Can you try chunky-canvas-resampling-steps-residues and chunky-canvas-resampling-steps - they both use #round. Then also chunky-canvas-resampling-nearest-neighbor and chunky-canvas-resampling-bilinear, which use those methods.

@chrisseaton The first benchmark I tried had regressed to 20% chunky-canvas-resampling-steps-residues so I added back the java specialization and a fallback to ruby specialization which brought the benchmark back to 100%. I didn't try the other three benchmarks but I assumed they would be okay if the first regression was fixed.