00:00:00 --- log: started retro/12.10.26
06:50:00 --- quit: erider (Changing host)
06:50:00 --- join: erider (~chatzilla@unaffiliated/erider) joined #retro
06:50:39 <erider> hi all 
06:56:59 --- quit: erider (Quit: ChatZilla 0.9.89 [Firefox 15.0.1/20120907231835])
11:23:57 --- join: tangentstorm (~michal@108-218-151-22.lightspeed.rcsntx.sbcglobal.net) joined #retro
11:25:48 --- join: timg (~timg@static-96-243-240-250.tampfl.fios.verizon.net) joined #retro
11:25:56 <timg> hi
11:26:17 <tangentstorm> heya :)
11:26:28 <timg> sup
11:26:40 <tangentstorm> just mistyping things i guess :)
11:30:50 <tangentstorm> actually, i'm driving myself nuts going around in circles
11:32:00 <timg> why
11:33:15 <tangentstorm> too many tasks on my plate and they all seem to depend on each other
11:33:19 <tangentstorm> gridlock i guess
11:39:58 <timg> ahh
11:48:54 --- join: Mat2 (~claude@91-65-144-133-dynip.superkabel.de) joined #retro
11:49:15 <Mat2> hello
11:53:30 <tangentstorm> heya Mat2 :)
11:53:45 <Mat2> hi tangentstorm
11:55:21 <Mat2> how it is going ?
11:56:02 <tangentstorm> trying to untangle my brain into some kind of project plan :)
11:57:07 <Mat2> oh, same as me; working on milestones
11:57:40 <Mat2> and study OpenCL
11:58:48 <tangentstorm> still no luck with the SSE instructions then?
12:00:11 <Mat2> no, but programming these Intel GPU on my netbook give nice results
12:01:22 <tangentstorm> huh. i wouldn't think an onboard graphics chip would really do much. interesting.
12:03:23 <Mat2> with help of some documentation from the X consortium I've found a way to processing large code blocks within the GPU shaders
12:04:39 <Mat2> not comparable to newer AMD or NVIDIA chips but it works :)
12:04:44 <tangentstorm> are you just doing all this for fun?
12:04:48 <Mat2> yes
12:04:57 <tangentstorm> that's cool :)
12:06:01 <tangentstorm> you should write about it though. you're doing some pretty cutting edge stuff. :)
12:07:36 <Mat2> I work on 2 documentations about the vm and its implementation beside coding
12:09:15 <timg> is the retro vm thing?
12:09:44 <Mat2> no navm
12:11:17 <Mat2> and retro on top of that (or more precisely an retro inspired, functional dialect)
12:11:50 <Mat2> http://www.kickstarter.com/projects/adapteva/parallella-a-supercomputer-for-everyone
12:11:58 <Mat2> have you red these funding ?
12:12:23 <tangentstorm> nope
12:12:57 <tangentstorm> was looking for your docs in the navm repo.... that's some weird looking C code :)
12:13:54 <Mat2> yes as my ngaro version in the experimental directory
12:15:04 <timg> ngaro.. that's what its called
12:16:06 <timg> a supercomputer for everyone you say
12:16:23 <Mat2> the ngaro version uses replicated switch threading and navm compiles an opcode trace into direct threaded code on the fly
12:17:02 <Mat2> and that depends on some ... lesser used C features
12:17:04 <tangentstorm> parallella looks pretty slick.
12:18:08 <tangentstorm> i don't understand all the threading terminology. it just looks like you're not really doing C so much as programming in the C preprocessor :)
12:18:13 <Mat2> sorry, what mean slick ?
12:18:41 <timg> "cool"
12:19:33 <Mat2> ah ok. That is true but without the code would be just unreadable (and would it be only for me)
12:20:54 <Mat2> If you want to see really cryptic, unreadable and preprocess dependent code you should study the code for the engine of gforth
12:21:00 <Mat2> *g*
12:21:16 <tangentstorm> haha.. i haven't looked at it
12:21:19 <tangentstorm> but...
12:23:21 <tangentstorm> trying to find the initial source code for J ( iverson's successor to APL )
12:23:36 <tangentstorm> someone asked on the mailing list if the J source code was encrypted.
