install cuda, cudnn, tensorrt

1. cuda-11
2. cudnn-8.2.1
3. tensorrt 8.2.1.8

get and patch tensorrt-oss

git clone https://github.com/NVIDIA/TensorRT/
pushd TensorRT
git submodule update --init --recursive
git checkout 156c59ae86d454fa89146fe65fa7332dbc8c3c2b
git submodule update
git apply ../tensorrt.diff
popd

build tensorrt

pushd TensorRT
mkdir -p build && cd build
CUDACXX=/opt/anaconda3/envs/cuda-11/bin/nvcc cmake .. -DTRT_LIB_DIR=/opt/anaconda3/envs/cuda-11/lib -DTRT_OUT_DIR=`pwd`/out
make
popd

get pretrianed model

get models from https://mega.nz/folder/X5UD1LDY#8ZI-gAq6AkpcUbE2z4n6RA and save models to model directory

run

make run-mnist make run-googlenet make run-mobilenet make run-resnet

run with int8

1. turn on CPPFLAGS += -DINT8 in Makefile
2. make clean
3. make run-mnist