RayMakie/Hikari renderer experiment: volume-mode animation at 3.5 s/frame - #39
Draft
SebastianM-C wants to merge 4 commits into
Draft
RayMakie/Hikari renderer experiment: volume-mode animation at 3.5 s/frame#39SebastianM-C wants to merge 4 commits into
SebastianM-C wants to merge 4 commits into
Conversation
…der scripts Two-process split because Hikari (StructArrays 0.6) and SciMLBase 3 (RecursiveArrayTools 4 → StructArrays 0.7) cannot resolve into one env: precompute_frames.jl (animation env) dumps plain-array frame payloads; thomson_ray.jl (raymakie env) renders them via Lava Vulkan on the W7900. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
First stills on the W7900 (SW BVH, 64 spp, 1280x960): look is close to the RPR reference out of the box, but the flash frame (both stripe stacks in scene, index-1.5 Dielectric on OPEN iso-surface sheets) took 2327 s — suspected TIR loops burning the depth-16 budget. HW RT DEVICE_LOSTs on the same scene (fine on the small Materials demo). Sweep isolates interface model × depth. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…_s grid Two upstream bugs pinned with mwe_medium.jl: (1) RGBGridMedium with σ_s_grid=nothing renders the whole bounding box as uniform fog; an explicit zero grid restores emission-only behavior. (2) A Dielectric boundary makes shadow rays treat the volume box as opaque — NullMaterial interface only (documented in RayMakie plots/volume.jl). Pulse medium gets its own floor (0.45, the mesh ±0.5 isolevel) so the envelope core doesn't saturate. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
61-frame flash clip at 1280x960/64spp in ONE process: 3.5 s/frame steady (RPR HybridPro: 10-15 s + 1.7 GB/frame leak). HW RT works on the volume scene but is slower (6.3 s) — volume cost dominates; SW BVH is the production path. Mesh style stays blocked upstream (209 s BLAS build, 2282 s observable mesh swap). Verdict: adopt for the volume look after a lookdev pass; upstream issues listed in NOTES.md. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
Evaluates RayMakie + Hikari (Julia-native spectral path tracing over Vulkan/Lava, from the RayDemo repo's dev branches) as a replacement for the RPRMakie/HybridPro animation pipeline. Adds
animation/raymakie/: a render-only env, a physics-side payload precompute (precompute_frames.jl, runs in theanimationenv), the renderer (thomson_ray.jl, mesh|volume styles, scene reuse via observables,EDM_RAY_*knobs), measurement sweeps, andNOTES.mdwith the full verdict.Why a two-process split
Hikari pins StructArrays 0.6 while SciMLBase 3 needs StructArrays 0.7 (via RecursiveArrayTools 4) — the physics stack and the ray stack cannot resolve into one env. Payloads cross the boundary as plain tuples/arrays.
Headline numbers (W7900, RADV, 1280×960)
Volume mode renders the radiation cube + analytic pulse as true emissive media (
RGBGridMediumfrom raw Julia arrays) — the thing RPR segfaults on — and the emission genuinely lights the room. Static mesh-style port reproduces the locked look closely on first try (stills inanimation/raymakie/, vsrpr_frames_v3).Upstream bugs found (reproducers included)
RGBGridMedium(σ_s_grid=nothing)renders the whole bounding box as uniform fog (mwe_medium.jl); explicit zero grid fixes ithw_accel=trueDEVICE_LOSTs on multi-million-triangle BLAS builds (fine on small scenes)Verdict
Adopt for the volume look after a human lookdev pass (mapping knobs are in place; cores still trend white and grading isn't matched to the locked RPR level). Keep
rpr_frames_v3as the shipped mesh-style render. Details + remaining work inanimation/raymakie/NOTES.md.Headless recipe (no X session):
kwin_wayland --virtual --socket=wayland-edm+WAYLAND_DISPLAY=wayland-edm JULIA_GLFW_PLATFORM=wayland(GLFW.jl hardcodes X11 on Linux).🤖 Generated with Claude Code