Skip to content

Add CUDA 12.8 images #488

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
chhwang merged 6 commits into from
Mar 29, 2025
Merged

Add CUDA 12.8 images #488

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
1cb5c72
Add CUDA 12.8 images
chhwang Mar 28, 2025
c6988fa
Update CodeQL to use CUDA 12.8
chhwang Mar 28, 2025
749f5c7
Build command for different archs
chhwang Mar 28, 2025
88563dc
fix
chhwang Mar 28, 2025
4248093
revert
chhwang Mar 28, 2025
0f22585
Merge branch 'main' into chhwang/cuda12.8
chhwang Mar 29, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 2 additions & 2 deletions .azure-pipelines/integration-test-rocm.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -78,7 +78,7 @@ jobs:
script: |
set -e
export PATH=/usr/local/mpi/bin:$PATH
sudo /usr/local/mpi/bin/mpirun --allow-run-as-root -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN -x LD_PRELOAD="$(pwd)/build/apps/nccl/libmscclpp_nccl.so" \
sudo mpirun --allow-run-as-root -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN -x LD_PRELOAD="$(pwd)/build/apps/nccl/libmscclpp_nccl.so" \
-x ALLREDUCE_SMALL_MSG_BOUNDARY=32K -x ALLREDUCE_LARGE_MSG_BOUNDARY=1M ./rccl-tests/build/all_reduce_perf -b 1K -e 1G -f 2 -d half -G 20 -w 10 -n 100
workingDirectory: '$(System.DefaultWorkingDirectory)'

Expand All @@ -90,7 +90,7 @@ jobs:
script: |
set -e
export PATH=/usr/local/mpi/bin:$PATH
sudo /usr/local/mpi/bin/mpirun -np 8 --bind-to numa --allow-run-as-root -x LD_PRELOAD=$(pwd)/build/apps/nccl/libmscclpp_nccl.so -x NCCL_DEBUG=WARN \
sudo mpirun -np 8 --bind-to numa --allow-run-as-root -x LD_PRELOAD=$(pwd)/build/apps/nccl/libmscclpp_nccl.so -x NCCL_DEBUG=WARN \
-x ALLREDUCEPKT_IP_JSON_FILE=./msccl-users/execution-files/allreduce_mi300_packet.json \
-x ALLREDUCE_IP_JSON_FILE=./msccl-users/execution-files/allreduce_mi300_sm_mscclpp.json \
-x ALLREDUCE_SMALL_MSG_BOUNDARY=32K -x ALLREDUCE_LARGE_MSG_BOUNDARY=1M ./rccl-tests/build/all_reduce_perf \
Expand Down
2 changes: 1 addition & 1 deletion .azure-pipelines/integration-test.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -45,7 +45,7 @@ jobs:
sudo apt-get update -y
sudo apt-get install pssh -y
curl -sL https://aka.ms/InstallAzureCLIDeb | sudo bash

- task: DownloadSecureFile@1
name: SshKeyFile
displayName: Download key file
Expand Down
2 changes: 1 addition & 1 deletion .github/workflows/codeql-analysis.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -28,7 +28,7 @@ jobs:
fail-fast: false
matrix:
language: [ 'cpp', 'python' ]
version: [ 'cuda11.8', 'cuda12.2' ]
version: [ 'cuda11.8', 'cuda12.8' ]

steps:
- name: Checkout repository
Expand Down
2 changes: 1 addition & 1 deletion .github/workflows/mscclpp-lang.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,7 +15,7 @@ jobs:
strategy:
fail-fast: false
matrix:
version: [ 'cuda11.8', 'cuda12.2' ]
version: [ 'cuda11.8', 'cuda12.8' ]

steps:
- uses: actions/checkout@v4
Expand Down
2 changes: 1 addition & 1 deletion docker/base-dev-x.dockerfile
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,7 +10,7 @@ RUN apt-get update && \
lcov \
vim \
&& \
apt-get autoremove && \
apt-get autoremove -y && \
apt-get clean && \
rm -rf /var/lib/apt/lists/* /tmp/*

Expand Down
43 changes: 18 additions & 25 deletions docker/base-x.dockerfile
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -25,41 +25,34 @@ RUN apt-get update && \
python3-setuptools \
python3-wheel \
sudo \
wget \
&& \
apt-get autoremove && \
apt-get clean && \
rm -rf /var/lib/apt/lists/* /tmp/*
wget

