Take 3: RFA: re-instate struct_equiv code

Daniel Berlin dberlin@dberlin.org
Wed Feb 8 18:07:00 GMT 2006


On Wed, 2006-02-08 at 17:56 +0000, Joern RENNECKE wrote:
> Daniel Berlin wrote:
> 
> >  
> >
> >By relying on broken life information and claiming it's "just fine"?
> >  
> >
> Depends on what you call broken.  We have established that some registers
> might be flagged as live that are not live any more, that this would be 
> expensive
> to fix, and that the algorithm can work with this overly-conservative life
> information without generating bad code.

It is only expensive to fix *iff* you keep it in the loop.

If you simply run this a few times per compilation, instead of at every
cleanup_cfg call, you can have correct and precise register liveness
information, and probably get *better* results than running it where it
is now with conservative info.


> 



More information about the Gcc-patches mailing list