integrated immediate uses - alpha patch

Andrew MacLeod amacleod@redhat.com
Fri Oct 15 13:37:00 GMT 2004


On Fri, 2004-10-15 at 08:32, Giovanni Bajo wrote:
> Andrew MacLeod wrote:
> 
> > I havent figure out how to avoid the BREAK macro tho. unless we find
> > another way to implement the safety aspect.  Perhaps something will
> > occur to me to someone else eventually.
> 
> I looked into your patch but there are things still unclear to me (eg. how do
> you plan to find out in verify_* if there is a guard a linked?), but maybe you
> can use attribute((cleanup))?
> 

The guard doesn't need either the 'stmt' or 'use' field set, so setting
them both to NULL in the guard node would mark it as a guard.

I haven't done that as yet because of the overly aggressive verify
checking I was doing to get this implementation up and running (I was
running verify links after every single linking at one point to make
sure everything was kosher). I didnt want it crapping out whenever it
saw a guard when an iterator was in use.


Andrew

oh, one other side effect I forgot is that now, given an arbitrary
use_operand_p, you can get the stmt it belongs to via USE_STMT (use_p).
The stmt is an attribute of the use, not the iterator.



More information about the Gcc-patches mailing list