- Measure how record layout changes memory efficiency for field-oriented GPU work.
- Does struct-of-arrays outperform array-of-structs for the same logical particle update?
aossoa
- Run the same logical kernel over identical particle data in both layouts.
- Keep the fields touched, arithmetic, and output contract fixed across both variants.
- Median GPU time by layout.
- Useful-payload GB/s by layout.
- Speedup of
soarelative toaos.
- Field-wise access usually favors SoA because threads read contiguous values from the same field.
- Any AoS win would need to be explained by a different access shape or reduced address-generation cost.