Add support for paddle trt

csrc/fastdeploy/backends/paddle/paddle_backend.cc

@@ -19,6 +19,32 @@ namespace fastdeploy {
 void PaddleBackend::BuildOption(const PaddleBackendOption& option) {
   if (option.use_gpu) {
     config_.EnableUseGpu(option.gpu_mem_init_size, option.gpu_id);
+    if (option.enable_trt) {
+#ifdef ENABLE_TRT_BACKEND
+      auto precision = paddle_infer::PrecisionType::kFloat32;
+      if (option.trt_option.enable_fp16) {
+        precision = paddle_infer::PrecisionType::kHalf;
+      }
+      config_.EnableTensorRtEngine(option.trt_option.max_workspace_size, option.trt_option.max_batch_size, 3, precision, false);
+      std::map<std::string, std::vector<int>> max_shape;
+      std::map<std::string, std::vector<int>> min_shape;
+      std::map<std::string, std::vector<int>> opt_shape;
+      for (auto iter = option.trt_option.max_shape.begin(); iter != option.trt_option.max_shape.end(); ++iter) {
+        auto min_iter = option.trt_option.min_shape.find(iter->first);
+        auto opt_iter = option.trt_option.opt_shape.find(iter->first);
+        FDASSERT(min_iter != option.trt_option.min_shape.end(), "Cannot find " + iter->first + " in TrtBackendOption::min_shape.");
+        FDASSERT(opt_iter != option.trt_option.opt_shape.end(), "Cannot find " + iter->first + " in TrtBackendOption::opt_shape.");
+        max_shape[iter->first].assign(iter->second.begin(), iter->second.end());
+        min_shape[iter->first].assign(min_iter->second.begin(), min_iter->second.end());
+        opt_shape[iter->first].assign(opt_iter->second.begin(), opt_iter->second.end());
+      }
+      if (max_shape.size() > 0) {
+        config_.SetTRTDynamicShapeInfo(min_shape, max_shape, opt_shape);
+      }
+#else
+    FDWARNING << "The FastDeploy is not compiled with TensorRT backend, so will fallback to GPU with Paddle Inference Backend." << std::endl;
+#endif
+    }
   } else {
     config_.DisableGpu();
     if (option.enable_mkldnn) {

csrc/fastdeploy/backends/paddle/paddle_backend.h

@@ -22,6 +22,10 @@
 #include "fastdeploy/backends/backend.h"
 #include "paddle_inference_api.h"  // NOLINT
 
+#ifdef ENABLE_TRT_BACKEND
+#include "fastdeploy/backends/tensorrt/trt_backend.h"
+#endif
+
 namespace fastdeploy {
 
 struct PaddleBackendOption {
@@ -34,6 +38,12 @@ struct PaddleBackendOption {
 
   bool enable_log_info = false;
 
+  // Enable TensorRT backend by Paddle2TRT
+  bool enable_trt = false;
+#ifdef ENABLE_TRT_BACKEND
+  TrtBackendOption trt_option;
+#endif
+
   int mkldnn_cache_size = 1;
   int cpu_thread_num = 8;
   // initialize memory size(MB) for GPU

csrc/fastdeploy/fastdeploy_runtime.cc

@@ -209,6 +209,16 @@ void RuntimeOption::SetPaddleMKLDNNCacheSize(int size) {
   pd_mkldnn_cache_size = size;
 }
 
+void RuntimeOption::EnablePaddleToTrt() {
+  FDASSERT(backend == Backend::TRT, "Should call UseTrtBackend() before call EnablePaddleToTrt().");
+#ifdef ENABLE_PADDLE_BACKEND
+  backend = Backend::PDINFER;
+  pd_enable_trt = true;
+#else
+  FDASSERT(false, "The FastDeploy is not compiled with Paddle Inference backend, cannot EnablePaddleToTrt now.");
+#endif
+}
+
 void RuntimeOption::SetTrtInputShape(const std::string& input_name,
                                      const std::vector<int32_t>& min_shape,
                                      const std::vector<int32_t>& opt_shape,
@@ -312,6 +322,22 @@ void Runtime::CreatePaddleBackend() {
   pd_option.gpu_id = option.device_id;
   pd_option.delete_pass_names = option.pd_delete_pass_names;
   pd_option.cpu_thread_num = option.cpu_thread_num;
+#ifdef ENABLE_TRT_BACKEND
+  if (option.pd_enable_trt) {
+    pd_option.enable_trt = true;
+    auto trt_option = TrtBackendOption();
+    trt_option.gpu_id = option.device_id;
+    trt_option.enable_fp16 = option.trt_enable_fp16;
+    trt_option.enable_int8 = option.trt_enable_int8;
+    trt_option.max_batch_size = option.trt_max_batch_size;
+    trt_option.max_workspace_size = option.trt_max_workspace_size;
+    trt_option.max_shape = option.trt_max_shape;
+    trt_option.min_shape = option.trt_min_shape;
+    trt_option.opt_shape = option.trt_opt_shape;
+    trt_option.serialize_file = option.trt_serialize_file;
+    pd_option.trt_option = trt_option;
+  }
+#endif
   FDASSERT(option.model_format == Frontend::PADDLE,
            "PaddleBackend only support model format of Frontend::PADDLE.");
   backend_ = utils::make_unique<PaddleBackend>();

csrc/fastdeploy/fastdeploy_runtime.h

@@ -63,6 +63,8 @@ struct FASTDEPLOY_DECL RuntimeOption {
   // use tensorrt backend
   void UseTrtBackend();
 
+  void EnablePaddleToTrt();
+
   // use openvino backend
   void UseOpenVINOBackend();
 
@@ -119,8 +121,9 @@ struct FASTDEPLOY_DECL RuntimeOption {
 
   // ======Only for Paddle Backend=====
   bool pd_enable_mkldnn = true;
-  bool pd_enable_log_info = false;
   int pd_mkldnn_cache_size = 1;
+  bool pd_enable_trt = false;
+  bool pd_enable_log_info = false;
   std::vector<std::string> pd_delete_pass_names;
 
   // ======Only for Trt Backend=======