# Install OFED
ARG OFED_VERSION=5.2-2.2.3.0
RUN cd /tmp && \
ARCH=$(uname -m) && \
OS_VERSION=$(lsb_release -rs) && \
OS_VERSION=ubuntu${OS_VERSION} && \
wget -q https://content.mellanox.com/ofed/MLNX_OFED-${OFED_VERSION}/MLNX_OFED_LINUX-${OFED_VERSION}-${OS_VERSION}-x86_64.tgz && \
tar xzf MLNX_OFED_LINUX-${OFED_VERSION}-${OS_VERSION}-x86_64.tgz && \
MLNX_OFED_LINUX-${OFED_VERSION}-${OS_VERSION}-x86_64/mlnxofedinstall --user-space-only --without-fw-update --without-ucx-cuda --force --all && \
wget -q https://content.mellanox.com/ofed/MLNX_OFED-${OFED_VERSION}/MLNX_OFED_LINUX-${OFED_VERSION}-${OS_VERSION}-${ARCH}.tgz && \
tar xzf MLNX_OFED_LINUX-${OFED_VERSION}-${OS_VERSION}-${ARCH}.tgz && \
MLNX_OFED_LINUX-${OFED_VERSION}-${OS_VERSION}-${ARCH}/mlnxofedinstall --user-space-only --without-fw-update --without-ucx-cuda --force --all && \
rm -rf /tmp/MLNX_OFED_LINUX-${OFED_VERSION}*

# Install OpenMPI
ENV OPENMPI_VERSION=4.1.5
RUN cd /tmp && \
export ompi_v_parsed="$(echo ${OPENMPI_VERSION} | sed -E 's/^([0-9]+)\.([0-9]+)\..*/\1.\2/')" && \
wget -q https://download.open-mpi.org/release/open-mpi/v${ompi_v_parsed}/openmpi-${OPENMPI_VERSION}.tar.gz && \
tar xzf openmpi-${OPENMPI_VERSION}.tar.gz && \
cd openmpi-${OPENMPI_VERSION} && \
./configure --prefix=/usr/local/mpi && \
make -j && \
make install && \
cd .. && \
rm -rf /tmp/openmpi-${OPENMPI_VERSION}*
# Install OpenMPI (should be done after the OFED installation) & clean apt cache
RUN apt-get update && \
apt-get install -y --no-install-recommends \
libopenmpi-dev \
&& \
apt-get autoremove -y && \
apt-get clean && \
rm -rf /var/lib/apt/lists/* /tmp/*

ARG EXTRA_LD_PATH=/usr/local/cuda-12.1/compat:/usr/local/cuda-12.1/lib64
ENV PATH="/usr/local/mpi/bin:${PATH}" \
LD_LIBRARY_PATH="/usr/local/mpi/lib:${EXTRA_LD_PATH}:${LD_LIBRARY_PATH}"
# OpenMPI short link (for compatibility with old images)
RUN ln -s /usr/lib/x86_64-linux-gnu/openmpi /usr/local/mpi

RUN echo PATH="${PATH}" > /etc/environment && \
echo LD_LIBRARY_PATH="${LD_LIBRARY_PATH}" >> /etc/environment
ARG EXTRA_LD_PATH=
ENV LD_LIBRARY_PATH="${EXTRA_LD_PATH}:${LD_LIBRARY_PATH}"
RUN echo LD_LIBRARY_PATH="${LD_LIBRARY_PATH}" >> /etc/environment

ENTRYPOINT []
WORKDIR /
6 changes: 5 additions & 1 deletion docker/build.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -9,6 +9,7 @@ baseImageTable=(
["cuda12.2"]="nvidia/cuda:12.2.2-devel-ubuntu20.04"
["cuda12.3"]="nvidia/cuda:12.3.2-devel-ubuntu20.04"
["cuda12.4"]="nvidia/cuda:12.4.1-devel-ubuntu22.04"
["cuda12.8"]="nvidia/cuda:12.8.1-devel-ubuntu22.04"
["rocm6.2"]="rocm/rocm-terminal:6.2.1"
)

Expand All @@ -23,13 +24,14 @@ extraLdPathTable=(
declare -A ofedVersionTable
ofedVersionTable=(
["cuda12.4"]="23.07-0.5.1.2"
["cuda12.8"]="24.10-1.1.4.0"
)

