Sync to upstream nccl-test master-20251216

.github/workflows/pre-check.yml

@@ -0,0 +1,29 @@
+on:
+  pull_request:
+  workflow_dispatch:
+
+jobs:
+  build-only:
+    runs-on: ubuntu-22.04
+    steps:
+      - uses: actions/checkout@v4
+      - uses: docker/setup-buildx-action@v3
+      - name: Free disk space
+        run: |
+          sudo rm -rf /usr/share/dotnet
+          sudo rm -rf /usr/local/lib/android
+          sudo rm -rf /opt/ghc
+          sudo rm -rf /opt/hostedtoolcache
+          sudo docker system prune -af || true
+          df -h
+      - name: Build run package
+        run: |
+          docker buildx build \
+            -f docker/Dockerfile.cuda13.x.ubuntu22.04 \
+            --platform linux/amd64 \
+            --target package \
+            --output type=local,dest=dist \
+            .
+      - name: List artifacts
+        run: |
+          ls -lh dist

.github/workflows/release.yml

@@ -0,0 +1,61 @@
+name: release
+
+on:
+  push:
+    tags:
+      - "v*"
+
+permissions:
+  contents: write
+
+jobs:
+  build-and-release:
+    name: build-run (${{ matrix.name }})
+    runs-on: ubuntu-22.04
+
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - name: cuda13-ubuntu22.04
+            dockerfile: docker/Dockerfile.cuda13.x.ubuntu22.04
+
+          # - name: cuda12-ubuntu20.04
+          #   dockerfile: docker/Dockerfile.cuda12.x.ubuntu20.04
+
+    steps:
+      - name: Checkout source
+        uses: actions/checkout@v4
+
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+
+      - name: Free disk space
+        run: |
+          sudo rm -rf /usr/share/dotnet
+          sudo rm -rf /usr/local/lib/android
+          sudo rm -rf /opt/ghc
+          sudo rm -rf /opt/hostedtoolcache
+          sudo docker system prune -af || true
+          df -h
+
+      - name: Build run package
+        run: |
+          BUILD_DATE=$(date +%Y%m%d)
+
+          docker buildx build \
+            --platform linux/amd64 \
+            --progress=plain \
+            -f ${{ matrix.dockerfile }} \
+            --build-arg BUILD_DATE=${BUILD_DATE} \
+            --output type=local,dest=dist/${{ matrix.name }} \
+            .
+
+          echo "Produced files:"
+          ls -lh dist/${{ matrix.name }}
+
+      - name: Upload run to GitHub Release
+        uses: softprops/action-gh-release@v2
+        with:
+          files: |
+            dist/${{ matrix.name }}/*.run

