Schematic comparison of LLVM lld call-chain clustering (hfsort/C3) and cache-directed sorting. Both use call profiles and section sizes. C3 follows the most likely predecessor under size and density constraints, then orders remaining clusters by density. CDSort compares both concatenation orientations using frequency and distance locality gains, and prioritizes beneficial merges. This animation is explanatory, not execution of LLVM or a timing benchmark. Both can select the same order. Browserbase reverted its faster-linking hfsort experiment after a startup regression. Counts and durations are schematic, not measured. The animation plays automatically and holds its result. Replay to watch again.

A64 BB128 BC64 BD96 B900800700hfsort / C³Likely predecessor + density limitsB → C is C’s strongest incoming edgeABCDSize limit · density degradationABCDRemaining clusters → density ordercdsortScore candidate merges for localityCompare [A B] + [C D] vs reverseABCDCDABABCDBest positive gain → update neighbors