From CG community
Revision as of 12:00, 15 June 2021 by Chat Log (talk | contribs) (Created page with "<img src=/a/48688403697191> Cybersick: I love you <img src=/a/48688403697191> Cybersick: :heart_eyes: File:Default_avatar.png Angelburgie22: wrong chat dude <img src=/a...")
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to navigation Jump to search

Cybersick: I love you

Cybersick: :heart_eyes:

Default avatar.png Angelburgie22: wrong chat dude

N_Flamel: lmao

N_Flamel: lmao

v1nam: lol

Default avatar.png henriq-88: Is it possible to avoid certain clashes? Don't like the shortest code ones, makes me write terrible code lol

Default avatar.png zhoubou: Not on public ones, no.

Totomns: in shortest you have time to practice languages like ruby, python, js, or bash

Totomns: there is a code golf section on stackexchange, you should look

Ludwigdn: ^ +1

kovi: 1st uttt game won vs ka'liso timeout

karliso: nice

kovi: nice would be a real vs. top2

kovi: real win

kovi: and there it is!

DomiKo: wow GG

kovi: thx

DomiKo: top10 in such a short time very nice

Default avatar.png Angecide: wow kovi ur bot suddenly got very strong

kovi: fixed a bug in mcts yesterday and now some heurstics work

eulerscheZahl: now poor smits has to amend his book

kovi: yay, vs. rebless as well

Default avatar.png Angecide: do you do any books?

kovi: no

DomiKo: heurstics are teccles and that one with not giving free move?

kovi: yes

kovi: another one, so karliso back to #1

Default avatar.png Angecide: do you do rollout from each expanded children?

Default avatar.png Angecide: casue the numbers look low, I guess maybe just very good heuristics

kovi: only if the same parent is chisen

kovi: almost 50% vs top2, wtf

MSmits: gj kovi

MSmits: btw, contrary to popular belief, books do almost nothing in uttt, unless someone else uses it and you hardcounter them

MSmits: or someone is very deterministic and thus also counterable

MSmits: I tested my bot vs top 10 without any book and it basically does 40% win vs top 2, 50% vs tomatoes and > 50% below

Default avatar.png Angecide: interesting, I imagined if you always encounter the same openings, perhaps it could be helpful, but I can see how just using faster bot + heuristics can achieve the same

MSmits: most of the action happens due to mistakes after ply 20 I believe

MSmits: the only way to make a book go that deep is to have an opponent that is very predictable

MSmits: there was a book war on uttt for a while, i won that and now everyone stopped using them, more or less

MSmits: and we're all better for it :)

MSmits: I guess miarem still has one, but it's not countering anyone anymore because all the bot versions it countered are gone

kovi: top10 was lucky. leaderboard of 2 game inparity, still depends on fast early advancement. if i reach top20 i can get 1:1 vs. top10 and reach high. i still resubmit because i have enough new ideas. i am thinking about some weaken-teccles thing. also i still wasnt able to enable the wider win position assessment

MSmits: what do you mean weaken-teccles?

MSmits: like... capping it to early game ?>

MSmits: that's probably wise

kovi: currently i give up center a lot as p2

MSmits: also... i wouldn't worry so much about the submit. It really is not very reliable. Set up a CG bench test pool. Even if you use the same number of games as a submit, the test will be better because you don't waste games vs weaker players

MSmits: so you wouldnt even be taxing the server any worse than you do now

MSmits: giving up the center as p2 is not necessarily a bad thing

MSmits: p1 already has a huge advantage there

MSmits: the sacrifices you need to make to still win in the center might not be worth it

MSmits: In general, losing miniboards can be a good thing if the opponent needed to play far more squares there than you did

MSmits: (means he has less elsewhere)

kovi: valid points

MSmits: it depends a lot on the game too... uttt is hard to analyze as a human

kovi: escpecially after a week

MSmits: a week?

MSmits: oh you mean you've played uttt for a week?

kovi: i start it a week ago. deliberately skipping all chatting before

MSmits: ahh ok

Default avatar.png RachitSingh: hi all

Default avatar.png RachitSingh: I am here to learn python and get expertise in it

MSmits: well you can get into uttt fairly quickly. It has serious diminishing returns though

MSmits: RachitSingh welcome

kovi: hello RachitSingh

MSmits: you did pretty well i'd say kovi, comparable to how quickly karliso got good

Default avatar.png RachitSingh: greetings and thanks

MSmits: cant speak for my own speed at this, i couldnt even code very well when i started this. It's a good starter multi

kovi: well, i have switched to uttt to make my mcts working. not sure if i want to finetune it more than top10

MSmits: well it's pretty addictive and the fact that it keeps coming up in chat risks drawing you back in

MSmits: but it's probably better not to get stuck on it

kovi: true. for most contest it is easier to give up after contest ends and spending a few days

MSmits: kovi, did you use a lookup for the big board? Like a 18 bit?

MSmits: I am trying to cut down on lookups, but i wonder if it's worth turning that one into a calculated value instead of looking it up

kovi: currently i use two lookups for global

MSmits: ok so, lookup p1, lookup p2

MSmits: that works I suppose

kovi: right...actually three

kovi: p1, p2, done

MSmits: ah

kovi: not utilizing draw yet

MSmits: I have a single lookup for this

kovi: (could be done - p1-p2)

MSmits: uint8_t bigBoardStatus[65536];

kovi: makes sense

MSmits: it's 18 bit, but needs 2 bit per result

MSmits: i mean 2 out of the uint8_t, so one lookup is actually 4 lookups out of which i shift 1 result

MSmits: makes it smaller

MSmits: you can do more complicated stuff in the lookup this way, like count drawn boards and such

kovi: i havent spent much on optimizing, but there seems to be single chokepoint

MSmits: what is it?

kovi: no single point

kovi: sorry

MSmits: no right

MSmits: you're right

MSmits: many little things

kovi: playout reach select/expand cost

MSmits: the big board lookup is about 10% of the total runtime though

MSmits: for me

MSmits: but it does a lot so...

kovi: keep single board heuristic reduced the cost of 9table action collection

MSmits: 9table action?

MSmits: whats that?

kovi: giving free pick

MSmits: oh ok

MSmits: yeah thats an expensive part

MSmits: one you can optimize though

MSmits: and there's 2 main ways to random too

MSmits: single random with loop or double random (random board, random square)

MSmits: double random is faster, but it always reduced my winrate a little bit

kovi: i know, i may add that even thought it is not correct

MSmits: could be different for everyone though, as with a lot of uttt things. Stuff interacts

MSmits: best to try it yourself


MSmits: gotta go for a bit, hf

Default avatar.png FunPlanet: is there a codingame api? (suppose not)

Default avatar.png LeeHari: no

eulerscheZahl: yes

eulerscheZahl: but not documented

eulerscheZahl: unofficial:

eulerscheZahl: how to get the requests manually:

MetehanBayraktar: guys do you know how to print like this in c++ /// 1 2 3 4/// I mean with spaces ?

MetehanBayraktar: not like cout << 1 << " "<< 2

MetehanBayraktar: I mean output an array

MetehanBayraktar: int a [4] = {1,2,3,4} how to output this like 1 2 3 4 without space at the end

Andriamanitra: a for loop is probably your best bet

eulerscheZahl: cout << a[0] for (int i = 1; i < 4; i++) cout << " " << a[i];

MetehanBayraktar: man thats happen space at the end

MetehanBayraktar: hmm I understand what you mean ok thank you

eulerscheZahl: much easier in other languages

eulerscheZahl: like ' '.join(array) in python

eulerscheZahl: or string.Join(" ", array) in C#

Andriamanitra: or array*" " in ruby

Default avatar.png ErrorRazor: Did someone get 100% in this puzzle?

