Yidong Ma
Results
5
issues of
Yidong Ma
Need to save training checkpoints after several epoch so next training could start from this one
**Concisely describe the proposed feature** Low level optimization to utilize hardware's instructions like mad, rcp, log2, exp2 which are only one-cycle, as well as use scalar instead of vector etc....
feature request
discussion
advanced optimization
Profile on solver with cProfiler and find that, copy houdini data to solver is slow, vice versa. In following case, solver substep take 0.2s, but the whole frame is 0.6s....
We have some limitations in current implementation - ti.init() can only be called inside Python SOP - We cannot init every frame otherwise the overhead is too big. So solver...
seems not support obstacles in surface environment?