Modify MarkTrtEngineOutputs API (#56858)

* Modify MarkTrtEngineOutputs API

paddle/fluid/inference/analysis/argument.h

@@ -241,6 +241,7 @@ struct Argument {
   DECL_ARGUMENT_FIELD(tensorrt_workspace_size, TensorRtWorkspaceSize, int64_t);
   DECL_ARGUMENT_FIELD(tensorrt_min_subgraph_size, TensorRtMinSubgraphSize, int);
   DECL_ARGUMENT_FIELD(trt_mark_output, TRTMarkOutput, bool);
+  DECL_ARGUMENT_FIELD(trt_mark_output_with_id, TRTMarkOutputWithId, bool);
   DECL_ARGUMENT_FIELD(trt_output_tensor_names,
                       TRTOutputTensorNames,
                       std::vector<std::string>);

paddle/fluid/inference/analysis/ir_pass_manager.cc

@@ -163,6 +163,8 @@ void IRPassManager::CreatePasses(Argument *argument,
       pass->Set("min_subgraph_size",
                 new int(argument->tensorrt_min_subgraph_size()));
       pass->Set("mark_output", new bool(argument->trt_mark_output()));
+      pass->Set("mark_output_with_id",
+                new bool(argument->trt_mark_output_with_id()));
       pass->Set(
           "output_tensor_names",
           new std::vector<std::string>(argument->trt_output_tensor_names()));

paddle/fluid/inference/analysis/ir_passes/tensorrt_subgraph_pass.cc

@@ -376,29 +376,30 @@ std::string TensorRtSubgraphPass::CreateTensorRTOp(
   std::vector<int> origin_outputs_dtype;
   std::map<std::string, int> map_origin_outputs_dtype;
 
-  // Whether to mark Outpus
+  // Mark TensorRT output nodes as trt outputs
   auto mark_output = Get<bool>("mark_output");
   auto output_tensor_name =
       Get<std::vector<std::string>>("output_tensor_names");
-  VLOG(1) << "mark Output: " << mark_output;
+  auto mark_output_with_id = Get<bool>("mark_output_with_id");
 
-  if (mark_output == 1) {
+  if (mark_output) {
     VLOG(1) << "begin to mark output ...";
     for (auto node : subgraph) {
       if (node->NodeType() == Node::Type::kOperation) {
-        if (node->Op()->Outputs().count("Xshape")) continue;
         for (auto *x : node->outputs) {
           if (std::count(parameters.begin(), parameters.end(), x->Name()) > 0)
             continue;
-          if (!output_tensor_name.empty() &&
-              std::count(output_tensor_name.begin(),
+          std::string name_with_id = x->Name() + std::to_string(x->id());
+          if (((!mark_output_with_id && std::count(output_tensor_name.begin(),
                                                    output_tensor_name.end(),
-                         x->Name())) {
-            VLOG(1) << "output " << x->Name() << " has been marked";
-            std::string output_name_withid =
-                x->Name() + std::to_string(x->id());
+                                                   x->Name()) > 0) ||
+               (mark_output_with_id && std::count(output_tensor_name.begin(),
+                                                  output_tensor_name.end(),
+                                                  name_with_id) > 0)) &&
+              !x->outputs.empty()) {
+            VLOG(3) << "output " << x->Name() << " has been marked";
             output_names.insert(x->Name());
-            output_names_with_id.insert(output_name_withid);
+            output_names_with_id.insert(name_with_id);
             origin_name_output_rank[x->Name()] = x->Var()->GetShape().size();
             trt_outputs.insert(x);
             map_origin_outputs_dtype[x->Name()] =

paddle/fluid/inference/api/analysis_config.cc

@@ -461,6 +461,7 @@ AnalysisConfig::AnalysisConfig(const AnalysisConfig &other) {
   CP_MEMBER(tensorrt_min_subgraph_size_);
   CP_MEMBER(tensorrt_precision_mode_);
   CP_MEMBER(trt_mark_output_);
+  CP_MEMBER(trt_mark_output_with_id_);
   CP_MEMBER(trt_output_tensor_names_);
   CP_MEMBER(trt_disabled_ops_);
   CP_MEMBER(trt_use_dla_);
@@ -762,8 +763,10 @@ void AnalysisConfig::EnableTensorRtEngine(int64_t workspace_size,
 }
 
