May be helpful for people trying to find optimal CFLAGS so we can really find out if source-based
distributions have this as an advantage.
Things it really needs to get to the next level:
- More "real world" application behavior - disk reads/writes, multimedia, device i/o, graphics, etc
- multiple runs accumulating stats - right now it runs each combination once, so if a big email comes
in during a test, the results might be skewed
- some intelligence about what flags are incompatible/redundant with others
- ditto for choosing flags not appropriate for the hardware
On that last point, I didn't have any flag combos that produced un-runnable code, even thought it
ran through all the -march possiblities (Intel and AMD) that it could think of. I would say this means
it needs some code that would be helped/broken by architecture/cpu-specific flags.
Steps to Reproduce:
Created attachment 15454 [details]
Interesting, assigning to me...
Last release 3 years ago.