Andriamanitra: not yet but give me 4 minutes

1400179: No I got 0 0 00 00%

Default avatar.png ErrorRazor: lol

Default avatar.png ErrorRazor: I was just wondering if you could explain this part in the expected output: "0 000000 0 0000"

Default avatar.png Maxim251: How I can resize wondows, that I can see game info and standard output ?

Default avatar.png ErrorRazor: Looks like two consecutive blocks of ones to me

Default avatar.png Maxim251: I can move midlle one to left and right, but I cant move down, up..

Default avatar.png ErrorRazor: You can expand the game info window

Default avatar.png Maxim251: Oh yes, I see...

Default avatar.png Maxim251: and window for the code? I am on laptop and I dont see to much in coding window

Default avatar.png ErrorRazor: Yeah that can be expanded as well

Default avatar.png Maxim251: but I want keep game info and window code..

Default avatar.png Maxim251: I want move only bottom part litlle more to the bottom

Andriamanitra: @ErrorRazor i don't see "0 000000 0 0000" anywhere in the expected outputs

Default avatar.png ErrorRazor: Test case 4

Default avatar.png zhoubou: >>> testcase.find("0 000000 0 0000") -1

Default avatar.png ErrorRazor: I wonder why it shows this then

Default avatar.png zhoubou: Is that your debug output?

Default avatar.png ErrorRazor:

Default avatar.png zhoubou: Weird

Default avatar.png ErrorRazor: Very weird

Andriamanitra: which test case is it?

Default avatar.png ErrorRazor: 4

Default avatar.png zhoubou: 4

Default avatar.png ErrorRazor: I think it also shows my output wrong

Default avatar.png ErrorRazor: nvm

Default avatar.png Maxim251: how I can stop program after first line? that game will not play while loop? Is that posible

Default avatar.png zhoubou: exit()?

Default avatar.png zhoubou: In Python

Default avatar.png Maxim251: C++

Default avatar.png zhoubou: I think C++ has exit too

Default avatar.png Maxim251: Ok, I check

Default avatar.png Sony541: if it's part of "main" function you can use "return"

Default avatar.png Maxim251: No call function exit(). I think this is not that

esfes: dude

esfes: the one that needed help with the cuck norris problem

Default avatar.png Maxim251: I can Try... But in main function there is while loop,

esfes: do you still need help?

Andriamanitra: there definitely is exit() function in c++, might need to include stdlib.h though

Default avatar.png Sony541: return will still work in "while" loop :)

esfes: it should be in a function tho

esfes: return must be in a function

Default avatar.png Sony541: Yeah, he said it's the case

esfes: oh, i didn't read that lol

esfes: sorry

Default avatar.png ErrorRazor: esfes yeah I'm still here

esfes: do you need help?

Default avatar.png Maxim251: No game is still running.

esfes: wdym?

Default avatar.png ErrorRazor: Well, I was just wondering why the expected output is displayed incorrectly

esfes: wrong person lol

Westicles: exit(1);

Default avatar.png zhoubou: Correct response though esfes :P

Default avatar.png Sony541: Yeah, or "return 0;" :)

esfes: remember that it expects a 7 bit binary representation

Default avatar.png Sony541: Didn't write on C++ for 7 years :D

Default avatar.png zhoubou: Yeah, but why is the expected output wrong? esfes

esfes: you must append 0 before to make it a 7 bit if its can be represented in less bits

esfes: its not wrong for me it worked

Default avatar.png zhoubou: Look at his screenshot

Default avatar.png ErrorRazor: It's not wrong, the IDE shows it wrong

Default avatar.png Maxim251: IS still running, Ok never mind, I will do like olways, stop playing and rewind messages to beginning

esfes: show me your code

Default avatar.png Maxim251: Thanks for help

esfes: zhoubou

Andriamanitra: @Maxim251 oh i understand what you were trying to do now, the game won't stop even if your code stops, i don't think there's any way to abort a game mid-way through

Default avatar.png zhoubou: ? esfes

esfes: send a screen shot of your game

Default avatar.png zhoubou: My code works fine

esfes: of your code*

esfes: you sure?

Default avatar.png zhoubou: Yeah, it passes all test cases

Default avatar.png ErrorRazor: That screen shot was from me

esfes: oh lol

esfes: Error show your code please

Default avatar.png zhoubou: There is something wrong with your binary conversion I reckon

Default avatar.png zhoubou: But I don't get why expected output is wrong

Default avatar.png ErrorRazor: Yeah I know my solution is wrong

Default avatar.png ErrorRazor: It's not the point

esfes: what was the problem?

esfes: if the solution is wrong it would show a problem with your answer

Default avatar.png zhoubou: It's showing a wrong solution

esfes: the expected output is not wrong

esfes: how do you know that?

Default avatar.png zhoubou: As per screenshot

struct: it is wrong

Default avatar.png ErrorRazor: It's not? 2 other people checked and got a different expected output

Default avatar.png ErrorRazor: Now 3

esfes: how did you identified the wrong part?

Default avatar.png zhoubou: There are 2 strings of 1s

struct: There is a red circle on it

Default avatar.png zhoubou: And I copy pasted a real testcase and used str.find, and it didn't find what was in the screenshotž

Default avatar.png zhoubou: screenshot*

esfes: what if you copy his solution and check on your ide

esfes: and if its right he will just submit

Default avatar.png zhoubou: Hmm?

esfes: you are getting the right test output right?

Default avatar.png zhoubou: Yeah

esfes: copy his code test if its right and if it is he will just submit

esfes: it tests server side before confirming the solution at the submit point

Default avatar.png ErrorRazor: Well here's my code if you want to try to reproduce it

Default avatar.png ErrorRazor: I know it's long

esfes: its wrong at the end for me too

Default avatar.png zhoubou: Oh I can reproduce it

Default avatar.png zhoubou: It's language related then

Default avatar.png zhoubou: Or something

esfes: or maybe the is a bug with longer strings

Default avatar.png zhoubou: And the testcase is different...

esfes: wdym?

Default avatar.png zhoubou: Let me see one more time

Default avatar.png zhoubou: Oh nvm

esfes: i think there is a bug that only shows on longer strings

esfes: or there is a bug with the c++ checker

Default avatar.png ErrorRazor: When I search for that part with ctrl+f, it doesn't find it in my console output

esfes: compile that program on your pc

esfes: and check the string

esfes: send us the string when you get the output on your pc

esfes: we will check it

Default avatar.png zhoubou: Oh you're right, it isn't in stdin

Default avatar.png zhoubou: stdout*

esfes: wdym?

Default avatar.png zhoubou: It isn't printed, but it's shown as his solution

esfes: wdym it isn't printed?

Default avatar.png ErrorRazor: By my program

esfes: cout is a print in c++?

esfes: i am a noob when it comes to c++

Default avatar.png ErrorRazor: Yeah stands for character output

esfes: so it is printed right?

esfes: or am i missing something?

Default avatar.png ChristianPR: ive just started on codinggame, Just beat the Othello bot and im still in league 2? how do I move up?

Default avatar.png ChristianPR: or does it take time?

