The Nineteenth Byte

Downgoat

00:29

@quartata how does one implement generators? Is it like one giant switch(){} statement with the current position in the function and the callee stores the function's scope?

Also @ASCII-only should generator functions return a Generator/Iterator class?

Don't want to go the Java route and have a Iterable and Interator interfaces/classes since those are euck

ASCII-only

@Downgoat should return lazy list maybe

Downgoat

@ASCII-only lazy list?

ASCII-only

Yeah

Downgoat

elaborate on semantics?

also @ASCII-only for generator function do we want generator func myGenerator or func* myGenerator

ASCII-only

00:51

Hmmmm

Star only if that's syntax for spread too

Downgoat

well spread operator will probably only be for array concat

so not sure if should do ... or *

because ... is inclusive range atm

lirtosiast

@ASCII-only I am

I've wanted for a while to make a golfing language with readable tokens that compresses

In the first version all tokens will take a multiple of 0.5 bytes, but I hope to eventually move to arithmetic encoding

ASCII-only

01:15

@lirtosiast mthmtca is a thing though

@lirtosiast arithmetic?

lirtosiast

@ASCII-only Program is stored as an arbitrary precision real number between 0 and 1; you predict probabilities for each token, assign them to intervals, decode the token, and zoom in on the interval

@ASCII-only ooh I didn't know about that

ASCII-only

@lirtosiast so... Like Huffman coding

Why choose arithmetic coding though

ais523

arithmetic coding should be theoretically unbeatable in the long run, out of any coding scheme

anything more advanced than that is a compression scheme

things like Huffman coding only have a chance of winning if your input is very short, because they can optimise the last few bits of the output

Quintec

02:06

@Downgoat What language is this

Pavel

@Quintec VSL

ASCII-only

02:23

@ais523 ???

ais523

@ASCII-only Huffman coding is basically arithmetic coding except that every command has to be rounded to a whole number of bits

so arithmetic coding does as well or better

ASCII-only

Ah, fair enough

quartata

02:46

@Downgoat you asked me this question before

Apr 25 at 0:10, by Downgoat

@quartata Like I was thinking I could do it with having a state struct and then jumping to correct part of code but I'm not sure if there is better way

Downgoat

huh

though how to then generate code for for loop is question :/

and also how to then deal with lambda capture :/

quartata

click through

since you are already using libc you have ucontext_t

fcontext from boost is optimized for x86 but is much faster

however i also brought up inline generators which is an important point for consideration

and some syntax considerations, since Python and JS both have bad syntax for this

you should think first whether you care about full vs semi coroutines though

additionally as i mentioned you could pick through the C output from the Nim compiler to learn more practically

Downgoat

04:31

SO says to use srand(time(0)) instead of rand() is this really a better idea? since iirc time() is accurate to second which means if you call multiple times in same second youd get same random number

Pavel

@Downgoat You call srand once, it sets the seed for future rand calls.

Downgoat

@Pavel is that necessary though for non-crypto applications

Pavel

@Downgoat Well otherwise you get the same seed every time, so your random doesn't end up being random: Try it online! (Run it multiple times, notice the number not changing)

Downgoat

wait wtf

Pavel

Now with the srand call: Try it online!

Downgoat

04:36

issue is if program is called within same second of another instance still get same number :/

Pavel

This isn't an issue for most programs.

Like how often does that happen.

Downgoat

forking

ASCII-only

@Pavel very

@Downgoat you didn't know?

Pavel

@ASCII-only Have you ever written a random-dependent program that can reasonably expect to be called multiple times a second? srand(time(0)) is reasonable for most purposes

ASCII-only

@Pavel but not all...

And that's the important thing

Time is just not good enough for a mainstream language

Downgoat

04:42

@ASCII-only no i thought was something like /dev/random

ASCII-only

Especially if gettimeofday is in the same header

Downgoat

also dont want to stream from there because stupid windows

ASCII-only

@Downgoat possibly. But yeah it's not cross platform

Pavel

@Downgoat Rand will, by the C specification, give the same values for the same seed.