GHCR="ghcr.io/microsoft/mscclpp/mscclpp"
TARGET=${1}

print_usage() {
echo "Usage: $0 [cuda11.8|cuda12.1|cuda12.2|cuda12.3|cuda12.4|rocm6.2]"
echo "Usage: $0 [cuda11.8|cuda12.1|cuda12.2|cuda12.3|cuda12.4|cuda12.8|rocm6.2]"
}

if [[ ! -v "baseImageTable[${TARGET}]" ]]; then
Expand Down Expand Up @@ -64,9 +66,11 @@ if [[ ${TARGET} == rocm* ]]; then
--build-arg EXTRA_LD_PATH=${extraLdPathTable[${TARGET}]} \
--build-arg TARGET=${TARGET} \
--build-arg ARCH="gfx942" .
docker rmi ${GHCR}-common:base-${TARGET}
else
echo "Building CUDA base image..."
docker tag ${GHCR}-common:base-${TARGET} ${GHCR}:base-${TARGET}
docker rmi --no-prune ${GHCR}-common:base-${TARGET}
fi

docker build -t ${GHCR}:base-dev-${TARGET} \
Expand Down
27 changes: 14 additions & 13 deletions test/deploy/run_tests.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,45 +1,46 @@
set -e
HOSTFILE=/root/mscclpp/test/deploy/hostfile_mpi
export PATH=/usr/local/mpi/bin:$PATH

function run_mscclpp_test()
{
echo "=================Run allgather_test_perf on 2 nodes========================="
/usr/local/mpi/bin/mpirun --allow-run-as-root -np 16 --bind-to numa -hostfile ${HOSTFILE} \
mpirun --allow-run-as-root -np 16 --bind-to numa -hostfile ${HOSTFILE} \
-x MSCCLPP_DEBUG=WARN -x LD_LIBRARY_PATH=/root/mscclpp/build:$LD_LIBRARY_PATH \
-npernode 8 /root/mscclpp/build/test/mscclpp-test/allgather_test_perf -b 1K -e 1G -f 2 -k 0 -o /root/mscclpp/output.jsonl

# For kernel 2, the message size must can be divided by 3
/usr/local/mpi/bin/mpirun --allow-run-as-root -np 16 --bind-to numa -hostfile ${HOSTFILE} \
mpirun --allow-run-as-root -np 16 --bind-to numa -hostfile ${HOSTFILE} \
-x MSCCLPP_DEBUG=WARN -x LD_LIBRARY_PATH=/root/mscclpp/build:$LD_LIBRARY_PATH \
-npernode 8 /root/mscclpp/build/test/mscclpp-test/allgather_test_perf -b 3K -e 3G -f 2 -k 2 -o /root/mscclpp/output.jsonl

/usr/local/mpi/bin/mpirun --allow-run-as-root -np 16 --bind-to numa -hostfile ${HOSTFILE} \
mpirun --allow-run-as-root -np 16 --bind-to numa -hostfile ${HOSTFILE} \
-x MSCCLPP_DEBUG=WARN -x LD_LIBRARY_PATH=/root/mscclpp/build:$LD_LIBRARY_PATH \
-npernode 8 /root/mscclpp/build/test/mscclpp-test/allgather_test_perf -b 1K -e 1G -f 2 -k 3 -o /root/mscclpp/output.jsonl

echo "==================Run allreduce_test_perf on 2 nodes========================="
/usr/local/mpi/bin/mpirun --allow-run-as-root -np 16 --bind-to numa -hostfile ${HOSTFILE} \
mpirun --allow-run-as-root -np 16 --bind-to numa -hostfile ${HOSTFILE} \
-x MSCCLPP_DEBUG=WARN -x LD_LIBRARY_PATH=/root/mscclpp/build:$LD_LIBRARY_PATH \
-npernode 8 /root/mscclpp/build/test/mscclpp-test/allreduce_test_perf -b 1K -e 1G -f 2 -k 0 -o /root/mscclpp/output.jsonl

/usr/local/mpi/bin/mpirun --allow-run-as-root -np 16 --bind-to numa -hostfile ${HOSTFILE} \
mpirun --allow-run-as-root -np 16 --bind-to numa -hostfile ${HOSTFILE} \
-x MSCCLPP_DEBUG=WARN -x LD_LIBRARY_PATH=/root/mscclpp/build:$LD_LIBRARY_PATH \
-npernode 8 /root/mscclpp/build/test/mscclpp-test/allreduce_test_perf -b 1K -e 1G -f 2 -k 1 -o /root/mscclpp/output.jsonl

/usr/local/mpi/bin/mpirun --allow-run-as-root -np 16 --bind-to numa -hostfile ${HOSTFILE} \
mpirun --allow-run-as-root -np 16 --bind-to numa -hostfile ${HOSTFILE} \
-x MSCCLPP_DEBUG=WARN -x LD_LIBRARY_PATH=/root/mscclpp/build:$LD_LIBRARY_PATH \
-npernode 8 /root/mscclpp/build/test/mscclpp-test/allreduce_test_perf -b 1K -e 1M -f 2 -k 2 -o /root/mscclpp/output.jsonl

