Overview
Explore the efficient handling of sensor data in modern automotive systems for self-driving vehicles in this conference talk. Learn about the basics of dmabuf, dma fence, dma heap, and reserved memory Linux kernel frameworks, and discover how to implement them to address the requirements of heterogeneous data processing pipelines. Gain insights into using dmabuf and reserved memory frameworks for allocating buffers from preferred memory, and understand a proposed solution for sharing buffers between different accelerators to set up a zero-copying data pipeline. Delve into the use of dma fence for synchronization between devices without userspace application intervention, and learn how userspace applications can monitor data execution using dma fence. The talk covers topics such as buffer exporters, driver APIs, DmaHip, CMAHip, buffer arrays, pipeline buffer allocation, and user notifications, providing a comprehensive overview of streamlining sensor data handling in autonomous vehicle systems.
Syllabus
Introduction
Agenda
Sensor Data Pipeline Example
Dmabuf Overview
DNA Buff
Dmabuf
Dmabuf Example
Buffer Exporter
Buffer Exporter List
Driver API
DmaHip
CMAHip
CMABuff
Arrays
Pipeline
Buffer Allocation
Reserved Memory
User Notifications
Data Pipeline
Taught by
Linux Foundation