12:23:42 <Mat2> I have give up after reading some comments about philosophical questions why some fixes work
12:24:06 <Mat2> hmm, as I know there exist an open source variant of J
12:25:38 <tangentstorm> http://www.jsoftware.com/papers/AIOJ/onepage.htm
12:25:44 <tangentstorm> yeah J is under the GPL now
12:26:25 <tangentstorm> but the code all looks like that, apparently :)
12:26:33 <Mat2> I'm a big fan of APL but J is somewhat unreadable to me
12:28:14 <tangentstorm> yeah. same here. i spent a couple weeks on it. it's easy to type but hard to remember what the symbols mean
12:29:40 <Mat2> by the way, I'm curious how the sources look like
12:30:42 <tangentstorm> for j? that's what i was linking... http://www.jsoftware.com/jwiki/Essays/Incunabulum has it in gray
12:30:48 <tangentstorm> that's the core of the j interpreter.
12:33:05 <Mat2> oh no, that's call threading, the slowest of all possible threading variants in C
12:35:39 <tangentstorm> https://github.com/openj/core  is the current source. no idea where anything is in that mess, though :)
12:36:18 <tangentstorm> too clever for me
12:38:48 <Mat2> it's also call threading
12:39:25 <tangentstorm> huh
12:39:48 <tangentstorm> are you making a concatenative apl?
12:41:23 <Mat2> hmm, yes , i've implementing all basic operators like iota
12:42:20 <Mat2> with one difference, these will operate on a fixed array type (so forth style)
12:44:22 <tangentstorm> neat
12:44:42 <tangentstorm> i think that's easier to understand
12:45:18 <Mat2> and much better to implement (if one care about performance)
12:48:41 <Mat2> these J sources are as unreadable as J code itself, probably that is the result of some code translation tool
12:49:39 <Mat2> I suspect J is self hosting
12:49:59 <tangentstorm> nope :)
12:50:25 <tangentstorm> they write that by hand.
12:50:37 <tangentstorm> they just try to make C look as close to J/APL as possible
12:52:32 <Mat2> ok, I have found the basic primitives which are coded in assembly
12:52:56 <Mat2> now I understand
12:53:42 <Mat2> *lol* there had used the loop opcodes
12:54:49 <Mat2> that's some ~20 clocks extra each iteration on nearly all Intel and AMD cpu's newer 1982
12:57:30 <Mat2> ok, these interpreter is just a loop which executes code though double indirected pointers into C functions (that's indirect call threading)
12:58:18 <Mat2> as each C function has some big entry and exit code this method is very inefficient
12:59:48 <tangentstorm> you mean like it has to copy all the parameter data to the stack?
13:00:30 <Mat2> for most functions yes
13:00:59 <Mat2> Therefore, the programmers had no choice as to optimize the most used functions in assembler
13:02:25 <tangentstorm> do you think forth can be faster than C because of that?
13:04:19 <Mat2> you mean forth would generate faster code than a C compiler ?
13:05:05 <Mat2> an optimized forth interpreter would be for sure much faster than any C based interpreter can be
13:07:37 <Mat2> I think stack based code can be compiled as efficient as C code with less effort
13:09:08 <Mat2> because forth code can be seen as some special SSA form whereby an efficient C compiler depends on some very expensive algorithms to transform C sources into SSA for optimizations
13:09:51 <Mat2> and the language syntax is not well suited for this
13:10:53 * tangentstorm looks up SSA :)
13:12:10 <tangentstorm> http://en.wikipedia.org/wiki/Static_single_assignment_form
13:13:32 <tangentstorm> SISAL was a language out of lawrence livermore labs here in the US... It was an early pascal-like language, but maybe similar to what you're talking about.
13:13:59 <Mat2> interesting, I don't know this language
13:15:03 <tangentstorm> http://en.wikipedia.org/wiki/SISAL
13:15:18 <tangentstorm> 1983... long since abandoned.
13:16:26 <Mat2> nice language :)
13:17:09 <tangentstorm> now that i understand a bit more about what you're doing, maybe i'll try to incorporate some of that into retropascal ... make a metrosisal :)
13:17:56 <Mat2> I think that would be a good idea
13:18:57 <tangentstorm> i need to put these steps in order first, though :)
13:24:26 <Mat2> make sure you plan well :)
13:25:17 <tangentstorm> my problem is my tasks all depend on each other :/
13:27:44 <Mat2> try to isolate some tasks build all other around them
13:29:38 <Mat2> for a compiler, I would try to separate the syntax parsing from code generation (ending up implementing a frontend and backend for code generation)
13:29:40 <tangentstorm> yeah. i'm drawing it all out as a graph. 
