Virtual memory support is a relatively hot topic in GPUs now. And one of the problems is that, unlike CPUs that have had this solved for a long time, GPUs have not had support for precise exception handling or efficient context switching. It's one thing to do address translation; it's entirely another to handle the case when the virtual memory is completely absent from physical memory because it's been swapped to disk. In cases like this, the time to resolve the memory access is so colossal that we need a way to completely suspend execution and restart it at a later time. Meanwhile, we would use the GPU for something else. This isn't like a cache miss where the shader engine can just wait it out because the delay is maybe only a thousand or so cycles.
I've added two more links for related papers. Unfortunately, one is available as yet _only_ via ACM. If you find a mirror somewhere, please give me the link. Thanks! https://sourceforge.net/p/openshader/wiki/gpu_links_and_articles/ -- Timothy Normand Miller, PhD http://www.cse.ohio-state.edu/~millerti Open Graphics Project _______________________________________________ Open-graphics mailing list [email protected] http://lists.duskglow.com/mailman/listinfo/open-graphics List service provided by Duskglow Consulting, LLC (www.duskglow.com)
