offsetof magic

you could in theory use a local and subtract field offset with offset of the local

65 Replies

Unsafe.SkipInit(out Vertex tmp);
Vertex* p = &tmp;
int offsetToField1 = &p->field1 - p;
int offsetToField2 = &p->field2 - p;

Unsafe.SkipInit(out Vertex tmp);
Vertex* p = &tmp;
int offsetToField1 = &p->field1 - p;
int offsetToField2 = &p->field2 - p;

und•3y ago

that looks weird I also can do an enormously stupid thing and use C++/CLI for native things like that but that sucks in the first place why isn't marshal.offsetof good enough for that though?

TechPizza•3y ago

because Marshal is for marshalling and isn't as reliable better?

Vertex tmp;
Vertex* p = &tmp;
long offsetToField1 = (byte*)&p->field1 - (byte*)p;
long offsetToField2 = (byte*)&p->field2 - (byte*)p;

Vertex tmp;
Vertex* p = &tmp;
long offsetToField1 = (byte*)&p->field1 - (byte*)p;
long offsetToField2 = (byte*)&p->field2 - (byte*)p;

at least this one compiles

und•3y ago

🤔 that's still weird how C# lacks this basic functionality

Supine•3y ago

This is Paradox but in reality it's caused by runtime doing marshalling and breaking stuff

und•3y ago

so which one is the most reliable for something like that also having attributes for that kind of thing would be much better

TechPizza•3y ago

wait what? how would attributes help here FieldOffset is an attribute btw

und•3y ago

using attributes to avoid touching GL attribute setups more than once still it doesn't solve the offset problem that's coorect

TechPizza•3y ago

that didn't really say much

und•3y ago

[VertexAttribute(0, VertexAttribPointerType.Float, 3)]
public readonly Vector3D<float> Position;

[VertexAttribute(0, VertexAttribPointerType.Float, 3)]
public readonly Vector3D<float> Position;

also it seems marshal.offsetof goes well with unsafe structures

TechPizza•3y ago

yes such an attribute would work

Supine•3y ago

But you gotta implement it by yourself

und•3y ago

I know right but the offset problem remains

TechPizza•3y ago

having an attribute like would eliminate offset problems though?

Supine•3y ago

But i think it will cause lower performance because of reflection stuff

TechPizza•3y ago

getting the attributes once at startup should be trivial enough

und•3y ago

how though? could you provide an example or whatever I really am not that deep into reflection

TechPizza•3y ago

you'd just use the data you have in the attribute

und•3y ago

the offset still needs to be a thing

TechPizza•3y ago

[VertexAttribute(index: 0, type: VertexAttribPointerType.Float3)] public Vector3 Pos;
[VertexAttribute(index: 1, type: VertexAttribPointerType.Float4)] public Vector4 Color;

[VertexAttribute(index: 0, type: VertexAttribPointerType.Float3)] public Vector3 Pos;
[VertexAttribute(index: 1, type: VertexAttribPointerType.Float4)] public Vector4 Color;

why?

und•3y ago

because of glVertexAttribFormat it still requires a relative offset (that's a true modern gl function)

TechPizza•3y ago

those attributes provide all offsets you'd need though wait why does glVertexAttribFormat even exist oh it's just to make amends of old shit, cool

und•3y ago

there's this cool thing in 4.5 core called direct state access after 4.5 you are able to fill up buffers, textures and VAOs without binding them ie access by handle not by binding

TechPizza•3y ago

i can already tell you that i could successfully use DSA without glVertexAttribFormat

und•3y ago

you have to though it's virtually impossible to define attribute format in DSA without VertexAttribFormat

TechPizza•3y ago

🤨 so can you tell me how those attributes are insufficent for a relative offset?

und•3y ago

relative offset is in bytes

TechPizza•3y ago

yes... and?

und•3y ago

I still need to have a byte offset and I literally don't know how could it provide me with this functionality

TechPizza•3y ago

get the size of the field type?

und•3y ago

then what about alignment what if a structure has different alignment?

TechPizza•3y ago

how would it have a different alignment

Supine•3y ago

I use dsa

und•3y ago

[StructLayout(Pack = X)] I see the man of culture

Supine•3y ago

It shouldn't Disable it then

und•3y ago

😳 fair enough

Supine•3y ago

Pog

TechPizza•3y ago

you can add a nullable offset in VertexAttribute for special cases

und•3y ago

but anyways I want to have a convenient way to retreive the byte offset of a certain field just adding sizes together doesn't seem like a good way to go

Supine•3y ago

Remember: do not use gl.BlitNamedFramebuffer. it's broken. (Depends on drivers)

und•3y ago

how is it broken though

TechPizza•3y ago

how else would one do it other than manually defining offsets

Supine•3y ago

It doesn't work. Works only if destination is window's backbuffer (0)

und•3y ago

ah so a hack program (jk lmao a pipeline) that is okay so pointer fuckery can do the thing right?

Supine•3y ago

Unsafe.SkipInit(out Structure s);
nuint offsetOfField1 = &s.field1 - &s;

Unsafe.SkipInit(out Structure s);
nuint offsetOfField1 = &s.field1 - &s;

Yeah

und•3y ago

allow me to describe it in the way it's done in C's offsetof and lets figure out how to do it using reflection so we have a null pointer to a structure we take the address of its field and that's it, we have the offset

TechPizza•3y ago

reflection won't help ya

und•3y ago

can I get an address of a field using reflection?

TechPizza•3y ago

nope

und•3y ago

well that just sucks, can I get an adress of a field using attributes?

TechPizza•3y ago

nope

und•3y ago

sucks even more

TechPizza•3y ago

that's why i recommended this you could even make the type: nullable and it would default based on the field type you'd just add whether you want the attribute normalized or not

und•3y ago

what about FieldDefinition.GetOffset

TechPizza•3y ago

hmm could work? literally never seen it before and Returns the field layout offset, or -1 if it is not available. is not reassuring

Supine•3y ago

Here's how struct lays in memory: