Skip to content

Perf/hier render fastpath#5

Draft
giovanni-mariano wants to merge 8 commits into
mainfrom
perf/hier-render-fastpath
Draft

Perf/hier render fastpath#5
giovanni-mariano wants to merge 8 commits into
mainfrom
perf/hier-render-fastpath

Conversation

@giovanni-mariano

Copy link
Copy Markdown
Owner

No description provided.

giovanni-mariano and others added 8 commits June 14, 2026 09:37
…aycast

  Solid rendering traced every ray to t_max through the full model depth even
  after hitting the first visible material. Add a first-hit render path that
  stops at the first material segment past the clip entry, plus a nested-universe
  neighbor-walk (using a new persistent surface->cell map) that replaces the
  per-step root-to-leaf relocation. Both are gated to the render entry points;
  transport/x-ray output stays byte-identical to baseline.
Add alea_find_cells_grid_raycast(): a drop-in for alea_find_cells_grid()
that traces one in-plane ray per image row and maps the resulting cell
segments to column spans (O(cell-crossings/row) instead of one point
query per pixel). Backed by a new buffer-reuse, neighbor-walk-off entry
alea_raycast_hier_fast_segments_nocache() whose per-pixel cell matches
the canonical point lookup. Includes a benchmark harness
(examples/c/slice_raycast_bench.c).

Experimental, additive, not wired into mc_plotter: benchmarking shows it
only beats the already row-coherent point-query fill at very high
resolution (~8K+) on dense models; it wins large on simple/coarse
geometry. Kept as a tool + cross-check (it surfaced a lattice descent
bug in the point-query fill).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant