0715ab3c10
This PR adds optional support for loading nccl with `dlopen` as an alternative of compile time linking. This is to address the size bloat issue with the PyPI binary release. - Add CMake option to load `nccl` at runtime. - Add an NCCL stub. After this, `nccl` will be fetched from PyPI when using pip to install XGBoost, either by a user or by `pyproject.toml`. Others who want to link the nccl at compile time can continue to do so without any change. At the moment, this is Linux only since we only support MNMG on Linux.
25 lines
849 B
Bash
Executable File
25 lines
849 B
Bash
Executable File
#!/bin/bash
|
|
|
|
set -euo pipefail
|
|
|
|
source tests/buildkite/conftest.sh
|
|
|
|
echo "--- Run Google Tests with CUDA, using a GPU"
|
|
buildkite-agent artifact download "build/testxgboost" . --step build-cuda
|
|
chmod +x build/testxgboost
|
|
tests/ci_build/ci_build.sh gpu nvidia-docker \
|
|
--build-arg CUDA_VERSION_ARG=$CUDA_VERSION \
|
|
--build-arg RAPIDS_VERSION_ARG=$RAPIDS_VERSION \
|
|
--build-arg NCCL_VERSION_ARG=$NCCL_VERSION \
|
|
build/testxgboost
|
|
|
|
echo "--- Run Google Tests with CUDA, using a GPU, RMM enabled"
|
|
rm -rfv build/
|
|
buildkite-agent artifact download "build/testxgboost" . --step build-cuda-with-rmm
|
|
chmod +x build/testxgboost
|
|
tests/ci_build/ci_build.sh gpu nvidia-docker \
|
|
--build-arg CUDA_VERSION_ARG=$CUDA_VERSION \
|
|
--build-arg RAPIDS_VERSION_ARG=$RAPIDS_VERSION \
|
|
--build-arg NCCL_VERSION_ARG=$NCCL_VERSION \
|
|
build/testxgboost --use-rmm-pool
|