freenode/clasp - IRC Chatlog

0:21:52 stassats Bike: a compatible calling convention, for starters

0:22:29 Bike compatible between what?

0:22:42 Bike oh, you mean the existing calling convention makes tail calls impossible?

3:08:19 xristos ** NICK Guest50557

4:19:23 drmeister Hello

4:20:17 drmeister Bike: I added a make_list initializer - it should reduce the depth of recursion when compiling lists in fasl files.

4:20:37 Bike how does it work? it's kind of tricky to handle all the different kinds of list.

4:20:48 drmeister I'm testing this new code. RIP-relative addressing all the way now.

4:21:05 drmeister Hmm - tricky you say.

4:21:22 drmeister I check if it's a proper list and then generate a make_list call.

4:21:24 Bike i mean, with dotted and circular lists.

4:21:56 drmeister https://www.irccloud.com/pastebin/m1YNP2TW/

4:22:54 Bike seems sensible, assuming i understand what that does correctly

4:23:48 drmeister Things have changed considerably in this file.

4:24:17 drmeister It doesn't immediately generate code anymore. It generates a list of run-all-xxxx objects.

4:24:42 drmeister They get compiled into code after the top level function and all the little functions that it spawns are compiled.

4:25:04 Bike oh, do you not have to reference-literal the length or is that an immediate

4:25:23 drmeister The load-time-values vectors should be depreciated now.

4:25:49 drmeister The length of the list? That's in there. The (length cons)

4:26:22 Bike yeah, i mean it's in there as (length cons) rather than (load-time-reference-literal (length cons) t)

4:26:47 drmeister llvm can handle integers and strings straight up.

4:26:56 Bike so, immediate, gotcha

4:27:14 drmeister Here's the code that's generated now:

4:27:30 drmeister https://gist.github.com/drmeister/d786c194b0a350eb201c193f64fbb7cc

4:27:58 drmeister Features to note - values aren't written to or loaded from a vector anymore everything is done with llvm registers.

4:29:03 drmeister The literals and load-time-values are written into a block of memory called [VALUES-TABLE].xxx - entries are accessed with RIP-relative addressing from the code.

4:30:01 drmeister The cc_register_roots does that work. All of the values that need to be written into the VALUES-TABLE are passed on the stack to cc_register_roots, it writes them into the memory and then makes them GC roots.

4:31:01 drmeister With Boehm it will simply copy the values into a block of Boehm managed UNCOLLECTABLE memory - it's a bit of duplication but there is no way to declare roots outside of boehm managed memory.

4:31:24 drmeister With MPS it's better, the VALUES-TABLE is added to the GC as a block of roots.

4:31:46 drmeister Every top level form gets its own block of roots.

4:33:06 drmeister Aaaaand it works.

4:33:09 beach Good morning everyone!

4:33:12 Bike neat.

4:33:13 drmeister Hi beach

4:40:59 drmeister Ok, it seems to work fine: (defun foo-list () '(1 2 3 4 5 6 7))

4:41:01 drmeister Generates this:

4:41:15 drmeister https://www.irccloud.com/pastebin/QKZDqM8S/

4:42:12 drmeister Would this be a circular list: (defun foo-list () '#1=(1 2 3 4 5 6 7 #1#))

4:42:36 Bike it's still a proper list. try '#1=(1 2 3 4 5 6 7 . #1#)

4:43:30 drmeister Shoot - I don't have the reader macros in place yet. I'm running in the interpreter.

4:43:40 drmeister I have to get COMPILE working next - then I can compile aclasp

5:05:42 loke drmeister: What is aclasp?

6:16:47 drmeister loke: clasp boots through a few stages. aclasp is the second stage. It's got some of Common Lisp but lacks CLOS and FORMAT and LOOP

6:24:32 Bike wait, second? what's the first?

6:24:41 drmeister The interpreter

6:26:11 Bike oh.

6:29:14 drmeister I have a ways to go before I get COMPILE working with the new code.

6:29:24 drmeister I'm heading to bed - it's a busy day tomorrow.

6:29:55 drmeister It's gonna be yuge.

6:37:04 Bike don't remind me

7:12:15 Shinmera Can't wait to stop having to hear about bloody politics.

7:17:36 jackdaniel it's not bloody yet ;)

12:12:52 frgo Hello all.