Default avatar.png ErrorRazor: If you've submitted your code and you got first place, then you just need to wait

Default avatar.png ChristianPR: cool, thanks

Default avatar.png ErrorRazor: Anyway, I'm gonna go get groceries. Don't lose your sanity over this bug guys, the puzzle still works, only the error is displayed incorrectly

esfes: try to run it on your pc

esfes: like compile it

struct: ChristianPR you need to be above the boss

struct: Its not easy

Default avatar.png ChristianPR: hmmm, how come there are lots in league 1 but lower than the bot?

struct: Their bot is not better than the boss

struct: Stronger bots get promoted

Default avatar.png ChristianPR: o i thought i just needed to beat the boss hahah

jacek: oh my

jacek: maybe next week will be othello puzzle :V

eulerscheZahl: 1 minute left for blockout

eulerscheZahl: hurry up!

kovi: not approved?

jacek: needs 2 approves

eulerscheZahl: nope

kovi: omg

jacek: and i dont have 20lvl on my alt

eulerscheZahl: on a sidenote i'm having fun with topcoder this time


eulerscheZahl: wait what?

PatrickMcGinnisII: eulerscheZahl, ha

eulerscheZahl: i was kidding with the hurry up

kovi: too bad :)

PatrickMcGinnisII: it did say 1 minute

eulerscheZahl: oh, I forgot the cover image

kovi: duh, who approved that without ci

kovi: another swap in top1-2

PatrickMcGinnisII: illedan's fault

eulerscheZahl: join us on topcoder, we swap every few minutes

eulerscheZahl: something is wrong with the platform

kovi: still?

eulerscheZahl: "Thank you both. Yes it could be a problem with multiple testing machines. We will look into it." and that guy from topcoder could start at CG right away

kovi: wow, top3?

eulerscheZahl: i think siman is overrated atm, that submit had like 80 points before

eulerscheZahl: weird system

eulerscheZahl: and still 3 days to go

Default avatar.png Angecide: I am a bit confused about topcoder, which games/puzzles are u working on and which leaderboard?

jacek: jewel

eulerscheZahl: leaderboard:

eulerscheZahl: task:

eulerscheZahl: swap any 2 cells. 3 in a row (not diagonally) remove that row and can cause a chain reaction

eulerscheZahl: points for long lasting chains and low rows

Default avatar.png Angecide: ohh that looks interesting

kovi: bugged relative scoring? (happened a lot with the "new" - two years old system)

eulerscheZahl: yeah, it's their relative scoring again. not updating the total best properly, thus giving wrong partial scores

JLukeSkywalker: what do i do if i made a clash, it got approved, then someone copy-pasted it and got theirs approved a couple days later?

JLukeSkywalker: do we just let both exist?

eulerscheZahl: real 1-1 copy-paste?

jacek: oO

eulerscheZahl: the only thing we can do is getting staff involved. the question: is it worth bothering them for it?

jacek: for the 5%?

kovi: anyone using something different than uct? maybe something faster?

Default avatar.png Angecide: u mean like a different equation or algorithm?

jacek: do you do lookup table for your log and sqrt?

struct: is the lookup even worth it?

kovi: first try failed on too big numbers

MSmits: I used to do lookup table and I never succeeded in making it faster without that. But I threw it away anyway.

MSmits: mostly because i had to cap my rollouts

MSmits: it's just annoying

jacek: oO

MSmits: kovi there are many different ways to change the selection mechanism a little bit. Never seen much of an effect in uttt

jacek: and i never did UCB-tuned for example

MSmits: i just use basic uct

kovi: yeah, i just noticed that sqrt is a chokepoint now

jacek: math formula :unamused:

jacek: sqrt has its own instruction

MSmits: oh kovi, sec, there's a way to speed that up if you havent already

Default avatar.png Angecide: u can do sqrtf instead of sqrf, keeping it as float instead of double, I think that performes faster

kovi: i just did that :)


MSmits: thats what i do in othello, not in uttt because i add a bonus to value, so it doesnt work then

MSmits: I use these as well:

MSmits: inline float fastlogf(const float& x) { union { float f; uint32_t i; } vx = { x }; float y = vx.i; y *= 8.2629582881927490e-8f; return(y - 87.989971088f); } inline float fastsqrtf(const float& x) { union { int i; float x; }u; u.x = x; u.i = (1 << 29) + (u.i >> 1) - (1 << 22); return(u.x); } inline float rsqrt_fast(float x) { return _mm_cvtss_f32(_mm_rsqrt_ss(_mm_set_ss(x))); }

MSmits: it's not my code so i dont mind sharing

kovi: oh, interesting

MSmits: just some crap off stack overflow or some other google

MSmits: it is just a tiny bit slower than caching your sqrts :P

MSmits: for me it was anyway

MSmits: oh and using AVX to calculate 8 sqrts in parallel didnt help me either

MSmits: in case you wanted to try that

kovi: i havent thought about that

kovi: i used a big array for vparent, vchild pairs

kovi: calculation the whole formula at once

kovi: but it was too big

1400179: Feels like 2019 again

MSmits: well new people keep trying the game so it's gonna keep coming up reCurse :)

jacek: just before final episode of mlp?

Default avatar.png Angecide: 2019 was the peak of uttt?

1400179: Dunno, could be 2018 too, time flies

MSmits: nah uttt peaks several times a year

kovi: yeah, sorry as i said today...i totally tried to avoid uttt discussions all along

MSmits: it's just someone starts talking about it and then a bunch of players join in

1400179: Wasn't complaining actually, sorry it seemed that way

MSmits: it's fine kovi, don't worry about it

jacek: but you alawys complained about uttt

MSmits: dont bite reCurse :P

kovi: at least with all these resubmit, it seem that recurse is better vs. me than karliso

MSmits: the other way around for me it seems

1400179: A boost working my way? No way

kovi: and i might push msmits to #3

MSmits: let me strip my book and resubmit, it's not like there's any point to it now

MSmits: dont think its doing anything

MSmits: think this is the right version

MSmits: damn those submits are fast

MSmits: I am used to community submits

MSmits: oh nvm it's the 10x + swap

MSmits: that makes it seem fast

jacek: community?

MSmits: contrib

jacek: uttt is contrib?

MSmits: no, thats why it surprised me the games came so fast

MSmits: i havent submitted a non-contrib for a long time

kovi: wow, i badly hurt jacek placement

jacek: i just resubmited

kovi: oh, nvm

MSmits: good... more games

jacek: now lets csb do the promotion

MSmits: dont touch csb!

MSmits: it will wreck the server :P

MSmits: uttt games are weird in that a draw + win is just as good as win + win

MSmits: with swaps i mean

reCurse: How is that weird

MSmits: it's not wrong i mean

MSmits: it looks weird

MSmits: i guess with the older games, swaps were more common, like in tron

reCurse: You mean back when they were trying? :P

reCurse: Sorry lost my zen

MSmits: the good old days :)

jacek: no longer amd?

jacek: you bought that intel?

reCurse: Wonder if you ever lose your trolling

MSmits: doubt it

eulerscheZahl: a worthy follower of Cyberpunk

struct: new intel seems to be good for the winter

reCurse: Besides gotta keep some energy for tomorrow :p

jacek: hm?

MSmits: it's not cyberpunk, jacek does not fill the right side of my screen

eulerscheZahl: true, no flooding

eulerscheZahl: and jacek doesn't even know cyberpunk, does he?

MSmits: probably not

jacek: the game?

reCurse: He's waiting for the cyberpony mod

kovi: blame msmits for another swap

MSmits: just lost to him again

MSmits: and drew karliso

MSmits: so its ok

MSmits: again draw karliso

struct: How do i see contests leaderboards?

