introduce libcugraph wheels (#4804)

Replaces #4340, contributes to rapidsai/build-planning#33. Proposes packaging `libcugraph` as a wheel, which is then re-used by `cugraph-cu{11,12}` and `pylibcugraph-cu{11,12}` wheels. ## Notes for Reviewers ### Benefits of these changes * smaller wheels (see "Size Changes" below) - *no more `pylibcugraph` and `cugraph` both holding copies of libcugraph.so* * faster compile times - *no more re-compiling RAFT, thanks to rapidsai/raft#2531 - *no more recompiling libcugraph.so in both `pylibcugraph` and `cugraph` wheel builds* * other benefits mentioned in rapidsai/build-planning#33 ### Wheel contents `libcugraph`: * `libcugraph.so` (shared library) * cuGraph headers * vendored dependencies (`fmt`, `spdlog`, CCCL, `cuco`) `pylibcugraph`: * `pylibcugraph` Python / Cython code and compiled Cython extensions `cugraph`: * `cugraph` Python / Cython code and compiled Cython extension ### Dependency Flows In short.... `libcugraph` contains `libcugraph.so` and `libcugraph_c.so` dynamic libraries and the headers to link against it. * Anything that needs to link against cuGraph at build time pulls in `libcugraph` wheels as a build dependency. * Anything that needs cuGraph's symbols at runtime pulls it in as a runtime dependency, and calls `libcugraph.load_library()`. For more details and some flowcharts, see rapidsai/build-planning#33 (comment) ### Size changes (CUDA 12, Python 3.12, x86_64) | wheel | num files (before) | num files (this PR) | size (before) | size (this PR) | |:---------------:|------------------:|-----------------:|--------------:|-------------:| | `libcugraph` | --- | 1762 | --- | 903M | | `pylibcugraph` | 190 | 187 | 901M | 2M | | `cugraph` | 315 | 313 | 899M | 3M | |**TOTAL** | **505** | **2,262** | **1,800M** | **908M** | *NOTES: size = compressed, "before" = 2025-01-13 nightlies* *This is a cuGraph-specific slice of the table from rapidsai/raft#2531. See that PR for details.* ### How I tested this These other PRs: * rapidsai/devcontainers#435 * rapidsai/cugraph-gnn#110 Authors: - James Lamb (https://github.com/jameslamb) - Ralph Liu (https://github.com/nv-rliu) - Bradley Dice (https://github.com/bdice) Approvers: - Brad Rees (https://github.com/BradReesWork) - Bradley Dice (https://github.com/bdice) URL: #4804
rapidsai · Jan 18, 2025 · 8ebff3b · 8ebff3b
1 parent 2848cd4
commit 8ebff3b
Show file tree

