Fast cycle-approximate MPSoC simulation based on synchronization time-point prediction |
| |
Authors: | Jinyong Jung Sungjoo Yoo Kiyoung Choi |
| |
Affiliation: | (1) Seoul National University, Seoul, South Korea;(2) Device and Solution Network, Samsung Electronics Inc., Soowon, South Korea |
| |
Abstract: | In this paper, we propose techniques for fast cycle-approximate multi-processor SoC simulation with timed transaction level
models and OS models. Cycle-approximate simulation with an abstract model is widely used for fast validation of a multi-processor
SoC in early design stages. However, the performance gain of abstract-level simulation is limited by the overhead of synchronizing
multiple concurrent processor/module simulators, which is inevitable in timed simulation. To reduce the synchronization overhead,
we adopt the synchronization time-point prediction method, which consists of two phases: static code analysis and dynamic
scheduling of synchronizations. In the static analysis phase before simulation, it estimates minimum execution time from every
point in the code to the nearest synchronization point. Then, during simulation, it pessimistically predicts the synchronization
time-points based on the estimates. The proposed approach targets fast cycle-approximate simulation of a system with delay
annotated SW code and transaction level models of HW with dynamic behavior. We present, in this paper, techniques to analyze
such abstract models of SW and HW and schedule minimal number synchronizations during cycle-approximate simulation of the
models. Experiments show that the approach achieves orders of magnitude higher performance in cycle-approximate multi-processor
SoC simulation. |
| |
Keywords: | Multiprocessor system-on-chip Cycle-approximate simulation Transaction-level model OS model |
本文献已被 SpringerLink 等数据库收录! |
|