Finally, a DirectX 10 video driver for Irrlicht

[Nadro]

Currently I don't have an access to Windows, but as I remember RTT example works fine for me some time ago.

[robmar]

Yes, mantle sounds interesting. How will MS make dx12 45% faster without changing the architecture? Maybe we should just do a mantle driver for irrlicht!

[Granyte]

Currently I don't have an access to Windows, but as I remember RTT example works fine for me some time ago.

yes the rtt example works fine because it use only one single render target as soon as there is a second render target of an other size it won't work

[Nadro]

Yes, mantle sounds interesting. How will MS make dx12 45% faster without changing the architecture? Maybe we should just do a mantle driver for irrlicht!

You don't need Mantle, because you can get the same performance with OGL4.

[robmar]

That would be good, but with Open GL on Irrlicht, and OGL4 on the system, I get much poorer performance that with D3D9 + HLSL. OGL uses much more memory and the frame buffer post processing using direct API ext calls is really so much slower than D3D9.

[Nadro]

OGL4 in system isn't important in this case, because in Irrlicht we still use OpenGL 2.1. You can decrease OpenGL memory consumption by disable KeepImage in COpenGLTexture (it will reduce memory usage ~2x). FBO shouldn't be slower than D3D9 RTT, anyway it's depend on eg. GPU driver quality, how it optimize your post process shader etc, maybe your shader is inefficient. Are you shure that lower performance is related to FBO and not for other things?

BTW. When we'll release OGL3.x/OGL4.x drivers it will not magically increase perf in existing apps, but it will allow Irrlicht users to write optimized for newer GPUs render path, however really good knowledge about shaders and how modern GPUs works is require for that.

[robmar]

Well my shaders are almost identical GLSL to HLSL, its really very similar both, and the code is based on standard examples from MS and NVidia.

On D3D9 I use StretchRect and the OGL ext. equivalent for bitblt, and the D3D9 version runs faster.

Of course this could be as you say because Irrlicht uses OGL 2 not 4...

As to irrlicht users writing "newer GPU render paths", I don´t think many have that level of knowledge and will just use what the drivers have.

Of course no one uses OGL for games consoles as far as I know, and I guess that´s because of speed and memory foorprint.

If we had a driver for OGL 4 on Irrlicht we would know of course, but we don´t, and I guess we ain´t gonna see one for a long time if ever. Or DX 10 driver even...

[Nadro]

If we had a driver for OGL 4 on Irrlicht we would know of course, but we don´t, and I guess we ain´t gonna see one for a long time if ever. Or DX 10 driver even...

OpenGL 3.x is almost ready to commit, after merge it with FVF branch we will just need some new features to get 4.x in Irrlicht.

[robmar]

Seriously I´ll send you a crate of beer when 4.0 is done! and two crates when DX10 or 11 is done!

[Granyte]

I'm pretty damn unable to recreate the bug I get with my main project and geees the debug output is criptic all it say I parameter corruption on

Code: Select all

 
        Context->OMSetRenderTargets(1, &CurrentBackBuffer, NULL);

I really don't guet what is going on

finaly I got it to work by changing the line to have a null depth buffer on render to texture and not asking for zbuffer clear this makes no sense


EDIT: i think i found it, it seem this had to do with some multi threaded device access DX11 really don't like but dx 9 and irrlicht did not care about i'm back to an other bug inside my project

[robmar]

So no depth stencil testing support? Was that used in the current driver?

[Granyte]

it was present but dx11 would not stop crying until i changed the depth texture to NULL but after i removed it, it started crying about something else

ill try to take it back in there was also a bug about the texture being paired with a larger depth stencil then the texture i fixed it to so maybe with the bug in my application and the pairing bug fixed it will work

i'll cheq after i'm done with my class

[robmar]

Good luck with that! If you need anyone to do any drudge work let me know!

[Granyte]

It depends what you mean by drudge work i mean get coding with the driver test it because right now there are nearly no project using it so it might fail in scenario we don't expect because our project don't use it that way.

[robmar]

Okay, how many of the standard Irrlicht examples work with it now?