MSmits: finally some wins vs supson, hes pretty tough also

struct: Its so confusing

struct: I cant get there

MSmits: where?

struct: Any past contest leaderboard

MSmits: what do you mean by "get there"

struct: navigate the page

MSmits: weird

struct: and get to the page where it shows the leaderboard

reCurse: Random thought experiment: do you think a better way to solve draws (e.g. in uttt) would be to make the one who won quicker the winner instead?

jacek: would that favor p1 more?

kovi: interesting...add that into mcts?

MSmits: mmh

reCurse: No in a pairing

reCurse: Not a single draw game

Default avatar.png Angecide: winning faster sounds really cool

reCurse: If pairing ends in W-L then one who won quickest gets the series

MSmits: i do wonder if that would just make the leaderboard more volatile

reCurse: Depends

reCurse: Would stronger bots win faster reliably? Or lose longer?

reCurse: If it's more reliable then it's less volatile

MSmits: yeah you could be right

reCurse: Might pervert the game itself a bit though

reCurse: I don't know

MSmits: but the source of the draws is mostly the p1 advantage i think. You're applying a patch when its better to attack the root cause i think

reCurse: Or is the right word in english subvert

jacek: hentai the game?

reCurse: The root solution is to delete uttt

reCurse: So next best thing

MSmits: best thing would be steal option

reCurse: More of a game design thought process

reCurse: I hate that

MSmits: it's such an easy solution

reCurse: It's so lazy

MSmits: yes, easy and lazy

MSmits: it works

reCurse: Pushes the burden of balance on the player

reCurse: It's stupid

reCurse: I think it's horrible design

MSmits: if you dont have time for extensive playtest, it's great

MSmits: sure, playtest with real balance would be better

reCurse: And you don't even prove there exists a neutral state

reCurse: You just hope

reCurse: It's garbage

MSmits: it's impossible to make it worse, and because there are 15 unique moves, it is certain one of them is at least better if not nearly balanced

MSmits: I think top board middle edge was nearly 50% on meta mcts

MSmits: with the center square being close to 70%

reCurse: If you're going to rely on meta mcts as an estimate

struct: just make the 1st move random

reCurse: You can rely on meta mcts as an estimate to not garbage design

reCurse: Design it so it starts as a neutral state

MSmits: but my point is, it can only improve things and it is very likely to. Especially if it is uncertain which starting move is most balanced, that leads to more openings and causes books to be even less attractive

reCurse: It's only improving things because you failed your job as a designer

MSmits: you can frame it that way, sure, doesnt invalidate my point :0

reCurse: I think relying on that is an admission of failure

reCurse: I can't figure out if the game I designed is even balanced so do it for me

MSmits: well it's a simple fact that interesting boardgames are sometimes hard to balance

reCurse: Sure but it's a copout

MSmits: sure

reCurse: And if there exists no neutral states you didn't improve much

MSmits: well, removed full book, based on 2 million+ games. Same rank on submit. That's how useful books are on uttt :)

reCurse: Or did you just hit under the margin of the leaderboard reliability

reCurse: I know where I put my money on

MSmits: nah, i did cg bench tests. It's just that books are only useful when hardcountering players.

reCurse: Ah ok

MSmits: on uttt that is

reCurse: You only mentioned submit so

MSmits: ah yes, i guess its not real proof

MSmits: I'm completely redoing my bot, getting rid of most lookups, going back to simple binary representation and such.

reCurse: If you can hardcounter kar a bit more

reCurse: Would appreciate

MSmits: I have a book version that wins vs his p2 100%

MSmits: his current version

MSmits: never submitted it

reCurse: Well I didn't mean counterbook but ok

MSmits: nah i am done with that

MSmits: he can just submit a new version

MSmits: then the effect is gone

MSmits: i like that noone books now

MSmits: afaik

reCurse: No one?

MSmits: tomatoes doesnt

MSmits: karliso doesnt

MSmits: the rest is lower so... dunno

MSmits: miarem still has his, but its not good

reCurse: 17?!

MSmits: he always got his rank from countering players

MSmits: his bot was weak

reCurse: I see

MSmits: everyone has new versions now

reCurse: I could improve mine now but meh

reCurse: It's all about the checkers

MSmits: improve your bot or improve your book?

MSmits: i'm assuming bot

jacek: im assuming book

reCurse: My last clean training could inch the rank 1 but not very convincingly

jacek: checkers?

struct: did you manage to use the dtw db?

MSmits: it's my belief karliso is not that far from perfect play

reCurse: Yeah

struct: nice

reCurse: My new training with 8piece db does a very good job at landing a good position

reCurse: But doesn't know how to win it

reCurse: That's next

MSmits: ah the WLD db

reCurse: I got wld + dtw thanks to struct

MSmits: how so

MSmits: did he find something?

reCurse: Yeah

reCurse: It's 40gb lol

MSmits: thats cool

MSmits: wow!

reCurse: It's totally where the edge is

reCurse: Analyzing positions with my clean bot

reCurse: It was often good but has no idea how to win

reCurse: Lots of shuffling necessary

MSmits: yes, just like every other bot

MSmits: except yours is more likely to win i guess

reCurse: All except db dr

reCurse: There it's all draws

reCurse: Few rare wins

reCurse: I could tweak to 75% winrate but then the others take a drop

reCurse: Like tric drops to 35% for whatever reason

MSmits: tric trac does some great ab bots

reCurse: True but the normal params give me 70% iirc

struct: Do you want the link too MSmits?

reCurse: So seeing it drop that much is surprising

MSmits: struct of course :)

MSmits: reCurse probably down to determinism again.


struct: Scroll to the end

reCurse: According to game lengths it shouldn't be that much deterministic

reCurse: But maybe

reCurse: I'm really surprised how an innocuous move can change a win in a draw

MSmits: cant beleive i missed that struct. I am sure i saw that page

MSmits: thanks

struct: np

MSmits: reCurse i am not at all surprised. At least in othello it happens all the time

MSmits: complexity is similar

reCurse: Yeah but othello is messy and unreadable

MSmits: true, AI doesnt care about that though

reCurse: Yeah but my human eyes care about it

reCurse: So they can see surprises when otherwise it's generally readable

MSmits: mine too... checkers is so much better in that regard

BetterPolu: Hey, i saw someone reading the test file in python with open()

BetterPolu: but now i can't find the contest anymore. Does anyone know, what the test file is called?

MSmits: I saw a squirrel the other day

BetterPolu: oh nice :)

reCurse: Did you run after it? If so you may be a dog.

MSmits: lol

MSmits: woof

MSmits: not sure what you're talking about BetterPolu

BetterPolu: hmm, with open("file.txt","w") you can open files in python. And since I am playing a lot of Clash of Code, it would be useful to read the entire file at once for shortest for example. However, if I don't know the file name, I can't open it :/

MSmits: you saw someone use open in a clash?

eulerscheZahl: let's just switch to bash and run "ls"


BetterPolu: yes MSmits

MSmits: better listen to eulerscheZahl

eulerscheZahl: so it's the in<xxx>.txt files

MSmits: he's a guru literally. My title is just eyecandy

eulerscheZahl: and on the out files you have the solution. but it says permission denied

eulerscheZahl: hey, i'm the one playing the candy crush contest right now ;)

MSmits: with a bot?

eulerscheZahl: of course

BetterPolu: thx, eulerscheZahl. But for the testing process you are giving only one test file, right?

eulerscheZahl: topcoder optim contest

MSmits: ahh i see, well good luck with the candy then :)

eulerscheZahl: i see multiple files BetterPolu

Westicles: back to lvl 55, congrats

eulerscheZahl: and this time honestly (kind of)

eulerscheZahl: i jokingly said "hurry up, 1 minute remaining to approve Blockout"

eulerscheZahl: now it's approved

MSmits: lol

Westicles: oh good, didn't deserve a 3rd loop

eulerscheZahl: was indeed 1 minute before the codinbot would have kicked it again

BetterPolu: btw huge fan of eulerscheZahl and MSmits lol

struct: just dd in bash

BetterPolu: how did you get so pro

struct: and you get the entire file

struct: on ruby I usually do

jacek: oh my

struct: p `dd`

struct: and rest of the code after

eulerscheZahl: is cat broken?

struct: no

MSmits: mmh i am only "pro" in a very limited type of botcoding... not general at all. euler can do everything

eulerscheZahl: too much of the honors

BetterPolu: oh, ok cool. Unfortunately I am not so good with either ruby or bash

struct: sometimes i also use tail -1 to get last line

eulerscheZahl: but can you also get the last line using head?

