|
00:10
hohoho joined
00:43
hohoho joined
00:46
tokuhirom joined
01:47
ilbot3 joined
02:48
tokuhirom joined
04:16
vendethiel joined
04:49
tokuhirom joined
|
|||
| dalek | Heuristic branch merge: pushed 70 commits to MoarVM/even-moar-jit by bdw | 05:51 | |
|
06:32
hohoho joined
06:41
tokuhirom joined
06:47
Ven joined
06:54
Ven_ joined
07:11
Ven joined
07:14
FROGGS joined
07:21
Ven_ joined
07:25
zakharyas joined
07:35
TimToady joined
07:49
leont joined
08:06
pyrimidi_ joined
08:40
tokuhirom joined
08:50
Ven joined
09:47
tokuhirom joined
09:48
brrt joined
|
|||
| brrt | \o #moarvm | 09:53 | |
| i apologise for breaking the build yesterday | |||
| connectivity in the bus was rather spotty, and when i noticed the error i couldn't correct it anymore | |||
| FROGGS++ for fixing, naturally :-) | 09:58 | ||
| FROGGS | hi brrt | ||
| :o) | |||
| brrt | vendethiel: yeah, comparing gcc and v8 is not100% fair in that way | ||
| jnthn | .oO( The wifi on the bus goes up and down, up and down... ) |
||
| brrt | but it was an.. interesting result, shall we say? | ||
| at any rate, nobody is really asking from dynamic languages to be as fast as C or fortran | 10:04 | ||
| nwc10 | oh, I think you're wrong there. Plenty of folks want the moon, on a stick, now, for free. | 10:16 | |
| without worrying about whether they are strong enough to hold it up. | |||
| no-one sane and reasonable. :-) | |||
| See www.chiark.greenend.org.uk/~sgtatha...stick.html | 10:17 | ||
| and the image linked there with "here" :-) | |||
| FROGGS | :D | ||
| brrt | lol | 10:18 | |
| brrt lunch & | |||
| arnsholt | "May be useful for compatibility testing" =D =D =D | 10:20 | |
| brrt | ok, yes, unrealistic expectations abound | 10:47 | |
| nwc10 | eg " | 10:53 | |
| RPerl compiles your low-magic Perl 5 code to run hundreds of times faster than interpreted Perl, with full backward compatibility! | |||
| " | |||
| brrt | have i seen that? | 10:54 | |
| nwc10 | I found it here: www.kickstarter.com/projects/wbras...-rperl-v12 | ||
| I read it on the Internet, it must be true. | |||
| brrt | oh lord, restricted dynamic language subsets | ||
| also, hundreds of times faster? | 10:55 | ||
| is this ehm.. for real? | |||
| nwc10 | he's really working on it. And I believe that he really believes that given enough time, he can acchieve that. | 10:56 | |
| brrt | hmmmm | ||
| how restricted are we talking about? | |||
| hmm, no i don't think that can be done, at all | 10:57 | ||
| i mean, handoptimized assembly can't reach 100s of times faster than just perl | 10:58 | ||
| nwc10 | as best I'm aware of, no-one else working on anything comparable thinks like he does | ||
| jnthn | brrt: Perhaps in some cases it maybe can, but in general I agree | 10:59 | |
| brrt | i'm not sure whether that is a good thing for him... | ||
| nwc10 | I know what I really think, but this is a logged channel. | ||
| brrt | i'm thinking of the tests of the potential benefits for the tiler i did last year. i got a factor 6 speedup and that was some pretty good code that would have to do some pretty good optimization before the compiler would ever do that | 11:00 | |
| (from the old JIT, that is) | |||
| now the old JIT will give you another factor 4 speedup on integer / numeric code, so that brings you at most at 24 | 11:01 | ||
| and that is tight code where the benefit from JIT compilation is optimal | |||
| most code looks nothing like that | |||
| lizmat | fwiw, the way I understand rperl, is that it is a real compiler, not a JIT | 11:02 | |
| as in, compile / optimize everything *before* you start running | 11:03 | ||
| jnthn | I think we call it "Ahead Of Time", not "real" ;) | ||
| lizmat | hehe... yeah, sorry, /me is old school :-) | ||
| jnthn | .oO( The JIT who wanted to be a real compiler :D ) |
||
