[VTA] retain more debug info in loop-ivopts

[Richard Guenther]

On Fri, Sep 4, 2009 at 10:33 AM, Alexandre Oliva<aoliva@redhat.com> wrote:
> On Sep  3, 2009, Richard Guenther <richard.guenther@gmail.com> wrote:
>
>> On Thu, Sep 3, 2009 at 9:07 AM, Alexandre Oliva<aoliva@redhat.com> wrote:
>>> On Sep  1, 2009, Alexandre Oliva <aoliva@redhat.com> wrote:
>
>>> In a last-minute (ok, make that last couple of hours ;-) review of the
>>> patch, I realized tree-ssa-loop-ivopts.c was working too hard, dropping
>>> uses of an SSA DEF early, instead of letting them be propagated into
>>> debug stmts by the later-introduced code in release_ssa_name.
>
>>> This patch reverts that (hopefully) unnecessary change.  Ok to install,
>>> if it passes regression testing?
>
>> Ok.
>
> That patch didn't work, because we removed DEFs in the wrong order.
> Sometimes uses would survive in debug stmts.
>
> Removing dominating DEFs after dominated ones fixes the problem.
> Unfortunately it adds some overhead, and we have to incur that overhead
> even for non-debug builds, otherwise there's codegen divergence :-(
>
> Is this still ok if it passes?  It passed bootstrap-debug-lean on
> x86_64-linux-gnu, but it's still building libraries, and then it will
> start testing.

Ugh.  No.

Why again does simply relying on release_ssa_name not work?
Shouldn't propagate_var_def_into_debug_stmts deal with the situation
and not propagate?  That's why we have the find_released_ssa_name stuff
there, no?

Richard.
>
>
> --
> Alexandre Oliva, freedom fighter    http://FSFLA.org/~lxoliva/
> You must be the change you wish to see in the world. -- Gandhi
> Be Free! -- http://FSFLA.org/   FSF Latin America board member
> Free Software Evangelist      Red Hat Brazil Compiler Engineer
>
>