A normalised, independent performance characterisation of the wavefront alignment family on CPUs and GPUs

Wavefront alignment, an irregular workload, spans CPUs, GPUs, FPGAs, ASICs and processing-in-memory, each benchmarked in isolation. We re-measure every implementation we could run under one gap-affine objective, on four CPU and four GPU configurations, across synthetic and real data, indexed by alignment score. On the x86 cores tested, efficiency declines as score rises, consistent with on-chip access latency, not bandwidth saturation, and retained wavefront footprint is a major determinant of multithreaded scaling. For WFA-GPU, traceback dominates end-to-end time; reporting convention alone moves apparent speedup 35×. From these we derive an operating-space map showing which platform suits which reads, and why.