TheRock/rocm-systems/projects/hipfile/util/docker/DOCKERFILE.ais_ci_rocky

# Copyright (c) Advanced Micro Devices, Inc. All rights reserved.
#
# SPDX-License-Identifier: MIT

ARG ROCKY_MAJOR_VERSION=9
ARG ROCKY_VERSION=${ROCKY_MAJOR_VERSION}.6
FROM rockylinux/rockylinux:${ROCKY_VERSION}-minimal AS base
LABEL org.opencontainers.image.source=https://github.com/ROCm/hipFile
# Brings ROCKY_VERSION in-scope
ARG ROCKY_MAJOR_VERSION
ARG ROCKY_VERSION
ARG ROCM_VERSION=7.2.2

# If the target ROCm Version on the package repository is non-standard, you can forcibly
# set the ROCm version used in the package repo URL.
# Ex. Package 7.0_alpha installs 7.0.0
ARG ROCM_PKG_REPO_OVERRIDE

# Update repo and install program dependencies.
# microdnf is too stripped down, install full dnf.
RUN microdnf install -y \
        dnf \
        epel-release && \
    dnf config-manager --set-enabled crb

# Setup AMD/ROCm Repositories (using AMD RHEL repo)
# Method 1: Using amdgpu_install.rpm - buggy
# The ROCm repo works in mysterious ways in that someone visiting from
# a browser can download a certain version, but a utility like curl will
# not be allowed to download it. Sometimes we may need to just forcibly
# specify the version to make this work...
# ARG AMDGPU_INSTALL="amdgpu-install-6.4.60402-1.el9.noarch.rpm"
# RUN : "${AMDGPU_INSTALL:=$( \
#         curl -s https://repo.radeon.com/amdgpu-install/latest/rhel/${ROCKY_VERSION}/ | \
#         grep -oP '(?<=href=")[^"]+' | \
#         grep amdgpu-install)}" && \
#     echo "AMDGPU_INSTALL is: ${AMDGPU_INSTALL}" && \
#     curl https://repo.radeon.com/amdgpu-install/latest/rhel/${ROCKY_VERSION}/${AMDGPU_INSTALL} \
#         -o amdgpu_install.rpm && \
#     dnf install -y ./amdgpu_install.rpm
# Method 2: Manually create repo file.

# ROCM_PKG_REPO_VERSION normalizes the ROCm version used in the package repo URL.
# If the PATCH component is 0, the PATCH of the version is dropped. Otherwise no
# modification is made.
RUN ROCM_PKG_REPO_VERSION="$(echo ${ROCM_PKG_REPO_OVERRIDE:-$ROCM_VERSION} | sed -E 's|([^.]+\.[^.]+)\.0$|\1|')"; \
    cat <<EOF > /etc/yum.repos.d/rocm.repo
[ROCm-${ROCM_VERSION}]
name=ROCm${ROCM_VERSION}
baseurl=https://repo.radeon.com/rocm/el${ROCKY_MAJOR_VERSION}/${ROCM_PKG_REPO_VERSION}/main
enabled=1
priority=50
gpgcheck=1
gpgkey=https://repo.radeon.com/rocm/rocm.gpg.key
EOF

# Development Dependencies
# clang version == 19
# llvm-dev == llvm-devel
# Defaults to gcc11. On 9.6 gcc14 seems to be present.
#   Use gcc13 to match Ubuntu24.
RUN dnf makecache && \
    dnf install -y \
        clang \
        cmake \
        doxygen \
        gcc-toolset-13 \
        gdb \
        git \
        hip-devel \
        libmount-devel \
        llvm-devel \
        python3.12-devel \
        rpm-build

# Point amdclang++ at the gcc-toolset-13 libstdc++.
# clang locates its C++ standard library by scanning a fixed set of prefixes
# (essentially /usr) for the newest GCC; it does NOT consult PATH. The default
# el9 GCC (11) is older than the GCC 13 used on Ubuntu24/SUSE, and
# gcc-toolset-13 lives under /opt/rh, which clang's scanner doesn't search on
# this toolchain. Append --gcc-install-dir to the ROCm config file that
# amdclang++ auto-loads so it selects the gcc-toolset-13 libstdc++.
RUN echo '--gcc-install-dir=/opt/rh/gcc-toolset-13/root/usr/lib/gcc/x86_64-redhat-linux/13' \
        >> /opt/rocm/lib/llvm/bin/rocm.cfg

# Configure system linker for ROCm
RUN cat <<EOF > /etc/ld.so.conf.d/rocm.conf
/opt/rocm/lib
/opt/rocm/lib64
EOF
RUN ldconfig

# Add ROCm bin directory to the path
ENV PATH=/opt/rocm/bin:${PATH}

# Environment Build Arg for ROCm
ENV ROCM_VERSION="${ROCM_VERSION}"

# .rc and .profile files are not automatically read on non-login,
# non-interactive shells. Tell Bash to read a minimal .profile
# when a shell of any type is opened.
# Uncomment to prefer newer version of GCC.
RUN echo 'source /opt/rh/gcc-toolset-13/enable' >> ~/.bashrc
RUN echo 'source ~/.bashrc' >> ~/.bash_profile
ENV BASH_ENV="~/.bashrc"

# AIS Code Repository Mount Point
VOLUME /mnt/ais

# Generic Mount point for container to access AIS-capable storage.
VOLUME /mnt/ais-fs

# ---- Stage 2: NVIDIA (extends base with CUDA SDK) ----
# The HIP runtime must be present to use hipFile with NVIDIA GPUs,
# which is why this stage inherits from base rather than a bare CUDA image.
FROM base AS nvidia
ARG ROCKY_MAJOR_VERSION=9

# Install a minimal CUDA SDK
RUN curl https://developer.download.nvidia.com/compute/cuda/repos/rhel${ROCKY_MAJOR_VERSION}/x86_64/cuda-rhel${ROCKY_MAJOR_VERSION}.repo \
        -o /etc/yum.repos.d/cuda-rhel${ROCKY_MAJOR_VERSION}.repo && \
    dnf makecache

# https://gitlab.com/nvidia/container-images/cuda/blob/master/dist/12.6.1/ubuntu2404/devel/Dockerfile
# Attempt a "minimal" CUDA Dev install.
# libcufile-dev-X.Y == libcufile-devel-X.Y
# libnvidia-compute-570-server == nvidia-driver-cuda + kmod-nvidia-open-dkms
RUN dnf module enable -y nvidia-driver:570-open && \
    dnf install -y \
        cuda-minimal-build-12-9 \
        kmod-nvidia-open-dkms \
        libcufile-devel-12-9 \
        nvidia-driver-cuda

# Add CUDA bin directory to the path
ENV PATH=/usr/local/cuda/bin:${PATH}