MSmits: BetterPolu i can do C#, a bit of c++ and python

MSmits: no ruby or bash or clashing very well either

MSmits: most of us dont clash

MSmits: we write bigger programs for the other types of competition

BetterPolu: why tho?

MSmits: it's not interesting to us to fight short 15 min battles

BetterPolu: ohh, ok. Maybe i will try the other types too

MSmits: you can spend weeks on a single one and keep finding new ideas and improving

MSmits: much more rewarding


struct: I think that siman is the same from the contests

MSmits: mmh that name is wrong. It sounds like he wants you to golf in style

MSmits: but I'm sure he wants you to golf better

BetterPolu: hmm maybe i should learn ruby

Default avatar.png adilmarghadi1: hi

Zenoscave: MSmits it makes sense to me. perhaps translation issues?

VladimirAngelov: you can do 45

Zenoscave: pewpew

MSmits: maybe... i was thinking more like this: If there is a python "style guide" it's not about making your python faster and shorter, it's just about making your compatible with the rest of the industry so they can read it. I interpreted it in that way.

Zenoscave: I read ruby golf-style guide. not ruby golf -style-guide

Zenoscave: slight nuance i guess

MSmits: sure, but it's still style, that is more about form, not substance, could be translation I dunno

Zenoscave: For me the golf-style means a golf form of coding. synonymous to my eyes

MSmits: ohh ok, sure that makes sense

Zenoscave: What are you working on now MSmits? I've been away for too long not sure whats popular now

MSmits: i might be the wrong guy to ask, i always relapse into older games

Zenoscave: Same ;)

MSmits: unfortunately i am rewriting uttt again :P

Zenoscave: I should revisit. Minimax is slow for it

MSmits: not slow, just less effective

MSmits: hard to eval

Zenoscave: No, mine is also slow lol

MSmits: ahh ok :)

MSmits: I did an early playout termination version

MSmits: basically mcts with no random rollout, just eval

Zenoscave: Where is the compete tab? I cna't find it

MSmits: worked ok, got around rank 20-30 back then

Zenoscave: did the layout change?

MSmits: if you try from here its the same:


MSmits: they are messing with it yes

Zenoscave: gotcha

struct: Still need to finish fireworks ai

MSmits: go go :)

kovi: ok, now i see what you meant by uttt and heuristics

MSmits: stop helping people with your google mastry and you gamedesign skills and start botting

MSmits: how so kovi?>

kovi: i slowly weakened my heuristic which pushed me from 100->50 and now i can reach 10 without it

MSmits: not sure what you mean by heuristic though?

MSmits: or weakening it

MSmits: do you mean doing it in stages of the game and not in other stages?

kovi: using it less/more

MSmits: like teccles in 1-20, avoid free move 1-25, that sort o fthing

MSmits: right

kovi: only one thing is consistent

kovi: utilizing early detect for non 3inaro wins still hurt

MSmits: yeah

kovi: i think i wiill settle with a top10 and move along

MSmits: makes sense

egaetan: Daporan ?

MSmits: well think of it this way, even if you somehow get top 3 or something, it's always temporary in uttt. This is the game that never stops :)

MSmits: eh hi egaetan

MSmits: are you looking for him specifically?

DomiKo: I'm so jealous of your work kovi

egaetan: yes

egaetan: appearing in sopra sterai challenge

MSmits: havent seen him here in a long time

egaetan: last minutes

reCurse: Didn't he get the memo about the prize? :rofl:

reCurse: :popcorn:

struct: pm him on forum egaetan

Zenoscave: oh no recurs e brought out poopcorn

MSmits: :poop: :corn: ?

reCurse: This is going to be entertaining

Zenoscave: My keyboard is duplicating strokes again

MSmits: and making everything funnier

Zenoscave: yuup

kovi: im really thinking about nn

kovi: when the heuristics are so...weird

MSmits: the heuristics in uttt are not weird, just a bit harder to see

kovi: what is the difference in finetuning a few

struct: Do you have experience with it kov i?

kovi: yeah, i meant that

kovi: not yet struct

MSmits: btw, there is no reason you cant still use heuristics with a nn

kovi: thx domiko, but really it is just fast mcts + solver. not much else left

MSmits: the few heuristics do make the difference

kovi: simple heuristics are probably better encapsulated by nn

kovi: better finetuned

MSmits: sure, but if a heuristic is 100% reliable, you're lightening the load on the nn, by taking it out maybe.

kovi: true

MSmits: I'm just thinking about limiting training time and maybe the size of the network and such

MSmits: personally i would fit as much domain knowledge in there if i could. It's not how robo and re curse do it

MSmits: Jacek does this sometimes i think

kovi: yeah, but if domain knowledge is simple our own nn's discoveries?

reCurse: Oh and it's not even that clear on the page that it's prizes for fr only

MSmits: kovi yeah obviously i dont have experience to prove it. Just thinking out loud

kovi: and what if it is only 95% reliable...well trained nn can realize the 5% (pun intended)

MSmits: thats true

MSmits: though... I have not seen a nn on CG that evaluated perfectly

MSmits: so it might not beat a nearly perfect heuristic

MSmits: I think they mostly shine with complex patterns you can't separate with simple heuristics. Most games have those

kovi: the problem is that heuristics overlap

kovi: and if you cant separate them (optimize them)

MSmits: you mean eval heuristics now

MSmits: thats true

MSmits: pruning heuristics are different though

kovi: i meant heuristics in more complex games

MSmits: ah ok

MSmits: it's hard to compare a NN to a "regular" bot anyhow. They're so different

MSmits: regular bot uses much more brute force

MSmits: as in sims, treesize, depth etc

MSmits: pb 4 told me the other day his csb has 3k sims, compared to my 300k and it's a lot stronger

kovi: yeah heavy rollouts

MSmits: super heavy

MSmits: I do heavy rollouts in a lot of games and they're nothign like this

MSmits: at most they maybe halve my simcount

MSmits: kovi what did you do for board representation?

kovi: i willl be interesting that in chess

kovi: 2xuint16

MSmits: for each miniboard?

kovi: yes

MSmits: I was thinking of trying this:

MSmits: uint16_t boards16[2][16] = { 0 };

Zenoscave: why not 0x88 it?

MSmits: can union for som avx magic maybe

kovi: hmm

kovi: but you step 1

kovi: at once

MSmits: step 1?

kovi: i dont see how avg help on that

MSmits: yeah i figured maybe for some of the heavier heuristics in the rollout

MSmits: not to apply moves

MSmits: i hadnt given it much thought yet

kovi: i c. that may makes sense if you find out how to use it

MSmits: i did find a way to wincheck with avx

MSmits: not sure yet if it is faster than a lookup

struct: you want to play on 16 mini boards at once?

