#!/bin/bash
set -e

# a function to install apt packages only if they are not installed
function apt_install() {
    if ! dpkg -s "$@" >/dev/null 2>&1; then
        if [ "$(find /var/lib/apt/lists/* | wc -l)" = "0" ]; then
            apt-get update
        fi
        apt-get install -y --no-install-recommends "$@"
    fi
}

apt_install \
    gnupg2 \
    curl \
    ca-certificates

apt-key adv --fetch-keys https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/7fa2af80.pub
apt-key adv --fetch-keys https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/3bf863cc.pub
echo "deb https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64 /" >/etc/apt/sources.list.d/cuda.list
echo "deb https://developer.download.nvidia.com/compute/machine-learning/repos/ubuntu1804/x86_64 /" >/etc/apt/sources.list.d/nvidia-ml.list
apt-get update

CUDA_VERSION=${CUDA_VERSION:-10.1.243}
CUDA_PKG_VERSION=${CUDA_PKG_VERSION:-10-1=$CUDA_VERSION-1}

# For libraries in the cuda-compat-* package: https://docs.nvidia.com/cuda/eula/index.html#attachment-a
apt_install \
    "cuda-cudart-${CUDA_PKG_VERSION}" \
    cuda-compat-10-1

ln -s cuda-10.1 /usr/local/cuda

# Required for nvidia-docker v1
echo "/usr/local/nvidia/lib" >>/etc/ld.so.conf.d/nvidia.conf &&
    echo "/usr/local/nvidia/lib64" >>/etc/ld.so.conf.d/nvidia.conf

## PATH & LD_LIBRARY_PATH are wrong?
PATH=${PATH:-/usr/local/nvidia/bin:/usr/local/cuda/bin:${PATH}}
LD_LIBRARY_PATH=${LD_LIBRARY_PATH:-/usr/local/nvidia/lib:/usr/local/nvidia/lib64}

# nvidia-container-runtime
NVIDIA_VISIBLE_DEVICES=${NVIDIA_VISIBLE_DEVICES:-all}
NVIDIA_DRIVER_CAPABILITIES=${NVIDIA_DRIVER_CAPABILITIES:-"compute,utility"}
NVIDIA_REQUIRE_CUDA=${NVIDIA_REQUIRE_CUDA:-"cuda>=10.1 brand=tesla,driver>=384,driver<385 brand=tesla,driver>=396,driver<397 brand=tesla,driver>=410,driver<411"}

## Tensorflow config for cuda runtime.
## Adapted from:

CUDA=10.1
CUDNN=7.6.4.38-1
LIBNVINFER=6.0.1-1
LIBNVINFER_MAJOR_VERSION=6

#SHELL ["/bin/bash", "-c"]
# Pick up some TF dependencies
# There appears to be a regression in libcublas10=10.2.2.89-1 which
# prevents cublas from initializing in TF. See
# https://github.com/tensorflow/tensorflow/issues/9489#issuecomment-562394257
apt_install \
    build-essential \
    cuda-command-line-tools-10-1 \
    libcublas10=10.2.1.243-1 \
    cuda-nvrtc-10-1 \
    cuda-cufft-10-1 \
    cuda-curand-10-1 \
    cuda-cusolver-10-1 \
    cuda-cusparse-10-1 \
    curl \
    libcudnn7=${CUDNN}+cuda10.1 \
    libfreetype6-dev \
    libhdf5-serial-dev \
    libzmq3-dev \
    pkg-config \
    software-properties-common \
    unzip

# Install TensorRT if not building for PowerPC
apt_install \
    libnvinfer${LIBNVINFER_MAJOR_VERSION}=${LIBNVINFER}+cuda${CUDA} \
    libnvinfer-plugin${LIBNVINFER_MAJOR_VERSION}=${LIBNVINFER}+cuda${CUDA}
apt-get clean

# For CUDA profiling, TensorFlow requires CUPTI.
LD_LIBRARY_PATH=${LD_LIBRARY_PATH:-/usr/local/cuda/extras/CUPTI/lib64:/usr/local/cuda/lib64:$LD_LIBRARY_PATH}

# Link the libcuda stub to the location where tensorflow is searching for it and reconfigure
# dynamic linker run-time bindings
ln -s /usr/local/cuda/lib64/stubs/libcuda.so /usr/local/cuda/lib64/stubs/libcuda.so.1
echo "/usr/local/cuda/lib64/stubs" >/etc/ld.so.conf.d/z-cuda-stubs.conf
ldconfig

## Add nvtop
#/rocker_scripts/install_nvtop.sh

## Add tensorflow-gpu==1.15 dependencies on the CUDA 10.0 libraries:
/rocker_scripts/install_tf1_cuda_10_0.sh

# Clean up
rm -rf /var/lib/apt/lists/*