initial creation of libcugraph_etl.so (#1885)

The new renumbering implementation will require C++ integration directly with cudf.  In order to facilitate that, but also support our customers that won't need cudf, this PR will create a separate library (`libcugraph_etl.so`) which will ultimately link with `libcudf.so` and contain the ETL portions of cugraph that require cudf features.

This way our other libcugraph customers that don't need to reference the new library will not need to install all of the cudf dependencies.

To seed this, the PR also includes a proposed API for the new renumbering capability.

Authors:
  - Chuck Hastings (https://github.com/ChuckHastings)
  - Rick Ratzel (https://github.com/rlratzel)

Approvers:
  - Seunghwa Kang (https://github.com/seunghwak)
  - https://github.com/chirayuG-nvidia
  - Rick Ratzel (https://github.com/rlratzel)
  - Jordan Jacobelli (https://github.com/Ethyling)

URL: #1885

build.sh

@@ -18,16 +18,18 @@ ARGS=$*
 # script, and that this script resides in the repo dir!
 REPODIR=$(cd $(dirname $0); pwd)
 LIBCUGRAPH_BUILD_DIR=${LIBCUGRAPH_BUILD_DIR:=${REPODIR}/cpp/build}
+LIBCUGRAPH_ETL_BUILD_DIR=${LIBCUGRAPH_ETL_BUILD_DIR:=${REPODIR}/cpp/libcugraph_etl/build}
 
-VALIDARGS="clean uninstall libcugraph cugraph pylibcugraph cpp-mgtests docs -v -g -n --allgpuarch --buildfaiss --show_depr_warn --skip_cpp_tests -h --help"
+VALIDARGS="clean uninstall libcugraph libcugraph_etl cugraph pylibcugraph cpp-mgtests docs -v -g -n --allgpuarch --buildfaiss --show_depr_warn --skip_cpp_tests -h --help"
 HELP="$0 [<target> ...] [<flag> ...]
  where <target> is:
    clean            - remove all existing build artifacts and configuration (start over)
    uninstall        - uninstall libcugraph and cugraph from a prior build/install (see also -n)
    libcugraph       - build libcugraph.so and SG test binaries
+   libcugraph_etl   - build libcugraph_etl.so and SG test binaries
    cugraph          - build the cugraph Python package
    pylibcugraph     - build the pylibcugraph Python package
-   cpp-mgtests      - build libcugraph MG tests. Builds MPI communicator, adding MPI as a dependency.
+   cpp-mgtests      - build libcugraph and libcugraph_etl MG tests. Builds MPI communicator, adding MPI as a dependency.
    docs             - build the docs
  and <flag> is:
    -v               - verbose build mode
@@ -36,17 +38,17 @@ HELP="$0 [<target> ...] [<flag> ...]
    --allgpuarch     - build for all supported GPU architectures
    --buildfaiss     - build faiss statically into cugraph
    --show_depr_warn - show cmake deprecation warnings
-   --skip_cpp_tests - do not build the SG test binaries as part of the libcugraph target
+   --skip_cpp_tests - do not build the SG test binaries as part of the libcugraph and libcugraph_etl targets
    -h               - print this text
 
- default action (no args) is to build and install 'libcugraph' then 'cugraph' then 'docs' targets
+ default action (no args) is to build and install 'libcugraph' then 'libcugraph_etl' then 'pylibcugraph' then 'cugraph' then 'docs' targets
 
  libcugraph build dir is: ${LIBCUGRAPH_BUILD_DIR}
 
  Set env var LIBCUGRAPH_BUILD_DIR to override libcugraph build dir.
 "
 CUGRAPH_BUILD_DIR=${REPODIR}/python/build
-BUILD_DIRS="${LIBCUGRAPH_BUILD_DIR} ${CUGRAPH_BUILD_DIR}"
+BUILD_DIRS="${LIBCUGRAPH_BUILD_DIR} ${LIBCUGRAPH_ETL_BUILD_DIR} ${CUGRAPH_BUILD_DIR}"
 
 # Set defaults for vars modified by flags to this script
 VERBOSE_FLAG=""
@@ -119,9 +121,11 @@ if hasArg uninstall; then
     # uninstall libcugraph
     if [[ "$INSTALL_PREFIX" != "" ]]; then
         rm -rf ${INSTALL_PREFIX}/include/cugraph
-        rm -rf ${INSTALL_PREFIX}/include/cugraph_c
         rm -f ${INSTALL_PREFIX}/lib/libcugraph.so
+        rm -rf ${INSTALL_PREFIX}/include/cugraph_c
         rm -f ${INSTALL_PREFIX}/lib/libcugraph_c.so
+        rm -rf ${INSTALL_PREFIX}/include/cugraph_etl
+        rm -f ${INSTALL_PREFIX}/lib/libcugraph_etl.so
     fi
     # This may be redundant given the above, but can also be used in case
     # there are other installed files outside of the locations above.
@@ -187,6 +191,27 @@ if buildAll || hasArg libcugraph; then
     cmake --build "${LIBCUGRAPH_BUILD_DIR}" -j${PARALLEL_LEVEL} --target ${INSTALL_TARGET} ${VERBOSE_FLAG}
 fi
 
