Explore how a classic image-labeling algorithm moves from SIMD to MIMD, and what that means for real-world performance.
This book examines the challenges of mapping parallel image-processing tasks to shared-memory, multi-processor systems. It uses the Shiloach– Vishkin connected components algorithm as a concrete example to show trade-offs in synchronization, task flow, and data access on realistic hardware.
The discussion moves beyond theory to practical design decisions. You’ll see how processors can be concentrated where work occurs, how lists of tasks evolve, and how asynchronous execution can affect results. The text also compares different hardware models and analyzes time bounds and parallelization in concrete terms.
- How to frame task flow in a MIMD setting and manage data dependencies.
- Techniques for enqueuing tasks, forming edge and vertex lists, and handling synchronization.
- Impacts of asynchronous vs. synchronous shortcutting on tree height and speed.
- Quantitative comparisons of different architectures for connected component labeling.
Ideal for readers of parallel image processing, computer vision, and high-performance computing who want a grounded look at how classic algorithms adapt to modern hardware.