Alpha test BEFORE fragment procesing

[Coluna]

Hi; In my app, the alpha channel in difuse textures is used as a mask (like fences, leaves...). When i use

glEnable(GL_ALPHA_TEST);
glAlphaFunc(GL_GREATER,0.5f);

with GLSL shaders, it didnt work, all pixels r drawned...but when i use it with nv_reg_combiners or ati_fragment_shader, it works...whats happening?

[Korval]

Alpha testing happens after fragment processing. You should make sure that the alpha you're outputting is correct.

[olmeca]

When I use Alpha Test + GLSL + Pbuffers on my 9600 it needs a lot of time to compute. But if I use a depth buffer with depth test, there is no such performance drop. Why does this happen?
Alpha test should be the most simplest test (no writing to a depth buffer, ... )

[Zengar]

I think you should perform alpha test manually...

[Humus]

Originally posted by olmeca:
When I use Alpha Test + GLSL + Pbuffers on my 9600 it needs a lot of time to compute. But if I use a depth buffer with depth test, there is no such performance drop. Why does this happen?
Alpha test should be the most simplest test (no writing to a depth buffer, ... )

How much is "lot of time" in this context?

[Humus]

Originally posted by Coluna:
Hi; In my app, the alpha channel in difuse textures is used as a mask (like fences, leaves...). When i use

glEnable(GL_ALPHA_TEST);
glAlphaFunc(GL_GREATER,0.5f);

with GLSL shaders, it didnt work, all pixels r drawned...but when i use it with nv_reg_combiners or ati_fragment_shader, it works...whats happening?

Consider using alpha-to-coverage instead of alpha testing.
http://www.humus.ca/index.php?page=3D&ID=61

[zed]

excellent as always humus (grrr now ive gotta go and play with this, as if i dont have anuf to do already)

[Korval]

Consider using alpha-to-coverage instead of alpha testing.

And what is the cost of this technique?

[Humus]

In the demo it's about 10-15% slower than alpha testing.