Optimize prepare stage for frames that did not change

[janhohenheim]

Disclaimer: I don't know how much impact this actually has on the performance.

[EmiOnGit]

Actually this is a pretty good change :) If you actually want you can optimize it even more by splitting the function into two functions. One that is just preparing the uniform buffers prepare_uniform_buffers and one just for the instance buffer prepare_instance_buffer They should be independent, but beside that it would be also more organized. However, you don't have to ofc.

[janhohenheim]

I see what you mean, but wouldn't they both have mut access to GrassCache and thus not run in parallel? Edit: ah, but they could have other early return guards, I see!

[EmiOnGit]

ah, but they could have other early return guards, I see!

The other return guards doesn't matter as much because these 2 functions are the only two that could run in parallel (The only in the RenderStage::Prepare at least currently) This means we'd need to split the buffers in separate resources or use unsafe code. I'm against the second option and unsure about the first

[EmiOnGit]

It is definitely better readable though

[janhohenheim]

How about this for the moment? This way, we at least distinguish the cases of config changed vs grass changed

[EmiOnGit]

Ah also the functions are named the wrong way around. prepare_instance_buffer is currently preparing all uniform buffers . Happens easily if you copy paste to fast :smile:

[janhohenheim]

hehehe, good catch. Fixed.

[EmiOnGit]

Nice, looks almost good now. If you simplify the for loop mentioned above I'll merge. Have you looked into the issues btw?

[janhohenheim]

Which loop did you mention? Do you mean like this? Checking the issues now 👍

[EmiOnGit]

Yes now it's perfect