Hi, It seems Intel has recently released a set of patches to zlib: https://github.com/jtkukunas/zlib This adds a few optimizations and new modes; in effect, it seems the new “medium” mode gets you get about 70-80% faster compression for 1-2% larger files (you can tweak this, of course). It remains compatible with regular zlib, both ways. Some of this would only be appropriate for modern x86 (e.g., the use of the crc32 instruction for hashing), but some of it is general speedups that would be useful to zlib as a whole. I know reconciling this so that we get speedups for x86 and no regressions for other platforms isn't a small task, but wishlist bugs don't need to be small, do they? :-)
Sorry about that; I do normally have it at 76, but this was sent from reportbug, not mutt. I have no idea what the plan for upstreaming this is. I fear they are rather nonexistent. /* Steinar */
There has been some traffic on the list regarding optimisations recently though I've not checked to see if it's the same work.
OK. I tried looking for this, but unfortunately the list archives seem to be down for a while, so it's hard from the outside. :-) /* Steinar */
On Sat, Oct 04, 2014 at 02:24:39PM +0200, Steinar H. Gunderson wrote: Please fix your mail client to word wrap at less than 80 columns (ideally 72 or something to allow for quoting) since it makes things much easier to read. I trust Intel will be contributing these upstream; I'm really reluctant to take out of tree patches due to the potential security implications (this is the same issue that means we don't use any of the out of tree assembly). zlib is network viaible so it's very important that the code is correct.
Currently both zlib and Intel zlib are no longer actively developed, however there is zlib-ng with optimization for many architectures: https://github.com/zlib-ng/zlib-ng