Parser
Installation
sudo apt-get update && apt-get install -y protobuf-compiler libprotoc-dev git python3 python3-pip cmake libopencv-dev
git clone -b onnx https://gitlab.cs.fau.de/radl/deeplearning
cd deeplearning
git clone https://github.com/onnx/onnx.git
cd onnx
git submodule update --init --recursive
export CMAKE_ARGS="-DONNX_USE_PROTOBUF_SHARED_LIBS=ON"
pip install wheel
pip install -e .
// In my case: export BASE_DIR=/home/cip/2020/ez46emyj/Documents/deeplearning
/usr/bin/protoc onnx/onnx.proto --proto_path=. --proto_path=/usr/include/google/protobuf --cpp_out $BASE_DIR/onnx/onnx"
cd $BASE_DIR
make
./inferenceBuilding
make METHOD=0 ROUNDS=100Use METHOD=0 for Cuda implementation and METHOD=1 for CPU implementation. Set the amound of rounds using the ROUNDS argument.
Run
./inferenceQuantization using cuda_fp16.h
Memory Load GiB/s Performance Images/s
https://ion-thruster.medium.com/an-introduction-to-writing-fp16-code-for-nvidias-gpus-da8ac000c17f