MSmits: if that were possible, sure struct

MSmits: but the jumps make it impossible i think

MSmits: hey struct, do you want to judge my wincheck for uttt?

MSmits: see if it can be improved?

MSmits: it's avx

MSmits: it's a small function

struct: Im not an avx guru, but I can check sure


MSmits: not sure if it's not faster to just look it up

struct: not sure if there is an operation that does this

struct: (result64[0] | result64[1])

MSmits: i looked

MSmits: could not find a check for "any 1's in vector"

ZarthaxX: what is that ?

ZarthaxX: you want to do an extract and compare?

MSmits: i want to know if it's all zeroes

ZarthaxX: the 128 ?

MSmits: or that there is at least 1 bit set

MSmits: yes

ZarthaxX: ah

ZarthaxX: but u can compare it with a 0 mask

MSmits: then i would need a constant 0 vector?

struct: yeah

ZarthaxX: right

ZarthaxX: or do some packing

MSmits: mmh would that be faster?

MSmits: because the result would not be boolean would it?

ZarthaxX: i haven been doing a project for uni, we used SSE, AVX and AVX-512

ZarthaxX: maybe i can help you :P

MSmits: cool

ZarthaxX: yes it's not boolean

MSmits: so there's more ops needed then

ZarthaxX: maybe you are looking for a better compare

ZarthaxX: mmm

MSmits: mine is just two uint16_t or-ed and then > 0

MSmits: seems hard to beat

MSmits: uint64_t

MSmits: i mean

ZarthaxX: what?

MSmits: i split the 128i in two

MSmits: then or them together

MSmits: and check if the result is nonzero

ZarthaxX: but how are you splitting it?

MSmits: union

ZarthaxX: ah that thing

MSmits: static union { __m128i result; uint64_t result64[2]; };

ZarthaxX: i mean, it's not cheap

MSmits: the union is expensive?

MSmits: does it even cost anything?

ZarthaxX: because if you load the 128 into a register, and do the ops, then u gotta store it to access via array index

kovi: wow, double win vs. rec

ZarthaxX: if you want to do that in the register itself, you have to do an extract

ZarthaxX: of the upper part

ZarthaxX: so in one case you access memory, the other you do an extract in the register which is faster

MSmits: hmm this happens under the hood you mean?

ZarthaxX: right

ZarthaxX: union just helps when working with stuff in memory, but you are manipulating the data in between

ZarthaxX: if you can do all of these things in the register itself it's obv better

ZarthaxX: you can check how that translates btw by yourself

MSmits: so when i do a union and i access it by uint64_t it copies it to a different register?

ZarthaxX: it may do different things based on optimizaction

MSmits: hm ok

ZarthaxX: i would just check how it is compiled

ZarthaxX: you know this ?

MSmits: ah yes

ZarthaxX: you can paste your code and check with the different flags of optim

ZarthaxX: i think it's just better to check that

MSmits: compilation failed

ZarthaxX: well the thing u pasted before was not legal, missing types

MSmits: what types?

ZarthaxX: the thing is that union is just for you to make it easier to code, but internally you dont know how it's managed

ZarthaxX: idk what are you pasting in there smits :P


darkhorse64: try _mm_testz_si128 (result, result)

MSmits: i saw that

MSmits: but not sure what it does darkhorse64

MSmits: Compute the bitwise AND of 128 bits (representing integer data) in a and b, and set ZF to 1 if the result is zero, otherwise set ZF to 0. Compute the bitwise NOT of a and then AND with b, and set CF to 1 if the result is zero, otherwise set CF to 0. Return the ZF value.

struct: bits at a & bits at b

MSmits: whats the CF?

ZarthaxX: Carry flag

darkhorse64: if it returns 1, all bits are 0

ZarthaxX: you jsut care of the ZF anyway

MSmits: so does this do exactly what i want then?

struct: int _mm_test_all_zeros (__m128i a, __m128i mask)

MSmits: i saw that one too, but it didnt do what i want

ZarthaxX: yeah it should

ZarthaxX: if there is a 1 with the and it survives doing the testz result result

struct: why doesnt it work?

MSmits: well it requires a mask

MSmits: why a mask?

kovi: why there is no flag for draws in last games?

ZarthaxX: it's just a var name MSmits

ZarthaxX: if you check it, it's the same instruction under the hood

struct: cout << _mm_test_all_zeros(Vec8s(0), Vec8s(1)) << endl;

struct: returns 1

struct: 0 & 1 == 0


struct: I think godbolt has this library one sec

ZarthaxX: nice

kovi: dapo win sopra?

MSmits: mmh that one isnt working yet

ZarthaxX: what?

MSmits: i need to do an "and" for each 16 bit

MSmits: not for the whole thing at once

MSmits: so and + equality cant be done at once

ZarthaxX: im confused

MSmits: _mm_testz_si128

MSmits: this cant replace the "and"

ZarthaxX: but it does the and

MSmits: oh i see what i am doing wrong

MSmits: i need to and, then check for comparison with the masks

MSmits: then check for zero

MSmits: (board & mask) == mask

MSmits: thats the win check i do 8 times

ZarthaxX: mmm

MSmits: the thing struct suggested might be a better fit

ZarthaxX: struct suggestion is the same instruction..


MSmits: _mm_testz_si128 and _mm_test_all_zeros

struct: upsa

struct: wait

struct: its wrong


ZarthaxX: MSmits yeah same

MSmits: dont i need to "not" that struct?

MSmits: now it gives false when you win right?

ZarthaxX: struct why did u change the instr?

struct: return !_mm_test_all_ones(result);

MSmits: seems to work with a simple test

MSmits: cerr << CheckWin(5) << endl; cerr << CheckWin(7) << endl;

MSmits: 1 0

MSmits: 7 is 3 in a row

MSmits: 0b111

MSmits: and i didnt "not" it yet

struct: Im not sure the first one worked ZarthaxX

ZarthaxX: but it's the same instruction

ZarthaxX: whatever

MSmits: mmh

MSmits: why would there be 2 different ones

MSmits: if they are the same?

ZarthaxX: there are many cases where diff instrinsics use the same instruction

struct: I think all ones instruction is not great


ZarthaxX: why?

MSmits: so guys... do you think this function would be faster than a wincheck[512] lookup?

struct: zarthax for loop

struct: Im not sure, isnt wincheck a cache miss everytime?

ZarthaxX: pseudo code might not be what it happens

ZarthaxX: you can set a register with all ones in 1 op

struct: Maybe you are right

ZarthaxX: as the instruction shows right there

ZarthaxX: Instruction: pcmpeqd xmm, xmm

            ptest xmm, xmm

struct: it says its faster

MSmits: it's 64 bytes struct... not a big array

ZarthaxX: you set all ones with pcmpeqd

ZarthaxX: then ptest your a with the all ones from before

MSmits: wait no

MSmits: 512 bytes

ZarthaxX: that's small

MSmits: 64 if i compress it to 1 bit per result

reCurse: That's not 1 op

ZarthaxX: i said 2 ops

ZarthaxX: as it states there

reCurse: Ah

MSmits: reCurse what do you think, parallel calculation or lookup?

ZarthaxX: ah missing the not in a maybe

struct: MSmits even normal loop performed the same for me as lookup

ZarthaxX: struct btw the other day i found out a wrong explanation of an intrinsic

ZarthaxX: :D

MSmits: really struct

ZarthaxX: hate that documentation

MSmits: it's hard to test if its not a bottleneck

struct: But my uttt code is poor

MSmits: or did you do a clean test?

struct: clean?

