1. Automatically generate high-performance TensorRT plugins for unsupported operators or replacing inefficient kernels.
2. End-to-end command line tool. No requirement for any CUDA programming knowledge. Users only need to provide the ONNX model and assign the node names or types to auto-generate TensorRT plugin.
3. The performance of auto-generated TensorRT plugins in real cases:
   • Performance comparation with hand-written kernels
   • Optimization for TensorRT's original kernels

• ONNX Operators supported by TPAT-1.0

nvidia-docker build .

nvidia-docker run -itd --gpus all -v <TPAT path dir>:/root <Image_ID> /bin/bash

nvidia-docker exec -it <Container_ID> /bin/bash

CUDA_PATH: local CUDA installation path
TRT_LIB_PATH: local TensorRT installation path

cd examples
python test_onehot_dynamic_direct.py

• tpat_onehot.so is stored in python/trt_plugin/lib/

• LLVM >= 9.0.1, (LLVM==9.0.1 recommended)
• GCC >= 7.3.0, (GCC==7.4.0 recommended)
• TensorRT

• numpy pycuda onnx onnxruntime onnx_graphsurgeon xgboost jinja2 ctypes tornado cloudpickle psutil

NOTE: these necessary packages are recorded in requirements.txt

• tensorflow-gpu==1.15
• tf2onnx
• torch
• pytest

NOTE: these optional packages are required by Example and UnitTest

git clone -b master https://github.com/nvidia/TensorRT TPAT
cd TPAT
git submodule update --init --recursive

mkdir build && cp cmake/config.cmake build
#Edit build/config.cmake to customize the compilation options
set(USE_LLVM /usr/local/llvm/bin/llvm-config)
set(USE_CUDA ON)
#gcc compiler is required to support C++14
cd build && cmake .. 
make -j
#TVM Python package
export TVM_HOME=/path/to/tvm
export PYTHONPATH=$TVM_HOME/python:${PYTHONPATH}

Modify python/trt_plugin/Makefile according to your environment setup.

CUDA_PATH: local CUDA installation path
TRT_LIB_PATH: local TensorRT installation path

TPAT provides a Python function and command line for usage.

onnx2plugin(
	input_model_path, 
	output_model_path, 
	node_names=None, 
	node_types=None, 
	plugin_name_dict=None,
	dynamic_bs=False, # if True, this operator support dynamic batchsize
	min_bs=1,
	max_bs=256,
	opt_bs=128
	)

• input_model_path[required] : input onnx model including nodes which require TRT plugin
• output_model_path[required] : output onnx model where the corresponding node types are replaced by plugin names. The output onnx model can be directly converted to TRT with onnx parser and built plugin dynamic library.
• node_names : list of node names for autogen
• node_types : list of node types for autogen
• plugin_name_dict : dict of {plugin_name: node_name} for autogen
• dynamic_bs : if True, TPAT will generate plugin that supported dynamic batch, if False, generated plugin only support fixed shapes but has better performance.
• min_bs: the minium batch size in range of dynamic batch.
• max_bs: the maxium batch size in range of dynamic batch.
• opt_bs: the optimize batch size in range of dynamic batch.

NOTE: For node_names, node_types, plugin_name_dict, at least one of them should be provided

# Separate different ops with spaces
python3 Onnx2Plugin.py -i input.onnx -o output.onnx -n op_name1 op_name2 -dynamic=true -min=1 -max=512 -opt=256
python3 Onnx2Plugin.py -i input.onnx -o output.onnx -t op_type1 op_type2 -dynamic=false
python3 Onnx2Plugin.py -i input.onnx -o output.onnx -p '{"op_name1": "plugin_name1", "op_name2": "plugin_name2"}'

• -i[required]: input_model_path
• -o[required]: output_model_path
• -n: node_names
• -t: node_types
• -p: plugin_name_dict
• -dynamic: dynamic_bs
• -min: min_bs
• -max: max_bs
• -opt: opt_bs

• trt_plugin/src contains {plugin_name}.cu and {plugin_name}.h
• trt_plugin/lib contains {plugin_name}.so

• trt_plugin/src contains tpat_{node_name}.cu and tpat_{node_name}.h
• trt_plugin/lib contains tpat_{node_name}.so

• Example : example_tensorflow.py
• UnitTest : test_tapt.py

• Support mutiple nodes for autogen
• Support boolean input/outputs
• Able to reuse plugins

• Only support dynamic BatchSize
• Opeartors with int8/float16/double inputs/outputs are not supported

• Support ONNX subgraph for autogen
• Support direction conversion from TensorFlow and PyTorch