TensorRT 6.0 ONNX Parser Release

CMakeLists.txt

@@ -44,7 +44,7 @@ set(CMAKE_CXX_STANDARD 11)
 
 # Enable compiler warnings
 if ( CMAKE_COMPILER_IS_GNUCC )
- set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wall")
+ set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wall -Wno-deprecated-declarations -Wno-unused-function -Wno-unused-but-set-variable")
 endif()
 if ( MSVC )
  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /W4")
@@ -59,9 +59,9 @@ set(RUNTIME_LINKER_SCRIPT ${ONNX2TRT_ROOT}/libnvonnxparser_runtime.version)
 #--------------------------------------------------
 # Version information
 #--------------------------------------------------
-set(ONNX2TRT_MAJOR 0)
-set(ONNX2TRT_MINOR 1)
-set(ONNX2TRT_PATCH 0)
+set(ONNX2TRT_MAJOR 6)
+set(ONNX2TRT_MINOR 0)
+set(ONNX2TRT_PATCH 1)
 
 #--------------------------------------------------
 # Build configurations, global to all projects
@@ -159,6 +159,7 @@ if(${CMAKE_VERSION} VERSION_LESS ${CMAKE_VERSION_THRESHOLD})
  -lineinfo \
  -g \
  --expt-extended-lambda \
+ -Xcompiler -Wno-deprecated-declarations \
  ${GENCODES} \
  ")
 
@@ -174,6 +175,7 @@ else()
  -lineinfo \
  -g \
  --expt-extended-lambda \
+ -Xcompiler -Wno-deprecated-declarations \
  ${GENCODES} \
  ")
 

ModelImporter.cpp

@@ -132,6 +132,21 @@ Status importInputs(ImporterContext* importer_ctx,
  ASSERT_INPUT(!tensors->count(input.name()), ErrorCode::kINVALID_GRAPH,input.name());
  tensors->insert({input.name(), tensor});
  }
+
+ // According to the ONNX spec: initializers do not have to be specified as agraph input.
+ // In order for these initializers to be populated down to TRT, we need to add them to the tensors list.
+ for (auto initializer : initializer_map)
+ {
+ const std::string initializer_name = initializer.first;
+ if (!tensors->count(initializer_name))
+ {
+ const auto& initializer_weight = *initializer.second;
+ ShapedWeights weights;
+ ASSERT(convert_onnx_weights(initializer_weight, &weights), ErrorCode::kUNSUPPORTED_NODE);
+ tensors->insert({initializer_name, weights});
+ }
+ }
+
  return Status::success();
 }
 

NvOnnxParser.h

@@ -26,9 +26,9 @@
 #include "NvInfer.h"
 #include "NvOnnxParserTypedefs.h"
 
-#define NV_ONNX_PARSER_MAJOR 0
-#define NV_ONNX_PARSER_MINOR 1
-#define NV_ONNX_PARSER_PATCH 0
+#define NV_ONNX_PARSER_MAJOR 6
+#define NV_ONNX_PARSER_MINOR 0
+#define NV_ONNX_PARSER_PATCH 1
 
 static const int NV_ONNX_PARSER_VERSION = ((NV_ONNX_PARSER_MAJOR * 10000) +
  (NV_ONNX_PARSER_MINOR * 100) +

README.md

@@ -6,9 +6,11 @@ See also the [TensorRT documentation](https://docs.nvidia.com/deeplearning/sdk/#
 
 ## Supported TensorRT Versions
 
-Development on the Master branch is for the latest version of [TensorRT (5.1)](https://developer.nvidia.com/nvidia-tensorrt-download)
+Development on the Master branch is for the latest version of [TensorRT 6.0](https://developer.nvidia.com/nvidia-tensorrt-download)
 
-For versions < 5.1, clone and build from the [5.0 branch](https://github.com/onnx/onnx-tensorrt/tree/v5.0)
+For version 5.1, clone and build from the [5.1 branch](https://github.com/onnx/onnx-tensorrt/tree/5.1)
+
+For version < 5.1, clone and build from the [5.0 branch](https://github.com/onnx/onnx-tensorrt/tree/v5.0)
 
