Processing Tiles in Parallel

This section explores how PVA optimizes DMA for tile-based image processing through hardware-accelerated data transfers, efficient memory management, and parallel processing capabilities.

You first learn the fundamentals of Raster Data Flow (RDF), including how to configure it for managing data transfers, trigger tile transfers, and allocate memory for individual image tiles. Building on this, the tutorials guide you through optimizing performance using RDF double buffering, a technique that enables parallel data processing and DMA transfers to hide latency.

Finally, you’ll explore cuPVA’s scheduling APIs to further accelerate your applications by distributing tasks across multiple VPU engines and synchronizing their execution, leveraging task-level parallelism for complex image processing pipelines like contrast stretching with histogram-based dynamic range adjustment.

This section includes the following tutorials:

• Raster Data Flow (RDF) and Tiles

• RDF Double Buffering

• Scheduling APIs