A response on the TBB forum has revealed that the cache_aligned_allocator will use the scalable_allocator when present. With all the possible choices of memory allocators, I’m a bit puzzled as to which ones to use where. This is something that will have to be examined in greater detail in future. The scalable_allocator seems like the best choice right now, since the cache does not seem to be a major bottleneck at the moment. Each data structure will have unique memory requirements, depending upon its role within the simulation. It is most likely that each component will have to be carefully considered, and the appropriate allocator chosen. But this is something that can happen later… and perhaps facilities will be provided to the user to adjust these defaults themselves at runtime to tune their particular simulation.
Posted by AJ Guillon (January 9, 2008)