13:30:09 <tangentstorm> it's not just the compiler... i'm making a course on how to bootstrap a system
13:31:11 <tangentstorm> i could write it all by hand right now, but i'm trying to leave a trail so that students can see how the system grows from almost nothing
13:31:59 <Mat2> I see
13:32:06 <tangentstorm> i think i'm going to start with a tool for visualizing ngaro machine code in a web browser
13:32:36 <tangentstorm> and then maybe try to shrink the instruction set to something like what you're using
13:33:03 <tangentstorm> but then you'd be able to just interactively hand-code and explore the image
13:33:53 <tangentstorm> like you'd see the ram, and you could color-code each square by hand to mark things... and you could set each cell to  appear as a character, a number, or an opcode
13:34:10 <tangentstorm> so you could build small images by hand
13:35:12 <Mat2> I think that would also allow CRC and others to improve retro the easy way. Ngaro suffers from a visual simulator like you describe
13:37:31 <tangentstorm> i think i can speed up the javascript stuff considerably along the way, too.
13:38:20 <Mat2> would be nice
13:38:26 <tangentstorm> this russian guy showed me another system, that's like a russian version of forth
13:39:06 <tangentstorm> but instead of gotos, the instruction set uses gosub for everything.
13:39:33 <tangentstorm> so instead of :    LABEL top-of-loop  do stuff here   JMP top-of-loop
13:40:02 <Mat2> russian coders have great experiences with forth inspired languages
13:40:15 <foucist> hmm, do you think the parallella kickstarter will make the funding?
13:40:16 <tangentstorm> it just has an opcode that does "while loop" and one that does "if"
13:40:25 <foucist> 25 hours away to raise another 65k..
13:40:37 <foucist> somehow i don't think it's gonna make it
13:40:41 <tangentstorm> guy's probably frantic :/
13:40:42 <foucist> $750k is pretty damned high
13:40:48 <Mat2> I hope so
13:41:00 <foucist> how about a forthchip kickstarter :P
13:41:13 <foucist> or a kickstarter to develop something that uses the greenarray chips
13:41:30 <tangentstorm> i want to do that
13:42:10 <Mat2> the GA144 is a nice cpu but these 64 core beast offer 1 gB ram
13:42:28 <tangentstorm> impomatic suggested a guy who might be interested in designing an arduino kind of board for ga144
13:43:27 <foucist> ga144 is dirt cheap though
13:44:37 <tangentstorm> i've been thinking about using a sort of cloud model.. where there's one big processor in the middle for handling ram, and lots of little processors around the edges
13:44:42 <Mat2> yes and the ephiphany core consume 5 Watt (idle)
13:44:53 <foucist> i think you can buy a pile of ga144 for $100 ? can't remember
13:45:01 <tangentstorm> like... i'm thinking 256 words of ram max
13:45:26 <tangentstorm> they're $20 per chip, but they have no staff so they only sell in 10-packs
13:45:54 <foucist> well bulk orders drop the prices way down
13:46:04 <foucist> was talking to the ceo awhile ago 
13:46:05 <tangentstorm> you can get one by itself + a little mounting board for $30 or so from another company
13:46:12 <foucist> aeh
13:46:13 <foucist> yeah*
13:46:41 <tangentstorm> that greg guy? he seems cool
13:47:50 <foucist> i had this idea for hosting an AI/ML contest where people would write a program that would generate code for the ga144 for efficient sha256 hashing
13:48:03 <foucist> i figure coding the ga144 directly would be a hassle :P
13:48:08 <foucist> might as well make it an ai/ml problem
13:48:08 <foucist> heh
13:48:36 <tangentstorm> bitcoin, huh? :)
13:48:39 <foucist> not sure what it would take to get a contest like that going..  oculd offer a prize or something
13:48:44 <foucist> yeah heh
13:48:59 <foucist> altho it might not beat the custom chips the bitcoiners are making these days
13:49:33 <tangentstorm> probably not, but do they cost $20 a pop? :)
13:49:39 <Mat2> I wonder where to store the immediate data for sha256 hashing ?!???
13:49:55 <tangentstorm> what do you mean?
