You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
We currently don't compress rep-def buffers at all when mini-block encoding. This is pretty wasteful as it requires 4-bytes per value. We should, at the very least, make sure we are byte packing and, for mini-block, it would probably be suitable to use bitpacking.
I think the main challenge for bit-packing is that we only have the "bitpack in 1024-element chunks" variant of bit-packing and not a simple "bitpack this buffer" interface.
This is not strictly required for the beta and could be done as part of the stabilization.
The text was updated successfully, but these errors were encountered:
We currently don't compress rep-def buffers at all when mini-block encoding. This is pretty wasteful as it requires 4-bytes per value. We should, at the very least, make sure we are byte packing and, for mini-block, it would probably be suitable to use bitpacking.
I think the main challenge for bit-packing is that we only have the "bitpack in 1024-element chunks" variant of bit-packing and not a simple "bitpack this buffer" interface.
This is not strictly required for the beta and could be done as part of the stabilization.
The text was updated successfully, but these errors were encountered: