libera/#commonlisp - IRC Chatlog

3:16:00 spacefrogg_ ** NICK spacefrogg

6:45:24 glenneth2 Good morning o/

6:46:46 Josh_2 Mornin' glenneth2

6:47:05 glenneth2 Hey Josh_2

6:47:14 glenneth2 How's it going?

6:48:55 Josh_2 Another day another dolla

6:49:22 glenneth2 True story

7:18:08 glenneth ** NICK Guest5255

7:18:08 glenneth_ ** NICK glenneth

7:18:26 glenneth ** NICK Guest3577

7:18:26 glenneth_ ** NICK glenneth

7:19:55 glenneth_ ** NICK glenneth

7:20:11 glenneth ** NICK Guest2624

7:20:11 glenneth_ ** NICK glenneth

7:26:47 glenneth_ ** NICK glenneth

7:27:02 glenneth ** NICK Guest6521

7:27:02 glenneth_ ** NICK glenneth

8:53:58 jackdaniel another wip snapshot https://turtleware.eu/static/paste/ff4fdf04-pre-fastgf-faster.png

8:55:12 beach Looks great!

8:55:58 jackdaniel thanks (still based on static tables, but exploiting class stamps to defer checking whether the instance is obsolete)

8:56:10 beach Er, "pre-fastgf"?

8:56:58 jackdaniel I've mentioned it the other day, first I'm optimizing table-based dispatch (for workloads where only bytecodes compiler is available), and after that I'm adding fastgf

8:57:27 beach So all this is before any fastgf implementation?

8:57:29 jackdaniel for instances where the native compiler is available at runtime

8:57:45 jackdaniel yes (but using class stamps)

8:57:51 beach Interesting!

8:58:33 jackdaniel I expect that fastgf will give even better results (for warm cache)

8:58:48 beach I would expect that as well.

9:07:27 jackdaniel and I think that the biggest improvement will come from inlining effective methods and not necessarily searching for them

9:07:48 jackdaniel but that prediction needs verification

9:21:47 beach Have effective methods already been inlined in this version?

9:22:38 jackdaniel no, and they can't be

9:22:47 beach Oh, I see.

9:23:16 damperfi ** NICK Guest3706

9:26:38 Guest3706 ** NICK damperfi

9:54:48 pl jackdaniel: do you have access to LW Pro to checkout prewarmed CLOS too?

9:58:59 jackdaniel I have hobbyist edition; also I'm not that much interested in benchmarking lw

9:59:09 jackdaniel unless it is in a close proximity of ecl

9:59:41 jackdaniel but, my benchmarks do perform warmup

9:59:50 jackdaniel before measuring all methods are invoked for all benchmarked combinations

10:12:36 pl My interests in that are solely in "make sure open source implementations are fast as hell"

10:13:14 pl And either ACL or LW always claimed interesting option of dumping image with prewarmed CLOS

10:14:29 glenneth__ ** NICK glenneth

10:14:45 glenneth_ ** NICK glenneth

10:51:31 divlamir_ ** NICK divlamir

11:01:39 jackdaniel deducing pre-existing bugs by simply reading code is very satisfying

11:16:15 divlamir_ ** NICK divlamir

11:27:12 glenneth_ ** NICK glenneth

11:27:29 glenneth_ ** NICK glenneth

11:29:21 Lord_of_Life_ ** NICK Lord_of_Life

11:42:38 jackdaniel relation of readers and slot-value-using-class are like a spooky action at distance :)

11:45:46 phoe how so?

11:46:53 jackdaniel if we expect, that readers behave as if they went through slot-value-using-class, then specializing the latter may invalidate optimized readers

11:47:24 jackdaniel the was also recently a discussion, whether it is allowed to specialize the instance in slot-value-using-class (assuming that the class is EQ standard-class)

11:49:00 phoe yes

11:49:19 jackdaniel yes its spooky or yes its allowed or yes there was such a disucssion?

11:59:17 glenneth ** NICK Guest4627

11:59:17 glenneth_ ** NICK glenneth

11:59:33 glenneth ** NICK Guest691

11:59:33 glenneth_ ** NICK glenneth

11:59:42 divlamir_ ** NICK divlamir

12:01:51 glenneth ** NICK Guest1521

12:01:51 glenneth_ ** NICK glenneth

12:02:07 glenneth_ ** NICK glenneth

12:10:23 glenneth ** NICK Guest3969

12:10:23 glenneth_ ** NICK glenneth

12:10:40 glenneth_ ** NICK glenneth

12:24:00 phoe yes, it's spooky; yes, there was such a discussion; yes, such invalidation is necessary

12:25:48 glenneth ** NICK Guest3495

12:25:48 glenneth__ ** NICK glenneth