Perpetual Adaptation of Software to Hardware: An Extensible Architecture for Providing Code Optimization as a Central System Service
Thomas Kistler, Michael Franz
Abstract
We present an open architecture for just-in-time code generation and dynamic code optimization that is flexible, customizable, and extensible. While previous research has primarily investigated functional aspects of such a system, architectural aspects have so far remained unexplored. In this paper, we argue that these properties are important to generate optimal code for a variety of hardware architectures and different processor generations within processor families. These properties are also important to make system-level code generation useful in practice.
Create a lesson
Related papers
mold: A Massively Parallel Linker
Rui Ueyama
AdaSprite: Resource-efficient Online Co-Adaptation for V2I Systems Under Large-scale Data Drifts
Lehao Wang, Zhiwen Yu, Sicong Liu et al.
Global Simulation-Guided Dynamic Operator Scheduling for Efficient Multi-Tenant Model Serving
Weinan Liu, Zeyuan Ding, Dian Ding et al.
From LLM Inference to Agentic Workloads: Characterization and Implications for Serving Systems
Chaokun Chang, Yukun Zhou, Kaihua Fu et al.
CoRun: Padding is Simple and Efficient for Deterministic LLM Inference
Shiju Zhao, Jiacheng Yang, Qihang Chen et al.
A Bounded Reclaim Actuator for PSI-Guided Compressed Memory: A Controlled Ablation
Abhiyan Dhakal, Sanjog Sigdel