On Dynamic Scheduling for the GPU and its Applications in Computer Graphics and Beyond.

Area of Science:

• Computer Science
• Parallel Computing
• High-Performance Computing

Background:

• Processor clock-rate increases have plateaued, necessitating parallel computing architectures like Graphics Processing Units (GPUs) for enhanced processing power.
• Harnessing the full potential of GPUs for complex algorithms remains a significant challenge in modern computing.

Purpose of the Study:

• To develop components enabling a wider range of algorithms to execute efficiently on GPUs.
• To advance parallel computing techniques for dynamic algorithms and complex computational tasks.

Main Methods:

• Implementation of novel processing models for dynamic algorithms with varying degrees of parallelism.
• Development of a versatile task scheduler featuring efficient work queues and dynamic priority scheduling.
• Introduction of efficient dynamic memory management strategies tailored for GPU architectures.

Main Results:

• Significant speed-ups in image generation through rendering algorithms that allocate more processing power to critical image regions.
• The first GPU-based grammar evaluation system capable of real-time city generation and rendering, overcoming previous memory and time constraints.
• Demonstrated substantial performance improvements for mesh processing algorithms on GPUs through advanced parallelization and scheduling strategies.

Conclusions:

• The developed components and scheduling strategies effectively enhance the efficiency of diverse algorithms on GPUs.
• This work significantly advances the state-of-the-art in GPU computing for fields such as rendering, geometric modeling, and mesh processing.
• The research provides a foundation for broader adoption of GPUs in computationally intensive scientific and engineering applications.