Devito-3.2

This release builds upon v3.1.0 including all features from Devito v3.2-beta as well as the following changes:

• Domain-allocation switch: now Operators run over the entire Grid, no matter what the stencil radius is.
• "Min-max" API to iterate over subregions of the Grid.
• A set of examples showing how to use Devito for linear algebra calculations
• Better seismic examples, more resampling.
• Parallelization of loops with reduction dependences with OpenMP atomics
• Simpler, better, more robust processing for Operator arguments
• Significantly more powerful IterationSpace
• Smarter clusterization

Devito-3.2-beta

• Improved organisation of the Intermediate Representation classes (ir submodules)
• Moved Clusterization to IR from DSE
• Better type hierarchy (reuse code from superclasses, SparseFunction -> [SparseFunction, SparseTimeFunction], ...)
• Better Dimension hierarchy, now wired into the main type hierarchy
• Drop "Global reverse flag" hack
• Drop "Global modulo flag" hack
• Revisited argument processing for Operator.apply
• Devito now explicitly provides its own Eq and solve modules
• Initial support for sub-domains
• Update the YASK backend, towards feature completeness
• Lots of changes in preparation for the domain-allocation switch

Devito-3.1

This release includes several new pieces of functionality and some significant API changes:

• Updated to Python-3 (exclusively) and SymPy-1.1
• Renamed symbol types: DenseData->Function, TimeData->TimeFunction, PointData->SparseFunction
• Introduction of a Grid class that provides the defaulting behaviour for space and time dimensions
• Enable completely generic operators without explicit substitution of spacing variables
• Addition of explicit start/end values for dimension iterations for operator invocation
• Removed fixed dimensions with hard-coded loop bounds
• Improvements to YASK backend resulting in functional acoustic operators (both forward and adjoint)
• Addition of new seismic example notebook using Dask for shot-level parallelism

Minor changes:

• Now using nbval to test notebooks
• Separating Function and Symbol types in internal hierarchy
• Make Constant a Symbol, not Function
• Added ability to switch backends at runtime
• Lots of internal refactoring...

Devito-3.0.3

A utility release to ensure a stable version before a set of planned API changes. Includes changes and some signficant progress on:

• Prototype YASK-backend
• Improved arguments handling
• Seismic examples and tutorials
  • RickerSource and demo_model utilities
• New TTI formulation (shifted and centred)
• Anaconda install
• Autotuning fixes
• Compiler-debug model

Devito-3.0.2

Minor update with various fixes and some internal re-structuring:

• Fix potential memory leak and removed unused memmap interface
• Fix reverse-time iteration bug
• Fix summary printing in DSE/DLE
• YASK integration phase 2 (data movement through YaskGrids)
• Generalized loop blocking (includes several bug fixes)
• Additional linear convection example notebook with smooth initial condition

Devito-3.0.1

Minor release that contains several bug-fixes and additional documentation:

• Fixes and test for the ForeignOperator interface
• Improved argument derivation in Operator.apply()
• Improved expression scheduling in Operator
• New tutorial notebook on seismic modelling
• New introductory tutorial based on CFD examples
• Various documentation fixes throughout

Devito-3.0

This is the base release for Devito version 3.0. It primarily provides a new and improved Operator API as well as two backend engines that perform a sequence of configurable compiler passes during the code generation stages to increase the performance of the generated code:

• DSE - Devito Symbolic Engine: Provides performance optimization passes that manipulate SymPy equations to decrease the number of floating-point operations performed when evaluating a grid point.
• DLE - Devito Loop Engine: Manipulates the loop structure of the generated code based on an abstract syntax tree (AST) and provides typical loop-level optimisations, including data alignment, explicit SIMD vectorization and thread-level parallelism through OpenMP pragmas, and loop blocking.

Stable v2.0.1 pre-release

Initial tag of stable v2.0.1 pre-release.