Add TensorRT inspector into Paddle-TRT (#38362)

AUTHORS.md

@@ -83,3 +83,4 @@
 | jeng1220 | Bai-Cheng(Ryan) Jeng (NVIDIA) |
 | mingxu1067 | Ming Huang (NVIDIA) |
 | zlsh80826 | Reese Wang (NVIDIA) |
+| leo0519 | Leo Chen (NVIDIA) |

paddle/fluid/inference/analysis/argument.h

@@ -219,6 +219,7 @@ struct Argument {
                       bool);
   DECL_ARGUMENT_FIELD(tensorrt_allow_build_at_runtime,
                       TensorRtAllowBuildAtRuntime, bool);
+  DECL_ARGUMENT_FIELD(tensorrt_use_inspector, TensorRtUseInspector, bool);
 
   DECL_ARGUMENT_FIELD(use_dlnne, UseDlnne, bool);
   DECL_ARGUMENT_FIELD(dlnne_min_subgraph_size, DlnneMinSubgraphSize, int);

paddle/fluid/inference/analysis/ir_pass_manager.cc

@@ -156,6 +156,7 @@ void IRPassManager::CreatePasses(Argument *argument,
       pass->Set("gpu_device_id", new int(argument->gpu_device_id()));
       pass->Set("use_static_engine", new bool(use_static_engine));
       pass->Set("model_from_memory", new bool(argument->model_from_memory()));
+      pass->Set("use_inspector", new bool(argument->tensorrt_use_inspector()));
 
       // tuned trt dynamic_shape
       pass->Set("trt_shape_range_info_path",

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

@@ -265,6 +265,7 @@ void TensorRtSubgraphPass::CreateTensorRTOp(
   op_desc->SetAttr("parameters", params);
   op_desc->SetAttr("allow_build_at_runtime", allow_build_at_runtime);
   op_desc->SetAttr("shape_range_info_path", shape_range_info_path);
+  op_desc->SetAttr("use_inspector", Get<bool>("use_inspector"));
 
   // we record all inputs' shapes in attr to check if they are consistent
   // with the real inputs' shapes retrieved from scope when trt runs.
@@ -375,6 +376,7 @@ void TensorRtSubgraphPass::CreateTensorRTOp(
   trt_engine->SetWithInterleaved(Get<bool>("with_interleaved"));
   trt_engine->SetUseDLA(Get<bool>("trt_use_dla"));
   trt_engine->SetDLACore(Get<int>("trt_dla_core"));
+  trt_engine->SetUseInspector(Get<bool>("use_inspector"));
 
   trt_engine->SetWithErnie(
       graph->Has(framework::ir::kEmbEltwiseLayernormPass) &&

paddle/fluid/inference/api/analysis_config.cc

@@ -194,6 +194,7 @@ AnalysisConfig::AnalysisConfig(const AnalysisConfig &other) {
   CP_MEMBER(trt_allow_build_at_runtime_);
   CP_MEMBER(collect_shape_range_info_);
   CP_MEMBER(shape_range_info_path_);
+  CP_MEMBER(trt_use_inspector_);
   // Dlnne related
   CP_MEMBER(use_dlnne_);
   CP_MEMBER(dlnne_min_subgraph_size_);
@@ -427,6 +428,8 @@ void AnalysisConfig::EnableTensorRtDLA(int dla_core) {
   trt_dla_core_ = dla_core;
 }
 
+void AnalysisConfig::EnableTensorRtInspector() { trt_use_inspector_ = true; }
+
 void AnalysisConfig::Exp_DisableTensorRtOPs(
     const std::vector<std::string> &ops) {
   trt_disabled_ops_.insert(trt_disabled_ops_.end(), ops.begin(), ops.end());

paddle/fluid/inference/api/analysis_predictor.cc

@@ -615,6 +615,7 @@ void AnalysisPredictor::PrepareArgument() {
         config_.tuned_tensorrt_dynamic_shape());
     argument_.SetTensorRtAllowBuildAtRuntime(
         config_.trt_allow_build_at_runtime());
+    argument_.SetTensorRtUseInspector(config_.trt_use_inspector_);
   }
 
   if (config_.dlnne_enabled()) {

paddle/fluid/inference/api/paddle_analysis_config.h

@@ -521,6 +521,9 @@ struct PD_INFER_DECL AnalysisConfig {
   ///
   bool tensorrt_dla_enabled() { return trt_use_dla_; }
 
+  void EnableTensorRtInspector();
+  bool tensorrt_inspector_enabled() { return trt_use_inspector_; }
+
   void EnableDlnne(int min_subgraph_size = 3);
   bool dlnne_enabled() const { return use_dlnne_; }
 
@@ -807,6 +810,7 @@ struct PD_INFER_DECL AnalysisConfig {
   bool trt_allow_build_at_runtime_{false};
   // tune to get dynamic_shape info.
   bool trt_tuned_dynamic_shape_{false};
+  bool trt_use_inspector_{false};
 