Hide file tree

Showing 28 changed files with 558 additions and 241 deletions.
diff --git a/.github/workflows/build.yaml b/.github/workflows/build.yaml
@@ -67,7 +67,29 @@ jobs:
       node_type: "gpu-v100-latest-1"
       run_script: "ci/build_docs.sh"
       sha: ${{ inputs.sha }}
+  wheel-build-libcugraph:
+    secrets: inherit
+    uses: rapidsai/shared-workflows/.github/workflows/wheels-build.yaml@branch-25.02
+    with:
+      build_type: ${{ inputs.build_type || 'branch' }}
+      branch: ${{ inputs.branch }}
+      sha: ${{ inputs.sha }}
+      date: ${{ inputs.date }}
+      script: ci/build_wheel_libcugraph.sh
+      node_type: cpu32
+  wheel-publish-libcugraph:
+    needs: wheel-build-libcugraph
+    secrets: inherit
+    uses: rapidsai/shared-workflows/.github/workflows/wheels-publish.yaml@branch-25.02
+    with:
+      build_type: ${{ inputs.build_type || 'branch' }}
+      branch: ${{ inputs.branch }}
+      sha: ${{ inputs.sha }}
+      date: ${{ inputs.date }}
+      package-name: libcugraph
+      package-type: cpp
   wheel-build-pylibcugraph:
+    needs: wheel-build-libcugraph
     secrets: inherit
     uses: rapidsai/shared-workflows/.github/workflows/wheels-build.yaml@branch-25.02
     with:
@@ -76,7 +98,6 @@ jobs:
       sha: ${{ inputs.sha }}
       date: ${{ inputs.date }}
       script: ci/build_wheel_pylibcugraph.sh
-      node_type: cpu32
   wheel-publish-pylibcugraph:
     needs: wheel-build-pylibcugraph
     secrets: inherit
@@ -87,8 +108,9 @@ jobs:
       sha: ${{ inputs.sha }}
       date: ${{ inputs.date }}
       package-name: pylibcugraph
+      package-type: python
   wheel-build-cugraph:
-    needs: wheel-publish-pylibcugraph
+    needs: wheel-build-pylibcugraph
     secrets: inherit
     uses: rapidsai/shared-workflows/.github/workflows/wheels-build.yaml@branch-25.02
     with:
@@ -107,3 +129,4 @@ jobs:
       sha: ${{ inputs.sha }}
       date: ${{ inputs.date }}
       package-name: cugraph
+      package-type: python
diff --git a/.github/workflows/pr.yaml b/.github/workflows/pr.yaml
@@ -23,6 +23,7 @@ jobs:
       - conda-python-build
       - conda-python-tests
       - docs-build
+      - wheel-build-libcugraph
       - wheel-build-pylibcugraph
       - wheel-tests-pylibcugraph
       - wheel-build-cugraph
@@ -149,14 +150,22 @@ jobs:
       arch: "amd64"
       container_image: "rapidsai/ci-conda:cuda11.8.0-ubuntu22.04-py3.10"
       run_script: "ci/build_docs.sh"
-  wheel-build-pylibcugraph:
+  wheel-build-libcugraph:
     needs: checks
     secrets: inherit
     uses: rapidsai/shared-workflows/.github/workflows/wheels-build.yaml@branch-25.02
+    with:
+      # build for every combination of arch and CUDA version, but only for the latest Python
+      matrix_filter: group_by([.ARCH, (.CUDA_VER|split(".")|map(tonumber)|.[0])]) | map(max_by(.PY_VER|split(".")|map(tonumber)))
+      build_type: pull-request
+      script: ci/build_wheel_libcugraph.sh
+  wheel-build-pylibcugraph:
+    needs: wheel-build-libcugraph
+    secrets: inherit
+    uses: rapidsai/shared-workflows/.github/workflows/wheels-build.yaml@branch-25.02
     with:
       build_type: pull-request
       script: ci/build_wheel_pylibcugraph.sh
-      node_type: cpu32
   wheel-tests-pylibcugraph:
     needs: [wheel-build-pylibcugraph, changed-files]
     secrets: inherit

diff --git a/.pre-commit-config.yaml b/.pre-commit-config.yaml
@@ -59,7 +59,7 @@ repos:
       - id: sphinx-lint
         args: ["--enable=all", "--disable=line-too-long"]
   - repo: https://github.com/rapidsai/dependency-file-generator
-    rev: v1.16.0
+    rev: v1.17.0
     hooks:
         - id: rapids-dependency-file-generator
           args: ["--clean"]
diff --git a/build.sh b/build.sh
@@ -1,6 +1,6 @@
 #!/bin/bash
 
-# Copyright (c) 2019-2024, NVIDIA CORPORATION.
+# Copyright (c) 2019-2025, NVIDIA CORPORATION.
 
