Hey
At the moment I'm drawing images which size is non power of 2 by this way:
D3DXIMAGE_INFO SrcInfo; D3DXIMAGE_INFO SrcInfo;
HRESULT hr = D3DXCreateTextureFromFileEx(d3dDevice, "data/savail.png", D3DX_DEFAULT_NONPOW2, D3DX_DEFAULT_NONPOW2, 1, 0, D3DFMT_A8R8G8B8, D3DPOOL_MANAGED, D3DX_FILTER_NONE, D3DX_DEFAULT, 0xFFFFFFFF, &SrcInfo, NULL, &texture);
This solution however works only if the video adapter has the feature of drawing non power of 2 sizes. I'm wondering if there's another way to draw sprites in their real size (non power of 2) no matter of adapter, that will work everytime. For example how was it done in Opengl? There we create texture and blit it on quad, then display it and the displayed image had always real sizes... Is there sth like this in DirectX?
I will be very grateful for any answer!
Most video adapters should have non-power-of-2 support nowadays, and it's quite reliable (in both detection and implementation) in D3D so the days when it was a concern are fading, but if you still need to support legacy hardware I would look to pad the texture.
What this means is figuring out a power-of-2 size equal to or above each texture dimension, creating an empty texture at that size, writing the data into a subrectangle of the texture, then adjusting the texture coords to get the final correct rectangle for drawing with. The main case where this is problematic is if you need to repeat the texture in any direction, but since you're talking about sprites I guess you don't.
This approach is easiest when you're loading texture data manually. You're obviously not, so a way around it would be to create the power-of-2 sized empty texture as I described above, then load the file using D3DXCreateTextureFromFileEx into a D3DPOOL_SCRATCH texture, which isn't bound by format or size restrictions. Then use IDirect3DTexture9::GetSurfaceLevel to get the underlying surface for this and for your power-of-2 sized texture, then D3DXLoadSurfaceFromSurface to copy the appropriate texture rectangle from your scratch texture to your real one, and finally Release both surfaces and your scratch texture.
Direct3D has need of instancing, but we do not. We have plenty of glVertexAttrib calls.
Thanks a lot for fast answer. But if you say that nowadays non power of 2 shouldn't be a problem in most adapters then I think im going to stay with this ^^
The last time I tried that function, it would load a non-power of 2 texture and if non-power of 2 wasn't supported then the loader would stretch the image to the next highest power of 2. For the user interface stuff we were using it for, it tended to make the UI images look a little blurry - since the loader would stretch it out and then it would get scaled back down to the proper size at render time.
I don't have the source handy at the moment, so I can't verify what the flags were or anything...
You mean that if the adapter hasn't supported non power of 2, you were scaling the image from power of 2 size to its real size? That seems very simple and good idea but doesn't it distort images? And would that affect somehow collision detection? I have no idea about collision yet so sorry if that's very stupid question ;p
Better description of what I said above (colours changed so you can see more clearly what's happening):
The original sprite is 60x54 and I've put it into a 64x64 texture. Then I'd change the uv coords to (60.0/64.0) and (54.0/64.0) instead of 1,1. How you put it into the 64x64 texture is just a matter of using the right API calls with the right params; worst case you could even open the PNG using GDI+ and write the data directly into a locked texture rectangle.
Direct3D has need of instancing, but we do not. We have plenty of glVertexAttrib calls.
hmm and which solution is mostly used in newest games etc? puting texture on power of 2 texture and then displaying or just the function D3DXCreateSpriteEx? Which would you recommend to me?