/dev/random is "real" random generated from hardware entropy

Downgoat

@ASCII-only i could try to use gettimeofday but issue is that, that uses long int which I don't accurately know the sizeof for VSL

quartata

04:43

dont use rand anyways

Downgoat

@Pavel that's srand

@quartata pls recommend alternative

Pavel

@Downgoat srand just supplies the seed for rand

And yeah as quartata points out almost all implementations of rand are pretty shite

ASCII-only

@Downgoat xorshift :P

@Pavel I thought that was urandom? Unless random is same just not cryptographically secure

quartata

i mean for actual C++ work the new random header is good

Pavel

@ASCII-only urandom and random use the same pool of random data, but random blocks when it runs out and urand starts generating pseudorandom data instead

quartata

04:45

but youre not linking to that

Downgoat

@quartata C++ bindgen to borkable for use in core stl and also yeah dont want to link to that

quartata

frankly you can make your own thats better than rand()

theres a few choices

Pavel

I think random is already better than rand but IDK how portable it is

quartata

vigna.di.unimi.it/xorshift/xoroshiro128plus.c

xoroshiro implementation

hg.openjdk.java.net/jdk8/jdk8/jdk/file/tip/src/share/classes/…

Java's splittablerandom

lirtosiast

oh I have a question about RNGs

if we want a really good noncrypto one and are lazy, can we just xor like 3 different RNGs together?

Or does that not improve the quality

quartata

04:49

Mathematica deadass used to use Rule 30 as a PRNG

Pavel

ISAAC is cryptographically secure: burtleburtle.net/bob/rand/isaacafa.html

@quartata You mean, the cellular automata?

lirtosiast

Yep

Pavel

Not too surprised tbqh

quartata

it actually sucks but

Downgoat

@quartata O_o does this actually work well

looks suspiciously simple

quartata

04:51

its one of the stronger non crypto

Downgoat

o ok

prob just gonna link to openssl for crypto rand anyways

Pavel

I wonder how fast it is compared to ISAAC

ISAAC gives performance information in CPU cycles and I'm not sure what to do with that information

quartata

the lowest bit on xoroshiro is biased IIRC

so for coinflips you check the sign instead

Pavel

@quartata Do you have a source on the Rule 30 thing

I want to tell people about it

quartata

youtube.com/watch?v=P7kX7BuHSFI&t=1860 timestamp included

wolfram has a CA fetish

one bit per iteration lol

the other columns are no good

Post Left Garf Hunter

04:58

I don't know what CA means and I assume that link would tell me

so I refuse to click that link

it is more fun to guess

Downgoat

Cellular automaton

A cellular automaton (pl. cellular automata, abbrev. CA) is a discrete model studied in computer science, mathematics, physics, complexity science, theoretical biology and microstructure modeling. Cellular automata are also called cellular spaces, tessellation automata, homogeneous structures, cellular structures, tessellation structures, and iterative arrays.A cellular automaton consists of a regular grid of cells, each in one of a finite number of states, such as on and off (in contrast to a coupled map lattice). The grid can be in any finite number of dimensions. For each cell, a set of cells...

oh sorry

Post Left Garf Hunter

Ah well

Downgoat

@quartata what to seed this with tho

Pavel

@quartata I got to meet Conway and Wolfram together once

Downgoat

@Pavel what where how

Pavel

04:59

There was a conference

Post Left Garf Hunter

I met wolfram once. It was at a hackathon where I programmed a spinning orange

Pavel

It wasn't like 1 on 1 or anything

(1 on 2?)

Downgoat

@Pavel ( ͡° ͜ʖ ͡°)

quartata

@Downgoat docs.microsoft.com/en-us/windows/desktop/api/bcrypt/… for win32, /dev/random everywhere else

or dont care and use time

Pavel

@quartata Why does it take 4 arguments tho

quartata

05:04

overengineered like everything else in this hell userland

it can run it through a prng too i think

Downgoat

o crap xoroshiro128 needs for loop which needs iterators which needs interfaces which i haven't implemented yet oops

while loops exists but yuck

quartata

wait ???

write it in IR lol, its a hot path

wait why would it need an iterator

Pavel

@quartata foreach, not for

Downgoat

@quartata dont want to have C-style for

@quartata I would but can't do inline IR because LLVM doesn't expose any API to parse or directly insert IR

quartata

then unroll the loop

Downgoat

05:08

:|

quartata

what

feersum

A for-loop can be rewritten as a recursive function :)

quartata

frankly im surprised that implementation doesnt

i never assume the compiler will unroll

@feersum ok now this is epic

Downgoat

@feersum don't want to create 64 new call frames every time i call random function though llvm will prob unroll

quartata

@Downgoat it wont, not tail recursive

uhh

I dont think

Downgoat

05:10

@quartata why wouldn't it? I am just doing for i = 0; i < 64; i++ as recursive

Pavel

@quartata Do you just unroll all your loops by hand

Downgoat

real men unroll their loops by hoof

feersum

With GCC I've found it will unroll a recursive function 15 times or so

So you might get a recursive function, but only calling itself 4 times

Don't know about clang.

quartata

@Pavel in low level yeah

ive seen some bad bad compilers

Downgoat

@quartata wait crap mean tco it

Pavel

05:13

have you tried not using bad bad compilers

Downgoat

though i have a decorator that will force it

quartata

@Pavel not usually a choice

feersum

It's already tied to clang.