13:50:10 <foucist> Mat2: is it that big?  
13:50:26 <Mat2> yes
13:50:42 <foucist> well i'm guessing that there's a way to split up the hash amongst the cores
13:50:54 <Mat2> I mean each F18 core has so less memory
13:51:59 <foucist> there was a paper on a similar sort of hashing thing coded in ga144
13:52:23 <foucist> http://www.greenarraychips.com/home/documents/pub/AP001-MD5.html
13:52:50 <foucist> A couple of things about the Green Arrays architecture dominate this implementation of MD5. One is the fact that MD5 is a 32 bit algorithm and GA architecture is 18 bits. The other is that each node of a Green Arrays chip is limited to 64 18 bit words of memory for both program and data.
13:52:55 <foucist> To address the first item we will perform all the 32 bit arithmetic and logic in parallel, 16 bits per partner node.
13:53:18 <foucist> that's interesting, mimic 32 bit chip by having two nodes work together
13:53:54 <tangentstorm> that's not how i was going to do it :)
13:54:04 <foucist> tangentstorm: how were you gonna do it?
13:55:02 <tangentstorm> basically one processor per bit, 18 hashes at a time
13:55:08 <Mat2> hmm, I think it would be possible building clusters of 4 cores
13:57:15 --- quit: timg (Read error: Connection reset by peer)
13:58:12 <Mat2> some for the tables some for processing
13:58:18 <foucist> tangentstorm: genius
13:58:24 <foucist> GENIUS!
13:58:52 <foucist> tangentstorm: have you coded much colorforth?
13:59:08 <tangentstorm> not really
13:59:49 <tangentstorm> i worked through the course and read through the html dump of all the blocks
14:00:03 <foucist> the course?
14:00:05 <foucist> there's a course?
14:00:11 <tangentstorm> for programming the greenarray chips
14:00:16 <tangentstorm> video course
14:00:54 <Mat2> foucist: You will find all courses on youtube
14:00:56 <tangentstorm> http://school.arrayforth.com/
14:01:16 <tangentstorm> you can log in and it'll grade your answers and stuff
14:02:06 <tangentstorm> i could write the code but i don't have any way to use the chip, so i put it on hold.
14:13:28 <Mat2> the folk from the forth gesellschaft (forth-ev) seem not so amused about the GA144, I do not know why
14:14:28 <Mat2> but there would be the chance to get a development board, so I would try to find a workable SSH256 implementation
14:18:15 <tangentstorm> now i'm thinking i should just learn to use my fpga... i bet could outperform the ga144 for a specific task like this if i put my mind to it
14:19:37 <tangentstorm> not entirely sure the return would be worth it in terms of time spent building the thing, but it would be fun :)
14:19:56 <Mat2> the problem for me is: An evaluation board from green arrays would cost me inclusive all costs for transport and taxes ~400 Euro
14:21:27 <Mat2> an FPGA board cost something around 100 Euro
14:21:56 <Mat2> and an PIC32 based board less than 40
14:22:08 <Mat2> (with 128 kB ram)
14:22:55 <tangentstorm> it would be easier to just hire someone to solder the greenarray chips onto a board with a little connector that you could plug into something else
14:24:03 <tangentstorm> or.. learn how to assemble it myself i guess... but i don't even have time for the stuff i'm already doing :)
14:24:48 --- join: kumul (~kumul@173.215.130.73) joined #retro
14:24:59 <tangentstorm> hey kumul
14:25:26 <Mat2> hi kumul
14:26:25 <Mat2> same here
14:27:17 <Mat2> there have $711,877 at current :)
14:27:58 <foucist> oh, school.arrayforth.com that's pretty nice
14:28:23 <foucist> Mat2: i think most forth people are afraid of colorforth & variants though?
14:28:30 <foucist> hence forth-ev ppl not liking ga144 ?
