Alex Purdy aspurdy

Lee, Mark, et al. "Vectorized production path tracing." Proceedings of High Performance Graphics. ACM, 2017.

Production path-tracer from Dreamworks. Engineered to take make heavy use of SIMD units throughout the entire rendering pipeline. Uses embree for ray interesection, focuses on discussion around speedup to shading/integration. Outlines queing / sorting of rays/samples to fill vector lanes and conversion to AOSOA formats. Makes use of Intel TBB and ISPC. Uses wavefront-based / bread-first approach to processing samples. Compares against scalar depth-first implementation.

@inproceedings{lee2017vectorized,
  title={Vectorized production path tracing},
  author={Lee, Mark and Green, Brian and Xie, Feng and Tabellion, Eric},
  booktitle={Proceedings of High Performance Graphics},
 pages={10},

	// warp-aggregated atomic multi bucket increment based on CUDA samples, modified to support compute capability < 7.0
	#include "helper_cuda.h"
	#include <cooperative_groups.h>
	#include <vector>

	#define NUM_ELEMS 10000000
	#define NUM_THREADS_PER_BLOCK 512

	using namespace cooperative_groups;