[PATCH] Avoid creating (mem (debug_implict_ptr)) (PR debug/47283)

Richard Guenther richard.guenther@gmail.com
Thu Jan 20 16:32:00 GMT 2011


On Thu, Jan 20, 2011 at 5:23 PM, Jakub Jelinek <jakub@redhat.com> wrote:
> Hi!
>
> MEMs for something that is not addressable just confuse the aliasing
> code (expectedly) and furthermore result in unnecessarily large
> location description (DW_OP_GNU_implicit_pointer followed by
> DW_OP_deref*).  This patch in that case just uses the
> COMPONENT_REF/ARRAY_REF etc. handling code, which will expand it as a SUBREG
> or similar.
>
> Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?

Ok.

Thanks,
Richard.

> 2011-01-20  Jakub Jelinek  <jakub@redhat.com>
>
>        PR debug/47283
>        * cfgexpand.c (expand_debug_expr): Instead of generating
>        (mem (debug_implicit_ptr)) for MEM_REFs use COMPONENT_REF
>        etc. handling.
>
>        * g++.dg/debug/pr47283.C: New test.
>
> --- gcc/cfgexpand.c.jj  2011-01-03 09:54:28.000000000 +0100
> +++ gcc/cfgexpand.c     2011-01-20 09:59:09.000000000 +0100
> @@ -2567,6 +2567,13 @@ expand_debug_expr (tree exp)
>
>       if (TREE_CODE (exp) == MEM_REF)
>        {
> +         if (GET_CODE (op0) == DEBUG_IMPLICIT_PTR
> +             || (GET_CODE (op0) == PLUS
> +                 && GET_CODE (XEXP (op0, 0)) == DEBUG_IMPLICIT_PTR))
> +           /* (mem (debug_implicit_ptr)) might confuse aliasing.
> +              Instead just use get_inner_reference.  */
> +           goto component_ref;
> +
>          op1 = expand_debug_expr (TREE_OPERAND (exp, 1));
>          if (!op1 || !CONST_INT_P (op1))
>            return NULL;
> @@ -2605,6 +2612,7 @@ expand_debug_expr (tree exp)
>
>       return op0;
>
> +    component_ref:
>     case ARRAY_REF:
>     case ARRAY_RANGE_REF:
>     case COMPONENT_REF:
> --- gcc/testsuite/g++.dg/debug/pr47283.C.jj     2011-01-20 10:11:09.000000000 +0100
> +++ gcc/testsuite/g++.dg/debug/pr47283.C        2011-01-20 10:13:16.000000000 +0100
> @@ -0,0 +1,58 @@
> +// PR debug/47283
> +// { dg-do compile }
> +
> +template <typename T> inline const T &
> +f1 (const T &a, const T &b)
> +{
> +  if (a < b)
> +    return b;
> +  return a;
> +};
> +
> +struct A
> +{
> +  A (int w, int h) { a1 = w; }
> +  A f2 (const A &) const;
> +  int a1, a2;
> +};
> +
> +inline A
> +A::f2 (const A &x) const
> +{
> +  return A (f1 (a1, x.a1), f1 (a2, x.a2));
> +};
> +
> +struct B
> +{
> +  A f3 () const;
> +  void f4 (const A &) { b2 = 5 + b1; }
> +  int b1, b2;
> +};
> +
> +struct C
> +{
> +};
> +
> +struct D
> +{
> +  virtual C f5 (const C &) const;
> +};
> +
> +struct E
> +{
> +  C f6 () const;
> +  int f7 () const;
> +  virtual B f8 (const C &) const;
> +  A f9 () const;
> +  virtual void f10 ();
> +  struct F { D *h; } *d;
> +};
> +
> +void
> +E::f10 ()
> +{
> +  const C c = d->h->f5 (f6 ());
> +  B b = f8 (c);
> +  b.f4 (b.f3 ().f2 (f9 ()));
> +  f7 ();
> +}
>
>        Jakub
>



More information about the Gcc-patches mailing list