Devito: Fast Finite Difference Computation from Symbolic Specification

Devito is a tool to perform optimised finite difference (FD) computation from high-level symbolic problem definitions. Starting from symbolic equations defined in SymPy, Devito employs automated code generation and just-in-time (JIT) compilation to execute FD kernels on multiple computer platforms.

Devito is part of the OPESCI seismic imaging project. A general overview of Devito features and capabilities can be found here, including a detailed API documentation.

Quickstart

The recommended way to install Devito uses the Conda package manager for installation of the necessary software dependencies. Please install either Anaconda or Miniconda using the instructions provided at the download links.

To install the editable development version Devito, including examples, tests and tutorial notebooks, please run the following commands:

git clone https://github.com/opesci/devito.git
cd devito
conda env create -f environment.yml
source activate devito
pip install -e .

If you don't want to use the Conda environment manager, Devito can also be installed directly from GitHub via pip:

pip install --user git+https://github.com/opesci/devito.git

Examples

At the core of the Devito API are the so-called Operator objects that allow users to create efficient FD kernels from SymPy expressions. Examples of how to configure operators are provided:

• A set of introductory notebook tutorials introducing the basic features of Devito operators can be found under examples/cfd. These tutorials cover a range of well-known examples from Computation Fluid Dynamics (CFD) and are based on the excellent introductory blog "CFD Python:12 steps to Navier-Stokes" by the Lorena A. Barba group. To run these, simply go into the tutorial directory and run jupyter notebook.
• A set of tutorial notebooks for seismic inversion examples is currently under construction in examples/acoustic/tutorials. We will add to these in the near future to provide more complex examples of how to use Devito operators for seismic imaging algorithms.
• Example implementations of acoustic forward, adjoint, gradient and born operators for use in full-waveform inversion (FWI) methods can be found in examples/acoustic.
• An advanced example of a Tilted Transverse Isotropy forward operator for use in FWI can be found in examples/tti.
• A benchmark example for the acoustic and TTI forward operators can be found in examples/benchmark.py

Compilation

Devito's JIT compiler engine supports multiple backends, with provided presets for the most common compiler toolchains. By default, Devito will use the default GNU compiler g++, but other toolchains may be selected by setting the DEVITO_ARCH environment variable to one of the following values:

• gcc or gnu - Standard GNU compiler toolchain
• clang or osx - Mac OSX compiler toolchain via clang
• intel or icpc - Intel compiler toolchain via icpc
• intel-mic or mic - Intel Xeon Phi using offload mode via the pymic package

Thread parallel execution via OpenMP can also be enabled by setting DEVITO_OPENMP=1.

For a full list of the available environment variables and their possible values, simply execute:

from devito import print_defaults
print_defaults()

Performance optimizations

Devito supports two classes of code optimizations, which are essential in a wide range of real-life kernels:

• Flop-count optimizations - They aim to reduce the operation count of an FD kernel. These include, for example, code motion, factorization, and detection of cross-stencil redundancies. The flop-count optimizations are performed by routines built on top of SymPy, which logically belong to the Devito Symbolic Engine (DSE), a sub-module of Devito.
• Loop optimizations - Examples include SIMD vectorization and parallelism (via code annotations) and loop blocking. These are performed by the Devito Loop Engine (DLE), a sub-module consisting of a sequence of compiler passes manipulating abstract syntax trees. Some existing stencil optimizers are being integrated with the DLE: one of these is YASK.

Devito supports automatic auto-tuning of block sizes when loop blocking is enabled. Enabling auto-tuning is simple: it can be done by passing the special flag autotune=True to an Operator. Auto-tuning parameters can be set through the special environment variable DEVITO_AUTOTUNING.

For more information on how to drive Devito for maximum run-time performance, the interested reader is invited to refer to examples/PERFORMANCE.md.