| lizmat | from a time where you had interpreters or compilers, and the two would never mix :-) | 11:04 | |
| jnthn | But yeah, I think it's been fairly well demonstrated that if you want to optimize dynamic languages, dynamic analyses tend to beat static analyses. | ||
| The paper on basic block versioning somebody linked here not so long ago did a nice job of showing that. | 11:05 | ||
| nwc10 | the way I understand Rperl (without digging into the backend source code) is that it's effectively a Perl (subset) to C++ transpiler | ||
| that then takes advantage of all the power of a C++ compiler | |||
| brrt | uhuh | ||
| like compilation speed ^^ | |||
| nwc10 | so, as long as you know all your types in advance | ||
| lizmat | what *i* find interesting about rperl, is its semi-intimate knowledge of the Perl 5 grammar :-) | ||
| brrt | jnthn: didn't read that, could you link it again? | 11:06 | |
| jnthn | uh, don't know I have it handy | ||
| lizmat | which might become useful for v5 again in the future | ||
| jnthn looks | |||
| brrt: Pretty sure it's pointersgonewild.com/2015/09/24/bas...esult-yet/ | |||
| brrt: Of note the bit around "To put things in perspective, I decided to compare this result with what’s achievable using a static type analysis. I devised a scheme to give me an upper bound on the number of type tests a static analysis could possibly eliminate. " | 11:07 | ||
| brrt | oh, cool | ||
| jnthn | Which I thought was an interesting way to look at it. | 11:08 | |
|
11:12
Ven joined
|
|||
| brrt | many interest, yes | 11:18 | |
| in our case, it seems that this could already work at the spesh level? | 11:19 | ||
| i'm asking so i know i don't have to do it ^^ | |||
| jnthn | :P | ||
| Yeah, spesh level I'd say | |||
| brrt | cool | 11:20 | |
|
11:48
tokuhirom joined
11:56
Ven joined
|
|||
| brrt | what's funny is that this: rperl.org/the_low_magic_perl_commandments.html is about as far as 'perl' can get from perl | 11:59 | |
| psch | brrt: "you can write java in any language" comes to mind | 12:02 | |
| brrt | supposing the benchmarks are real, it's pretty impressive though | 12:03 | |
| ah, that was harsh, i do believe the benchmarks are real | |||
| nwc10 | I have no reason to doubt the benchmarks | 12:19 | |
| I'm just not convinced that this is a better trade off than using Inline::CPP | 12:20 | ||
|
13:05
hohoho joined
|
|||
| brrt | jnthn: on closer examinatin, that seems almost to simple to work | 13:08 | |
| (reading further) | 13:11 | ||
| authors themselves admit it's pretty similar to trace compilation | 13:12 | ||
|
13:13
hohoho joined,
hohoho_ joined
|
|||
| jnthn | brrt: Which "that" specifically? BBV? | 13:22 | |
| brrt | yes | ||
| and they rely very much on compiling-on-demand | |||
| which of course we don't do | 13:23 | ||
| i wonder whether tracing would be simpler for us to implement, or something like this | |||
| jnthn | Well, given we already have a CFG, the idea of a BB-level analysis seemed at least worth a look :) | 13:24 | |
| [Coke] | I would be interested if anyone can duplicate the run times they were seeing. I was unable to get rperl working on my OSX box. | 13:25 | |
| (but I think that was mostly on me.) | |||
| brrt | or OSX :-P | 13:26 | |
|
13:50
tokuhirom joined
14:12
Ven joined
14:51
tokuhirom joined
15:16
FROGGS joined
16:18
Ven joined
16:53
tokuhirom joined
17:16
Peter_R joined
17:53
tokuhirom joined
|
|||
| FROGGS | wow, I get a double free in t/spec/S17-lowlevel/lock.t | 18:23 | |
| hoelzro | o/ #moarvm | 18:47 | |
| [Coke] | hio | 18:51 | |
| nwc10 | \o hoelzro | ||
| hoelzro | o/ [Coke], nwc10 | 18:52 | |
|
19:44
retupmoca joined
19:55
tokuhiro_ joined
20:05
dalek joined
21:56
tokuhiro_ joined
22:23
leont joined
22:50
tokuhiro_ joined
23:30
diakopter joined
23:31
dalek joined
23:35
Util joined
23:39
hohoho joined
|
|||