/usr/local/mpi/bin/mpirun --allow-run-as-root -np 16 --bind-to numa -hostfile ${HOSTFILE} \
mpirun --allow-run-as-root -np 16 --bind-to numa -hostfile ${HOSTFILE} \
-x MSCCLPP_DEBUG=WARN -x LD_LIBRARY_PATH=/root/mscclpp/build:$LD_LIBRARY_PATH \
-npernode 8 /root/mscclpp/build/test/mscclpp-test/allreduce_test_perf -b 3K -e 3G -f 2 -k 3 -o /root/mscclpp/output.jsonl

/usr/local/mpi/bin/mpirun --allow-run-as-root -np 16 --bind-to numa -hostfile ${HOSTFILE} \
mpirun --allow-run-as-root -np 16 --bind-to numa -hostfile ${HOSTFILE} \
-x MSCCLPP_DEBUG=WARN -x LD_LIBRARY_PATH=/root/mscclpp/build:$LD_LIBRARY_PATH \
-npernode 8 /root/mscclpp/build/test/mscclpp-test/allreduce_test_perf -b 3K -e 3G -f 2 -k 4 -o /root/mscclpp/output.jsonl

echo "==================Run alltoall_test_perf on 2 nodes========================="
/usr/local/mpi/bin/mpirun --allow-run-as-root -np 16 --bind-to numa -hostfile ${HOSTFILE} \
mpirun --allow-run-as-root -np 16 --bind-to numa -hostfile ${HOSTFILE} \
-x MSCCLPP_DEBUG=WARN -x LD_LIBRARY_PATH=/root/mscclpp/build:$LD_LIBRARY_PATH \
-npernode 8 /root/mscclpp/build/test/mscclpp-test/alltoall_test_perf -b 1K -e 1G -f 2 -k 0 -o /root/mscclpp/output.jsonl

Expand All @@ -51,28 +52,28 @@ function run_mscclpp_test()
function run_mp_ut()
{
echo "============Run multi-process unit tests on 2 nodes (np=2, npernode=1)========================="
/usr/local/mpi/bin/mpirun -allow-run-as-root -tag-output -np 2 --bind-to numa \
mpirun -allow-run-as-root -tag-output -np 2 --bind-to numa \
-hostfile ${HOSTFILE} -x MSCCLPP_DEBUG=WARN -x LD_LIBRARY_PATH=/root/mscclpp/build:$LD_LIBRARY_PATH \
-npernode 1 /root/mscclpp/build/test/mp_unit_tests -ip_port mscclit-000000:20003

echo "============Run multi-process unit tests on 2 nodes (np=16, npernode=8)========================="
/usr/local/mpi/bin/mpirun -allow-run-as-root -tag-output -np 16 --bind-to numa \
mpirun -allow-run-as-root -tag-output -np 16 --bind-to numa \
-hostfile ${HOSTFILE} -x MSCCLPP_DEBUG=WARN -x LD_LIBRARY_PATH=/root/mscclpp/build:$LD_LIBRARY_PATH \
-npernode 8 /root/mscclpp/build/test/mp_unit_tests -ip_port mscclit-000000:20003
}

function run_pytests()
{
echo "==================Run python tests================================"
/usr/local/mpi/bin/mpirun -allow-run-as-root -tag-output -np 16 --bind-to numa \
mpirun -allow-run-as-root -tag-output -np 16 --bind-to numa \
-hostfile ${HOSTFILE} -x MSCCLPP_DEBUG=WARN -x LD_LIBRARY_PATH=/root/mscclpp/build:$LD_LIBRARY_PATH \
-x MSCCLPP_HOME=/root/mscclpp -npernode 8 bash /root/mscclpp/test/deploy/pytest.sh
}

function run_py_benchmark()
{
echo "==================Run python benchmark================================"
/usr/local/mpi/bin/mpirun -allow-run-as-root -np 16 --bind-to numa \
mpirun -allow-run-as-root -np 16 --bind-to numa \
-hostfile ${HOSTFILE} -x MSCCLPP_DEBUG=WARN -x LD_LIBRARY_PATH=/root/mscclpp/build:$LD_LIBRARY_PATH \
-mca pml ob1 -mca btl ^openib -mca btl_tcp_if_include eth0 -x NCCL_IB_PCI_RELAXED_ORDERING=1 -x NCCL_SOCKET_IFNAME=eth0 \
-x CUDA_DEVICE_ORDER=PCI_BUS_ID -x NCCL_NET_GDR_LEVEL=5 -x NCCL_TOPO_FILE=/opt/microsoft/ndv4-topo.xml \
Expand Down
Loading