feat(ci): add package generation

.docker/Dockerfile.dnf

@@ -0,0 +1,44 @@
+#
+# Copyright (c) 2024, Amazon.com, Inc. or its affiliates. All rights reserved.
+#
+# See LICENSE.txt for license information
+#
+
+ARG FAMILY=fedora
+ARG VERSION=rawhide
+ARG VARIANT=cuda
+ARG CUDA_DISTRO
+ARG AWS_BUILD
+ARG ENABLE_POWERTOOLS
+
+# Install EFA-installer deps.
+FROM ${FAMILY}:${VERSION} AS builder
+ARG CUDA_DISTRO
+ARG ENABLE_POWERTOOLS
+ENV CUDA_DISTRO=${CUDA_DISTRO}
+ENV ENABLE_POWERTOOLS=${ENABLE_POWERTOOLS}
+# Add NVIDIA repo for CUDA builds.
+COPY --from=efainstaller / /
+RUN --mount=type=cache,target=/var/cache/yum,sharing=locked \
+    --mount=type=cache,target=/var/cache/dnf,sharing=locked \
+    bash -c "cd /aws-efa-installer && dnf install -y gcc rpmdevtools rpmlint dnf-plugins-core util-linux && ./efa_installer.sh -n -l -k -d -y && rm -rf /aws-efa-installer" && \
+    dnf config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/${CUDA_DISTRO}/$(uname -m)/cuda-${CUDA_DISTRO}.repo && \
+    ( test "${ENABLE_POWERTOOLS}" = "1" && sed -i 's/enabled=0/enabled=1/' /etc/yum.repos.d/Rocky-PowerTools.repo || /bin/true ) && \
+    dnf -y update && dnf -y upgrade
+RUN rpmdev-setuptree
+
+FROM builder AS environment
+ARG VARIANT
+ARG AWS_BUILD
+ENV VARIANT=${VARIANT}
+ENV AWS_BUILD=${AWS_BUILD}
+COPY --from=srpm . .
+RUN yum search hwloc
+RUN echo "%with_${VARIANT} 1" >> ~/.rpmmacros
+RUN echo "%with_platform_aws ${AWS_BUILD}" >> ~/.rpmmacros
+RUN --mount=type=cache,target=/var/cache/yum,sharing=locked \
+    --mount=type=cache,target=/var/cache/dnf,sharing=locked \
+    dnf -y install cuda-cudart-devel-12-6 && dnf -y builddep *.src.rpm && rpmbuild --rebuild *.src.rpm
+
+FROM scratch
+COPY --from=environment /root/rpmbuild/RPMS/**/* /

