Andarwinux wrote:

> > But I still feel it will open a Pandora’s box. I really don’t want to see a 
> > situation decades from now still trapped in the original x86-64 without 
> > extensions.
> 
> Sure, but does that really justify intentionally reducing performance for 
> default builds? Maybe in other cases it does, but in this case the 
> performance difference between runtime check and -march is negligeable (and 
> even invisible because of the noise). Adding a runtime dispatch won't help 
> distros switch to -march builds, but it is for the valid reason that -march 
> makes no difference in this case. The compromise seems to me more between 
> improving performance (in the short-term at least) at the cost of 50 lines of 
> code, and doing nothing. The nothing route is justifiable by the fact this PR 
> may become somewhat obsolete in a few years from now, but at the same time 
> the runtime dispatch doesn't seem to have any real negative aspect to it.

Although this PR itself is insignificant, it will become a precedent - people 
might start adding runtime dispatch everywhere in LLVM - why not?



https://github.com/llvm/llvm-project/pull/180631
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

Reply via email to