This project is a Java layer bridging Java user and libcuspatial.so
Follow these instructions to create local builds and installation of librmm, libcudf and cudf-java, and libcuspatial
1.1) Follow instructions on this page for libcudf:
https://github.com/rapidsai/cudf/blob/branch-0.11/CONTRIBUTING.md#script-to-build-cudf-from-source
,
but append two more cmake config flags -DARROW_STATIC_LIB=ON -DBoost_USE_STATIC_LIBS=ON
,
changing
$ cmake .. -DCMAKE_INSTALL_PREFIX=$CONDA_PREFIX -DCMAKE_CXX11_ABI=ON
to
$ cmake .. -DCMAKE_INSTALL_PREFIX=$CONDA_PREFIX -DCMAKE_CXX11_ABI=ON -DARROW_STATIC_LIB=ON -DBoost_USE_STATIC_LIBS=ON
Make sure to make install
after make
. DO NOT use the build.sh
script to build libcudf.so
1.2) Follow the instructions in ${CUDF_HOME}/java/README.md
to build and install cuDF product JAR and test JAR
$ cd ${CUDF_HOME}/java
$ mvn clean install
NOTE:
- Following above steps you will have conda env
cudf_dev
which is required for subsequent builds. - libcudf.so will be installed under conda env
cudf_dev
. - libcudfJni.so will be built by maven pom.
- cuDF JARs will be installed to maven local.
export RMM_HOME=${pwd}/rmm
and git clone
source into $RMM_HOME
.
Follow instructions on this page:
https://github.com/rapidsai/rmm
, section Script to build RMM from source
$ cd ${RMM_HOME}/
$ ./build.sh librmm
NOTE:
- The build.sh script does installation by default.
- The destination is conda env
cudf_dev
.
Follow the instructions on this page for libcuspatial: https://github.com/rapidsai/cuspatial
$ cd ${CUSPATIAL_HOME}/
$ ./build.sh libcuspatial tests
NOTE:
- Require env
RMM_HOME
,CUDF_HOME
,CUSPATIAL_HOME
, and conda envcudf_dev
.
$ cd ${CUSPATIAL_HOME}/java
$ mvn clean install
NOTE:
- cuspatialJni.so will be built by maven pom.
- Unit tests of Cuspatial-java will run during the JAR build.