Daniel Jünger
Daniel Jünger
I would suggest we start with our cuco hashers and find a place that feels right for them in CCCL. We have a whole zoo of hashers [here](https://github.com/NVIDIA/cuCollections/blob/dev/include/cuco/hash_functions.cuh). If we...
Having the same build time checks all over the library isn't ideal. Maybe this is an opportunity to refactor this whole insertion logic into a separate struct `slot_inserter` that internally...
Something like this: https://godbolt.org/z/nvGGYaPWa This would also solve #547
CC @kkraus14
Thanks for the valuable insights, @felipeblazing ! I can address some of the points right away: > We would like to be able to serialize and deserialize the underlying data...
> Impala filters actually are partitioned Bloom filters - but partitioned within a small contiguous block, with each partition being the size of a single word. Very interesting point! Yes,...
Hey folks, I ran some new benchmarks on H100 and wanted to share the results: [bloom_filter_h100.csv](https://github.com/user-attachments/files/16808216/bloom_filter_h100.csv) I ran the outer product of all config dims so the result file is...
I just looked into how we could support functional parity with parquet/impala/etc. The key differences between all these implementations are: - how hash values are generated aka which hasher is...
> One change I might make: To put extent_ in a policy seems unusual to me. I'd prefer it as a parameter to some_filter_policy::block_idx Good point. The extent type is...