Skip to content

Latest commit

 

History

History
104 lines (78 loc) · 2.97 KB

README.md

File metadata and controls

104 lines (78 loc) · 2.97 KB

cuStrings

API Design

cuStrings API

Dependecies

The following has been used to build and test library.

  • Linux 16.04
  • CMake 3.12
  • gcc 5.4
  • CUDA 9.2, 10.0
  • NVIDIA driver 396.44+ (Pascal arch or better)

Building for Python

Follow this section to use the C++ Python module. This will build the string library dependencies automatically. The above dependencies can be setup using conda and the /dev_p37.yml file:

    conda env create -n devstr conda/environments/dev_p37.yml

This .yml file is based on the ones from cudf and may have more packages then is actually needed for this module.

CMake for Python module

The CMake build requires the conda environment in order to find the dependent Python libraries. Build by creating a build dir under /cpp and calling cmake:

    source activate devstr
    cd cpp
    mkdir build
    cd build
    cmake ..
    make -j

Output includes pyniNVStrings.so and libNVStrings.so Use it with nvstrings.py to run any of the .py files in the python/tests dir. Output also includes pyniNVCategory.so and libNVCategory.so Use them with nvcategory.py to run any of the .py files in the python/tests dir.

Makefile for Python module

The Makefile.with_python can be used to build standalone modules though they may be imcompatible with cuDF. Use the CMake as described above to ensure compatibility.

The Makefile.with_python requires environment variable PYNI_PATH to point to Python include and lib. For example:

    export PYNI_PATH=/home/user/anaconda3/envs/devstr
    
    # if you're using a RAPIDS container
    export PYNI_PATH=/conda/envs/cudf

The Python version must also be specified with the PYTHON_VERSION environment variable:

    export PYTHON_VERSION=3.7

    make -f Makefile.with_python

Output is pyniNVStrings.so and libNVStrings.so and pyniNVCategory.so and libNVCategory.so

Installing Python lib

The following will call cmake to build the modules and install/deploy them in the current environment:

    source activate devstr
    cd python
    python setup.py install

Building for C++

Follow this section to build just the C++ classes directly without the Python modules.

    cd cpp
    make -j

Output is libNVStrings.so and libNVCategory.so only. Use these along with NVStrings.h and NVCategory.h. Example test file /cpp/tests/csv.cu

Makefile for NVStrings C++ interface

    cd cpp
    make -j
    sudo make install

The make install copies the libNVStrings.so and libNVCategory.so to /usr/local/lib and NVStrings.h and NVCategory.h to /usr/local/include/custr.

Use #include <custr/NVStrings.h> in source files. Use -L/usr/local/lib -lNVStrings for linking.

Docker

There is also a Dockerfile that can be used to setup a build and test environment. It is based on the cuDF docker image mainly to keep a consistent conda enviornment.