Pavel

Oh right Visual compiler exists

Downgoat

I will be a proud goat tonight if VSL can type deduct (x << k) | (x >> (64 - k)) without crashing

uhh @quartata how do I pass a seed into: vigna.di.unimi.it/xorshift/xoroshiro128plus.c

so I change s0 or s1 or something else?

Pavel

05:18

It's blows my mind that this language is being written in JavaScript

Downgoat

@Pavel it is actually pretty fast because JS handles graph data structures pretty well

and also hot spot is really great for things like iterating trees because those get called like a bajillion times

feersum

Maybe it's fast because all your programs so far are one file with 10 lines?

quartata

@Downgoat the array s is the state

128-bit hence the name

Downgoat

aaaaaaahhhhhhhhh that makes more sense

quartata

two long longs

Downgoat

05:23

can i turbocharge this by using llvm 128bit stuff or would that not work anyway

quartata

it still needs the low/high separate in next

Downgoat

:| apparently JS cannot represent 0xdf900294d8f554a5 which the algorithm uses as it looses precision so now cannot use in VSL :|

2

quartata

um

uh

@everyone get a load of this guy trying to emit a 64-bit number in javascript

wheeze

feersum

Is there any German word for the feeling of being entertained by the misfortune of Javascript users

Downgoat

good news is wolfram alpha found factorization that is representable in 32 bit

Anonymous

05:38

@feersum JSchadenfreude

quartata

when you multiply them its just going to overflow ??

theyre doubles

you need the exact number, its a bit mask

Downgoat

@quartata they are multiply in LLVM

quartata

uhhhhhhhhh

have you considered

feersum

What kind of code are you even writing? VSL? LLMV IR textual representation? A library that generates binary IR?

Downgoat

  %0 = call i64 @"FSTCUInt64N*AlhsTCUInt64ArhsTCUInt64"(i64 422105777641, i64 38164317)
  store i64 %0, i64* @gNjumpConstantA

@feersum LLVM's C++ API

quartata

05:40

hes apparently using LLVM bindings which is why he cant just use a string

feersum

@Downgoat And this involves JS how...?

quartata

@Downgoat ah cool, i didnt want to use my eyeballs anyways

Downgoat

@feersum because JS talks to LLVM C++ api

quartata

@feersum his compiler is written in JS

so he literally cant emit the number because it gets stored in a double by JS before being truncated to an int

feersum

How is that even possible? It is an abomination.

Downgoat

05:41

still JS compiler code is neater than GCC source code

quartata

the bindings have to use a uintarray somewhere

canr you pass one

Downgoat

@quartata github.com/vsl-lang/llvm-node/blob/master/llvm-node.d.ts#L219

so no

@quartata what is the purpose/when do I use the jump() long_jump() functions

quartata

so the cast is done in C

you realize this language sucks right

Downgoat

it says for parallel computation but we'll all be old by the time I implement parallelism in VSL

@quartata uhh so its parsed as a string we parse string -> double in node.js and then we do double -> int32 -> uint64 in c++

Pavel

@quartata When you say "this language" do you mean JS, VSL, C, or all of them?

quartata

05:47

JS

Downgoat

I do but too late to rewrite this in php

quartata

@Downgoat so as it says its like it skips ahead 2^64 iterations

Pavel

Perl is like the one other language that can't figure out integers either

Even if you use integer you only get 32 bit ints

lirtosiast

@Pavel Conway and Wolfram both spoke at a math camp when I was young, but not on the same day

Downgoat

@quartata but why would i want to do that

Pavel

05:50

Can you like reverse-babel it to a sane language

Is that a thing

Downgoat

@Pavel me and @ASCII-only were considering making something like that

but no

technically i could convert to typescript and use an experimental typescript to llvm compiler and then try to use an llvm to c compiler

Pavel

Brilliant

quartata

a second thread can start from the same seed but jump ahead so they dont run the same sequence

is my guess

Downgoat

ok i am just going to not implement that part until i can get 64-bit ints working :P

feersum

> Node Bindings for LLVM. The API is mostly identical to the one of LLVM (when possible and not to cumbersom)

Guess this one was to cumbersom.

quartata

05:53

mm cucumbers

Downgoat

is there no cross-platform hardware rng

quartata

you know i love a good cucumber

@Downgoat lol no

Pavel

There's probably a boost thing for it

quartata

well

technically

random.org

Downgoat

what is windows alternative to /dev/urandom then?

