if a novice programmer can code something that runs in a safe language, then an experienced programmer can translate this program to optimized, high-performance semantics.
i theorize that the experienced programmer's job can be sufficiently automated, provided that we control the novice's language, without requiring silly differentiable algorithms.
@lritter the class of systems that are safe, but also slow, is very large. Safe and sloppy can also be at odds: What happens when the runtime runs out of memory. If it's a distributed system, what happens when execution time approaches network timeout?
If the safe language can put hard bounds on mem and cpu time, it seems you've taken a big step in optimizing.
@lritter@morten_skaaning really cool but I'd likely never get anything done, I OBSESS about performance and size enough as it is that I constantly have to stop myself from over-optimizing, having to wait to profile the program until I'm actually running and testing it is...probably for the best.
@raptor85@morten_skaaning i'd argue that it teaches you something for a while - eventually you'll run out of new things to learn, having adopted a set of best practices. then you see the cost "blind" if you will
@lritter@morten_skaaning to a point, but I know myself, I'll see that number, then keep re-implementing over and over to get it lower, and before you know it I've re-implemented parts of the standard library and am knee deep in asm blocks full of extended instructions and re-ordering everything to maximize the efficiency of keeping data in registers instead of on the heap....
@raptor85@lritter@morten_skaaning
.. and then you run that on different CPU (vendor/generation, not even arch) and realize that you'd have to optimize differently to make it run optimally there :-p
Add comment