How does a modern GPU access texture cache?

[Ronin_Samurai]

Hello all.

As an example, when an R350 is applying trilinear filtering to a texture, each filtered pixel requires two samples of 2x2 pixels, correct?

Does each of the rendering pipes access their own eight pixel sample, or is the one sample batch used by all the pipes, (or does each pipe even have their own exclusive caches, rather than a single unified larger cache)?.

Are DX9 textures held in cache as 32-bit integers or in floating point format?

Just curious.

 

[3dcgi]

Does each of the rendering pipes access their own eight pixel sample, or is the one sample batch used by all the pipes, (or does each pipe even have their own exclusive caches, rather than a single unified larger cache)?.

This can be different for different manufacturers, but you can be sure they only want to read from memory once. How this is accomplished could vary between GPUs. Maybe someone else can be more specific.

Are DX9 textures held in cache as 32-bit integers or in floating point format?

They are held in the proper format of the texture. I.e. an integer is stored as an integer and a float as a float.