+# Configure, build, and install libcugraph_etl
+if buildAll || hasArg libcugraph_etl; then
+    if (( ${BUILD_ALL_GPU_ARCH} == 0 )); then
+        CUGRAPH_CMAKE_CUDA_ARCHITECTURES="NATIVE"
+        echo "Building for the architecture of the GPU in the system..."
+    else
+        CUGRAPH_CMAKE_CUDA_ARCHITECTURES="ALL"
+        echo "Building for *ALL* supported GPU architectures..."
+    fi
+    mkdir -p ${LIBCUGRAPH_ETL_BUILD_DIR}
+     cd ${LIBCUGRAPH_ETL_BUILD_DIR}
+    cmake -DCMAKE_INSTALL_PREFIX=${INSTALL_PREFIX} \
+          -DCMAKE_CUDA_ARCHITECTURES=${CUGRAPH_CMAKE_CUDA_ARCHITECTURES} \
+          -DDISABLE_DEPRECATION_WARNING=${BUILD_DISABLE_DEPRECATION_WARNING} \
+          -DCMAKE_BUILD_TYPE=${BUILD_TYPE} \
+          -DBUILD_TESTS=${BUILD_CPP_TESTS} \
+          -DBUILD_CUGRAPH_MG_TESTS=${BUILD_CPP_MG_TESTS} --log-level=VERBOSE \
+          ${REPODIR}/cpp/libcugraph_etl
+    cmake --build "${LIBCUGRAPH_ETL_BUILD_DIR}" -j${PARALLEL_LEVEL} --target ${INSTALL_TARGET} ${VERBOSE_FLAG}
+fi
+
 # Build, and install pylibcugraph
 if buildAll || hasArg pylibcugraph; then
     cd ${REPODIR}/python/pylibcugraph

ci/cpu/build.sh

@@ -66,12 +66,14 @@ conda config --set ssl_verify False
 # BUILD - Conda package builds
 ###############################################################################
 
-gpuci_logger "Build conda package for libcugraph"
+gpuci_logger "Build conda package for libcugraph and libcugraph_etl"
 if [ "$BUILD_LIBCUGRAPH" == '1' ]; then
   if [[ -z "$PROJECT_FLASH" || "$PROJECT_FLASH" == "0" ]]; then
     gpuci_conda_retry build --no-build-id --croot ${CONDA_BLD_DIR} conda/recipes/libcugraph
+    gpuci_conda_retry build --no-build-id --croot ${CONDA_BLD_DIR} conda/recipes/libcugraph_etl
   else
     gpuci_conda_retry build --no-build-id --croot ${CONDA_BLD_DIR} --dirty --no-remove-work-dir conda/recipes/libcugraph
+    gpuci_conda_retry build --no-build-id --croot ${CONDA_BLD_DIR} --dirty --no-remove-work-dir conda/recipes/libcugraph_etl
     mkdir -p ${CONDA_BLD_DIR}/libcugraph/work
     cp -r ${CONDA_BLD_DIR}/work/* ${CONDA_BLD_DIR}/libcugraph/work
   fi

ci/cpu/upload.sh

@@ -31,6 +31,7 @@ fi
 gpuci_logger "Get conda file output locations"
 
 export LIBCUGRAPH_FILE=`conda build --no-build-id --croot ${CONDA_BLD_DIR} conda/recipes/libcugraph --output`
+export LIBCUGRAPH_ETL_FILE=`conda build --no-build-id --croot ${CONDA_BLD_DIR} conda/recipes/libcugraph_etl --output`
 export PYLIBCUGRAPH_FILE=`conda build --croot ${CONDA_BLD_DIR} conda/recipes/pylibcugraph --python=$PYTHON --output`
 export CUGRAPH_FILE=`conda build --croot ${CONDA_BLD_DIR} conda/recipes/cugraph --python=$PYTHON --output`
 
@@ -45,6 +46,10 @@ if [[ "$BUILD_LIBCUGRAPH" == "1" && "$UPLOAD_LIBCUGRAPH" == "1" ]]; then
   echo "Upload libcugraph"
   echo ${LIBCUGRAPH_FILE}
   gpuci_retry anaconda -t ${MY_UPLOAD_KEY} upload -u ${CONDA_USERNAME:-rapidsai} ${LABEL_OPTION} --skip-existing ${LIBCUGRAPH_FILE} --no-progress
+  test -e ${LIBCUGRAPH_ETL_FILE}
+  echo "Upload libcugraph_etl"
+  echo ${LIBCUGRAPH_ETL_FILE}
+  gpuci_retry anaconda -t ${MY_UPLOAD_KEY} upload -u ${CONDA_USERNAME:-rapidsai} ${LABEL_OPTION} --skip-existing ${LIBCUGRAPH_ETL_FILE} --no-progress
 fi
 
