =

paddle/fluid/inference/tests/api/trt_dynamic_shape_ernie_test.cc

@@ -29,11 +29,6 @@ void run(const AnalysisConfig& config, std::vector<float>* out_data) {
   int run_batch = 1;
   const int run_seq_len = 128;
 
-  std::vector<int64_t> tmp_input;
-  std::vector<float> tmp_four_input;
-  tmp_input.reserve(run_batch * run_seq_len);
-  tmp_four_input.reserve(run_batch * run_seq_len);
-
   int64_t i0[run_seq_len] = {
       1,    3558, 4,   75,  491, 89, 340, 313, 93,   4,   255,   10, 75,    321,
       4095, 1902, 4,   134, 49,  75, 311, 14,  44,   178, 543,   15, 12043, 2,