 # cugraph build script
 
@@ -175,11 +175,6 @@ SKBUILD_EXTRA_CMAKE_ARGS="${EXTRA_CMAKE_ARGS}"
 # Replace spaces with semicolons in SKBUILD_EXTRA_CMAKE_ARGS
 SKBUILD_EXTRA_CMAKE_ARGS=$(echo ${SKBUILD_EXTRA_CMAKE_ARGS} | sed 's/ /;/g')
 
-# Append `-DFIND_CUGRAPH_CPP=ON` to EXTRA_CMAKE_ARGS unless a user specified the option.
-if [[ "${EXTRA_CMAKE_ARGS}" != *"DFIND_CUGRAPH_CPP"* ]]; then
-    SKBUILD_EXTRA_CMAKE_ARGS="${SKBUILD_EXTRA_CMAKE_ARGS};-DFIND_CUGRAPH_CPP=ON"
-fi
-
 # If clean or uninstall targets given, run them prior to any other steps
 if hasArg uninstall; then
     if [[ "$INSTALL_PREFIX" != "" ]]; then

diff --git a/ci/build_wheel.sh b/ci/build_wheel.sh
@@ -1,10 +1,11 @@
 #!/bin/bash
-# Copyright (c) 2023-2024, NVIDIA CORPORATION.
+# Copyright (c) 2023-2025, NVIDIA CORPORATION.
 
 set -euo pipefail
 
 package_name=$1
 package_dir=$2
+package_type=$3
 
 source rapids-configure-sccache
 source rapids-date-string
@@ -29,22 +30,31 @@ python -m pip wheel \
 
 sccache --show-adv-stats
 
-case "${RAPIDS_CUDA_VERSION}" in
-    12.*)
-        EXCLUDE_ARGS=(
-            --exclude "libcublas.so.12"
-            --exclude "libcublasLt.so.12"
-            --exclude "libcurand.so.10"
-            --exclude "libcusolver.so.11"
-            --exclude "libcusparse.so.12"
-            --exclude "libnvJitLink.so.12"
-        )
-    ;;
-    11.*)
-        EXCLUDE_ARGS=()
-    ;;
-esac
+EXCLUDE_ARGS=(
+  --exclude "libraft.so"
+)
+
+# Avoid picking up dependencies on CUDA wheels that come through
+# transitively from 'libraft'.
+#
+# 'libraft' wheels are responsible for carrying a runtime dependency on
+# these based on RAFT's needs.
+EXCLUDE_ARGS+=(
+  --exclude "libcublas.so.12"
+  --exclude "libcublasLt.so.12"
+  --exclude "libcurand.so.10"
+  --exclude "libcusolver.so.11"
+  --exclude "libcusparse.so.12"
+  --exclude "libnvJitLink.so.12"
+)
+
+if [[ "${package_dir}" != "python/libcugraph" ]]; then
+    EXCLUDE_ARGS+=(
+      --exclude "libcugraph_c.so"
+      --exclude "libcugraph.so"
+    )
+fi
 
 mkdir -p final_dist
 python -m auditwheel repair -w final_dist "${EXCLUDE_ARGS[@]}" dist/*
-RAPIDS_PY_WHEEL_NAME="${package_name}_${RAPIDS_PY_CUDA_SUFFIX}" rapids-upload-wheels-to-s3 python final_dist
+RAPIDS_PY_WHEEL_NAME="${package_name}_${RAPIDS_PY_CUDA_SUFFIX}" rapids-upload-wheels-to-s3 "${package_type}" final_dist
diff --git a/ci/build_wheel_cugraph.sh b/ci/build_wheel_cugraph.sh
@@ -1,37 +1,25 @@
 #!/bin/bash
-# Copyright (c) 2023-2024, NVIDIA CORPORATION.
+# Copyright (c) 2023-2025, NVIDIA CORPORATION.
 
 set -euo pipefail
 
 package_dir="python/cugraph"
 
 RAPIDS_PY_CUDA_SUFFIX="$(rapids-wheel-ctk-name-gen ${RAPIDS_CUDA_VERSION})"
 
