🛠️ Installing sTiles

sTiles provides a C++/C/Fortran computational framework for sparse structured matrix factorizations. The recommended way to install sTiles is via CMake with optional support for GPU, LAPACKE variants, and mimalloc.

Requirements

CMake ≥ 3.22
GCC or Clang with C++ and Fortran support
BLAS and LAPACK (e.g., MKL or OpenBLAS)
Optionally: METIS, mimalloc, CUDA, LAPACKE headers

Quick Start

git clone https://github.com/esmail-abdulfattah/sTiles.git
cd sTiles/devel
mkdir build && cd build

cmake .. \
  -DCMAKE_BUILD_TYPE=Release \
  -DCMAKE_INSTALL_PREFIX=../../install \
  -DSTILES_GPU=OFF \
  -DUSE_MIMALLOC=ON \
  -DUSE_MKL_LAPACKE=OFF \
  -DLAPACKE_DIR=../../lapack-3.5.0/lapacke/include \
  -DMETIS_DIR=../lib/metis

cmake --build .
cmake --install .

Install Output

The install directory will contain only the essential components:

install/
├── include/
│   └── stiles.h
└── lib/
    ├── libstiles.a
    ├── libstiles.so
    └── libcoreblas.a

Installation Options

Option	Description	Default
`USE_MKL_LAPACKE`	Use Intel MKL-provided LAPACKE	`OFF`
`LAPACKE_DIR`	Path to generic LAPACKE headers	empty
`STILES_GPU`	Enable GPU CUDA kernels	`OFF`
`USE_MIMALLOC`	Use mimalloc allocator	`ON`
`METIS_DIR`	Path to METIS install (lib + include)	empty
`CMAKE_INSTALL_PREFIX`	Installation path	`/usr/local`

Examples

🔹 With MKL LAPACKE

cmake .. \
  -DUSE_MKL_LAPACKE=ON \
  -DMKLROOT=/opt/intel/oneapi/mkl/latest

🔹 With GPU and METIS

cmake .. \
  -DSTILES_GPU=ON \
  -DMETIS_DIR=../lib/metis

Uninstall

If you installed to a custom prefix (e.g., install/), simply delete that folder:

rm -rf ../../install

If you accidentally installed to /usr/local, you can remove:

sudo rm -f /usr/local/lib/libstiles.*
sudo rm -f /usr/local/lib/libcoreblas.a
sudo rm -f /usr/local/include/stiles.h

← Back to main page