quartata

05:54

i linked it

Pavel

@Downgoat docs.microsoft.com/en-us/windows/desktop/api/bcrypt/…

quartata

docs.microsoft.com/en-us/windows/desktop/api/bcrypt/…

Downgoat

link to your linkn?

oh

@Pavel does this use windows's calls to hardware RNG

Pavel

@Downgoat Yes

quartata

do not ask me Win32 api questions

Pavel

05:55

But yeah, Win32 is fucking terrifying

quartata

@Downgoat whatever sources of entropy it has

Pavel

Have you considered linking boost

quartata

no

Pavel

Has downgoat considered linking boost

Downgoat

@Pavel oh god no

who do you think i am

feersum

05:57

What about declaring random number generation to be the user's responsibliity?

Pavel

@Downgoat Someone who's writing a compiler for a practical language in JavaScript

Two of them

@Downgoat Why don't you write your random in VSL

VSL should probably be able to handle 64-bit integers

I screencapped this convo in another chat and now I'm being assaulted by a JS defender please advise

Downgoat

@Pavel tell him as the person who is writing a language in JS that is stance is irrational and that he should learn python

@Pavel I am but still need to figure out how to get seed

and accurate time is hard to get

Can't use gettimeofday as that uses a cstruct

I don't think I can assume sizeof(long) == sizeof(void*) either

quartata

@Pavel the "everything is a double" is a sore point for some reason

@Downgoat hm?

Downgoat

is clock() ^ time() a good enough random seed

wait wtf it can be a float

~~god what idiot made the c standard~~

feersum

Well you still get 32 bits, just in a different order, so whatever :)

Pavel

06:04

C, libc, and the actual standard we have now were made by different people

quartata

why cant you read from dev random and worry about Win32 when your compiler works

Pavel

Why is boost magically harder to link than any other library

quartata

you just read 16 bytes with the pointer to s

Pavel

<-- hasn't tried linking anything himself

feersum

Use a constant seed and tout reproducible executions as a language feature.

quartata

06:05

reversible computing

Downgoat

@quartata compiler work relatively fine except interface but doing things without linking to custom C files is pain

@quartata should I set s[0] and s[1] to same values or should I make them different

after 2038 hopefully I can assume time_t is 64-bit and whatever idiot used a computer with 32-bit time is already crashed anyway

6

ASCII-only

@Pavel at least both have bigint now

Pavel

@Downgoat brilliant long-term planning strategy "just wait until 2038 and the code will work"

Downgoat

:)

ASCII-only

@Downgoat maybe it'll just loop around. Maybe. So the people that use 32bit go back to 1970

Downgoat

06:15

private lazy let randS1: UInt64 = UInt64::clock()
private lazy let randS2: UInt64 = time()

I think this is how random works?

ASCII-only

@Pavel tbh it might not be finished before 2038

Downgoat

oh god LLVM unrolled

primo

@ASCII-only no look-ahead or look-behind

Pavel

@Downgoat Why id it private let and not let private

ASCII-only

@primo O_o

primo

06:17

the "magic" bit handles significantly more than 8 cases

Downgoat

@Pavel i have not seen any language in which it is let private and not private let

Pavel

private let just sounds weird

Downgoat

that is true

I could change it back to var

Pavel

"adjective verb noun" just sounds wrong compared to "verb adjective noun"

ASCII-only

@primo that's not magic anymore. That's sorcery

Downgoat

06:20

O_o seems like LLVM optimizes xor to branching instruction

ASCII-only

@primo what about $`/$' then

Downgoat

does C have any pow() function that takes int instead of floating points

ASCII-only

@Downgoat no

Downgoat

how to do 64-bit pow then

ASCII-only

the slow way: multiplying and squaring

Downgoat

06:34

ew

I don't want to chop 64-bit ints to doubles and don't want to use 80/128-bit floats because then things get weird with data layout :/

primo

@ASCII-only $' is utilized, yes

Downgoat

07:13

good news: you can now print numbers in vsl bad news: those numbers are in reverse

ASCII-only

@Downgoat lol

You need to allocate enough space and do double dabble or w/e

@Downgoat you don't have to?

