[PATCH] D18576: Initial implementation of optimization bisect

[Matthias Braun via llvm-commits]

MatzeB added a comment.

In http://reviews.llvm.org/D18576#386421, @andrew.w.kaylor wrote:

> I definitely agree that the pass manager, and by extension the OptBisect class, shouldn't know anything about the passes.
>
> I'd like to have some mechanism to ask the pass (or its pass registry entry) if it can be skipped.  That would be pretty trivial to fit into the legacy pass manager interface with a default response (I think it should default to skippable, but we can debate that).  I suppose a conceptually similar thing could be done with the new pass manager, but it feels more forced in that case in as much as it is one more bit of interface imposed on the pass concept.
>
> The one thing I'm sure of is that the way I'm currently doing it is wrong.


Maybe the passes that should be skipped for optnone functions are the same passes that can be disabled by OptBisect? Which would mean we could design a common helper functions for both replacing all the optnone checks with it.


Repository:
  rL LLVM

http://reviews.llvm.org/D18576