   // In CollectShapeInfo mode, we will collect the shape information of
   // all intermediate tensors in the compute graph and calculate the

paddle/fluid/inference/tensorrt/engine.cc

@@ -57,7 +57,6 @@ 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);
@@ -244,8 +243,10 @@ void TensorRTEngine::FreezeNetwork() {
 #endif
   }
 #if IS_TRT_VERSION_GE(8200)
-  infer_builder_config_->setProfilingVerbosity(
-      nvinfer1::ProfilingVerbosity::kDETAILED);
+  if (use_inspector_) {
+    infer_builder_config_->setProfilingVerbosity(
+        nvinfer1::ProfilingVerbosity::kDETAILED);
+  }
 #endif
 
 #if IS_TRT_VERSION_LT(8000)
@@ -411,6 +412,21 @@ void TensorRTEngine::freshDeviceId() {
   platform::SetDeviceId(device_id_);
 }
 
+void TensorRTEngine::GetEngineInfo() {
+#if IS_TRT_VERSION_GE(8200)
+  LOG(INFO) << "====== engine info ======";
+  std::unique_ptr<nvinfer1::IEngineInspector> infer_inspector(
+      infer_engine_->createEngineInspector());
+  auto infer_context = context();
+  infer_inspector->setExecutionContext(infer_context);
+  LOG(INFO) << infer_inspector->getEngineInformation(
+      nvinfer1::LayerInformationFormat::kONELINE);
+  LOG(INFO) << "====== engine info end ======";
+#else
+  LOG(INFO) << "Inspector needs TensorRT version 8.2 and after.";
+#endif
+}
+
 }  // namespace tensorrt
 }  // namespace inference
 }  // namespace paddle

paddle/fluid/inference/tensorrt/engine.h

@@ -580,17 +580,10 @@ class TensorRTEngine {
   }
 
   void SetProfileNum(int num) { max_profile_num_ = num; }
-  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
-  }
+
+  void GetEngineInfo();
+
+  void SetUseInspector(bool use_inspector) { use_inspector_ = use_inspector; }
 
  private:
   // Each ICudaEngine object is bound to a specific GPU when it is instantiated,
@@ -664,6 +657,7 @@ class TensorRTEngine {
   std::vector<std::unique_ptr<plugin::DynamicPluginTensorRT>> owned_pluginv2_;
 #endif
   std::mutex mutex_;
+  bool use_inspector_;
 };  // class TensorRTEngine
 
 // Add a layer__ into engine__ with args ARGS.

paddle/fluid/operators/tensorrt/tensorrt_engine_op.h

@@ -140,6 +140,7 @@ class TensorRTEngineOp : public framework::OperatorBase {
   bool enable_int8_;
   bool enable_fp16_;
   bool use_calib_mode_;
+  bool use_inspector_;
   std::string calibration_data_;
   std::string engine_key_;
   std::string calibration_engine_key_;
@@ -175,6 +176,7 @@ class TensorRTEngineOp : public framework::OperatorBase {
     shape_range_info_path_ = Attr<std::string>("shape_range_info_path");
     allow_build_at_runtime_ = Attr<bool>("allow_build_at_runtime");
     use_static_engine_ = Attr<bool>("use_static_engine");
+    use_inspector_ = HasAttr("use_inspector") && Attr<bool>("use_inspector");
     if (use_static_engine_) {
       model_opt_cache_dir_ = Attr<std::string>("model_opt_cache_dir");
     }
@@ -285,6 +287,9 @@ class TensorRTEngineOp : public framework::OperatorBase {
       return;
     }
     auto *trt_engine = GetEngine(scope, dev_place);
+    if (use_inspector_) {
+      trt_engine->GetEngineInfo();
+    }
     if (trt_engine->with_dynamic_shape()) {
       // get runtime input shapes.
       std::map<std::string, std::vector<int32_t>> runtime_input_shape;
@@ -331,7 +336,6 @@ class TensorRTEngineOp : public framework::OperatorBase {
             anc = &scope;
           }
           PrepareTRTEngine(*anc, trt_engine);
-
           // update shape_range_info_pbtxt
           if (!shape_range_info_path_.empty()) {
             inference::UpdateShapeRangeInfo(

paddle/fluid/pybind/inference_api.cc

@@ -615,6 +615,10 @@ void BindAnalysisConfig(py::module *m) {
       .def("enable_tensorrt_dla", &AnalysisConfig::EnableTensorRtDLA,
            py::arg("dla_core") = 0)
       .def("tensorrt_dla_enabled", &AnalysisConfig::tensorrt_dla_enabled)
+      .def("enable_tensorrt_inspector",
+           &AnalysisConfig::EnableTensorRtInspector)
+      .def("tensorrt_inspector_enabled",
+           &AnalysisConfig::tensorrt_inspector_enabled)
       .def("tensorrt_engine_enabled", &AnalysisConfig::tensorrt_engine_enabled)
       .def("enable_dlnne", &AnalysisConfig::EnableDlnne,
            py::arg("min_subgraph_size") = 3)