-# Download the pylibcugraph wheel built in the previous step and make it
+# Download the libcugraph and pylibcugraph wheels built in the previous step and make them
 # available for pip to find.
-#
-# ensure 'cugraph' wheel builds always use the 'pylibcugraph' just built in the same CI run
-#
-# using env variable PIP_CONSTRAINT is necessary to ensure the constraints
-# are used when creating the isolated build environment
-CPP_WHEELHOUSE=$(RAPIDS_PY_WHEEL_NAME="pylibcugraph_${RAPIDS_PY_CUDA_SUFFIX}" rapids-download-wheels-from-s3 /tmp/pylibcugraph_dist)
+LIBCUGRAPH_WHEELHOUSE=$(RAPIDS_PY_WHEEL_NAME="libcugraph_${RAPIDS_PY_CUDA_SUFFIX}" rapids-download-wheels-from-s3 cpp /tmp/libcugraph_dist)
+PYLIBCUGRAPH_WHEELHOUSE=$(RAPIDS_PY_WHEEL_NAME="pylibcugraph_${RAPIDS_PY_CUDA_SUFFIX}" rapids-download-wheels-from-s3 python /tmp/pylibcugraph_dist)
 
-echo "pylibcugraph-${RAPIDS_PY_CUDA_SUFFIX} @ file://$(echo ${CPP_WHEELHOUSE}/pylibcugraph_*.whl)" > ./constraints.txt
-export PIP_CONSTRAINT="${PWD}/constraints.txt"
-
-PARALLEL_LEVEL=$(python -c \
-  "from math import ceil; from multiprocessing import cpu_count; print(ceil(cpu_count()/4))")
-case "${RAPIDS_CUDA_VERSION}" in
-  12.*)
-    EXTRA_CMAKE_ARGS=";-DUSE_CUDA_MATH_WHEELS=ON"
-  ;;
-  11.*)
-    EXTRA_CMAKE_ARGS=";-DUSE_CUDA_MATH_WHEELS=OFF"
-  ;;
-esac
+cat >> ./constraints.txt <<EOF
+libcugraph-${RAPIDS_PY_CUDA_SUFFIX} @ file://$(echo ${LIBCUGRAPH_WHEELHOUSE}/libcugraph_*.whl)
+pylibcugraph-${RAPIDS_PY_CUDA_SUFFIX} @ file://$(echo ${PYLIBCUGRAPH_WHEELHOUSE}/pylibcugraph_*.whl)
+EOF
 
-export SKBUILD_CMAKE_ARGS="-DDETECT_CONDA_ENV=OFF;-DFIND_CUGRAPH_CPP=OFF${EXTRA_CMAKE_ARGS}"
-export SKBUILD_BUILD_TOOL_ARGS="-j${PARALLEL_LEVEL};-l${PARALLEL_LEVEL}"
+# Using env variable PIP_CONSTRAINT is necessary to ensure the constraints
+# are used when creating the isolated build environment.
+export PIP_CONSTRAINT="${PWD}/constraints.txt"
 
-./ci/build_wheel.sh cugraph ${package_dir}
+./ci/build_wheel.sh cugraph ${package_dir} python
 ./ci/validate_wheel.sh ${package_dir} final_dist
diff --git a/ci/build_wheel_libcugraph.sh b/ci/build_wheel_libcugraph.sh
@@ -0,0 +1,34 @@
+#!/bin/bash
+# Copyright (c) 2025, NVIDIA CORPORATION.
+
+set -euo pipefail
+
+package_name="libcugraph"
+package_dir="python/libcugraph"
+
+rapids-logger "Generating build requirements"
+matrix_selectors="cuda=${RAPIDS_CUDA_VERSION%.*};arch=$(arch);py=${RAPIDS_PY_VERSION};cuda_suffixed=true"
+
+rapids-dependency-file-generator \
+  --output requirements \
+  --file-key "py_build_${package_name}" \
+  --file-key "py_rapids_build_${package_name}" \
+  --matrix "${matrix_selectors}" \
+| tee /tmp/requirements-build.txt
+
+rapids-logger "Installing build requirements"
+python -m pip install \
+    -v \
+    --prefer-binary \
+    -r /tmp/requirements-build.txt
+
+rapids-logger "Done build requirements"
+
+# build with '--no-build-isolation', for better sccache hit rate
+# 0 really means "add --no-build-isolation" (ref: https://github.com/pypa/pip/issues/5735)
+export PIP_NO_BUILD_ISOLATION=0
+
+RAPIDS_PY_CUDA_SUFFIX="$(rapids-wheel-ctk-name-gen ${RAPIDS_CUDA_VERSION})"
+
+./ci/build_wheel.sh libcugraph ${package_dir} cpp
+./ci/validate_wheel.sh ${package_dir} final_dist
diff --git a/ci/build_wheel_pylibcugraph.sh b/ci/build_wheel_pylibcugraph.sh
@@ -1,24 +1,23 @@
 #!/bin/bash
-# Copyright (c) 2023-2024, NVIDIA CORPORATION.
+# Copyright (c) 2023-2025, NVIDIA CORPORATION.
 