MSmits: yeah a test where it's not using 2% of your total calculation time

MSmits: meaning the error in any test is larger than the actual use of time for the lookup

MSmits: btw, i realize this is microoptimizing, but the point is for me to try and write a bot with less lookups :)

struct: yes

ZarthaxX: what is the final code then?


ZarthaxX: cant you do 1 less thing there?


MSmits: no, you need to 'and" first

ZarthaxX: mmmm

MSmits: basically its like this

ZarthaxX: you sure that doesnt work

MSmits: my board is 000111111

MSmits: say i have a mask of 000000111

YurkovAS: MSmits "_mm_set1_epi16(1)" may change to static

MSmits: I and them together

MSmits: to get 000000111

MSmits: then check if 000000111 is equal to 000000111

MSmits: if i didnt and first they wouldnt be equal

ZarthaxX: ah because u just care of some bits

MSmits: yeah

ZarthaxX: not exactly the same

MSmits: YurkovAS what would that do?

ZarthaxX: you are hard to plase smito

MSmits: :)

ZarthaxX: cant' you test all zeroes against itself anyway?

ZarthaxX: why the set1 there?

MSmits: hmm i worry about that part

MSmits: not sure that works

MSmits: doesnt it make all 16 bit values 1

MSmits: so you got 8 1's

ZarthaxX: that's the set yes

ZarthaxX: so?

MSmits: well if it is a mask, doesnt it check only the first bit now?

MSmits: ohhh

ZarthaxX: the cmpeq? no

MSmits: thats where it gets deposited anyway

MSmits: the cmpeq dumps a 1 bit

MSmits: the first

MSmits: so thats fine

ZarthaxX: cmpeq sets all 1s in the correct place

MSmits: yeah 8 of them

ZarthaxX: no i mean

ZarthaxX: a mask of 0xffff

ZarthaxX: in this case

ZarthaxX: if the cmp is true for a given position, it sets all bits to 1

MSmits: oh, well thats fine too

MSmits: as long as the first bit is 1

ZarthaxX: but you can do a test against itself, or not?

MSmits: how?

ZarthaxX: if there is a 1 then the and gives 1 and gives false

MSmits: ohh ok

MSmits: let me check it

ZarthaxX: aight

MSmits: brilliant, it works

ZarthaxX: nice

MSmits: thanks guys

struct: I got lost

ZarthaxX: the setr thing from the beginning i guess it just ends being a global variable or smth right

ZarthaxX: i hope


MSmits: we changed it like this struct

MSmits: ZarthaxX yeah i do that in other games too

MSmits: othello for example

MSmits: it was quite fast

ZarthaxX: so it's not done every time?

MSmits: nah it's a global constant

MSmits: static means the memory is always reserved

ZarthaxX: excellent, but you can see the whole code together

ZarthaxX: that's cool

ZarthaxX: never saw that usage

MSmits: yeah i learned that from finding code on othello

MSmits: someone shared how to write an othello bot

MSmits: and it included avx code

MSmits: I used avx succesfully all on my own for the first time in connect4

ZarthaxX: nice :O

MSmits: it's kinda fun to figure out, but also quite hard

ZarthaxX: yeah i remember the thing with the shifts, i never got that haha

MSmits: oh yeah thats it

ZarthaxX: avx512 is a whole new world, sadly now available

ZarthaxX: not*

ZarthaxX: what were you doing with the shifts again?

MSmits: I got it to the point that it stuck for me for a few hrs and then when i looked at it again i forgot and didnt get it anymore

MSmits: shifts were mostly about finding moves and flipping them

MSmits: in connect4 it's finding winning moves for me

ZarthaxX: okey

ZarthaxX: so you cant really explain what you did right

MSmits: in connect4?

ZarthaxX: yes

ZarthaxX: i didnt get what you parallelize

MSmits: oh, i parallelize looking in multiple directions

MSmits: say you place a token

MSmits: you can check in multiple directions if this gave you winning moves

MSmits: i do it two times

MSmits: because 8 directions

MSmits: and 63 bit

ZarthaxX: okey

struct: Main problem is to find the instructions

ZarthaxX: thought you couldnt because you had to shift in diff directions

MSmits: you can have a shift vector

MSmits: that shifts different amounts

ZarthaxX: i found out so many interesting instructions while doing my project

MSmits: there's 4 left shift amounts and 4 right shift amounts

MSmits: yeah i guess thats why you need 2

MSmits: aside from the fact that you dont have avx512

ZarthaxX: but you cant shift individually

MSmits: you can

ZarthaxX: i mean

ZarthaxX: different amounts

MSmits: yeah you can shift diffeent amounts

MSmits: just not left and right at the same time

ZarthaxX: but not all at the same time

MSmits: let me find the instruction sec

ZarthaxX: ah nvm

ZarthaxX: stap

ZarthaxX: lol

ZarthaxX: i just never used it like that hehe

MSmits: _mm256_srlv_epi64

ZarthaxX: i didnt use those shifts in my project i already forgot

struct: ZarthaxX you are ready for csb now

MSmits: it's quite strange though that you can

ZarthaxX: it's in 128 bits too MSmits but i get it

ZarthaxX: why so MSmits?

MSmits: it just seems weird, mostly parallizing seems to be doing the same thing to ever part

ZarthaxX: struct na :P

MSmits: but shifting differently is not doing the same

ZarthaxX: ah, you mean it's weird that you are not treating all the data the same way

MSmits: yeah

ZarthaxX: i see

MSmits: i mean i am fine using it, i dont have to know all of how it works

MSmits: just seemed weird

ZarthaxX: btw this instruction, not worth for u here because it just replaces the test all zeroes

ZarthaxX: _mm_movemask_epi8

ZarthaxX: when i found it out i was happy

ZarthaxX: MSmits the circuit logic for it is not hard, tho who cares about that :rofl:

MSmits: hey

MSmits: that thing is interesting

ZarthaxX: many usages

ZarthaxX: hehe

MSmits: thats like parallel __builtin_ctz

MSmits: isnt it?

ZarthaxX: i mean that just collapses the bits so you can apply that

MSmits: or clz

ZarthaxX: i used it to apply first one builtin

ZarthaxX: so similar

ZarthaxX: do you know blend?

MSmits: what does it do?

struct: Improvement: 172.457% over the for loop one

ZarthaxX: you have for example 2 regs with 16 bytes

MSmits: +172%

MSmits: ?

struct: 72% gain

ZarthaxX: and you use another reg of 16 bytes as a selector

struct: 100 = equal

MSmits: is that pdep/pext?

ZarthaxX: for each position you select the byte from one or the other register based on the most sig bit of the selector

MSmits: sounds similar to pdep/pext ZarthaxX

ZarthaxX: _mm_blendv_epi8

MSmits: got it struct, good gain

ZarthaxX: i will check those again

MSmits: is it just a mask selector or does it do something with the order?

ZarthaxX: no, its different as that stores contiguous

MSmits: ah yes

ZarthaxX: this preserves position

MSmits: is there a pdep/pext also?

ZarthaxX: i know those for 512

ZarthaxX: :(

MSmits: ah

MSmits: it's useful to know a lot of these functions

MSmits: often you just dont know what to even look for

ZarthaxX: pdep would be expand and petr compress i guess?

MSmits: pext is compress yes

ZarthaxX: yeah haha, and do you know shuffles?

