@cstross @blacklight I still optimize for memory when I code. It’s a hard habit to break.
Top-level
@cstross @blacklight I still optimize for memory when I code. It’s a hard habit to break. 4 comments
@blacklight @dan613 @alan @cstross Most processors are ill-suited to the highly dynamic list-oriented data structures that are prevalent in semantic/conceptual AI programs (as opposed to the matrix operations common to associative AI, e.g., LLMs). Until we have reliable processor-in-memory (PIM), developers will still need to optimize memory management. @dan613 @cstross @blacklight As you should. Memory management has been a long time source of dynamic runtime errors. I would cast a skeptical eye towards anyone who says compilers handle memory optimization just fine. @blacklight @dan613 @cstross I optimise for speed wherever there is a trade-off between the two. For example you could not store an intermediate result and recalculate it every time you need it. That would save memory but use more CPU cycles. |
@dan613 @cstross @blacklight The amount of time it's taken me to accept that compilers actually do a good job of optimization these days and thus that I can abandon various constructs to help them out has been considerable. Except ++var over var++. That' still sometimes a thing.