Skip to content

Latest commit

 

History

History
27 lines (19 loc) · 1.23 KB

build.md

File metadata and controls

27 lines (19 loc) · 1.23 KB

Steps to build XSTORE benchmark & tests

Prepare and build dependencies

  • ./magic.py config -f build-config.toml;
  • Get PyTorch c++, check from PyTorch c++;
  • cmake -DCMAKE_PREFIX_PATH=deps=/path/to/pytorch/share/cmake/Torch/ .;
  • Get Intel Math Kernel Library, check from MKL. Using the default steps for install. Usually, MKL is installed at /opt/intel/mkl. If that's not the case, please modify the mkl = ... entry inbuild-config.toml and rerun ./magic.py config -f build-config.toml; cmake -DCMAKE_PREFIX_PATH=deps=/path/to/pytorch/share/cmake/Torch/ .;
  • Build boost : make boost (May take sometime depends on network conditional and computation power);
  • Build jemalloc: cd ./deps/jemalloc; autoconf; cd ../../; make jemalloc;
  • cmake .

Build unittests

Most unittests are in separate modules (e.g., x_ml). For building and running unittests in each module, you can directly goto each module.

For instance, the following steps will run unittests in xkv_core, which implements the core storage of XStore.

  • cd xkv_core;
  • ./test.sh

Build benchmarks

  • make -j12