Load-balanced rendering on a general-purpose tiled architecture

Abstract

Commodity graphics hardware has become increasingly programmable over the last few years, but has been limited to a fixed resource allocation. These architectures handle some workloads well, others poorly; load-balancing to maximize graphics hardware performance has become a critical issue. I have designed a system that solves the load-balancing problem in real-time graphics by using compile-time resource allocation on general-purpose hardware. I implemented a flexible graphics pipeline on Raw, a tile-based multicore processor. The complete graphics pipeline is expressed using StreamIt, a high-level language based on the stream programming model. The StreamIt compiler automatically maps the stream computation onto the Raw architecture. The system is evaluated by comparing the performance of the flexible pipeline with a fixed allocation representative of commodity hardware on common rendering tasks. The benchmarks place workloads on different parts of the pipeline to determine the effectiveness of the load-balance. The flexible pipeline achieves up to twice the throughput of a static allocation.