5:42:39jackdanielFade: I will see how ironclad implements sha1
5:43:10jackdanielfwiw we could link with some well-optimized C library without dependencies statically and make ironclad use those; would that be still ironclad then? rather not
5:46:49jackdanielI already like ironclad's source code
6:06:35jackdanielit seems that biggest offender is gf dispatch here
6:07:06jackdanielafter that we have other functions with noticable piece of cake
6:10:06jackdanielsecond function there shows, that for some reason mod32+ is not inlined despite the eclaration
6:11:08jackdanielpthread_getspecific - I suspect we'd get rid of that in a single-threaded environment (it is responsible for managing environments across different threads)
6:12:10jackdanielrol32+ has the same issue as mod32+ - it should be inlined and I don't see the reason in source code why it is not
6:12:49jackdanieland that summirizes already >40% overhead
6:13:57jackdanielgetting inlining of said functions working would give us ~15% improvement (if not more). I'll have to investigate that before the release
6:27:46jackdanielif you have something to chip in, I've created a ticket here: https://gitlab.com/embeddable-common-lisp/ecl/issues/451