MSmits: i know yeah

ZarthaxX: well the compress and expand exists but just in 512

MSmits: havent used them

ZarthaxX: i see

MSmits: ah ok

ZarthaxX: if some day 512 is available here it will get wild

MSmits: pext is super useful to take a pattern from a board, compress it and then do a lookup with it

ZarthaxX: i remember u used that for yavalath

MSmits: i did and i used it elsewhere too

MSmits: you can also use pdep as a random selector

MSmits: create a random number

MSmits: then expand that to the pattern you're randoming from

ZarthaxX: mmmm

MSmits: int selectedBit = _pdep_u32(1 << rnd, movesPossible);

ZarthaxX: so you select random moves?

MSmits: for uttt you can do this

ZarthaxX: nice

MSmits: i'm not sure if pdep is maybe slow

MSmits: some people use lookups here too

ZarthaxX: besides benchmarking you can find info about the #clocks it takes

ZarthaxX: as in intrinsics

MSmits: ah is that compiled somewhere?

ZarthaxX: you have the intel manuals

MSmits: ahh ok

ZarthaxX: do you want to find pdep and pext?

ZarthaxX: mmm

MSmits: sure yeah

ZarthaxX: lemme check

ZarthaxX: sadly it doesnt appear, just the instruction info


ZarthaxX: you can do ctrl+f and pdep or pext

ZarthaxX: in the intrinsics thing it says the clocks based on cpu arquitecture, but in that manual it just doesnt appear

MSmits: ohh ok

struct: Doesnt the intel intrinsics guide tell the clock?

MSmits: well good know this exists

ZarthaxX: struct i just said that lmao

struct: ah sorry

ZarthaxX: MSmits found it,4067,4729,94,76,270,3622,3874,4152&text=pdep

MSmits: is it 1?


ZarthaxX: no, 3

MSmits: how does that compare with other instructions?

MadKnight: it's just terrible guys

ZarthaxX: read that about latency and throughput

MadKnight: just ignore me everyone

ZarthaxX: MSmits what do you mean?

ZarthaxX: hi mado :P

MSmits: well say popcount

MSmits: or ctz

MadKnight: no hi to me zart

MadKnight: just keep on going

MadKnight: okay while i'm here

MSmits: hi MadKnight, we're discussing avx instructions

MadKnight: yea i get it

ZarthaxX: MSmits do you mean comparing the clocks?

MadKnight: i'm looking into that link

MSmits: yeah

MSmits: i was just wondering if 3 cycles is particularly slow or fast

MSmits: dont have any signposts

ZarthaxX: tbh, when i searched through the manual, i saw avg 3

MadKnight: wow guys u just gotta learn every single low-level rule

MadKnight: okay i'm back out guys

ZarthaxX: i would say it's ok, but anyway uou are better off benchmarking

MadKnight: i dunno anything else

MSmits: will do MadKnight

ZarthaxX: just that sometimes you may find it cool to check combination of instructions that waste less

MSmits: yeah

MadKnight: ZarthaxX knows more than me

ZarthaxX: 1-3 is smth common

MadKnight: it's terrible guys

ZarthaxX: i found one that wasted 10 for example

MadKnight: i just found it out

MSmits: bit ops and addition are 1 i guess

MadKnight: i have to fix that

MadKnight: i have to learn new info

ZarthaxX: yes MSmits mainly

MSmits: multiplication is 4 ?

MadKnight: MSmits teach me

ZarthaxX: MadKnight update yourself

MadKnight: i need to learn more

MadKnight: i'm open to learning

MadKnight: MSmits

MSmits: depends if its int or float i guess

MadKnight: i just need to know more than ZarthaxX

MadKnight: no it's not in floats

ZarthaxX: im seeing 5 for mul

MadKnight: i only need ints

MSmits: ah ok

MadKnight: teach me any info MSmits

MadKnight: i gotta learn more than ZarthaxX

MSmits: impossible

MadKnight: oh right

MadKnight: we gotta PM

MadKnight: or else he will learn the same info

MSmits: no, not today, i should have been asleep 2 hrs ago

MadKnight: u are so smart MSmits

MSmits: have to get up in 5

MadKnight: wat

MadKnight: sleeping before 5am

MadKnight: so cringe

ZarthaxX: so are you going to sleep now smitty? :p

MadKnight: okay i need to find another way of learning

MSmits: yeah i think so ZarthaxX. Busy day tomorrow. Thanks a lot for the help

MadKnight: tomorrow is monday

MadKnight: who goes to sleep early before monday?

ZarthaxX: yw :)

MadKnight: cringe

MSmits: gn!

ZarthaxX: hey i didnt know there was a swap byte instruction

ZarthaxX: bb

MadKnight: someone said gn

MadKnight: cringe

MadKnight: oh wait i ruined it

MadKnight: i missed the "... gn to ... (something)

MadKnight: "

MadKnight: now it makes no sense and just looks dumb

MadKnight: gddamnity

Default avatar.png Rbee: its the time of the night where we lag

Default avatar.png Rbee: lets gooo

MadKnight: just ignore me

MadKnight: its so terrible

MadKnight: ill just disappear myself

MadKnight: with msmits

MadKnight: i'll just pretend i left for sleep

MadKnight: zairnaim

MadKnight: ZarthaxX

MadKnight: come

MadKnight: here

MadKnight: tell us

MadKnight: about your code

MadKnight: write an essay

MadKnight: in 20 messages

MadKnight: i need u to write 20 messages about it

MadKnight: don't ask why 20

MadKnight: just do it

MadKnight: it's actually an important number

ZarthaxX: i have cute graphics explaining parallel things

MadKnight: wat

ZarthaxX: u are crazy tonight mado

MadKnight: no way

MadKnight: what does "parallel things" mean ZarthaxX ?

ZarthaxX: wanna see my cute graphics with latex

ZarthaxX: i meant the AVX stuff

MadKnight: what kind of latex

ZarthaxX: latex the language for documents

MadKnight: the meme one or the normal one ?

MadKnight: oh right

MadKnight: the normal one

MadKnight: yea let's see it

MadKnight: gimme it

MadKnight: i'm still counting my 20 things ZarthaxX


MadKnight: what mask is this

ZarthaxX: the normal one LMAO

MadKnight: and what is this for

ZarthaxX: that's just showing how we used an instruction to do smth

ZarthaxX: load stuff from memory

ZarthaxX: it's crazy, you can select the bytes you want to load from memory

MadKnight: well we used each bit X byte

MadKnight: bit per byte

MadKnight: wow

MadKnight: so

MadKnight: what was it about ?

Default avatar.png CamiloRodriguezG: hey everyone

ZarthaxX: this is from a project we are doing with people from uni MadKnight

ZarthaxX: we implemented 2 algorithms that use dynamic programming using SSE, AVX and AVX512 stuff

3ataja: server errors, here we go again

Severed: for tic-tac-toe:

struct: Why not check for winning first?

Severed: yeah, made that change after typing that out

struct: you are still in the first league right?

struct: Also if you are first player, play always on the center

Severed: bronze i think, 9 games at once, hard to deal with random chance giving enemy same square 3 times in a row

Severed: yeah got that in there too, think only improvement i could make is to program in strategies instead of picking random choice if no winning moves or blocking moves available

struct: A heuristic used by top players is to force opponents to play on the same board that you just played, if it's empty

Severed: you can choose what board they play on?

struct: So if board on top right is empty, you should play on top right to force him to play there

Severed: oh, i see the pattern now

Severed: where i play decides what game they play

struct: yeah

Severed: thanks

struct: if the place where they should go is already full, then they can play anywhere

struct: full or completed by any player

Default avatar.png shuo502: Which live chat tool do you use

Severed: guessing not me since im clueless as to what youre talking about

Default avatar.png shuo502: qq