Automatic Generation of Interpolants for Lattice Samplings: Part II -- Implementation and Code Generation

TL;DR

This paper develops a code generation pipeline that automatically translates spline space algorithms into LLVM-IR, enabling efficient, architecture-tuned implementations with adjustable parameters for diverse computational platforms.

Contribution

It introduces a systematic framework for translating spline space algorithms into optimized, tunable code across multiple architectures, building on prior theoretical results.

Findings

Generated LLVM-IR code for spline spaces with performance tuning options

Demonstrated flexibility of the code generation pipeline across architectures

Evaluated parameter effects on computational performance

Abstract

In the prequel to this paper, we presented a systematic framework for processing spline spaces. In this paper, we take the results of that framework and provide a code generation pipeline that automatically generates efficient implementations of spline spaces. We decompose the final algorithm from Part I and translate the resulting components into LLVM-IR (a low level language that can be compiled to various targets/architectures). Our design provides a handful of parameters for a practitioner to tune - this is one of the avenues that provides us with the flexibility to target many different computational architectures and tune performance on those architectures. We also provide an evaluation of the effect of the different parameters on performance.