 set -euo pipefail
 
 package_dir="python/pylibcugraph"
 
-PARALLEL_LEVEL=$(python -c \
-  "from math import ceil; from multiprocessing import cpu_count; print(ceil(cpu_count()/4))")
+RAPIDS_PY_CUDA_SUFFIX="$(rapids-wheel-ctk-name-gen ${RAPIDS_CUDA_VERSION})"
 
-case "${RAPIDS_CUDA_VERSION}" in
-  12.*)
-    EXTRA_CMAKE_ARGS=";-DUSE_CUDA_MATH_WHEELS=ON"
-  ;;
-  11.*)
-    EXTRA_CMAKE_ARGS=";-DUSE_CUDA_MATH_WHEELS=OFF"
-  ;;
-esac
+# Download the libcugraph wheel built in the previous step and make it
+# available for pip to find.
+LIBCUGRAPH_WHEELHOUSE=$(RAPIDS_PY_WHEEL_NAME="libcugraph_${RAPIDS_PY_CUDA_SUFFIX}" rapids-download-wheels-from-s3 cpp /tmp/libcugraph_dist)
 
-export SKBUILD_CMAKE_ARGS="-DDETECT_CONDA_ENV=OFF;-DFIND_CUGRAPH_CPP=OFF${EXTRA_CMAKE_ARGS}"
-export SKBUILD_BUILD_TOOL_ARGS="-j${PARALLEL_LEVEL};-l${PARALLEL_LEVEL}"
+cat >> ./constraints.txt <<EOF
+libcugraph-${RAPIDS_PY_CUDA_SUFFIX} @ file://$(echo ${LIBCUGRAPH_WHEELHOUSE}/libcugraph_*.whl)
+EOF
 
-./ci/build_wheel.sh pylibcugraph ${package_dir}
+# Using env variable PIP_CONSTRAINT is necessary to ensure the constraints
+# are used when creating the isolated build environment.
+export PIP_CONSTRAINT="${PWD}/constraints.txt"
+
+./ci/build_wheel.sh pylibcugraph ${package_dir} python
 ./ci/validate_wheel.sh ${package_dir} final_dist
diff --git a/ci/release/update-version.sh b/ci/release/update-version.sh
@@ -1,5 +1,5 @@
 #!/bin/bash
-# Copyright (c) 2018-2024, NVIDIA CORPORATION.
+# Copyright (c) 2018-2025, NVIDIA CORPORATION.
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at
@@ -57,6 +57,7 @@ DEPENDENCIES=(
   dask-cuda
   dask-cudf
   libcudf
+  libcugraph
   libraft
   librmm
   pylibcugraph
@@ -109,4 +110,3 @@ find .devcontainer/ -type f -name devcontainer.json -print0 | while IFS= read -r
 done
 
