Closed dsentinel closed 5 years ago
Is the fpzzenc32 lossy?
All TurboPFor functions are lossless. There are two preprocessing transforms padfloat + paddouble that can be used to convert float or double reducing the precision.
Which one?
SIMD Compression and the Intersection of Sorted Integers
Hi, it will be nice, if you can post your pull request for clang-700.1.76 mac os 10.10.5 with the current TurboPFor version. I have no possibility for testing TurboPFor under a mac os.
use icapp to find out the best TurboPFor function for your data. You can use the float functions also for integers and vice versa. You can also use the float functions for double content or 16 bits functions for 32 bits contents.... There are some cases where you get better compression.
In general zigzag/delta compress better than xor.
see the references at the bottom of the readme