Low overhead path regeneration
- Authors: Frolov V.A.1,2, Galaktionov V.A.1
-
Affiliations:
- Keldysh Institute of Applied Mathematics RAS
- Moscow State University
- Issue: Vol 42, No 6 (2016)
- Pages: 382-387
- Section: Article
- URL: https://journals.rcsi.science/0361-7688/article/view/176468
- DOI: https://doi.org/10.1134/S0361768816060025
- ID: 176468
Cite item
Abstract
Monte Carlo Path Tracing is a core light transport technique which is used for modern methods (like BDPT, MLT, VCM and others). One of the main challenge of efficient GPU Path Tracing implementation is inefficient workload caused by paths of different lengths; few threads process the long paths, while other threads are idle. A work distribution technique called “Path Regeneration” is commonly used to solve this problem. We introduce a novel GPU implementation of path regeneration technique called “in place block based path regeneration.” In comparison to previous approaches our algorithm possesses two main advantages: it has lower self-cost and it does not move any per-ray data along threads in memory, thus, our algorithm can be easily integrated to any advanced path tracing technique (like BDPT, MLT and other) or photon mapping. We tested our solution with path tracing using both CUDA and OpenCL.
About the authors
V. A. Frolov
Keldysh Institute of Applied Mathematics RAS; Moscow State University
Author for correspondence.
Email: vfrolov@graphics.cs.msu.ru
Russian Federation, Moscow; Moscow
V. A. Galaktionov
Keldysh Institute of Applied Mathematics RAS
Email: vfrolov@graphics.cs.msu.ru
Russian Federation, Moscow