Multiprocessing system for path tracing of big data
A path tracing system in which the traversal task is distributed between one global acceleration structure, which is central in the system, and multiple local acceleration structures, distributed among cells, of high locality and of autonomous processing. Accordingly, the centrality of the critical resource of accelerating structure is reduced, lessening bottlenecks, while improving parallelism.
1. A multiprocessing system for path tracing of big data,
employing distributed acceleration structures, comprising:
multiple general-purpose processor,
multiple graphics processing units having a raster pipeline, and
data of a 3D scene;
wherein in run time for generating primary hit points
a) data structure of cells is generated;
b) the data of the 3D scene is mapped onto cells;
c) a global acceleration structure of cells is constructed;
d) each cell builds its own local acceleration structure for an autonomous rendering of its local sub-space data;
e) primary rays are generated by rendering the 3D scene with the raster pipeline;
f) the global acceleration structure of cells is traversed for each primary ray, to find a cell of intersection; and
g) a hit point is registered at the cell of intersection; and
for generating secondary hit points
a) autonomously at each cell secondary rays are created originating at each primary hit point, or at each incoming ray; and
b) autonomously at each cell the local acceleration structure is traversed by a secondary ray seeking for hits with local objects;
wherein in event of hit, radiance at a point of intersection is sampled, and diffuse radiance at ray's origin is calculated and registered; and
wherein in event of no hit, the secondary ray is moved to the global acceleration structure for traversal, seeking the next cell of hit for the secondary ray.
2. The system of claim 1 , wherein generating primary rays involve fluctuation of projected rays.
3. The system of claim 1 , wherein basic data elements of the global acceleration structure are cells.
4. The system of claim 1 , wherein basic data elements of the local acceleration structures are triangles.
5. The system of claim 1 , wherein data elements of a cell's local acceleration structures are the cell's triangles.
6. The system of claim 1 , wherein the global acceleration structure resides centrally in the system, accessed by the primary rays as well as by secondary rays that are moved out from local cells in case of no hit with local objects.
7. The system of claim 1 , wherein secondary rays that move between local and global structures, may be rearranged in packets.
8. The system of claim 1 , wherein the local acceleration structure resides in a cell, along with a local sub-space data.
9. The system of claim 1 , wherein the multiprocessing system comprises multiple CPUs.
10. The system of claim 1 , wherein the multiprocessing system comprises multiple GPUs.
11. The system of claim 1 , wherein the multiprocessing system comprises multiple CPUs and multiple GPUs.