Downgoat

@ASCII-only how else to do

@quartata do you know why LLVM icmp ugt still finds 1 << 31 > 0 as false. Shouldn't icmp ugt basically be icmp ne 0

H.PWiz

07:36

@primo I just found the (Python) solution to phi. Very cool

Downgoat

also how tf is 2147483648 converted to an FP 0x41E0000000000000 O_o

primo

@H.PWiz it's a common trick in ruby, rarely applicable to python though

H.PWiz

In ruby, because you can have multiple = in one expression?

primo

trailing assignment x=4*y=3 makes it a lot easier to do

maxb

07:54

Would there be any interest in doing fastest-code challenges with GPUs? I haven't seen any so far, but with the right problem I think it could be interesting. I have at least one idea for a challenge, perhaps related to my avatar.

New Main Posts

09:43

1

Q: Trifid Cipher (without keyword)

Introduction: I have loads of different ciphers stored in a document I once compiled as a kid, I picked a few of the ones I thought were best suitable for challenges (not too trivial, and not too hard) and transformed them into challenges. Most of them are still in the sandbox, and I'm not sure ...

New Main Posts

10:13

0

Q: Regex only base64 decoder

Write a regular expression substitution string which can base64 decode a string using only a regular expression engine built into the language. Eg: # Example in python import re print(re.sub(..., ..., "SGVsbG8gd29ybGQK")) # "Hello world" # ^^^ ^^^ # These count toward...

code-golf regular-expression

feersum

11:58

@maxb Doesn't seem like it would work well since GPU hardware is too diverse.

New Main Posts

0

Q: Fill the Bucket

Your task is to fill the bucket with numbers upto a given input. Rules Numbers occupy the leftmost position then rightmost, then leftmost and so on. After overflow, the numbers start to gather around the bucket in a similar manner. They occupy position diagonally. The examples should make it ...

code-golf

maxb

@feersum Of course I'd provide a test rig similar to every fastest-code challenge. But you're right, it's not as simple as regular programming when it comes to testing on your own before. It could definitely be a problem that not all people have Nvidia with CUDA. Though OpenGL should work for both AMD and Nvidia right?

feersum

@maxb Can you use OpenGL for general computation somehow? I thought it can only draw graphics.

maxb

@feersum I haven't used it personally, so I don't know unfortunately, I have only used CUDA. I'm just hoping that it could provide a hardware-independent way of participating in the challenge.

I feel that a CUDA specific challenge could also be interesting, but I'd understand if it's not allowed.

@feersum From some googling, it seems as if the problem I had envisioned has a solution using OpenCL, so judging by that it should be viable to create a challenge

maxb

12:45

I'll put a post in the sandbox, that way people can leave feedback there before an eventual challenge.

dzaima

13:46

ಠ_ಠ if i hadn't submitted the answer to todays AoC part 1 as x;y instead of x,y i would almost be on the leaderboard

ASCII-only

Lol

At least you could actually do it

dzaima

APLs ⌺ is a builtin for "do this for x×y subregions of the input" so that was helpul

ASCII-only

@feersum doesn't minxomat's turbo.js use WebGL to speed up computation

@feersum not like you normally program for the hardware anyway

@dzaima :|

dzaima

sadly it pads 0s around the input and is slow so didn't manage to do task 2 in the 15 mins i had

quartata

@Downgoat different

@Downgoat you know clock() is based on time since it started right

So your high is going to be 0

New Sandboxed Posts

15:00

0

A: Sandbox for Proposed Challenges

Buddhabrot (speed edition) Your goal is to generate an image like this one: (Sandbox: this image will be replaced by a valid solution image for the challenge) This is a render of a 2D histogram called Buddhabrot. The algorithm for generating it is very simple. If you have heard of (or writte...

Downgoat

15:43

@quartata why would exactly zero clock cycle have elapse since program start

Oliver

Is TIO down?

quartata

@Downgoat its not in cpu cycles

technically speaking the epoch is undefined, its only supposed to be compared to other clock_ts. but its going to be at about the same time you initialize the RNG since its static so

Skidsdev

@Oliver working for me

Geobits

Random Announcement: In roughly several months' time, there will be a Microbits in addition to Minibits.

22

Pavel

@Geobits What's a Microbits

And what's a Minibits

Geobits

15:56

Minibits is my son :D

Pavel

Ah

Congrats!

quartata

@Geobits are they like nesting dolls

so you have the microbits

Geobits

I certainly hope not lol

quartata

then minibits

then geobits

you just stack them

Erik the Outgolfer

that sounds very disturbing...

quartata

15:57

its just science

Pavel

I'm gonna pretend Geobits actually named them that because that sounds like something Geobits would do

Geobits

Stackable, yes. Nestable, no.

quartata

ohh ok so theyre not hollow

all right i get it now

Geobits

Well, I'm not hollow anyway, and I've seen some x-rays of Minibits due to various fractures, so I'm assuming the Micro version will also not be hollow, but with technology these days, who knows?

Transcript for