and that's where having the distinction between `Vector` and `Point` can be useful because a `Vector

and that's where having the distinction between

Vector

Vector

Vector

Vector and

Point

Point

Point

Point can be useful
because a

Vector

Vector

Vector

Vector is largely only beneficial with floating-point, due to the operations it has to support

while the core things people do with

Point

Point

Point

Point (or

Coordinate

Coordinate

Coordinate

Coordinate or ...) are less mathematical and can have core functionality filled in nicely still

Maxine•10/20/23, 11:52 PM

so, you're proposing having both as generic types?

Ttannergooding and that's where having the distinction between `Vector` and `Point` can be usef...

Aqua•10/20/23, 11:53 PM

this is why my generic math vectors only implement certain operations for floating point types, such as distance, and even length

tannergoodingOP•10/20/23, 11:53 PM

right. Vector2<T> where T : IFloatingPoint<T>Vector2<T> where T : IFloatingPoint<T>
and Point2<T> where T : INumber<T>Point2<T> where T : INumber<T>

and allow conversion between them for cases like Point2<int> -> Vector2<float>Point2<int> -> Vector2<float> for example

Maxine•10/20/23, 11:53 PM

i like that

Maxine•10/20/23, 11:53 PM

there isn't any way to make an implicit cast from Vector3<float> to System.Numerics.Vector3, is there

AAqua this is why my generic math vectors only implement certain operations for floati...

tannergoodingOP•10/20/23, 11:54 PM

I get that
but a fundamental component of a vectorvector is that it has a lengthlength
and so if you don't expose lengthlength, you don't have a vector

Aqua•10/20/23, 11:54 PM

yeah fair enough

tannergoodingOP•10/20/23, 11:54 PM

that's where I then go, what's the next closest thing where it does make sense for integers as well
and I arrive at

Point

Point

Coordinate

Coordinate

Aqua•10/20/23, 11:55 PM

a lot of my reasoning for using generic math in the first place was to cut down on the number of stuff i'd need to write, i need integer vectors a lot and i simply need them to represent them as a 3-dimentional point, and its easier for me to just use a vector than create a point struct that does the same thing
for silk however I can understand why the distinction could and probably should be made

MMaxine there isn't any way to make an implicit cast from Vector3<float> to System.Numer...

tannergoodingOP•10/20/23, 11:55 PM

not really, since casts can only be defined on the Vector<T>Vector<T> and so must support any Vector<T>Vector<T>, which means it might be lossless

this is where I'd define an .AsVector3().AsVector3() (or some qualified name that disambiguates here)

Maxine•10/20/23, 11:55 PM

yeah

Maxine•10/20/23, 11:55 PM

unfortunate, but workable

tannergoodingOP•10/20/23, 11:56 PM

with "extension everything", you might be able to define an implicit conversion that works like that
but I'm not sure if that's on the table or not

Maxine•10/20/23, 11:56 PM

yeah that'll be awesome when it comes out in 2036

tannergoodingOP•10/20/23, 11:57 PM

I think its in the working set for C# 13, so maybe sooner
but who knows, stuff gets pushed out all the time

Maxine•10/20/23, 11:57 PM

ooh

Ttannergooding with "extension everything", you might be able to define an implicit conversion ...

Aqua•10/20/23, 11:57 PM

i do have a question - is there any plan for built-in integer vector/point types in system.numerics? i assume not but doesn't hurt to ask

Perksey•10/20/23, 11:57 PM

who needs it when you have System.Drawing.Point when

Aqua•10/20/23, 11:58 PM

lol

AAqua i do have a question - is there any plan for built-in integer vector/point types...

tannergoodingOP•10/21/23, 12:01 AM

I still plan on pushing through Vector2<T> where T : IFloatingPointIeee754<T>Vector2<T> where T : IFloatingPointIeee754<T>
I'd really like to also support integers, but convincing API review that a Point2<T> where T : INumber<T>Point2<T> where T : INumber<T> is appropriate won't be as easy

I don't have enough time to do everything of course, so I'm not sure what will get done in 9 vs not
ISimdVector<TSelf, T>ISimdVector<TSelf, T> is the thing I really want to finish next (might be internal only until language features come along)
and the Arm64 has SVE work that will likely happen in .NET 9

I'm going to try and finish the design proposal for Vector2<T>Vector2<T> in December
and that will also come alongside Vector3/4<T>Vector3/4<T>, Quaternion<T>Quaternion<T>, Plane<T>Plane<T>, Matrix4x4<T>Matrix4x4<T>
and Complex<T>Complex<T>

plus the general TensorPrimitivesTensorPrimitives work

Maxine•10/21/23, 12:01 AM

@grifon.petre can you send over your generic math lib, just for me to look

PPerksey who needs it when you have System.Drawing.Point <:when:712347014456672438>

Perksey•10/21/23, 12:01 AM

(this is what Silk.NET.Windowing used before Silk.NET.Maths existed)