python/paddle/fluid/tests/unittests/ir/inference/CMakeLists.txt

@@ -75,6 +75,7 @@ set_tests_properties(test_trt_activation_pass PROPERTIES TIMEOUT 120)
 set_tests_properties(test_trt_conv_pass PROPERTIES TIMEOUT 120)
 #set_tests_properties(test_trt_multiclass_nms_op PROPERTIES TIMEOUT 200)
 set_tests_properties(test_trt_dynamic_shape PROPERTIES TIMEOUT 120)
+set_tests_properties(test_trt_inspector PROPERTIES TIMEOUT 60)
 if(WITH_NV_JETSON)
   set_tests_properties(test_trt_pool_op PROPERTIES ENVIRONMENT FLAGS_fraction_of_gpu_memory_to_use=0.1 TIMEOUT 450)
   set_tests_properties(test_trt_pool3d_op PROPERTIES ENVIRONMENT FLAGS_fraction_of_gpu_memory_to_use=0.1 TIMEOUT 450)

python/paddle/fluid/tests/unittests/ir/inference/inference_pass_test.py

@@ -122,6 +122,11 @@ class InferencePassTest(unittest.TestCase):
                     self.trt_parameters.precision,
                     self.trt_parameters.use_static,
                     self.trt_parameters.use_calib_mode)
+                if self.trt_parameters.use_inspector:
+                    config.enable_tensorrt_inspector()
+                    self.assertTrue(
+                        config.tensorrt_inspector_enabled(),
+                        "The inspector option is not set correctly.")
 
                 if self.dynamic_shape_params:
                     config.set_trt_dynamic_shape_info(
@@ -244,14 +249,21 @@ class InferencePassTest(unittest.TestCase):
         Prepare TensorRT subgraph engine parameters. 
         '''
 
-        def __init__(self, workspace_size, max_batch_size, min_subgraph_size,
-                     precision, use_static, use_calib_mode):
+        def __init__(self,
+                     workspace_size,
+                     max_batch_size,
+                     min_subgraph_size,
+                     precision,
+                     use_static,
+                     use_calib_mode,
+                     use_inspector=False):
             self.workspace_size = workspace_size
             self.max_batch_size = max_batch_size
             self.min_subgraph_size = min_subgraph_size
             self.precision = precision
             self.use_static = use_static
             self.use_calib_mode = use_calib_mode
+            self.use_inspector = use_inspector
 
     class DynamicShapeParam:
         '''

python/paddle/fluid/tests/unittests/ir/inference/test_trt_inspector.py

+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import sys
+import os
+import threading
+import time
+import unittest
+import numpy as np
+from inference_pass_test import InferencePassTest
+import paddle
+import paddle.fluid as fluid
+import paddle.fluid.core as core
+from paddle.fluid.core import PassVersionChecker
+from paddle.fluid.core import AnalysisConfig
+import subprocess
+
+
+class TensorRTInspectorTest(InferencePassTest):
+    def setUp(self):
+        self.set_params()
+        with fluid.program_guard(self.main_program, self.startup_program):
+            data = fluid.data(name="data", shape=[1, 16, 16], dtype="float32")
+            matmul_out = fluid.layers.matmul(
+                x=data,
+                y=data,
+                transpose_x=self.transpose_x,
+                transpose_y=self.transpose_y,
+                alpha=self.alpha)
+            out = fluid.layers.batch_norm(matmul_out, is_test=True)
+
+        self.feeds = {"data": np.ones([1, 16, 16]).astype("float32"), }
+        self.enable_trt = True
+        self.trt_parameters = InferencePassTest.TensorRTParam(
+            1 << 30, 1, 0, AnalysisConfig.Precision.Float32, False, False, True)
+        self.fetch_list = [out]
+
+    def set_params(self):
+        self.transpose_x = True
+        self.transpose_y = True
+        self.alpha = 2.0
+
+    def test_check_output(self):
+        if core.is_compiled_with_cuda():
+            build_engine = subprocess.run(
+                [sys.executable, 'test_trt_inspector.py', '--build-engine'],
+                stderr=subprocess.PIPE)
+            engine_info = build_engine.stderr.decode('ascii')
+            trt_compile_version = paddle.inference.get_trt_compile_version()
+            trt_runtime_version = paddle.inference.get_trt_runtime_version()
+            valid_version = (8, 2, 0)
+            if trt_compile_version >= valid_version and trt_runtime_version >= valid_version:
+                self.assertTrue('====== engine info ======' in engine_info)
+                self.assertTrue('====== engine info end ======' in engine_info)
+                self.assertTrue('matmul' in engine_info)
+                self.assertTrue('LayerType: Scale' in engine_info)
+                self.assertTrue('batch_norm' in engine_info)
+            else:
+                self.assertTrue(
+                    'Inspector needs TensorRT version 8.2 and after.' in
+                    engine_info)
+
+
+if __name__ == "__main__":
+    if '--build-engine' in sys.argv:
+        test = TensorRTInspectorTest()
+        test.setUp()
+        use_gpu = True
+        test.check_output_with_option(use_gpu)
+    else:
+        unittest.main()