ImageSharp thing

(Sorry for butting in) So I'm trying to load a texture from path, but I get this error:

public unsafe Texture(string path) {
    Image<Rgba32> image = (Image<Rgba32>) Image.Load(path);

    fixed (void* data = MemoryMarshal.GetReference(image.DangerousGetPixelRowMemory(0))) {
        Load(data, (uint) image.Width, (uint) image.Height);
    }

    image.Dispose();
}

public unsafe Texture(string path) {
    Image<Rgba32> image = (Image<Rgba32>) Image.Load(path);

    fixed (void* data = MemoryMarshal.GetReference(image.DangerousGetPixelRowMemory(0))) {
        Load(data, (uint) image.Width, (uint) image.Height);
    }

    image.Dispose();
}

35 Replies

Perksey•3y ago

will check this in a sec

bjr29OP•3y ago

ok ty

bjr29OP•3y ago

https://github.com/dotnet/Silk.NET/blob/f07a7d8e3f61c63a9dfa78adc3eaad74b280955d/examples/CSharp/OpenGL%20Tutorials/Tutorial%201.4%20-%20Textures/Texture.cs#L10 based off this

GitHub

Silk.NET/Texture.cs at f07a7d8e3f61c63a9dfa78adc3eaad74b280955d · d...

The high-speed OpenGL, OpenCL, OpenAL, OpenXR, GLFW, SDL, Vulkan, Assimp, and DirectX bindings library your mother warned you about. - Silk.NET/Texture.cs at f07a7d8e3f61c63a9dfa78adc3eaad74b280955...

TechPizza•3y ago

don't you need to call .AsSpan() or something similar on the return of DangerousGetPixelRowMemory?

bjr29OP•3y ago

I'll check created a new error and windows snipping tool just crashed

bjr29OP•3y ago

Perksey•3y ago

try this:

public unsafe Texture(string path)
{
    using var image = (Image<Rgba32>) Image.Load(path);
    fixed (void* data = image.DangerousGetPixelRowMemory(0).Span)
    {
        Load(data, (uint) image.Width, (uint) image.Height);
    }
}

public unsafe Texture(string path)
{
    using var image = (Image<Rgba32>) Image.Load(path);
    fixed (void* data = image.DangerousGetPixelRowMemory(0).Span)
    {
        Load(data, (uint) image.Width, (uint) image.Height);
    }
}

lol just realised @thomasmiz you were pinged onto this haha https://github.com/SixLabors/ImageSharp/issues/1739

bjr29OP•3y ago

AsSpan() doesn't exist btw, just Span

Perksey•3y ago

well then Span 😛

bjr29OP•3y ago

😛 Btw it works now thanks

Perksey•3y ago

roeyskoe•3y ago

The examples should be updated at some point, since ImageSharp 2.0 changed some stuff. I would also recommend reading this: https://docs.sixlabors.com/articles/imagesharp/memorymanagement.html And note that images larger than 4MB currently do not work, it is fixed for the next release: https://github.com/SixLabors/ImageSharp/issues/1983

bjr29OP•3y ago

ah ok I'll probably use jpg to get around that for now then if I need to

roeyskoe•3y ago

To my understanding that will not help, since 4MB limit is regarding raw image data. So at 4bytes/pixel => ~1000x1000 image is the limit.

bjr29OP•3y ago

Oh I see

roeyskoe•3y ago

But I would still recommend going with Imagesharp 2, since updating later will bring some extra work 😄 And I would expect that they will release a bugfix patch soon-ish.

bjr29OP•3y ago

Yeah, I don't want to rewrite code that's taken like 8 hours I'm writing a game engine and making everything very modular and simple as possible to use for the user Well just simple the old design was modular but it would take just too long to write

Perksey•3y ago

it's a shame that's a dangerous method now (well, i suppose it always was, but you know what i mean)

bjr29OP•3y ago

Yeah, does it affect anything or is it just a rename really?

Perksey•3y ago

well it just conveys that it's an unsafe method if you don't know what you're doing

bjr29OP•3y ago

Oh ok

Perksey•3y ago

@jbsp is there any non-dangerous method you know of to get row memory? i.e. one that copies if necessary. or perhaps a way to iterate through the pixel memory in use for an image? (this would be cool and if that's possible and not already in there i'd look into it!)

roeyskoe•3y ago

https://docs.sixlabors.com/articles/imagesharp/pixelbuffers.html This?

Perksey•3y ago

oh excellent (sorry for the ping james)

ThomasMiz•3y ago

Yeah, I've been checking in on those guys because I use them for loading images They want to make changes that could make images not be stored fully sequentially, which is a bit of a problem for us who want to use it to load textures

Perksey•3y ago

yeah sounds like it but hey, at least the think roey linked looks pretty nice. could either be used to flatten the buffer and do one textimage or do a bunch of subimage calls.

Kai•3y ago

I think this is actually a good thing - you really should not be flushing the whole image to GPU at once anyways. Forcing you to tell the driver which part you are uploading, uploading that, and letting it do the invalidation etc improves perf and memory usage.

ThomasMiz•3y ago

Really? I thought it was better to do it all at once

Kai•3y ago

why?

ThomasMiz•3y ago

Less GL calls, so it only needs to do any checks once

Perksey•3y ago

the driver will likely divvy it up internally for bus reasons anyways [an uninformed guess]

Kai•3y ago

GL I have no idea, but in vulkan it's for sure better. Not very informed about GL, but there needs to be some range of host-gpu shared memory that the texture is first loaded into, and then copied into GPU-only memory. There is usually only a small amount of host-gpu shared memory, so flushing a large texture all at once is not possible in the first place now in GL maybe it's best to give it all the data at once cause you don't get to know what these pieces should be sized, but from a driver & VK perspective you can't really flush the whole texture at once

ThomasMiz•3y ago

Why would it be more efficient for the user to send the texture in parts then? Wouldn't the driver be able to better optimize the sending if it knew everything from the start?

Perksey•3y ago

with opengl perhaps, with vulkan the idea is the driver doesn't need to know anything because it trusts the user

Kai•3y ago

^ this, you are the only one who can ever know how much host memory your program is going to need and what size of upload buffer you can "afford" the driver can just make some guesses if you upload a huge texture and the driver has decided to send it over 4mb at a time, then those 4mb are dedicated to this task for some time if you now are trying to also stream mesh info to the GPU and are requesting some some visible memory for that, the driver has already taken those 4mb out of it's budget plus it has to do some complicated logic to try and reduce fragementation, as there is no way for it to know for how long each bit is taken up, and compacting is hardly possible

Gaming

Programming

ImageSharp thing