This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH,PR43478] preserve dying REGs in debug temps
On Mon, May 3, 2010 at 5:24 PM, Jakub Jelinek <jakub@redhat.com> wrote:
> On Mon, May 03, 2010 at 11:55:42AM +0200, Steven Bosscher wrote:
>> Yes. Let's make GCC modular by making it one big module. Seriously, DF
>> is RTL only, so it should not have to know about trees.
>
> Well, as you know several RTX types contain tree arguments, so there aren't
> so strict boundaries in between the tree and rtl worlds.
Yes, I'm not so naive to think there should be an impenetrable wall
between RTXen and trees. I think this is true for almost all types in
coretypes.h.
But what I hope, is that we can limit the dependencies to knowing
about the existence of 'tree' in the RTL world and vice versa (i.e.
pointer to struct/union is known but not more), and handle the
interactions over more-or-less properly defined interfaces. Obviously
GCC is far from this ideal picture of mine (and hopefully others) at
the moment, but one has to start somewhere.
For example, next on my list is unincluding rtl.h in the tree-* files
as much as possible.
> Here is an updated patch that moves this tree/rtl stuff to a separate
> helper, and thus keeps df-problems almost tree handling free. ?Of course,
> DEBUG_EXPR_TREE_DECL macro gives a tree and gen_rtx_VAR_LOCATION takes
> a tree, but you don't need tree.h for that.
Thanks for taking care of this.
Ciao!
Steven