On Saturday, 21 May 2016 at 21:12:15 UTC, Walter Bright wrote:
So really, how good are you at fast code?
I recall being able to get a marginal boost in speed by assigning the first 1-2 bytes in preassigned buckets (256 & 65536 respectively), so you have an immediate 1-2 byte matches (skipping the chafe so to speak). Unfortunately memory-wise it's highly inefficient and requires 2 passes over the data, and probably doesn't play well with cache.
I assume this is related to compressing symbols thread? I mentioned possibly considering the LZO library.