.docker/Dockerfile.dpkg

@@ -0,0 +1,65 @@
+#
+# Copyright (c) 2024, Amazon.com, Inc. or its affiliates. All rights reserved.
+#
+# See LICENSE.txt for license information
+#
+
+ARG FAMILY=ubuntu
+ARG VERSION=latest
+ARG CUDA_DISTRO
+ARG DEBIAN_FRONTEND=noninteractive
+ARG AWS_BUILD
+
+FROM ${FAMILY}:${VERSION} AS build
+ARG CUDA_DISTRO
+ENV CUDA_DISTRO=${CUDA_DISTRO}
+ARG AWS_BUILD=0
+ENV AWS_BUILD=${AWS_BUILD}
+
+RUN --mount=type=cache,target=/var/cache/apt,sharing=locked \
+    --mount=type=cache,target=/var/lib/apt,sharing=locked \
+    apt-get update -y && apt-get install wget -y
+
+RUN wget https://developer.download.nvidia.com/compute/cuda/repos/${CUDA_DISTRO}/$(uname -m)/cuda-keyring_1.1-1_all.deb
+
+RUN --mount=type=cache,target=/var/cache/apt,sharing=locked \
+    --mount=type=cache,target=/var/lib/apt,sharing=locked \
+    dpkg -i cuda-keyring_1.1-1_all.deb
+
+COPY --from=efainstaller / .
+RUN --mount=type=cache,target=/var/cache/apt,sharing=locked \
+    --mount=type=cache,target=/var/lib/apt,sharing=locked \
+    bash -c "apt-get update -y && cd /aws-efa-installer && ./efa_installer.sh /efa_installer.sh -n -l -k -d -y && apt-get install -y autoconf automake libtool gcc g++ git libhwloc-dev make && rm -rf /aws-efa-installer"
+
+COPY --from=makedist / .
+RUN tar xvf ./aws-ofi-nccl*.tar.gz -C .
+RUN cd aws-ofi-nccl* && \
+	./configure --$(test "$ACCELERATOR" = "cuda" && echo "with-cuda=/usr/local/cuda" || echo "enable-neuron=yes") \
+	--prefix=/opt/amazon/libnccl-net-ofi$(test "$AWS_BUILD" -eq 0 || echo -n "-aws") \
+	--with-libfabric=/opt/amazon/efa \
+	--disable-tests \
+        --$(test "$AWS_BUILD" -eq 0 && echo -n "disable" || echo -n "enable")-platform-aws \
+	--with-mpi=no && make -j && make install
+
+FROM ubuntu:latest AS packager
+ARG FAMILY
+ARG VERSION
+ARG AWS_BUILD=0
+ENV AWS_BUILD=${AWS_BUILD}
+ENV FAMILY=${FAMILY}
+ENV VERSION=${VERSION}
+COPY --from=build /opt/amazon/ /opt/amazon/
+RUN find /opt/amazon/ | grep -E \.la$ | xargs rm
+RUN --mount=type=cache,target=/var/cache/apt,sharing=locked \
+    --mount=type=cache,target=/var/lib/apt,sharing=locked \
+    apt-get update -y && apt-get install -y ruby tar squashfs-tools binutils && gem install fpm
+RUN fpm \
+	-s dir -t deb \
+	--license Apache2.0 \
+	-p /libnccl-net-ofi$(test "$AWS_BUILD" -eq 0 || echo -n "-aws")-${FAMILY}-${VERSION}.deb \
+	--name nccl-net-ofi$(test "$AWS_BUILD" -eq 0 || echo -n "-aws") \
+	/opt/amazon/libnccl-net-ofi$(test "$AWS_BUILD" -eq 0 || echo -n "-aws")/=/opt/amazon/libnccl-net-ofi$(test "$AWS_BUILD" -eq 0 || echo -n "-aws")
+
+FROM scratch
+COPY --from=packager /libnccl-net-ofi* /
+

.docker/Dockerfile.efa

@@ -0,0 +1,15 @@
+#
+# Copyright (c) 2024, Amazon.com, Inc. or its affiliates. All rights reserved.
+#
+# See LICENSE.txt for license information
+#
+
+FROM alpine:latest AS efa_installer_extracted
+ARG EFA_INSTALLER_VERSION=latest
+ENV EFA_INSTALLER_VERSION=${EFA_INSTALLER_VERSION}
+RUN apk add tar curl
+RUN mkdir /libfabric
+RUN curl -s -L https://efa-installer.amazonaws.com/aws-efa-installer-${EFA_INSTALLER_VERSION}.tar.gz | tar -xvzf - -C /
+
+FROM scratch
+COPY --from=efa_installer_extracted /aws-efa-installer /aws-efa-installer

.docker/Dockerfile.makedist

@@ -0,0 +1,26 @@
+#
+# Copyright (c) 2024, Amazon.com, Inc. or its affiliates. All rights reserved.
+#
+# See LICENSE.txt for license information
+#
+
+ARG ACCELERATOR
+ARG BASE_IMAGE=ubuntu:22.04
+FROM ${BASE_IMAGE} AS distbuilder
+ARG ACCELERATOR
+ENV ACCELERATOR=${ACCELERATOR}
+RUN mkdir /aws-efa-installer
+COPY --from=efainstaller /aws-efa-installer /aws-efa-installer
+RUN --mount=type=cache,target=/var/cache/apt,sharing=locked \
+    --mount=type=cache,target=/var/lib/apt,sharing=locked \
+    bash -c "apt-get update -y && cd /aws-efa-installer && ./efa_installer.sh /efa_installer.sh -n -l -k -d -y && apt-get install -y autoconf automake libtool gcc git libhwloc-dev make && rm -rf /aws-efa-installer"
+COPY ../ /proj
+WORKDIR /proj
+RUN autoreconf -ivf
+RUN ./configure --with-libfabric=/opt/amazon/efa --$(test "$ACCELERATOR" = "cuda" && echo "with-cuda=/usr/local/cuda" || echo "enable-neuron=yes") --with-libfabric=/opt/amazon/efa
+RUN make dist
+RUN ls -lart
+RUN pwd
+
+FROM scratch
+COPY --from=distbuilder /proj/aws-ofi-nccl*.tar.gz /

