This index points to detailed lecture-note style plans for post-25 advanced investigations.
Canonical plan files now live next to their experiment reports under experiments/<id>/plan.md.
- Phase A: core primitives (radix sort, subgroup ops, occupancy modeling).
- Phase B: rendering data systems (tiled assignment, culling pipelines, GPU-driven blocks).
- Phase C: architecture depth (BVH layout, ray-friendly layouts, temporal coherence).
- Phase D: systems and platform depth (persistent queues, async overlap, cross-GPU validation).
- Advanced Investigation 01: Radix Sort on GPU: Multi-pass radix sorting for key-only and key-value data.
- Advanced Investigation 02: BVH Node Layout Experiments: Node representation impact on traversal efficiency.
- Advanced Investigation 03: Frustum Culling vs Clustered Culling: Comparative rendering-style visibility pipelines.
- Advanced Investigation 04: Tiled Light Assignment: Light-to-tile list construction strategies.
- Advanced Investigation 05: Persistent Threads and Work Queues: Dynamic scheduling for irregular workloads.
- Advanced Investigation 06: Subgroup Operations Study: Warp/wave-level intrinsics versus shared-memory patterns.
- Advanced Investigation 07: Async Compute Overlap: Overlapping compute, transfer, and independent stages.
- Advanced Investigation 08: Occupancy Modeling Against Vendor Guidance: Resource-pressure interpretation of measured trends.
- Advanced Investigation 09: Memory System Study for Ray-Friendly Layouts: Traversal-friendly layout behavior under coherent/incoherent query patterns.
- Advanced Investigation 10: Frame-to-Frame Coherence Studies: Temporal stability and ordering reuse effects.
- Advanced Investigation 11: GPU-Driven Pipeline Building Blocks: Compose culling, compaction, bucketing, and argument-like generation.
- Advanced Investigation 12: Reproducibility and Cross-GPU Comparison: Trend stability across architectures and vendors.