What I would like to be able to do, instead of -O1, -O2 -O3 flags to a compiler is for the "optimization level" to be measured in seconds. So -O1 would run the optimizer for 1 second and give me the best optimized code it could come up with in that time. -O3600 would let the machine think about it for an hour, using any and all heuristics and empirical tests and then giving me what it had at the end. Pre-release, I might want to run the optimizer for a week.
> Pre-release, I might want to run the optimizer for a week.
That's a dangerous model, as certain classes of bugs would only come out in the super-optimized version, but that version would presumably not get the same amount of testing as the regular builds.
Comments
What I would like to be able to do, instead of -O1, -O2 -O3 flags to a compiler is for the "optimization level" to be measured in seconds. So -O1 would run the optimizer for 1 second and give me the best optimized code it could come up with in that time. -O3600 would let the machine think about it for an hour, using any and all heuristics and empirical tests and then giving me what it had at the end. Pre-release, I might want to run the optimizer for a week.
> Pre-release, I might want to run the optimizer for a week.
That's a dangerous model, as certain classes of bugs would only come out in the super-optimized version, but that version would presumably not get the same amount of testing as the regular builds.