Enabling timing predictability in the presence of store buffers - IRIT - Institut de Recherche en Informatique de Toulouse Accéder directement au contenu
Communication Dans Un Congrès Année : 2023

Enabling timing predictability in the presence of store buffers

Résumé

We study the effect of store buffers on the timing predictability of processor pipelines. We show that the concurrency between the load unit and the store buffer to access the memory bus is an obstacle to timing predictability, even in simple scalar in-order pipeline designs. We then propose a gating mechanism that removes these shortcomings. We model subsets of pipelines that implement our mechanism using an established logic framework. Using the Coq proof assistant, we prove that it enables timing predictability. Finally, we show that the performance cost of our mechanism measured on an FPGA-synthesized RISC-V core is around 2% only on average, with a minimal increase in resource usage.
Fichier principal
Vignette du fichier
timing-predictability-store-buffers-rtns2023-authorversion.pdf (693.98 Ko) Télécharger le fichier
Origine : Accord explicite pour ce dépôt
licence : Copyright (Tous droits réservés)

Dates et versions

hal-04082519 , version 1 (26-04-2023)

Licence

Copyright (Tous droits réservés)

Identifiants

Citer

Alban Gruin, Thomas Carle, Christine Rochange, Pascal Sainrat. Enabling timing predictability in the presence of store buffers. 31st International Conference on Real-Time Networks and Systems (RTNS 2023), Jun 2023, Dortmund, Germany. pp.1-10, ⟨10.1145/3575757.3593653⟩. ⟨hal-04082519⟩
57 Consultations
60 Téléchargements

Altmetric

Partager

Gmail Facebook X LinkedIn More