Generality of the observed detrimental impact of loop blocking and sorting under view-based SoA conversion

Ascertain whether the observed detrimental effect of loop blocking and explicit data sorting, when using compiler-guided temporary SoA views, persists across different particle struct sizes, deeper memory hierarchies, and algorithms with significant long-range interactions.

Background

The results indicate that with compiler-generated views and temporary SoA conversions, attempts to exploit coalesced access via loop blocking and explicit sorting into small contiguous chunks are counterproductive; continuous ordering offers little benefit and small sub-chunks hurt performance.

The authors caution that it is not clear whether these observations generalize to other particle sizes, deeper memory hierarchies, or SPH variants with substantial long-range interactions, suggesting the need for further investigation.