.docker/Dockerfile.srpm

@@ -0,0 +1,18 @@
+#
+# Copyright (c) 2024, Amazon.com, Inc. or its affiliates. All rights reserved.
+#
+# See LICENSE.txt for license information
+#
+
+FROM fedora:rawhide AS packitimg
+RUN dnf install -y packit mock
+
+FROM packitimg AS srpm
+RUN mkdir /proj
+WORKDIR /proj
+COPY --from=src . .
+COPY --from=makedist . .
+RUN packit srpm
+
+FROM scratch
+COPY --from=srpm /proj/*.src.rpm /

.docker/Dockerfile.yum

@@ -0,0 +1,42 @@
+#
+# Copyright (c) 2024, Amazon.com, Inc. or its affiliates. All rights reserved.
+#
+# See LICENSE.txt for license information
+#
+
+ARG FAMILY=amazonlinux
+ARG VERSION=2
+ARG VARIANT=cuda
+ARG CUDA_DISTRO
+ARG AWS_BUILD
+
+# Install EFA-installer deps.
+FROM ${FAMILY}:${VERSION} AS builder
+ARG CUDA_DISTRO
+ENV CUDA_DISTRO=${CUDA_DISTRO}
+# Add NVIDIA repo for CUDA builds.
+COPY --from=efainstaller / /
+RUN --mount=type=cache,target=/var/cache/yum,sharing=locked \
+    --mount=type=cache,target=/var/cache/dnf,sharing=locked \
+    bash -c "cd /aws-efa-installer && yum install -y gcc rpmdevtools rpmlint yum-utils util-linux && ./efa_installer.sh -n -l -k -d -y && rm -rf /aws-efa-installer" && \
+    yum-config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/${CUDA_DISTRO}/$(uname -m)/cuda-${CUDA_DISTRO}.repo && \
+    yum update -y
+RUN rpmdev-setuptree
+
+FROM builder AS environment
+ARG VARIANT
+ARG AWS_BUILD
+ARG TOOLKIT_VERSION=12-6
+ENV VARIANT=${VARIANT}
+ENV AWS_BUILD=${AWS_BUILD}
+ENV TOOLKIT_VERSION=${TOOLKIT_VERSION}
+COPY --from=srpm . .
+RUN echo "%with_${VARIANT} 1" >> ~/.rpmmacros
+RUN echo "%with_platform_aws ${AWS_BUILD}" >> ~/.rpmmacros
+RUN echo "%_cuda_toolkit_version ${TOOLKIT_VERSION}" >> ~/.rpmmacros
+RUN --mount=type=cache,target=/var/cache/yum,sharing=locked \
+    --mount=type=cache,target=/var/cache/dnf,sharing=locked \
+    yum install -y cuda-cudart-devel-${TOOLKIT_VERSION} && yum-builddep -y *.src.rpm && rpmbuild --rebuild *.src.rpm
+
+FROM scratch
+COPY --from=environment /root/rpmbuild/RPMS/**/* /

.github/workflows/distcheck.yaml

@@ -27,6 +27,9 @@ jobs:
       matrix:
         sdk:
           - cuda
+        mode:
+          - cpp
+          - c
         amazonlinux:
           - al2023
           - al2
@@ -47,7 +50,7 @@ jobs:
             cudapackages: cuda-cudart-devel-12-3 cuda-driver-devel-12-3
 
     runs-on: codebuild-ghactions-${{ matrix.amazonlinux }}-${{ github.run_id }}-${{ github.run_attempt }}
-    name: ${{matrix.amazonlinux}}/${{ matrix.sdk }}/efa@${{ matrix.efainstaller }}/makeinstall
+    name: ${{matrix.amazonlinux}}/${{ matrix.sdk }}/efa@${{ matrix.efainstaller }}/makeinstall(${{ matrix.mode }})
     steps:
       # note, do not bump to v4: https://github.com/actions/checkout/issues/1590
       - uses: actions/checkout@v3
@@ -76,6 +79,7 @@ jobs:
           ./configure --prefix=/opt/aws-ofi-nccl --with-mpi=/opt/amazon/openmpi \
                --with-libfabric=/opt/amazon/efa \
                --with-cuda=/usr/local/cuda \
+               --enable-cpp=${{ matrix.mode == 'cpp' && 'yes' || 'no' }} \
                --enable-tests=no \
                --enable-platform-aws
 
