nah the reducing performance thing wasn't even to do with the generic math, from what i've read mult

nah the reducing performance thing wasn't even to do with the generic math, from what i've read multiplying 2 vectors together using simd is just a waste, where simd only has benefits if you're doing like matrix multiplication, or large amounts of vector multiplication at once

KKai props to you for recognizing that as old german writing

Lyris the Kitori•8/12/23, 10:33 AM

oh i was being sarcastic it's knife spelled with au and starting with the letter C, long story, but yeah i could tell it's old german, i studied german for a couple months awhile back

Lyris the Kitori•8/12/23, 10:34 AM

im quite good at recognizing languages

Kai•8/12/23, 10:34 AM

That's an Iceberg right there

AAqua nah the reducing performance thing wasn't even to do with the generic math, from...

Kai•8/12/23, 10:34 AM

oh sure it wasn't from that, I'm just suggesting that that's where it gets hard

Kai•8/12/23, 10:34 AM

SIMD also really only has benefits in loops so

AquaOP•8/12/23, 10:35 AM

ah
yeah
simd will be more challenging with generic math for sure

Kai•8/12/23, 10:35 AM

since just multiplying 4 numbers will just result in

LOAD
LOAD
LOAD
LOAD
LOAD
LOAD
LOAD
LOAD
MUL
MUL
MUL
MUL
STORE
STORE
STORE
STORE

LOAD
LOAD
LOAD
LOAD
LOAD
LOAD
LOAD
LOAD
MUL
MUL
MUL
MUL
STORE
STORE
STORE
STORE

Kai•8/12/23, 10:35 AM

so the muls are independent and processor pipeline will handle it

AquaOP•8/12/23, 10:35 AM

yep

AquaOP•8/12/23, 10:35 AM

i found just multiplying the 4 numbers together was much faster

AquaOP•8/12/23, 10:35 AM

but hey, worth experimenting

Lyris the Kitori•8/12/23, 10:36 AM

the solution is to fit all the memory you need for your app into 1 register, then you never have to deal with any of that

Kai•8/12/23, 10:36 AM

ah yes

Kai•8/12/23, 10:37 AM

public static void Main() { return 5; }

public static void Main() { return 5; }

Kai•8/12/23, 10:37 AM

that's what my apps also look like xD

AquaOP•8/12/23, 10:37 AM

top 10 games of all time

Lyris the Kitori•8/12/23, 10:37 AM

> void
> return 5

Kai•8/12/23, 10:40 AM

yes all my apps have compiler errors

Kai•8/12/23, 10:41 AM

problem?

Kai•8/12/23, 10:41 AM

https://i.kaij.tech/file/2023/08/1691836873_grim.png

KKai problem?

Lyris the Kitori•8/12/23, 10:41 AM

yes, method of type void can not have a return value

Kai•8/12/23, 10:41 AM

that's what the compiler keeps telling me

Kai•8/12/23, 10:41 AM

it's really mean :(

Kai•8/12/23, 10:42 AM

BEYLEY HAVE YOU SEEN THIS

Kai•8/12/23, 10:42 AM

https://media.discordapp.net/attachments/1040036623401894032/1139818088766914661/0e7672980d48e48c2d1373eb2505db5a.png

Lyris the Kitori•8/12/23, 10:43 AM

yes

Lyris the Kitori•8/12/23, 10:43 AM

UNUX

Kai•8/12/23, 10:43 AM

UNUX lmao

MMaxine you could have the best of both worlds by calling into system.numerics methods i...

Perksey•8/12/23, 1:34 PM

this is not true, the JIT does not recognise Vector4D and friends as TYP_SIMD thus it will not place our custom struct in VEX registers. This means that there is an enregistration penalty every time we want to do an operation using System.Numerics. That being said, it should still be possible and less expensive than regular float maths.

Perksey•8/12/23, 1:35 PM

There was some work being done to recognise user structs as TYP_SIMD which, once implemented, will allow us to have the same performance as System.Numerics

PPerksey this is not true, the JIT does not recognise Vector4D and friends as TYP_SIMD th...

Maxine•8/12/23, 8:18 PM

ahhhh

Maxine•8/12/23, 8:18 PM

i didn't even think of that, fair cop

khitiara•8/15/23, 4:05 AM

https://github.com/dotnet/Silk.NET/pull/1610 love submitting a PR on another PR's branch lmao

GitHub

Proper frame timing in BreakneckSleep by Khitiara · Pull Request #1...

Summary of the PR
Fixes BreakneckSleep in the existing CPU usage PR to store the last tick timestamp and compare to that instead of to when Sleep is called.
The branch still doesnt actually use Bre...

Perksey•8/15/23, 4:43 AM

reviewed. Appreciate those comments probably may come across as a bit infuriating or cold, and that’s partially because I myself have a cold at this time, but also if I’m like that you know I think you’re competent because I don’t tip-toe around pointing out flaws haha.

Maxine•8/15/23, 4:44 AM

the brace style

Furball•8/15/23, 4:45 AM

the only correct one

Maxine•8/15/23, 4:45 AM

you will suffer /j

Perksey•8/15/23, 4:45 AM

yeah little things like that I have to avoid bringing up in PR reviews because it becomes a waste of time unless it is egregious.

Maxine•8/15/23, 4:45 AM

it is a waste of time but you have to fix it eventually anyway

Maxine•8/15/23, 4:46 AM

i think you can set it up to fix automatically but i don't know how

Perksey•8/15/23, 4:46 AM

like you can tell when beyley has been in a file when there’s no space after a //, you can tell when devvoid has been in a file because of that brace style (though I think they’re all gone now and I don’t mind telling beyley off for it given that most of their PRs are too perfect that I don’t want to merge them first try)

khitiara•8/15/23, 4:47 AM

I did it in like 10 minutes on the github web ui lol, i want the blunt feedback. will correct and go further in the morning after actually cloning it

Maxine•8/15/23, 4:48 AM

question: why's BreakneckSleep a struct, especially a disposable one

Perksey•8/15/23, 4:48 AM

perf

Perksey•8/15/23, 4:48 AM

it’s a thin wrapper over a sleep primitive

Maxine•8/15/23, 4:48 AM

is it repeatedly allocated

Perksey•8/15/23, 4:49 AM

no if the user uses it properly

Maxine•8/15/23, 4:49 AM

then what perf benefit does it add to be a struct?

Perksey•8/15/23, 4:49 AM

alright you’ve got me there

Perksey•8/15/23, 4:49 AM

but this has no reason to be a class

nah the reducing performance thing wasn't even to do with the generic math, from what i've read mult

Similar Threads

Similar Threads

Similar Threads