A question about storage interface #773
-
|
Why we use storage to pass bucketSize argument? Why not directly use bucketSize as a template argument to pass to inner storage class? |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment
-
|
We do plan to further improve our interface, as the current setup is not very user-friendly. Our approach will mirror what we did for the bloom filter: introducing a policy strong type that bundles all performance-tuning parameters, such as CG size, bucket size, hash functions, and more, so users won’t need to define multiple strong types themselves. This work is planned as part of the migration to CCCL NVIDIA/cccl#4308. |
Beta Was this translation helpful? Give feedback.
BucketSizedefines how many slots each thread loads from storage at a time, it belongs to the storage characteristics. Using strong types instead of plain integers is just a design enhancement to improve code clarity.We do plan to further improve our interface, as the current setup is not very user-friendly. Our approach will mirror what we did for the bloom filter: introducing a policy strong type that bundles all performance-tuning parameters, such as CG size, bucket size, hash functions, and more, so users won’t need to define multiple strong types themselves. This work is planned as part of the migration to CCCL NVIDIA/cccl#4308.