i look at the directX samples in the samples folder
i thought about doin the rendering by myself, but i don't want to change anything in my framework for the user.
so how can i mix native directX and gamestudio?

This shows that you have no clue about performance and profiling. Optimizing is about to get the low hanging fruits, the few if comparisons and switches done internally by gamestudio don't have any great performance impact at all thus its not worth to spend ~10 hours writing and integrating your own renderer just to get 10 ms.

The only two things that are worth the effort in my opinion and should really speed things up would be occlusion culling and dynamic batching, which is both possible without touching the rendering itself.


This shows that you´ve got no idea how much 10ms are
If you want the final game to run at 30fps you´ve got like 33ms for each frame, so 10ms are nearly a third .
But the stuff described would probably even slow things down, and will never gain 10ms, so the point is of course still valid, that it is not worth to spend the time into your own rendering.

Edit: You should also use texture atlasses.
Edit2: I also don´t see how sprites are faster than models, I actually think that it is the other way around.

I agree that GStudio doesn't spend 10ms with those switches and ifs hidden in the renderer, I would guess in reality its about one ms max. Nonetheless, 10ms are nothing against the really low hanging fruits in this project and those are, like already mentioned multiple times, occlusion culling and sprite instancing.
I wouldn't bother to optimize the rendering when I can spend like 3 hours modifying existing occlusion culling functions and get a real performance boost out of it.

What are texture atlasses?

Its a technique to save texture bindings (which can be pretty expensive in terms of CPU cycles) by just having multiple textures inside one larger texture. You then tell the engine which polygons have to be drawn with which part of the texture.

The other advantage is, that it allows to create bigger batches when using dynamic batching, which will save many draw calls and to push basicly all polygons at once to the gpu, which I expect to be the biggest performance hit when doing something in minecraft style. The batching process of course also takes some time, but as you don´t need rotation on the blocks and even have a fixed size, you just have to apply the translation transformation, also you most probably don´t have to update the batch every frame but only if something changed, the trick then is only to find a good size for each batch, which however should be quite doable.

Well there's a few things to remember here. First, I already did mention implementing a hidden surface removal algorithm, so that base is covered.

Second, such a modification as the one I mentioned could actually increase the speed of his project by a substantial amount (although it depends upon the current complexity of the Acknex renderer). I can say this because I've made a few simple software renderers in the past (although admittedly they've never been more complex than the Doom engine), and I can say from experience that unrolling and simplifying these complex loops can indeed have an impact on performance, even if it's just bits at a time.

Anyway, he doesn't have to do something like that. But that is what I'd do (after implementing an HSR algorithm), because I tend to not trust machines I have little-to-no control over. For example, I prefer not to hand the keys to the compiler when it comes to building certain critical bits of a renderer; I like to do it myself. But that's just my personal preference.

EDIT: Hey JustSid, I just noticed your old avatar is back.

Okay I see, well thats where we decide different. I usually avoid volatile wherever possible and trust the compiler because imo the compiler can generate mor efficient code than any human (well, probably not the Lite-C compiler, but a "real" compiler ^^)
But I can't give any real examples and if this really matters beside my experience with LLVM and GCC.


Yup, mostly because the new one is on my server which currently doesn't answer any request -.-
The old one is currently hosted by Slin (many thanks bro!)

Another Mincraft clone doing pretty well !

It has some good lillte differences , like for example
Cubes with 45° slope than can be usefull to make arches or other things like stairs etc ...

I don't know if the code behind is free or not ??
Mythruna

Perhaps you could talk with the author ? it could be a better solution for this game type !!