 
 ## Supported Operators
@@ -19,8 +21,8 @@ Current supported ONNX operators are found in the [operator support matrix](oper
 
 ### Dependencies
 
- - [Protobuf](https://github.com/google/protobuf/releases)
- - [TensorRT](https://developer.nvidia.com/tensorrt)
+ - [Protobuf >= 3.8.x](https://github.com/google/protobuf/releases)
+ - [TensorRT 6.0](https://developer.nvidia.com/tensorrt)
 
 ### Download the code
 Clone the code from GitHub. 
@@ -31,7 +33,7 @@ Clone the code from GitHub.
 
 The TensorRT-ONNX executables and libraries are built with CMAKE. Note by default CMAKE will tell the CUDA compiler generate code for the latest SM version. If you are using a GPU with a lower SM version you can specify which SMs to build for by using the optional `-DGPU_ARCHS` flag. For example, if you have a GTX 1080, you can specify `-DGPU_ARCHS="61"` to generate CUDA code specifically for that card.
 
-See [here](https://developer.nvidia.com/cuda-gpus) for finding what maximum compute capability your specific GPU supports.
+See [here](https://developer.nvidia.com/cuda-gpus) for the compute capability matrix for your specific GPU.
 
  mkdir build
  cd build
@@ -56,6 +58,15 @@ See more usage information by running:
 
  onnx2trt -h
 
+### Python modules
+Python bindings for the ONNX-TensorRT parser are packaged in the shipped `.whl` files. Install them with
+
+ pip install <tensorrt_install_dir>/python/tensorrt-6.0.1.5-cp27-none-linux_x86_64.whl
+
+TensorRT 6.0 supports ONNX release 1.5.0. Install it with:
+
+ pip install onnx==1.5.0
+
 ## ONNX Python backend usage
 
 The TensorRT backend for ONNX can be used in Python as follows:
@@ -84,22 +95,15 @@ libnvonnxparser_runtime.so, which has its C++ API declared in this header:
 
  NvOnnxParserRuntime.h
 
-### Python modules
-Python bindings for the ONNX-TensorRT parser in TensorRT versions >= 5.0 are packaged in the shipped `.whl` files. Install them with
-
- pip install <tensorrt_install_dir>/python/tensorrt-5.1.6.0-cp27-none-linux_x86_64.whl
-
-For earlier versions of TensorRT, the Python wrappers are built using SWIG.
-Build the Python wrappers and modules by running:
+Important typedefs required for parsing ONNX models are declared in this header:
 
- python setup.py build
- sudo python setup.py install
+ NvOnnxParserTypedefs.h
 
 ### Docker image
 
 Build the onnx_tensorrt Docker image by running:
 
- cp /path/to/TensorRT-5.1.*.tar.gz .
+ cp /path/to/TensorRT-6.0.*.tar.gz .
  docker build -t onnx_tensorrt .
 
 ### Tests
@@ -118,4 +122,4 @@ You can use `-v` flag to make output more verbose.
 
 ## Pre-trained models
 
-Pre-trained models in ONNX format can be found at the [ONNX Model Zoo](https://github.com/onnx/models)
+Pre-trained models in ONNX format can be found at the [ONNX Model Zoo](https://github.com/onnx/models)

ShapedWeights.cpp

@@ -44,11 +44,16 @@ bool convertINT64(void* weightValues, const size_t nbWeights, std::vector<int32_
 }
 
 size_t ShapedWeights::count() const {
- if( this->shape.nbDims == 0 ) {
+ if( this->values == nullptr && this->shape.nbDims == 0 )
+ {
  return 0;
- } else {
+ }
+ else 
+ {
+ // TRT supports scalars, so 0D tensors should have a count of 1.
  size_t c = 1;
- for( int i=0; i<this->shape.nbDims; ++i ) {
+ for( int i=0; i<this->shape.nbDims; ++i )
+ {
  c *= this->shape.d[i];
  }
  return c;