14:28:44 <foucist> even crc is afraid of colorforth
14:28:51 <foucist> it's a sad state of affairs :P
14:29:04 <Mat2> I like colorforth :)
14:30:56 <foucist> Mat2: yeah, eval board for ga144 is too pricey for me too
14:31:28 <foucist> i agree about getting someone else to solder the ga144 to something that plugs into the computer
14:31:31 <Mat2> the problem I see is: Most members here seem to be associated with ANS standard support ;)
14:33:00 <Mat2> and colorforth solves some parsing problems and tend to be too fast to follow
14:35:00 <Mat2> the idea to intent colours for sources is questionable anyhow
14:37:49 <Mat2> I think the cheapest solution would be to program a GA144 simulator as soft core
14:38:44 <Mat2> that's a nice project, it is a good application after finishing my retro version
14:38:56 <Mat2> I think
14:52:20 <foucist> well, i've been pretty much anti-ANS from the beginning of my forth journey a good 12 years ago :P
14:52:37 <foucist> i mean, i felt it kind of distroyed some aspects of forth
14:52:48 <foucist> ruined the community a bit etc
14:53:28 <foucist> i think it makes more sense to standardize on whatever chuck moore or other high level forth guys are doing :P
14:53:47 <Mat2> I agree
14:55:54 <foucist> or rather, the ANS thing came far too early
14:55:58 <foucist> far too soon*
14:56:41 <foucist> Mat2: i ended up becoming a ruby coder after forth..  the sheer readability of ruby..  ahhhh
14:56:52 <foucist> a breath of the freshest mountain air
14:57:06 <foucist> would be nice to combine the two heh
14:58:02 <Mat2> *g*, ok I'm more someone who prefers a postfix APL like notation
14:58:32 <Mat2> (or mathematica)
15:00:05 <Mat2> at school I had learned programming in pascal and my first language was commodore basic and later assembler
15:00:22 <Mat2> (6502 and IA32)
15:09:19 <Mat2> it is after midnight here, I will go to bed, see you all
15:09:21 <Mat2> ciao
15:09:33 --- quit: Mat2 (Quit: Verlassend)
16:09:32 --- join: arescorpio (~arescorpi@222-206-17-190.fibertel.com.ar) joined #retro
16:37:06 --- join: timg (~timg@pool-71-180-13-77.tampfl.fios.verizon.net) joined #retro
16:37:11 <timg> whoa
16:37:16 <timg> this guy might actually do it
16:37:22 <timg> http://www.kickstarter.com/projects/adapteva/parallella-a-supercomputer-for-everyone
18:53:53 <docl> looks like they topped the goal already
18:57:51 <timg> yep
18:57:54 <timg> by 10k
18:58:00 <timg> I kindof want one
18:58:07 <timg> just don't know what I would do with it yet
18:58:19 <timg> already have a raspberry pi i barely use
19:21:17 --- quit: tangentstorm (Quit: leaving)
19:25:44 <foucist> hmm
19:25:55 <foucist> so $500 bucks for 72 cpus in a box
19:26:02 <foucist> hmm
19:26:06 <foucist> i bet bitcoin miners are buying this
19:26:17 <foucist> i bet at least half of the pledgers are bitcoiners
19:30:29 <timg> ha
19:30:57 <timg> i still haven't used one bitcoin
19:32:39 <foucist> heh
19:34:25 <foucist> funny, a 64 core parallela chip has the same glops as a ga144 chip
19:34:44 <foucist> ga144 has 96 gflops
20:19:23 --- quit: arescorpio (Read error: Connection reset by peer)
20:21:06 <kumul> im currently using an editor with no color support, and i often find myself writing pseudocode in pencil/paper, that sort of beats colorforth for me
20:22:11 <kumul> its funny, but no color, forces me to organize the code in a very neat way (or maybe its making me delusional)
20:25:30 <foucist> kumul: ah, yeah if there's no syntax hiilghting i guess that's true
20:25:39 <foucist> but i never had syntax hilighting with regular forth either
20:25:47 <foucist> are you talking about forth code or some other language?
20:36:50 <kumul> well, i mostly write shellscripts and forth. maybe no syntax highlighting just forces one to manually organize code a certain way, not necesarily a good thing just a time waster
20:39:12 <foucist> kumul: well, do you limit yourself to oneliners with forth?
20:39:16 <foucist> i pretty much do
20:39:20 <foucist> or did
20:39:40 <foucist> haven't touched forth in a long time
20:41:48 <kumul> if possible yes
22:06:02 --- join: Z_Mass (~zachary@bas1-cornwall24-1242467096.dsl.bell.ca) joined #retro
22:36:03 --- quit: Z_Mass (Quit: Leaving)
22:36:55 --- quit: kumul (Quit: WeeChat 0.3.9)
23:59:59 --- log: ended retro/12.10.26