libera/#sicl - IRC Chatlog

18:55:08 scymtym yeah, similar. maybe cycle through the values instead of calling 06RANDOM inside the loop? not sure if it matters but i could imagine 06RANDOM being significant in this context

18:55:37 yitzi Ok. I'll try that.

18:56:14 scymtym i'm not saying that explains any discrepancies. just a random observation

18:57:34 yitzi Updated the paste with a LOOP

18:58:12 yitzi With those changes I get 567 microseconds on master and 374 microseconds on quaviver.

19:00:07 yitzi On SBCL, that is.

19:07:38 craigbro with sbcl and latest main, I am getting a failure to find a sicl-environment as needed by sicl-boot-base. I see defsystems for "sicl-environment-((in|ex)trindic|shared)" but no "sicl-enviroment" as needed by Code/Boot/sicl-boot-base.asd

19:09:09 yitzi I don't think you can boot SICL right now. beach is currenly reworking that.

19:10:28 craigbro claro

19:32:27 craigbro I have a patch to Specification/codify to use /usr/bin/env bash (i'm on nix) -- if I were to submit a diff, would it be via github pr? it's trivial enough to just relate it here perhaps

19:32:46 craigbro not covered by the CL style-guide in the Specification doc, haha

19:50:09 paulapatience At that point I don't think bash is required for that script, plain /bin/sh should work

19:50:50 paulapatience craigbro: PRs are good, sometimes mentioning it here is enough too

21:10:41 scymtym yitzi: with your benchmark on my machine, i get https://plaster.tymoon.eu/view/4485#4486

21:26:26 scymtym ::notify yitzi i pushed sketches of the liebler optimizations into the liebler-optimizations branch in the quaviver repository. the main technique is to caculate backwards from the overflow checks with 06Q to derive overflow checks with 06EXPONENT since those will allow tighter bounds for more computatonns including 06ASH and arithmetic

21:26:26 Colleen scymtym: Got it. I'll let yitzi know as soon as possible.

23:18:18 yitzi scymtym: If you have time, can you try the benchmark with leibler from the main branch without your optimizations? I am curious how some of our recent math optimizations will affect the timings.

23:18:18 Colleen yitzi: scymtym said 1 hour, 51 minutes ago: i pushed sketches of the liebler optimizations into the liebler-optimizations branch in the quaviver repository. the main technique is to caculate backwards from the overflow checks with 06Q to derive overflow checks with 06EXPONENT since those will allow tighter bounds for more computatonns including 06ASH and arithmetic