Query linear sampling support(PCF, float tex)

[Nighthawk]

Is is possible to detect whether linear sampling is supported for:
a.) float16/32 textures
b.) Percentage closer filtering(PCF) for shadow comparisons

For example, the Ati X800 can sample them only with GL_NEAREST, and will report an error when trying GL_LINEAR. ATI X3850 can sample them correctly with GL_LINEAR.

Is there a specific extension or texture state to query the capability?

[Alfonse Reinheart]

No, there is not. The GL spec requires that these are supported, and it is a driver bug if they are not.

The best you can do is use the VENDOR string to try to detect which hardware you're running on. The wiki has an article with some tips on this.

[arekkusu]

You can just try it, see it if works, then choose a rendering path.

I.e. create a 2x1 texture with one black texel and one white texel. Draw a quad covering a 3x1 area. Now read back the middle pixel; was it grey?

[ZbuffeR]

Even better if it reports an error.

[Nighthawk]

Thank you for your answers. So there is no "clean" check.

The gl error occured when creating a FBO with a shadow comparison texture as target with GL_LINEAR(error=framebuffer incomplete on X800).

The check i use now: If (maxTextureSize>=8192 or nvidia_only_extension) => use PCF / float hardware filtering.

(PCF, float tex filtering and maxTextureSize=8192 are requirements of DX10, afaik)

[Alfonse Reinheart]

You could just check for GL 3.0+ or NVIDIA extensions. No need to test the texture size.