3 replies to this topic

[domok]

Hello

I'm working on a third person (kind of) shooter and until now I've made the targeting with the cross-hair with a ray test. This works well for the map, but not so well for the enemies because for the map I do a correct triangle test, but for the enemies only a ray ellipsoid test (a triangle test would be far too slow).

Now I thought it would be much simpler and faster to just get the z-buffer data. Is there a problem with that? I only need the z-buffer data of one pixel (the one in the centre where the cross-hair is).

How can this be done best with DirectX9? Making the z-Buffer lockable is probably not the best idea as I also need the stencil buffer..? Could I maybe do it with a pixel shader? Or is it just no good idea?

[Dawoodoz]

Reading data back from the GPU require locking the texture and that is seriously slow even with DirectX 11. When I read one pixel every render, PIX show that 50% of the time is used for waiting.

[Mussi]

I'm working on a third person (kind of) shooter and until now I've made the targeting with the cross-hair with a ray test. This works well for the map, but not so well for the enemies because for the map I do a correct triangle test, but for the enemies only a ray ellipsoid test (a triangle test would be far too slow).

What makes you say that? I think it could be the fastest solution. You could also use multiple primitives to represent an enemy and test against those.

[domok]

Ok so I will stay with the ray test.

But now I found another problem: When I cast a ray from the near clipping plane to the far clipping plane it can also occur that the intersection occurs behind the player so the bullet goes backwards. How can I solve this? Can I somehow find out the depth (z) value of where the player is so I don't cast from z=0 to z=1 but from like z = 0.5 to z=1..??

Thanks for the answers