 void AnalysisConfig::MarkTrtEngineOutputs(
-    const std::vector<std::string> &output_tensor_names) {
+    const std::vector<std::string> &output_tensor_names,
+    const bool mark_output_with_id) {
   trt_mark_output_ = true;
+  trt_mark_output_with_id_ = mark_output_with_id;
   trt_output_tensor_names_ = output_tensor_names;
 }
 

paddle/fluid/inference/api/analysis_predictor.cc

@@ -1392,6 +1392,7 @@ void AnalysisPredictor::PrepareArgument() {
     argument_->SetTensorRtMaxBatchSize(config_.tensorrt_max_batchsize_);
     argument_->SetTensorRtMinSubgraphSize(config_.tensorrt_min_subgraph_size_);
     argument_->SetTRTMarkOutput(config_.trt_mark_output_);
+    argument_->SetTRTMarkOutputWithId(config_.trt_mark_output_with_id_);
     argument_->SetTRTOutputTensorNames(config_.trt_output_tensor_names_);
     argument_->SetTensorRtDisabledOPs(config_.trt_disabled_ops_);
     argument_->SetTensorRtUseDLA(config_.trt_use_dla_);

paddle/fluid/inference/api/paddle_analysis_config.h

@@ -695,7 +695,8 @@ struct PD_INFER_DECL AnalysisConfig {
   /// \param output_tensor_names The name of the Tensor that needs to be marked
   ///
   void MarkTrtEngineOutputs(
-      const std::vector<std::string>& output_tensor_names = {});
+      const std::vector<std::string>& output_tensor_names = {},
+      const bool trt_mark_output_with_id = false);
   ///
   /// \brief Turn on the TensorRT memory optimization.
   ///
@@ -1237,6 +1238,7 @@ struct PD_INFER_DECL AnalysisConfig {
   bool trt_use_varseqlen_{false};
   bool trt_with_interleaved_{false};
   bool trt_mark_output_{false};
+  bool trt_mark_output_with_id_{false};
   std::vector<std::string> trt_output_tensor_names_{};
   std::string tensorrt_transformer_posid_{""};
   std::string tensorrt_transformer_maskid_{""};

paddle/fluid/pybind/inference_api.cc

@@ -896,7 +896,8 @@ void BindAnalysisConfig(py::module *m) {
            &AnalysisConfig::tensorrt_dynamic_shape_enabled)
       .def("mark_trt_engine_outputs",
            &AnalysisConfig::MarkTrtEngineOutputs,
-           py::arg("output_tensor_names") = std::vector<std::string>({}))
+           py::arg("output_tensor_names") = std::vector<std::string>({}),
+           py::arg("mark_output_with_id") = false)
       .def("enable_tensorrt_varseqlen", &AnalysisConfig::EnableVarseqlen)
       .def("tensorrt_varseqlen_enabled",
            &AnalysisConfig::tensorrt_varseqlen_enabled)

test/cpp/inference/api/trt_mark_trt_engine_outputs_test.cc

@@ -24,9 +24,11 @@ TEST(TensorRT, mark_trt_engine_outputs) {
   config.EnableUseGpu(100, 0);
   config.EnableTensorRtEngine(
       1 << 30, 1, 5, AnalysisConfig::Precision::kFloat32, false, false);
-  // The name of the tensor that needs to be marked, the default is empty (all
-  // marks)
-  std::vector<std::string> markOutput = {"fc_0.tmp_0", "fc_0.tmp_1"};
+  // The name of the tensor that needs to be marked
+  std::vector<std::string> markOutput = {"pool2d_0.tmp_0",
+                                         "elementwise_add_0.tmp_0",
+                                         "conv2d_5.tmp_0",
+                                         "batch_norm_6.tmp_2"};
   config.MarkTrtEngineOutputs(markOutput);
 
   std::vector<std::vector<PaddleTensor>> inputs_all;