Module chimp 

Chimp float encoding (VLDB 2022).

Chimp improves upon Gorilla by exploiting the observation that XOR values of successive floating-point values cluster their leading zeros. It uses a leading-zero bucket scheme to reduce the overhead of encoding the number of leading zeros.

Leading-zero buckets

Bucket	Range (leading zeros)	Stored as
0	0–7	3 bits
1	8–15	3 bits
2	16–23	3 bits
3	24–31	3 bits
4	32–39	3 bits
5	40–47	3 bits
6	48–55	3 bits
7	56–63	3 bits

Chimp128 ring-buffer optimisation

Chimp128Encoder / Chimp128Decoder implement the Chimp128 variant from the same VLDB 2022 paper. A ring buffer of the last 128 raw f64 bit-patterns is maintained. For each new value the encoder computes the XOR against the previous value and against every ring-buffer entry, picking whichever reference yields the most leading zeros. When a ring-buffer reference wins, a 7-bit index is stored so the decoder can reconstruct the same XOR base.

This improves compression by 5–15 % on highly periodic workloads (e.g. fixed-interval sensor data that cycles through a limited set of levels) at the cost of ~1 KiB of encoder/decoder state.

Structs

Chimp128Decoder

Chimp128 float decoder.

Chimp128Encoder

Chimp128 float encoder — ring-buffer variant (VLDB 2022, §4.2).

ChimpDecoder

Chimp float decoder.

ChimpEncoder

Chimp float encoder (VLDB 2022).