Intel® oneAPI Data Analytics Library Developer Guide and Reference

Developer Guide and Reference

  • 2021.1
  • 12/04/2020
  • Public Content
Contents

Build applications with oneDAL

See Get Started Guides for Windows* and Linux* for instruction on how to build applications for C++. This section contains instructions for building applications with oneDAL for SYCL*.

Applications on Windows* OS

  1. Download and install Intel® oneAPI Base Toolkit.
  2. In Microsoft Visual Studio* Integrated Development Environment (IDE), open or create a C++ project for your oneDAL application to build.
  3. In project properties:
    • Set Intel® oneAPI DPC++/C++ Compiler platform toolset:
      In General configuration properties, choose Platform Toolset property
    • Add oneDAL
      includes
       folder to Additional Include Directories.
    • Add folders with oneDAL and oneTBB libraries to Library Directories:
      In VC++ Directories, choose Library Directories property
    • Add oneDAL and OpenCL libraries to Additional Dependencies:
      In Linker configuration properties, choose Input.
  4. Add the appropriate libraries to your project based on oneDAL threading mode and linking method:
    Single-threaded (non-threaded) oneDAL
    Multi-threaded (internally threaded) oneDAL
    Static linking
    onedal_core.lib, onedal_sequential.lib
    onedal_core.lib, onedal_thread.lib
    Dynamic linking
    onedal_core_dll.lib
    onedal_core_dll.lib

Applications on Linux* OS

  1. Download and install Intel® oneAPI Base Toolkit.
  2. Set environment variables by calling
    <install dir>/setvars.sh
    .
  3. Build your application with clang++:
    • Add
      fsycl
       option to the command:
      -fsycl
    • Add
      ONEAPI_DAAL_USE_MKL_GPU_GEMM
       definition:
      -DONEAPI_DAAL_USE_MKL_GPU_GEMM
    • Add oneDAL
      includes
       folder:
      -I<install dir>/daal/latest/include
    • Add oneDAL libraries. Choose the appropriate oneDAL libraries based on oneDAL threading mode and linking method:
      Single-threaded (non-threaded) oneDAL
      Multi-threaded (internally threaded) oneDAL
      Static linking
      libonedal_core.a, libonedal_sequential.a
      libonedal_core.a, libonedal_thread.a
      Dynamic linking
      libonedal_core.so, libonedal_sequential.so
      libonedal_core.so, libonedal_thread.so
  • Add an additional oneDAL library:
    -foffload-static-lib=<install dir>/daal/latest/libintel64/libonedal_sycl.a
Examples
Dynamic linking, Multi-threaded oneDAL:
clang++ -fsycl -DONEAPI_DAAL_USE_MKL_GPU_GEMM my_first_daal_program.cpp -Wl,
--start-group -L<install dir>/daal/latest/lib/intel64 -lonedal_core -lonedal_thread.so -lpthread -ldl -lOpenCL -L<install dir>/tbb/latest/lib/intel64/gcc4.8 -ltbb -ltbbmalloc -foffload-static-lib=<install dir>/daal/latest/lib/intel64/libonedal_sycl.a -Wl,--end-group
Static linking, Single-threaded oneDAL:
clang++ -fsycl -DONEAPI_DAAL_USE_MKL_GPU_GEMM my_first_daal_program.cpp -Wl,
--start-group <install dir>/daal/latest/lib/intel64/libonedal_core.a <install dir>/daal/latest/lib/intel64/libonedal_sequential.a -lpthread -ldl -lOpenCL -foffload-static-lib=<install dir>/daal/latest/lib/intel64/libonedal_sycl.a -Wl,--end-group

Product and Performance Information

1

Performance varies by use, configuration and other factors. Learn more at www.Intel.com/PerformanceIndex.