[PATCH] Remove special-casing of PTR_IS_REF_ALL pointers from alias analysis
Richard Guenther
rguenther@suse.de
Wed Apr 30 14:33:00 GMT 2008
On Wed, 30 Apr 2008, Eric Botcazou wrote:
> > Sure. But that's the business of PTA to figure out. The current
> > code tries to be too much clever here (just change PTR_IS_REF_ALL
> > to include char * in C and see the execute torture scream at you).
>
> I don't think PTA can do what PTR_IS_REF_ALL was doing.
? What do you think PTR_IS_REF_ALL was doing?
> As for the failures
> in the testsuite, they could come from TYPE_REF_CAN_ALIAS_ALL itself, it
> doesn't play nice with inlining at the moment.
? I merely changed flow-insensitive aliasing to treat all pointed-to
alias set zero memory like TYPE_REF_CAN_ALIAS_ALL. This shows that
flow-insensitive aliasing for TYPE_REF_CAN_ALIAS_ALL pointers is broken.
TYPE_REF_CAN_ALIAS_ALL itself is not very well tested (two or three
testcases, ada only and not in a torture).
Richard.
More information about the Gcc-patches
mailing list