Parallelization of sub-dnn processing

CMakeLists.txt

@@ -0,0 +1,25 @@
+cmake_minimum_required(VERSION 3.0.0)
+enable_language(CUDA)
+find_package(OpenMP REQUIRED)
+find_package(Threads REQUIRED)
+
+set(THREADS_PREFER_PTHREAD_FLAG ON)
+
+find_package(OpenCV REQUIRED)
+find_package (gflags REQUIRED)
+include_directories(/usr/local/cuda/include include)
+link_directories(/usr/local/cuda/lib64 /usr/lib/aarch64-linux-gnu/nvidia/ /usr/local/cuda-12.2/targets/aarch64-linux/lib)
+find_package(CUDA REQUIRED)
+set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -O3")
+
+add_compile_options(-D LIGHTNET_STANDALONE)
+
+set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++17 -O3 -Wall ${OpenMP_CXX_FLAGS} -D USE_DLA_STANDALONE_MODE")
+#set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++17 -O0 -g -Wno-write-strings ${OpenMP_CXX_FLAGS}" -d USE_DLA_STANDALONE_MODE)
+
+
+include_directories (${gflags_INCLUDE_DIR} ./include ./include/nvsci_headers)
+
+add_executable(cudla_lightnet src/main.cc src/config_parser.cc  src/cudla_context_standalone.cc src/cudla_lightnet.cc)
+target_include_directories(cudla_lightnet PRIVATE  ${OpenCV_INCLUDE_DIRS})
+target_link_libraries(cudla_lightnet ${OpenCV_LIBS} gflags  boost_system  boost_filesystem z cuda cudart nvinfer nvscibuf nvscisync nvcudla cudla "stdc++fs" "stdc++" "-ldl")

include/cudla_context_standalone.h

@@ -70,7 +70,7 @@ class cuDLAContextStandalone
     //!
     //! \brief Construct infer context from loadable
     //!
-    cuDLAContextStandalone(const char *loadableFilePath);
+    cuDLAContextStandalone(const char *loadableFilePath, const int dla_id);
 
     //!
     //! \brief Deconstructor
@@ -172,6 +172,7 @@ class cuDLAContextStandalone
     uint64_t m_WaiterValue = 0;
 #endif
     cudlaTask m_Task;
+    int m_dla_id;
 };
 
 #endif

include/cudla_lightnet.hpp

@@ -94,7 +94,7 @@ class Lightnet
 {
 public:
 
-  Lightnet(ModelConfig &model_config, InferenceConfig &inference_config, std::string &engine_path, LightnetBackend backend);
+  Lightnet(ModelConfig &model_config, InferenceConfig &inference_config, std::string &engine_path, LightnetBackend backend, int dla_id);
 
   ~Lightnet();
 
@@ -246,8 +246,9 @@ class Lightnet
    */
   std::vector<BBoxInfo> subnet_bbox_;
 
+  int dla_id_;
 };
 
 }
 
-#endif
+#endif

include/nvsci_headers/nvsci_headers_README.txt

@@ -0,0 +1,15 @@
+#
+# SPDX-FileCopyrightText: Copyright (c) 2021 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: LicenseRef-NvidiaProprietary
+#
+# NVIDIA CORPORATION, its affiliates and licensors retain all intellectual
+# property and proprietary rights in and to this material, related
+# documentation and any modifications thereto. Any use, reproduction,
+# disclosure or distribution of this material and related documentation
+# without an express license agreement from NVIDIA CORPORATION or
+# its affiliates is strictly prohibited.
+
+* nvsci_headers.tbz2 contains NvSci headers and a sample nvsciipc.cfg.
+* nvsciipc.cfg has to be put in /etc/.
+   Use the below command to copy the file to /etc directory:
+   $sudo cp ./nvsciipc.cfg /etc/nvsciipc.cfg