This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: preparing patch for IPCP merge to SSA
> On 1/9/07, Razya Ladelsky <RAZYA@il.ibm.com> wrote:
> >Hi,
> >
> >This patch is a preceding patch to IPCP merge to SSA.
> >
> >It includes:
> >
> >1. Creating a new function, get_last_funcdef_no() that returns the value
> > of funcdef_no, and increases it by one.
>
> The function should be called get_next_funcdef_no() instead. Also
> simply
> return funcdef_no++;
> will do, no need for the temporary.
>
> > The newly allocated function in initialize_cfun uses it to get a new
> >funcdef_no .
> >2. Cleanup in tree_function_versioning.
>
> This change misses from the changelog. Can you use -c -p to generate
> patches please?
>
> The funcdef parts are ok if you make the suggested changes, if honza
> agrees to the cleanup part this is ok as well if you add the missing
> changelog part.
Yes, that part is OK - it is same as we do in inliner. We need to fully
cleanup after constant propagation or we eventually get ICEs on
unreachable blocks during dominator tree construction.
I would say that this shall be normally done by the passmanager - I plan
to add capability to drop function specific TODO flags from IPA passes
so this work can again be deferred to it.
Thank you,
Honza
>
> Thanks,
> Richard.
>
> >Thanks to Jan for his help.
> >
> >Bootstrapped and tested on i686 Linux.
> >Bootstrapped and tested on ppc Linux.
> >
> >Ok for mainline?
> >
> >Thanks,
> >Razya
> >
> >Changelog:
> >
> > * function.c (get_last_funcdef_no): New function.
> > * function.h (get_last_funcdef_no): Declare.
> > * tree-inline.c (initialize_cfun): Add initialization.
> >
> >
> >
> >
> >
> >