 sed_runner "s/:[0-9][0-9]\.[0-9][0-9]/:${NEXT_SHORT_TAG}/" ./notebooks/README.md
-sed_runner "s/branch-[0-9][0-9].[0-9][0-9]/branch-${NEXT_SHORT_TAG}/" ./docs/cugraph/source/nx_cugraph/nx_cugraph.md
diff --git a/ci/test_wheel_cugraph.sh b/ci/test_wheel_cugraph.sh
@@ -7,11 +7,13 @@ set -eoxu pipefail
 mkdir -p ./dist
 RAPIDS_PY_CUDA_SUFFIX="$(rapids-wheel-ctk-name-gen ${RAPIDS_CUDA_VERSION})"
 RAPIDS_PY_WHEEL_NAME="cugraph_${RAPIDS_PY_CUDA_SUFFIX}" rapids-download-wheels-from-s3 python ./dist
+RAPIDS_PY_WHEEL_NAME="libcugraph_${RAPIDS_PY_CUDA_SUFFIX}" rapids-download-wheels-from-s3 cpp ./local-libcugraph-dep
 RAPIDS_PY_WHEEL_NAME="pylibcugraph_${RAPIDS_PY_CUDA_SUFFIX}" rapids-download-wheels-from-s3 python ./local-pylibcugraph-dep
 
 # echo to expand wildcard before adding `[extra]` requires for pip
 python -m pip install \
     "$(echo ./dist/cugraph*.whl)[test]" \
-    ./local-pylibcugraph-dep/pylibcugraph*.whl
+    ./local-pylibcugraph-dep/pylibcugraph*.whl \
+    ./local-libcugraph-dep/libcugraph*.whl
 
 ./ci/test_wheel.sh cugraph
diff --git a/ci/test_wheel_pylibcugraph.sh b/ci/test_wheel_pylibcugraph.sh
@@ -6,10 +6,12 @@ set -eoxu pipefail
 # Download the packages built in the previous step
 mkdir -p ./dist
 RAPIDS_PY_CUDA_SUFFIX="$(rapids-wheel-ctk-name-gen ${RAPIDS_CUDA_VERSION})"
+RAPIDS_PY_WHEEL_NAME="libcugraph_${RAPIDS_PY_CUDA_SUFFIX}" rapids-download-wheels-from-s3 cpp ./local-libcugraph-dep
 RAPIDS_PY_WHEEL_NAME="pylibcugraph_${RAPIDS_PY_CUDA_SUFFIX}" rapids-download-wheels-from-s3 python ./dist
 
 # echo to expand wildcard before adding `[extra]` requires for pip
 python -m pip install \
-    "$(echo ./dist/pylibcugraph*.whl)[test]"
+    "$(echo ./dist/pylibcugraph*.whl)[test]" \
+    ./local-libcugraph-dep/libcugraph*.whl
 
 ./ci/test_wheel.sh pylibcugraph
diff --git a/ci/validate_wheel.sh b/ci/validate_wheel.sh
@@ -1,35 +1,17 @@
 #!/bin/bash
-# Copyright (c) 2024, NVIDIA CORPORATION.
+# Copyright (c) 2024-2025, NVIDIA CORPORATION.
 
 set -euo pipefail
 
 package_dir=$1
 wheel_dir_relative_path=$2
 
-RAPIDS_CUDA_MAJOR="${RAPIDS_CUDA_VERSION%%.*}"
-
-# some packages are much larger on CUDA 11 than on CUDA 12
-if [[ "${package_dir}" == "python/cugraph" ]] || [[ "${package_dir}" == "python/pylibcugraph" ]]; then
-    if [[ "${RAPIDS_CUDA_MAJOR}" == "11" ]]; then
-        PYDISTCHECK_ARGS=(
-            --max-allowed-size-compressed '1.5G'
-        )
-    else
-        PYDISTCHECK_ARGS=(
-            --max-allowed-size-compressed '975M'
-        )
-    fi
-else
-    PYDISTCHECK_ARGS=()
-fi
-
 cd "${package_dir}"
 
 rapids-logger "validate packages with 'pydistcheck'"
 
 pydistcheck \
     --inspect \
-    "${PYDISTCHECK_ARGS[@]}" \
     "$(echo ${wheel_dir_relative_path}/*.whl)"
 
 rapids-logger "validate packages with 'twine'"