docker/Dockerfile.cuda12.x.ubuntu20.04

@@ -0,0 +1,154 @@
+###########################
+# Build-time configuration
+###########################
+
+# Base OS and CUDA versions
+ARG UBUNTU_VERSION=20.04
+ARG CUDA_VERSION=12.8.1
+ARG CUDART_VERSION=12.8.90
+ARG CUDART_MAJOR_VERSION=12
+
+# NCCL versions
+ARG NCCL_PACKAGE_VERSION=2.27.7-1+cuda12.4
+ARG NCCL_SO_VERSION=2.27.7
+
+# OpenMPI versions
+# - MPI_VERSION: full OpenMPI version
+# - MPI_SERIES: major.minor series used in download URL
+ARG MPI_VERSION=4.1.8
+ARG MPI_SERIES=4.1
+
+# Build date (override at build time)
+ARG BUILD_DATE=20251221
+
+###########################
+# Build Stage
+###########################
+FROM nvidia/cuda:${CUDA_VERSION}-devel-ubuntu20.04 AS build
+
+# Re-declare build args for this stage (values are inherited)
+ARG UBUNTU_VERSION
+ARG CUDA_VERSION
+ARG CUDART_VERSION
+ARG NCCL_PACKAGE_VERSION
+ARG NCCL_SO_VERSION
+ARG MPI_VERSION
+ARG MPI_SERIES
+ARG BUILD_DATE
+
+ENV DEBIAN_FRONTEND=noninteractive
+WORKDIR /workspace
+
+# -------------------------
+# 1. Base build dependencies
+# -------------------------
+RUN mv /etc/apt/sources.list.d/cuda*.list /tmp/disabled-cuda.list || true && \
+    { apt-get -o Acquire::http::No-Cache=true update > build.log 2>&1 && \
+    apt-get install -y --no-install-recommends \
+        build-essential gcc g++ curl git wget ca-certificates \
+        make automake autoconf libtool pkg-config \
+        python3 python3-pip gzip xz-utils >> build.log 2>&1 && \
+    rm -rf /var/lib/apt/lists/* && rm -f build.log; } || (cat build.log && false)
+
+# -------------------------
+# 2. Install CUDA keyring and restore NVIDIA repository
+# -------------------------
+RUN { wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/cuda-keyring_1.1-1_all.deb > build.log 2>&1 && \
+    dpkg -i cuda-keyring_1.1-1_all.deb >> build.log 2>&1 && \
+    apt-get update >> build.log 2>&1 && rm -f build.log; } || (cat build.log && false)
+
+# -------------------------
+# 3. Install NCCL (pinned version)
+# -------------------------
+RUN apt-mark unhold libnccl2 libnccl-dev || true && \
+    { apt-get install -y --no-install-recommends \
+    libnccl2=${NCCL_PACKAGE_VERSION} \
+    libnccl-dev=${NCCL_PACKAGE_VERSION} > build.log 2>&1 && \
+    apt-mark hold libnccl2 libnccl-dev >> build.log 2>&1 && \
+    ldconfig >> build.log 2>&1 && \
+    rm -rf /var/lib/apt/lists/* && rm -f build.log; } || (cat build.log && false)
+
+# -------------------------
+# 4. Build OpenMPI from source
+# -------------------------
+RUN wget https://download.open-mpi.org/release/open-mpi/v${MPI_SERIES}/openmpi-${MPI_VERSION}.tar.gz > /dev/null 2>&1 && \
+    tar zxvf openmpi-${MPI_VERSION}.tar.gz > /dev/null 2>&1 && \
+    cd openmpi-${MPI_VERSION} && \
+    ./configure --prefix=/usr/local/sihpc --with-cuda=/usr/local/cuda > /dev/null  2>&1 && \
+    make -j$(nproc) > /dev/null 2>&1 && make install > /dev/null 2>&1 && \
+    rm -rf /workspace/openmpi-${MPI_VERSION} /workspace/openmpi-${MPI_VERSION}.tar.gz
+
+# -------------------------
+# 5. Build nccl-tests
+# -------------------------
+RUN { git clone --depth 1 --single-branch -b sicl https://github.com/scitix/nccl-tests.git > build.log 2>&1 && \
+    cd nccl-tests && \
+    make MPI=1 MPI_HOME=/usr/local/sihpc > build.log 2>&1 && rm -f build.log || (cat build.log && false); } && \
+    mkdir -p /usr/local/sihpc/libexec/nccl-tests && \
+    cp -rf build/*_perf /usr/local/sihpc/libexec/nccl-tests/ && \
+    mkdir -p /usr/local/sihpc/bin && \
+    cp scripts/nccl_perf /usr/local/sihpc/bin/nccl_perf && \
+    cp scripts/nccl_test /usr/local/sihpc/libexec/nccl-tests/nccl_test && \
+    cp scripts/env.sh /usr/local/sihpc/env.sh && \
+    cp scripts/install_sihpc /usr/local/sihpc/bin/install_sihpc && \
+    cp scripts/uninstall_sihpc /usr/local/sihpc/bin/uninstall_sihpc && \
+    rm -rf /workspace/nccl-tests
+
+# -------------------------
+# 6. Collect runtime libraries (strict selection)
+# -------------------------
+RUN { set -e && \
+    mkdir -p /usr/local/sihpc/lib > build.log 2>&1 && \
+    cp /usr/local/cuda/lib64/libcudart* /usr/local/sihpc/lib/ >> build.log 2>&1 && \
+    cp /usr/lib/x86_64-linux-gnu/libnccl.so* /usr/local/sihpc/lib/ >> build.log 2>&1 && \
+    rm -f build.log; } || (cat build.log && false)
+    # cp /lib/x86_64-linux-gnu/libltdl.so.7.3.1 /usr/local/sihpc/lib/ && \
+    # cp /usr/lib/x86_64-linux-gnu/libhwloc.so* /usr/local/sihpc/lib/ && \
+    # cp /usr/lib/x86_64-linux-gnu/libevent_core* /usr/local/sihpc/lib/ && \
+    # cp /usr/lib/x86_64-linux-gnu/libevent_pthreads* /usr/local/sihpc/lib/
+
+# -------------------------
+# 7. Fix library symlinks
+# -------------------------
+RUN cd /usr/local/sihpc/lib && \
+    rm -f libcudart.so libcudart.so.12 && \
+    ln -sf libnccl.so.2.27.7 libnccl.so.2 && \
+    ln -sf libnccl.so.2 libnccl.so && \
+    ln -sf libcudart.so.12.8.90 libcudart.so.12 && \
+    ln -sf libcudart.so.12 libcudart.so
+    # rm -f libevent_core-2.1.so.7 && \
+    # ln -sf libhwloc.so.15.1.0 libhwloc.so.15 && \
+    # ln -sf libhwloc.so.15.1.0 libhwloc.so && \
+    # ln -sf libevent_core-2.1.so.7.0.0 libevent_core-2.1.so.7 && \
+    # ln -sf libevent_core-2.1.so.7 libevent_core-2.1.so && \
+    # ln -sf libevent_pthreads-2.1.so.7.0.0 libevent_pthreads-2.1.so.7 && \
+    # ln -sf libevent_pthreads-2.1.so.7 libevent_pthreads-2.1.so && \
+    # ln -sf libltdl.so.7.3.1 libltdl.so.7 && \
+    # ln -sf libltdl.so.7 libltdl.so
+
+###########################
+# Package Stage
+###########################
+FROM ubuntu:20.04 AS package
+
+# Re-declare args for this stage (values are inherited)
+ARG UBUNTU_VERSION
+ARG NCCL_PACKAGE_VERSION
+ARG MPI_VERSION
+ARG BUILD_DATE
+
+# Expose versions/date as environment variables for runtime shell expansion
+ENV NCCL_PACKAGE_VERSION=${NCCL_PACKAGE_VERSION} \
+    MPI_VERSION=${MPI_VERSION} \
+    BUILD_DATE=${BUILD_DATE}
+
+COPY --from=build /usr/local/sihpc /usr/local/sihpc
+
+WORKDIR /
+RUN apt-get update && apt-get install -y --no-install-recommends makeself && \
+    SAFE_NCCL_PKG=$(printf '%s\n' "${NCCL_PACKAGE_VERSION}" | tr '+' '-') && \
+    PACKAGE_FILENAME="sicl-nccl${SAFE_NCCL_PKG}-ompi${MPI_VERSION}-ubuntu${UBUNTU_VERSION}-${BUILD_DATE}.run" && \
+    { makeself --gzip /usr/local/sihpc \
+    "${PACKAGE_FILENAME}" \
+    "SiHPC MPI + NCCL + NCCL-tests Portable Installer" \
+      ./bin/install_sihpc > build.log 2>&1 && rm -f build.log; } || (cat build.log && false)