Guillermo Garcia

Guillermo Garcia

Modern equality saturation systems excel at expression-level rewrites by exploring large spaces of equivalent programs without suffering from the phase-ordering problem. How- ever, these systems struggle to represent equivalence directly over arbitrary control-flow graphs, often requiring normal- ization into structured or tree-like forms before optimization can occur. We present the E-Path data structure, a prototype frame- work for equality-saturation-style optimization over control- flow graphs. Instead of representing congruence between individual expressions, the E-Path records equivalence be- tween instruction sequences embedded within a compiler intermediate representation. In this prototype, E-Path is in- stantiated over a restricted ANF-based control-flow graph used in a compiler backend, but the model itself is intended to be IR-agnostic. By treating instruction sequences as the fundamental unit of congruence, the E-Path enables non-destructive optimiza- tion of loops and other control-flow structures while preserv- ing multiple equivalent program variants simultaneously. This allows classical CFG optimizations to be expressed as rewrite-driven transformations without destructive mutation of the underlying graph.

Yuancheng Luo, Dmitriy Yamkovoy, Guillermo Garcia

Multichannel audio mixer and limiter designs are conventionally decoupled for content reproduction over loudspeaker arrays due to high computational complexity and run-time costs. We propose a coupled mixer-limiter-envelope design formulated as an efficient linear-constrained quadratic program that minimizes a distortion objective over multichannel gain variables subject to sample mixture constraints. Novel methods for asymmetric constant overlap-add window optimization, objective function approximation, variable and constraint reduction are presented. Experiments demonstrate distortion reduction of the coupled design, and computational trade-offs required for efficient real-time processing.