Alternative memory layout: rays grouped by direction rather than by probe position. Improves GPU texture cache coherence during merging (reads for same-direction rays across neighbouring probes become spatially contiguous). Also enables hardware bilinear interpolation during merging.
Combined with pre-averaging, allows the entire RC implementation in under 100 lines of shader code.
- Described in: GM Shaders Part 2
- Origin: Graphics Programming Discord
Related: pre-averaging