The Nineteenth Byte

« first day (4641 days earlier) ← previous day next day → last day (500 days later) »

OldSandboxPosts

01:00

Sandbox posts last active a week ago: Compile Japt to JavaScript, Golfy Fairy Chess Notation + Validation

noodle man

01:41

@lyxal no here copilot is giving you advice about how to let the compiler optimize your code and get rid of those expensive branching instructions

5 hours later…

mousetail

06:31

Is there a easy way to find the most recently deleted answer in the sandbox?

lyxal

You could scrape every page until you find a deleted post

You'd need the scraper to have 10k+ rep access though

If you had a code golf diamond it'd just be deleted:1 inquestion:2140 :p

Neil

06:57

@mousetail can't you just scroll to the bottom of the last page? Seems to be that the most recently deleted answer is codegolf.meta.stackexchange.com/a/26071

mousetail

07:15

@Neil Bottom of the page is the oldest deleted answer, the newest is pages and pages back

@lyxal Ok brb going to grab a diamond

@Neil That does seem to be the post thanks

lyxal

@mousetail sorting by date created puts it on the last page

mousetail

That makes sense thanks

Neil

@lyxal Ah right, I use that sort by default, since that used to mean that posts would never change page

if you don't want to change your sort order, open the sandbox in a logged-out window and find the last page of regular questions that way

Simd

08:09

Dim question, in codegolf.stackexchange.com/a/265865/116117 why write out the logs explicitly instead of using the log2 function?

Is there some advantage to writing them out explicitly?

2 hours later…

Simd

10:23

ah.. the painful weekend silence

lyxal

it'll do be is like that's how it is

Simd

your grammar has confused me!

lyxal

10:45

nyeh heh heh :p

Simd

do you understand why codegolf.stackexchange.com/a/265865/116117 explicitly writes out LOG3 etc?

why not just compute it there?

lyxal

it's probably faster to just have a constant than to perform computations

it is a fastest code challenge after all

Simd

but I don't mean do it in one of the loops. Is const double LOG3 = log2(3); not allowed?

I know no C++

1 hour later…

Simd

12:16

@mousetail your rust answer has exactly the same number of upvotes as the C++ answer

mousetail

12:58

Now I need exactly the same performance

The Empty String Photographer

13:37

@Simd Speed, again. Which is faster for you to evaluate, 9 or 3*3?

mousetail

@TheEmptyStringPhotographer It will be the exact same because constant folding

The Empty String Photographer

@mousetail ?

mousetail

Google constant folding

The Empty String Photographer

Hmm, ok…

But this wouldn’t apply to interpreted languages…

mousetail

It can, interpreted languages can do some optimization, if they have a bytecode or JIT or something

this is C though, so it's guarenteed to happen

>>> import dis
>>> dis.dis("3*3")
  1           0 LOAD_CONST               0 (9)
              2 RETURN_VALUE
>>>

Even in python 3*3 is exactly the same bytecode as 9

In C you can mark functions as constexprwhich means they can be evaluated at compile time for constant folding and other optimizations, in python it only applies to basic operators and maybe some basic builtin functions

Ginger

14:01

@mousetail ~~holy hell~~

mousetail

I love how chess has coopted major corporation into a meme

Simd

@TheEmptyStringPhotographer but they would only be computed once

@mousetail I didn't know that!

In [7]: dis.dis("math.log2(3)")
1 0 LOAD_NAME 0 (math)
2 LOAD_METHOD 1 (log2)
4 LOAD_CONST 0 (3)
6 CALL_METHOD 1
8 RETURN_VALUE

I guess python doesn't fold that constant

mousetail

Yea python doesn't normally fold functions

You could have hypothetically assigned something different to math.log so it would have a different result

so it can't be folded safely

Simd

14:17

@mousetail i guess it could know about its standard library. I mean it could know that math.log2 of an int computes a float

or is that too naive?

mousetail

No you don't know that, you could literally do math.log2 = print and it would work

then it would return None

Simd

ah... python :)

mousetail

That's why attempts like Mojo to turn python into a performant language are fundamentally hopeless

you can't optimize much

Simd

can the interpreter not tell that math.log2 has its original definition?

is mojo even available to the general public yet?

just installed mojo

I have no idea how to use it!

@mousetail the charcoal answer is only just behind yours and the C++ one!

mousetail

@Simd Not at the bytecode compilating stage, pypy might be able to it during incremental JIT but still it would need to insert a check every time which will be slow

Simd

14:31

@mousetail how come it can't be done at the bytecode compilation stage?

is it hard to tell if math.log2 has been redefined anywhere?

Seggan

@Simd because you dont have to do any linking

@Simd yes

suppose its been redefined in a library hidden somewhere, or with metaclass magic or whatnot

Simd

ah...

Seggan

in my language its even worse because i cannot assume that the operator + for numbers hasnt been overridden

so i cannot assume that 1 + 1 can get folded into 2 because it might be 11

Simd

@Seggan :) which one is that?

Seggan

Metis

Simd

14:43

Can constant folding work for log in C++?

Seggan

@Simd indeed it does

Simd

This is good

Does c++ have a function for bit length?

Still stackoverflow.com/a/74374791/1473517?

mousetail

15:00

In C++ you can mark any function as constant foldable with constexpr

doesn't mean the compiler will always do it though

m90

@Simd Yes, since C++20: std::bit_width (found from the duplicate question from your link)

mousetail

there is some weired edge cases around floats that make it sometimes really hard to guarentee the operation will have the same result at compile time and run time

2 hours later…

Simd

17:17

@m90 that's great!

@mousetail nice

5 hours later…

Simd

22:13

I really want to know how assuming the Riemann hypothesis is going to help with my challenge

Simd

22:25

Is there a C equivalent for constexpr to allow constant folding?