This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: df.c data reprezentation
- To: Jan Hubicka <jh at suse dot cz>
- Subject: Re: df.c data reprezentation
- From: Daniel Berlin <dan at cgsoftware dot com>
- Date: Tue, 25 Sep 2001 10:59:28 -0400
- Cc: gcc at gcc dot gnu dot org, rth at cygnus dot com, m dot hayes at elec dot canterbury dot ac dot nz, matzmich at cs dot tu-berlin dot de
On Tuesday, September 25, 2001, at 10:25 AM, Jan Hubicka wrote:
> Hi,
> Just note about the data reprezentation. Unlike our previous dataflow
> analyzers
> (REG_DEAD/UNUSED notes+loglinks), we don't use RTL to reprezent the
> dataflow here.
>
> Do you consider that as supperrior?
No.
> Main advantage of RTL based reprezentation is the
> fact that you can easilly attach the information to each insn and see
> it reprezented
> in the dumps?
>
However, dataflow is completely seperate from the representation.
> It would actually mean to change REF structure to RTX form and use
> expr_list for links.
> Insn info can use the existing loglink field with one new for the
> oposite chain.
>
> What do you think?
>
This is a very bad idea.
I know I was, and others were, specifically planning on reusing the df
module for dataflow on trees.
Right now, it only requires assigning each variable an id (trivial), and
adding routines that know how to record defs/uses for trees.
If you want to transform the info at the end of dataflow into stuff on
rtx, that's fine. But please don't make it do it in the actual routines,
or in the dataflow structures.
> Honza