Sean (Apr 10 2026 at 22:45): Sean (Apr 10 2026 at 22:45):
Sean (Apr 11 2026 at 15:14):If anyone would like to test this new interface out and provide feedback, please let me know and you can get a test binary (Windows only).
Erik (Apr 11 2026 at 15:15):is it csg, or precomputed? cuda/rocm?
Sean (Apr 11 2026 at 15:43):It's real-time LIBRT, so anything we can RT,.
Erik (Apr 11 2026 at 15:45):neat!
Sean (Apr 11 2026 at 15:52):Yeah, it's a pretty exciting advancement. I prototyped this back in 2019 but it was raw, multithreading was broken, no colors, a hacked proof-of-concept. This is way more polished.
It's Intel OSPRay (+ IMGGUI) + LIBRT + Qt. Loads LIBRT as an OSPRay plugin so that it can load a .g model. Then as OSPRay/Embree dispatches rays for rendering and encounters anything in our bounding volume, it hands off to LIBRT for rt_shootray() evaluation. LIBRT passes back hit info, normal, color, object IDs, etc, and OSPRay performs the optics based on the current renderer. OSPRay comes with three: a basic scivis renderer, an ambient occlusion renderer, and a path tracer (shown).
Sean (Apr 11 2026 at 15:58):Erik said:
is it csg, or precomputed? cuda/rocm?
Dusting off our opencl code path is on my short-term TODO for sure.. we got someone to finish up the boolean eval piece a couple years ago under GSoC but didn't have time to validate it. Then probably need to refactor that all to SPIR-V (i.e., get off opencl) so it can just be natively managed and not forked intersection code.
Sean (Apr 18 2026 at 15:15):Here is a brief poster overview of the work and the team at TAMU that has been helping us get this capability going.
Last updated: Apr 21 2026 at 01:28 UTC