Re: [Emc-developers] Link-time optimization in realtime objects
> ... > I think the only way to find out is to test it. > On the face of it, short functions, repeatedly used, should benefit from it. Then programming for Micro controller I quite often then there is no or little difference between different methods for other reasons check size of compiled code constructions to learn which one produce the smallest code. I would not be suprised if smaller code with fewer instructions also execute faster. Most of time difference in execution time and size is small and not very imortant though I still consider it good to learn otherwise optimizations might go the wrong way or constructions consuming quite large resources may be used without a good reason. What is missing is a tool to calculate worst case execution time. Nicklas Karlsson ___ Emc-developers mailing list Emc-developers@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/emc-developers
Re: [Emc-developers] Link-time optimization in realtime objects
Hi all, the link-time optimization as such is highly dependant on the target architecture (as well as the compiler/linker suite of course). While on smaller, more simple core architectures, e.g. small µCs, the effect can be quite large, on µP architectures with complex multi level caches, intelligent prefetch whatsoever, it's most of the times a matter of try and error whether the optimization is really helpful. If you now consider the code might be run on different processors, it renders the effort more on the waste than helpful side. But try it anyways and let us know what you think! Regards, Daniel Am Fr., 16. Aug. 2019 um 15:34 Uhr schrieb schoone...@gmail.com < schoone...@gmail.com>: > > On 16/08/19 13:04, Robert Ellenberg wrote: > > Hi All, > > > > Has anyone tried to use link-time optimization in the realtime object > > builds? I think the motion / TP build would be a good candidate for it, > so > > we could improve inlining of things like posemath. > > > > Best, > > Rob > > > > > Hi Rob, > > In an ideal world, the inline code in question should be inserted by the > compiler, > removing the call and stack shuffling thus producing a faster binary. > What the compiler actually does sometimes differs.;-) > > If the code is long, a couple of push/pops from the stack if negligible, > compared to the amount of binary code bloat you can get, sometimes at > the expense > of code sitting nicely in processor cache ready to be instantly called > again. > > I think the only way to find out is to test it. > On the face of it, short functions, repeatedly used, should benefit from > it. > > Nothing in defining inlines should affect rt operation AFAIK, > we have shed loads of inlines and defined functions in the revised rtai > / hal code in Machinekit. > (mostly wrapper functions and accessors etc. to cater for both legacy > code and the newer code) > > cheers > > > ___ > Emc-developers mailing list > Emc-developers@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/emc-developers > ___ Emc-developers mailing list Emc-developers@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/emc-developers
Re: [Emc-developers] Link-time optimization in realtime objects
On 16/08/19 13:04, Robert Ellenberg wrote: Hi All, Has anyone tried to use link-time optimization in the realtime object builds? I think the motion / TP build would be a good candidate for it, so we could improve inlining of things like posemath. Best, Rob Hi Rob, In an ideal world, the inline code in question should be inserted by the compiler, removing the call and stack shuffling thus producing a faster binary. What the compiler actually does sometimes differs.;-) If the code is long, a couple of push/pops from the stack if negligible, compared to the amount of binary code bloat you can get, sometimes at the expense of code sitting nicely in processor cache ready to be instantly called again. I think the only way to find out is to test it. On the face of it, short functions, repeatedly used, should benefit from it. Nothing in defining inlines should affect rt operation AFAIK, we have shed loads of inlines and defined functions in the revised rtai / hal code in Machinekit. (mostly wrapper functions and accessors etc. to cater for both legacy code and the newer code) cheers ___ Emc-developers mailing list Emc-developers@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/emc-developers
