At GDC, SemiAccurate sat down with Andrew Richards of Codeplay to talk about OpenCL’s past, present, and future. The results went in directions we had not been anticipating, especially surrounding the future directions of the language.
The interview, conducted in what started out as a quiet corner of Moscone South, turned in to a wonderland of people drilling concrete, workers on break deciding to talk loudly next to our cameras, and just about everything else you could imagine going wrong all happened to us. Our usual low standards were exceeded by the atmosphere’s, no question there, but at least the sound is leaps and bounds above the last time we tried it.
In spite of all this, the talk covered the present and future directions of OpenCL, including some aspects that the author hadn’t even considered. For example, how much have you considered the power costs of moving massively parallel data across a chip? Front ends for OpenCL other than C/C++? If those topics aren’t high on your worry list, you probably aren’t participating in the planning for OpenCL 2.0, or whatever it ends up being called.
In addition to these big ones, Andrew talked about performance portability, code obfuscation problems, intermediary data structures, and how this all affects tools for different platforms. Even some blindingly obvious things like 68000, Sega Genesis, and Z80 ports for OpenCL, along with possibly doing graphics in OpenCL were touched upon. Really. Watch the video, and then wonder about what we talked about after the interview was over. That’s when things got really weird.S|A