Re: [Algorithms] Texture mip streaming prioritization

[Josh G. <in...@gm...>]

Thanks for your thoughts,

>Alternatively - object X is Y meters from camera - what should I load.
This is essentially what my current calculation gives me. It describes what
I need to load in order to render the seen at it's highest image quality.
The problem though is then applying a memory constraint to that, that says,
I can't have everything I need.... So what will I choose to load?

I must admit, I am currently thinking down the line that Sebastian
described above:
>Assuming you only load one MIP level at a time (i.e. if MIP N is currently
loaded, you'll only consider N-1), maybe something like:
>(desired_mip - current_mip) * num_pixels / memory_usage, add tuning knobs
as needed...

I have the concept of the "Ideal" mip I would like loaded.
And I also think dealing with one mip at a time (from prioritisation point
of view at least) makes sense.
Though that doesn't necessarily mean only one mip is loaded at a time, just
one mip is considered for being loaded at a time.

I'm thinking if I start with my "target set" as equal to the "ideal set",
then execute the following:

while (memoryUsage of target set > budget)
{
     sort target texture set according to the "Priority" of having it's
current target mip loaded.
     where priority = (idealMip - currentTargetMip) * numPixelsUsingTexture
/ MemoryUsage

     Select texture with the lowest priority

     subtract 1 from it's currentTargetMip.
}

This would remove all the mips that will have the least benifit to the
scene.
Then I would task the resource system with making sure the target set gets
loaded.

Unfortunately the above algorithm looks like it'd be very serial, and
involve a bunch of sorts... But I think it might acheive good results?

Have any of you seen these implementations executed in a product or game
before? Was it successful? Tricky? worth the time?

Thanks for the comments and suggestions so far,

Josh