Texture array as filter for performance


Has anyone ever used a 3D Texture array as a means of filtering input data? I would like to load a 3D texture array with 1 or 0 to determine if a given x,y,z coordinate should or should not be used for computation respectively.

Is this something that is often used for GPGPU computing as a means of potentially improving performance or is this a wrong-headed concept/idea?

Thanks for any response.

So this is not something that is done often?

It is pretty much impossible to answer design questions without knowing the larger context. My recommendation would be to prototype some of the design alternatives you have in mind, then benchmark and profile them. That should provide you with a lot of good information tailored to your use case for a very moderate investment in time.

Experimenting with software in this way often results in new insights into long-existing problems. In my own experience over decades, I have found that the conventional wisdom of how to best accomplish particular tasks is correct about 80% of the time, while fresh unconventional approaches come out on top 20% of the time. If you are in the academic space, that might be good for a publication if the idea is in fact novel (do a literature search to find out).

1 Like

Thanks for the reply @njuffa

I will do a literature search and see what I come up with. I was just wondering if this is something that has been attempted before.


The universe of real-life CUDA applications is so big these days, it is impossible for any one person to have visibility into more than a tiny sliver of it. These forums don’t have millions of participants. And as with software in general, only a tiny portion of it is published about, and some information may be treated as some entity’s secret sauce not to be revealed to the general public.

When I have what I think is a novel idea, I usually first try create a prototype implementation based on it. Then when it looks promising, I dig into the literature. That way my mind is not subconsciously constrained by other people’s ideas. Half the time the literature search will show that someone else had the same (or at least a very similar) idea. In a few cases, the literature will show that somebody else had a superior variant of the same idea. Or an entirely different idea that I can combine with my idea for a superior solution.

Thanks @njuffa, good input.

I will give it a shot - prototype first, then literature review.