Figure 13. Top: Program runtime for selected benchmarks under different window scheduling methods, with and without SWIPER. Classical processor resources are unlimited. Vertical black lines indicate standard deviation over 10 randomized trials for smaller benchmarks. Top right: Relative differences in runtime when using auto-limited processor count (Section 5.3.4) for benchmarks with runtime longer than $10^4 \mu s$. Bottom: Relative runtimes normalized to default parallel window runtime.