v1+ (and if possible for older versions too, mainly 0.4+) LB_CULL vs. LB
jowens opened this issue · comments
John Owens commented
LB_CULL vs. LB
Impact of kernel fusion, maybe normalized over LB.
John Owens commented
@neoblizz : consider the graphs per primitive labeled "advance mode", e.g.:
https://gunrock.github.io/docs/#/analysis/results_bfs?id=vis_gunrock_primitives_dobfs_advance_mode
(the run called "runs per advance mode, measured on V100")
In this I made a few decisions:
- V100 only
- Separate out different primitive-specific options into different rows (here, idempotence/mark-pred/undirected)
- Plot MTEPS not normalized MTEPS
Obviously easier for me to leave what I made, but I want you to have something you find useful, so speak up if you want something different.