Maxine•10/21/23, 12:02 AM

no proposal yet, but i wonder if it'd be worth it to implement more methods on the Vector3<T> (like what Unity does) or just do the System.Numerics ones

tannergoodingOP•10/21/23, 12:03 AM

There is a proposal up already (via xoofx, from unity) for expanding Vector2/3/4 and Vector64/128/256/512<T> with the rest of the missing IFloatingPointIeee754<T> functions

and that will likely happen in .NET 9, since we're implementing them anyways for TensorPrimitivesTensorPrimitives

MMaxine @grifon.petre can you send over your generic math lib, just for me to look

Aqua•10/21/23, 12:04 AM

https://github.com/piegfx/Warp/tree/main/src/Warp.Math matrix and quaternion have varying levels of implementation (and a generic quaternion supporting int sounds frankly stupid so i'm not even sure i'll bother supporting it), but the rest i'm pretty sure is good

Maxine•10/21/23, 12:04 AM

thanks, tanner, i'll look into it

Aqua•10/21/23, 12:04 AM

i'm currently reworking my engine from scratch as i've changed the way i do things a bit, so this code will soon officially be MIT

Maxine•10/21/23, 12:04 AM

i won't copy your code, just your ideas :^)

Ttannergooding I still plan on pushing through `Vector2<T> where T : IFloatingPointIeee754<T>` ...

Aqua•10/21/23, 12:04 AM

oh right cool, that would be awesome to have in base .net

tannergoodingOP•10/21/23, 12:05 AM

yep! The TensorPrimitivesTensorPrimitives are all nice too, since its going to be the full set of math functions
but over Span<T>Span<T> rather than fixed size vectors

tannergoodingOP•10/21/23, 12:05 AM

and then expanded to basic BLAS/LAPACK functions as well

Maxine•10/21/23, 12:11 AM

if the goal is to vectorize all of the implementations, i'm going to be here for the rest of my life

MMaxine if the goal is to vectorize all of the implementations, i'm going to be here for...

tannergoodingOP•10/21/23, 12:12 AM

https://github.com/dotnet/runtime/blob/main/src/libraries/System.Numerics.Tensors/src/System/Numerics/Tensors/TensorPrimitives.netcore.cs#L10876-L10926
>.>

tannergoodingOP•10/21/23, 12:13 AM

(doesn't have acosh, asinh, or atanh atm; but has the other 3 + exp + log + a few others)

Aqua•10/21/23, 12:13 AM

what is a tensor primitive

Ttannergooding <https://github.com/dotnet/runtime/blob/main/src/libraries/System.Numerics.Tenso...

Maxine•10/21/23, 12:13 AM

holy cow, is this actually fast?

tannergoodingOP•10/21/23, 12:13 AM

it's "tensor operations" but directly over a Span<T>Span<T>

Aqua•10/21/23, 12:14 AM

i see

Maxine•10/21/23, 12:14 AM

i feel like for Vector3, using FPU operations in serial is still faster than doing all of this SIMD stuff

tannergoodingOP•10/21/23, 12:14 AM

tensors being just the abstract terminology encompassing scalar + vector + dense matrix + sparse matrix

Aqua•10/21/23, 12:14 AM

In mathematics, a tensor is an algebraic object that describes a multilinear relationship between sets of algebraic objects related to a vector space.

why is math

MMaxine i feel like for Vector3, using FPU operations in serial is still faster than doi...

tannergoodingOP•10/21/23, 12:14 AM

not even close

Ttannergooding not even close 😄

Maxine•10/21/23, 12:14 AM

really?

Maxine•10/21/23, 12:14 AM

for 3 components?

tannergoodingOP•10/21/23, 12:15 AM

yeah, the x87 FPU (assuming you're referring to fsinfsin and friends) has been effectively deprecated for 20 years
and SSE/SSE2 has been faster for nearly 25

for the scalar implementation, you're doing SIMD operations anyways, just scalar ones (so addssaddss rather than addpsaddps)

Maxine•10/21/23, 12:16 AM

i hate computers

tannergoodingOP•10/21/23, 12:16 AM

so the SIMD algorithm is basically the scalar one, just doing 3 at once
so it takes the same amount of time

Maxine•10/21/23, 12:16 AM

does this apply for all operators?

Maxine•10/21/23, 12:17 AM

i would think serial divide vs SIMD divide would not make a difference

tannergoodingOP•10/21/23, 12:17 AM

the one exception is you replace branching with ConditionalSelect
and so you still relegate the exceptional cases

for example, with LogLog we have the fast path which is the normal scalar algorithm
and only if we have a special case (zero, infinity, nan, subnormal) do we do the extra computation
https://github.com/dotnet/runtime/blob/main/src/libraries/System.Numerics.Tensors/src/System/Numerics/Tensors/TensorPrimitives.netcore.cs#L11004-L11047

and that's where having the distinction between `Vector` and `Point` can be useful because a `Vector

Similar Threads