Just some questions about TurboPFor Implementation about turbopfor-integer-compression HOT 3 CLOSED

1. Only functions with 256 in the name are using 256 integers, all other functions use 128.
   If the d functions compress better than d1, then your sequence is not strictly increasing (assuming turbopfordelta128 = vp4denc32).
   2.The bit width with the best compression is used in the coding. There are different build in algos in TurboPFor (vp4) functions: bitpacking, PFor, Variable byte, constant.
   3.All relevant bit sizes are examined, the best algo and the best bit width with maximal compression (regardless of speed) is used to encode all 128/256 integers in the block.
   Feel free to ask me about TurboPFor. I'm glad you have interest in it, especially in the most complex part (idxqry.c).
   After many years, TurboPFor is still SOTA. I have more ideas how to possibly improve it with novel algos, but I have actually no time for such a complex thing.

from turbopfor-integer-compression.

Thanks for answering all my questions :) I'd be interested in hearing about what ideas you have for improving these algorithms.

from turbopfor-integer-compression.

Note that for the vp4 functions, the compression is never worse than bitpacking.

from turbopfor-integer-compression.