I’ve been wondering about that… I’ve been wondering if it’s possible to read and write from and to texture memory.
I have seen some slides and presentations and tables which show tables claiming “texture” is read only.
Some sections of the guide mention reading and writing is possible but it produces undefined results (in combination with the cache ?)
However section “3.3.13 Graphics Interoperability” does mention the possibility of mapping and unmapping resources.
These are Driver API calls though (host side) and they could have high overhead.
The guide isn’t completely clear to me, also I am not much of a graphics programmer when it comes to OpenGL/DirectX, my interest in that is low. I am more interested in general computation.
But if textures can somehow be used to read and write reliably I might be interested in it, especially if the texture caching effect can lead to higher performance.
But so far reading and writing seems unreliable (at least in combination with the cache) ???
Also I started using cuda to get away from textures and such because it’s more weird/more complex/more akward to use for general computation :)
Otherwise I could just go back to dx9/opengl :) and my now obsolete 7900 gtx 512 MB with 70 GB/sec (?!) =D (<- but I know it’s probably obsolete, to little write storage space for render to multi texture. just 64 bytes per pixel if I remember correctly ).