Loo.py: transformation-based code generation for GPUs and CPUs

TL;DR

Loo.py is a Python-based system that simplifies generating high-performance code for diverse hardware by transforming array computations with a rich set of optimization techniques.

Contribution

It introduces a data model and transformation library for array computations, enabling flexible, high-performance code generation across CPUs and GPUs.

Findings

Supports a wide range of transformations like tiling, vectorization, and unrolling.

Facilitates gradual transition from prototype to optimized code.

Deep integration with numpy and PyOpenCL enhances usability.

Abstract

Today's highly heterogeneous computing landscape places a burden on programmers wanting to achieve high performance on a reasonably broad cross-section of machines. To do so, computations need to be expressed in many different but mathematically equivalent ways, with, in the worst case, one variant per target machine. Loo.py, a programming system embedded in Python, meets this challenge by defining a data model for array-style computations and a library of transformations that operate on this model. Offering transformations such as loop tiling, vectorization, storage management, unrolling, instruction-level parallelism, change of data layout, and many more, it provides a convenient way to capture, parametrize, and re-unify the growth among code variants. Optional, deep integration with numpy and PyOpenCL provides a convenient computing environment where the transition from prototype to high-performance implementation can occur in a gradual, machine-assisted form.