@@ -95,13 +99,15 @@ jobs:
         cc:
           - gcc
           - clang
+        mode:
+          - cpp
+          - c
         tracing:
           - lttng
           - none
         sdk:
           - cuda
           - neuron
-
         include:
           - cc-variant: latest
             cc: clang
@@ -110,7 +116,7 @@ jobs:
             cc: gcc
             cc-version: 13
 
-    name: u2204/${{ matrix.sdk }}/libfabric@git/${{matrix.cc}}(${{matrix.cc-variant}})/distcheck/
+    name: u2204/${{ matrix.sdk }}/libfabric@git/${{matrix.cc}}(${{matrix.cc-variant}})/distcheck(${{ matrix.mode }})/
     steps:
       - uses: actions/checkout@v4
       - name: Configure Neuron SDK Repository
@@ -196,10 +202,12 @@ jobs:
             ./configure --with-mpi=/opt/amazon/openmpi \
                         --with-libfabric=/opt/amazon/efa \
                         --with-cuda=/usr/local/cuda/ \
+                        --enable-cpp=${{ matrix.mode == 'cpp' && 'yes' || 'no' }} \
                         --enable-tests \
                         --enable-platform-aws
           else
             ./configure --with-libfabric=/opt/amazon/efa \
+                        --enable-cpp=${{ matrix.mode == 'cpp' && 'yes' || 'no' }} \
                         --enable-neuron \
                         --enable-platform-aws
           fi
@@ -226,6 +234,9 @@ jobs:
         cc:
           - gcc
           - clang
+        mode:
+          - cpp
+          - c
         sdk:
           - cuda
           - neuron
@@ -237,7 +248,7 @@ jobs:
             cc: gcc
             cc-version: 13
 
-    name: u2204/${{ matrix.sdk }}/${{matrix.cc}}(${{matrix.cc-variant}})/unit-tests/
+    name: u2204/${{ matrix.sdk }}/${{matrix.cc}}(${{matrix.cc-variant}})/unit-tests(${{ matrix.mode }})/
     steps:
       - uses: actions/checkout@v4
       - name: Configure Neuron SDK Repository
@@ -316,13 +327,15 @@ jobs:
             ./configure --with-mpi=/opt/amazon/openmpi \
                         --with-libfabric=/opt/amazon/efa \
                         --with-cuda=/usr/local/cuda/ \
+                        --enable-cpp=${{ matrix.mode == 'cpp' && 'yes' || 'no' }} \
                         --enable-tests \
                         --enable-debug \
                         --enable-platform-aws
           else
             ./configure --with-libfabric=/opt/amazon/efa \
                         --enable-neuron \
                         --enable-debug \
+                        --enable-cpp=${{ matrix.mode == 'cpp' && 'yes' || 'no' }} \
                         --enable-platform-aws
           fi
           make -j "$(nproc)"
@@ -346,7 +359,10 @@ jobs:
         sdk:
           - cuda
           - neuron
-    name: CodeChecker - ${{ matrix.sdk }}
+        mode:
+          - cpp
+          - c
+    name: CodeChecker - ${{ matrix.sdk }} ${{ matrix.mode }}
     steps:
       - uses: actions/checkout@v4
       - uses: actions/setup-python@v5
@@ -404,12 +420,14 @@ jobs:
             ./configure \
               --with-libfabric="/opt/amazon/efa" \
               --enable-neuron \
+              --enable-cpp=${{ matrix.mode == 'cpp' && 'yes' || 'no' }} \
               --enable-platform-aws
           else
             ./configure \
               --with-libfabric="/opt/amazon/efa" \
               --with-mpi="/opt/amazon/openmpi" \
               --with-cuda=/usr/local/cuda/ \
+              --enable-cpp=${{ matrix.mode == 'cpp' && 'yes' || 'no' }} \
               --enable-tests \
               --enable-platform-aws
           fi
@@ -428,7 +446,7 @@ jobs:
       - name: Save CodeChecker HTML output.
         uses: actions/upload-artifact@v4
         with:
-          name: CodeChecker Bug Reports for ${{ matrix.sdk }}
+          name: CodeChecker Bug Reports for ${{ matrix.sdk }} ${{ matrix.mode }}
           path: ${{ steps.codechecker.outputs.result-html-dir }}/*.html
 
       - name: CodeChecker Pass Or Fail?