 if [[ "$BUILD_CUGRAPH" == "1" ]]; then

ci/gpu/build.sh

@@ -91,7 +91,10 @@ if [[ -z "$PROJECT_FLASH" || "$PROJECT_FLASH" == "0" ]]; then
     gpuci_logger "Build from source"
     $WORKSPACE/build.sh -v clean libcugraph pylibcugraph cugraph
 else
-    export LIBCUGRAPH_BUILD_DIR="$WORKSPACE/ci/artifacts/cugraph/cpu/conda_work/cpp/build"
+    # ...cugraph/cpu/conda_work/... is the dir name when only 1 lib* library is
+    # present. For multiple libs (ie. libcugraph and libcugraph_etl), the
+    # "_work" dir is prefixed with the lib name.
+    export LIBCUGRAPH_BUILD_DIR="$WORKSPACE/ci/artifacts/cugraph/cpu/libcugraph_work/cpp/build"
 
     # Faiss patch
     echo "Update libcugraph.so"

ci/test.sh

@@ -57,8 +57,12 @@ fi
 if [[ -z "$PROJECT_FLASH" || "$PROJECT_FLASH" == "0" ]]; then
     cd ${CUGRAPH_ROOT}/cpp/build
 else
-    export LD_LIBRARY_PATH="$WORKSPACE/ci/artifacts/cugraph/cpu/conda_work/cpp/build:$CONDA_PREFIX/lib:$LD_LIBRARY_PATH"
-    cd $WORKSPACE/ci/artifacts/cugraph/cpu/conda_work/cpp/build
+    # ...cugraph/cpu/conda_work/... is the dir name when only 1 lib* library is
+    # present. For multiple libs (ie. libcugraph and libcugraph_etl), the
+    # "_work" dir is prefixed with the lib name.
+    export LIBCUGRAPH_BUILD_DIR="$WORKSPACE/ci/artifacts/cugraph/cpu/libcugraph_work/cpp/build"
+    export LD_LIBRARY_PATH="$LIBCUGRAPH_BUILD_DIR:$CONDA_PREFIX/lib:$LD_LIBRARY_PATH"
+    cd $LIBCUGRAPH_BUILD_DIR
 fi
 
 # Do not abort the script on error from this point on. This allows all tests to

conda/recipes/libcugraph/meta.yaml

@@ -34,6 +34,7 @@ requirements:
     - cmake>=3.20.1
     - cudatoolkit {{ cuda_version }}.*
     - librmm {{ minor_version }}.*
+    - cudf {{ minor_version }}.*
     - boost-cpp>=1.66
     - nccl>=2.9.9
     - ucx-proc=*=gpu

conda/recipes/libcugraph_etl/build.sh

@@ -0,0 +1,7 @@
+#!/usr/bin/env bash
+
+# Copyright (c) 2021, NVIDIA CORPORATION.
+
+# This assumes the script is executed from the root of the repo directory
+
+./build.sh libcugraph_etl -v --allgpuarch

conda/recipes/libcugraph_etl/meta.yaml

@@ -0,0 +1,48 @@
+# Copyright (c) 2021, NVIDIA CORPORATION.
+
+# Usage:
+#   conda build -c nvidia -c rapidsai -c conda-forge .
+{% set version = environ.get('GIT_DESCRIBE_TAG', '0.0.0.dev').lstrip('v') + environ.get('VERSION_SUFFIX', '') %}
+{% set minor_version =  version.split('.')[0] + '.' + version.split('.')[1] %}
+{% set cuda_version='.'.join(environ.get('CUDA', '9.2').split('.')[:2]) %}
+package:
+  name: libcugraph_etl
+  version: {{ version }}
+
+source:
+  git_url: ../../..
+
+build:
+  number: {{ GIT_DESCRIBE_NUMBER }}
+  string: cuda{{ cuda_version }}_{{ GIT_DESCRIBE_HASH }}_{{ GIT_DESCRIBE_NUMBER }}
+  script_env:
+    - CC
+    - CXX
+    - CUDAHOSTCXX
+    - PARALLEL_LEVEL
+    - VERSION_SUFFIX
+    - CCACHE_DIR
+    - CCACHE_NOHASHDIR
+    - CCACHE_COMPILERCHECK
+    - CMAKE_GENERATOR
+    - CMAKE_C_COMPILER_LAUNCHER
+    - CMAKE_CXX_COMPILER_LAUNCHER
+    - CMAKE_CUDA_COMPILER_LAUNCHER
+
+requirements:
+  build:
+    - cmake>=3.20.1
+    - cudatoolkit {{ cuda_version }}.*
+    - cudf {{ minor_version }}.*
+    - libcugraph {{ minor_version }}.*
+  run:
+    - {{ pin_compatible('cudatoolkit', max_pin='x.x') }}
+    - libcugraph {{ minor_version }}.*
+    - faiss-proc=*=cuda
+    - libfaiss 1.7.0 *_cuda
+
+about:
+  home: http://rapids.ai/
+  license: Apache-2.0
+  license_file: ../../../LICENSE
+  summary: libcugraph_etl library