taichi
taichi copied to clipboard
taichi-dev
How to use hash snode
It seems that Taichi docs missing how to use hash snode. I'm wondering if I can get an example about 'hash snode'. Thank you in advance!
The current version of Taichi doesn't support the hash SNode. It was in the original Taichi paper and we plan to recover it in the future. Could you briefly introduce your use case so that we can try to find some alternative solution for now? Thanks.
BTW you could also use the pointer SNode for now. You only need to activate those keys which are inserted into the hash table.
Thanks for your suggestion. However, I found pointer SNode performs much slower than dense SNode in traversal cycle, which is another troubles. Or the performance of pointer SNode depends on how to set up SNode, is there have some strategies to accelerate it? Thank you in advance!
For example, I use a hash value (pairing function) to identify the contact pair between particle i and particle j. Once particle i and particle j are in contact, the keys activate. The most important is, the system have about 900,000 particles, thus the hash value may become very large.
Just going to jump in here as I had a similar question:
As I understand it Hash SNodes aren't implemented (the reference to them in the docs might be an oversight), but how can you construct a sparse multidimensional structure then? The dynamic SNode can only be used for a single dimension right? So is there currently no way to create a VDB-like structure? The original paper describes one as such:
root.hash(ijk, 32).dense(ijk, 16).pointer().dense(ijk, 8).place(u, v, w);
What should be done to create a sparse, multi-level, three-dimensional and unbounded structure? Any alternatives? Also, why was the hashnode ever removed?
