Re: [patch][4.0] PR 21171

[Michael Matz <matz at suse dot de>]

Hi,

On Mon, 13 Jun 2005, Richard Earnshaw wrote:

> > http://gcc.gnu.org/ml/gcc-patches/2005-04/msg02750.html
> > 
> > OK for 4.0?
> > 
> > Zdenek
> 
> The use of volatile in this test was always a red herring.  The code
> should execute correctly even when the variables aren't volatile
> qualified.  Your patch (IIRC) only addresses the volatile issue so,

In tree-ssa, instructions of this form:
  ((T*)1000)->mem = x;

i.e. where an INTEGER_CST is dereferenced, will have their
has_volatile_ops flag set (see get_indirect_ref_operands) .  This has
nothing to do with the user specifying any volatile himself.  Zdeneks fix
will make IVops not substitute the integer constant used in such context
with some SSA_NAME (in which case the reason for setting has_volatile_ops
would be missing, so DCE will then remove the store).

For completeness: this bug also results in a miscompilation of grub, which 
is also fixed by the bug.

> while it might be useful in some real cases, doesn't really fix this
> particular problem.

Really?  Does the problem still actually occur with the patched compiler?


Ciao,
Michael.