A Framework for Optimizing CPU-iGPU Communication on Embedded Platforms
Event Type
Research Manuscript
Virtual Programs
Hosted in Virtual Platform
Embedded Software
Embedded Systems
DescriptionThe unified memory architecture in modern devices allows the explicit data copying between CPU and integrated GPU (iGPU) to be eliminated. However, to enable such a "no-copy" communication model, many devices either implement intricate cache coherence protocols or they may disable the GPU cache. This often leads to strong performance degradation. This paper presents a framework based on a performance model and a novel no-copy communication pattern that accurately estimates the potential performance of an application by considering different CPU-iGPU communication models (i.e., standard copy, unified memory, pinned "no-copy" memory) on a given programmable embedded device.