|
01:14
dalek joined
01:40
PerlJam joined,
dalek joined
01:45
[Coke] joined
|
|||
| timotimo | aha, aha. two instructions directly in sequence grabbing the value from one spesh slot, then the next one and putting them into the very same register | 02:23 | |
| without kicking one of the getspeshslot instructions out | |||
|
07:33
FROGGS joined
08:13
zakharyas joined
08:46
woolfy left
|
|||
| nwc10 | paste.scsys.co.uk/438182 -- SAN barfs on t/spec/S17-lowlevel/thread.t | 08:53 | |
| no, I don't know what changed, but the stack trace starts in get_lexotic_for_handler_idx | |||
|
09:19
kjs_ joined
09:51
kjs_ joined
|
|||
| timotimo | oh, i didn't even see hoelzro's patch up there | 11:01 | |
| dalek | arVM: 811b90c | (Timo Paulssen)++ | src/jit/graph.c: jitting write_fhs lets us jit many print methods/subs |
11:02 | |
| FROGGS | ohh nice! | 11:25 | |
| timotimo++ | |||
| timotimo | doesn't seem to win us terribly much | 11:26 | |
| we may want to have a specialization for dynamic variables like $*OUT | |||
| it seems like we're spending quite some time looking up $*OUT over and over | |||
| FROGGS | or $*EXECUTABLE | ||
| maybe | 11:27 | ||
| I dunno | |||
|
12:49
JimmyZ joined
|
|||
| JimmyZ | timotimo: ping | 12:49 | |
| hmm, It would be nice to have docs/opitmize.md | 13:01 | ||
| timotimo | hey JimmyZ | 13:21 | |
| JimmyZ | hello | ||
| timotimo | what kind of stuff would you like to see in that document? | 13:22 | |
| JimmyZ | I think the basic structure, and a bit examples about the idea of optimization like optimize_coerce | 13:24 | |
| timotimo | so it should focus on spesh? | ||
| JimmyZ | and how to debug ... | ||
| yeah | 13:25 | ||
| speaking of optimize_coerce, is it trying to change coerce_in to const_n and let the eliminate_dead_ins function remove the pre const_i instruction? | 13:27 | ||
| and before coerce_in, it always is const_i? | 13:28 | ||
| timotimo | of course not | 13:30 | |
| but there are many cases when a const_i is followed by a coerce_in | |||
| because our code-gen does that in a not so clever way | |||
| that's why it looks for "KNOWN_VALUE | 13:31 | ||
| " | |||
| JimmyZ | what's about the first question? | 13:36 | |
| timotimo | ah, yes, the const_i instruction should be removed by remove_:ead_ins | 13:40 | |
| JimmyZ | nice | 13:41 | |
| timotimo | well, hopefully :) | ||
| [Coke] wonders if nwc10 is seeing any of the failures coke & lizmat are getting with moar, jit or nonjit. | 14:14 | ||
| (coke & lizmat are on os x) | |||
| dalek | arVM: 26656ba | (Timo Paulssen)++ | src/spesh/dump.c: for logging purposes, a BB's address is a good unique ID |
14:30 | |
| nwc10 | [Coke]: no. only the one I reported. Everything else passes on "my" machine | 14:32 | |
| x86_64 CentOS | |||
| [Coke] | shame. :) | 14:35 | |
| dalek | arVM: 5dd519d | (Timo Paulssen)++ | src/spesh/ (3 files): dump a spesh graph's callsite (if any) |
14:51 | |
| JimmyZ | timotimo: I thinks it should use MVM_string_utf8_encode_C_string instead of MVM_string_utf8_encode | 15:08 | |
| timotimo | oh? i stole that from a few lines below in that same file %) | 15:09 | |
| JimmyZ | that may be wrongish, since MVM_string_utf8_encode returns non-null-end strings, and call strlen ( in append function ) to the non-null-end strings may ... | 15:12 | |
| timotimo | imgur.com/N1sbOYa | 15:18 | |
| FROGGS | what are the colors for? | 15:21 | |
| timotimo | readability | ||
| making the same register pop out visually | |||
| FROGGS | ahh, well, if you say so :o) | ||
| ahh | |||
| timotimo | it's far from perfect | 15:22 | |
| i'd really appreciate feedback | |||
| i'm not sold on the "greyscale for version number" idea either | |||
| JimmyZ | that's nice! | 15:23 | |
| timotimo | imgur.com/hFNJ0VV - maybe a bit better | 15:25 | |
| FROGGS | that's better, aye | 15:26 | |
| timotimo | imgur.com/KQskiOx - wow. | 15:27 | |
| JimmyZ | wow, but why there is a background color in the right shape :) | 15:29 | |
| timotimo | hm? | 15:30 | |
| you mean outside the triangle? | |||
| JimmyZ | yeah | ||
| timotimo | version of the register | 15:31 | |
| follows a greyscale | |||
|
15:33
vendethiel joined
|
|||
| JimmyZ | It a bit hurt my eye ... and others are awesome :P | 15:33 | |
| the background part :P | 15:34 | ||
| dalek | arVM: e2264ff | (Timo Paulssen)++ | tools/colorize_spesh.p6: tool to colorize registers in spesh logs |
||
| timotimo | feel free to play around with different "hashing algorithms" :P | ||
| JimmyZ | ;) | 15:38 | |
| timotimo | now i want a tool that turns the SSA form into a graphviz graph | 15:45 | |
| JimmyZ | that will be much more awesome with dotted arrow | 15:57 | |
| ;) | |||
| timotimo | dotted ... arrow? | 15:58 | |
| JimmyZ | to guide the next version | ||
| timotimo | ah, yes | 16:00 | |
| these would go through operators | |||
| JimmyZ | something like git lg | 16:01 | |
| :P | |||
| timotimo | i wonder if i can tell graphviz to connect two nodes as "messily" as necessary | 16:02 | |
| JimmyZ | .oO(Perl6::Cario...) |
16:03 | |
| timotimo | doesn't seem to exist | 16:06 | |
| [Coke] | moar/moar-jit still the failingest backends. | 16:59 | |
|
18:14
FROGGS joined
|
|||
| dalek | arVM: c0053cf | (Timo Paulssen)++ | tools/ (2 files): let update_ops.p6 also write a perl6 module with ops data |
19:22 | |
| FROGGS | timotimo: what are you going to do with MAST::Ops? | 20:17 | |
| timotimo | FROGGS: i need to know if the first argument of an op is a write or read register | 21:27 | |
| or a write/read | |||
| (like in the inc op, iirc) | |||
| FROGGS | ahh | 21:37 | |
|
22:58
lizmat joined
23:00
woolfy joined
23:04
woolfy joined
23:09
woolfy joined
23:11
woolfy1 joined
23:51
xiaomiao joined
|
|||