Re: [RFA] Use data cache for stack accesses

[Daniel Jacobowitz <drow@false.org>]

On Thu, Sep 03, 2009 at 08:38:30AM -0700, Doug Evans wrote:
> On Wed, Sep 2, 2009 at 1:42 PM, Tom Tromey<tromey@redhat.com> wrote:
> >>>>>> "Doug" == Doug Evans <dje@google.com> writes:
> >
> > Doug> * dwarf2loc.c (dwarf2_evaluate_loc_desc): Mark values on stack with
> > Doug> set_value_stack.
> >
> > I ran across this while merging the DW_OP_*_value patch.
> >
> > Do we really know that such values always come from the stack?  It seems
> > plausible to me that this is the case in practice, but aren't compilers
> > free to refer to any memory at all from a DWARF expression?
> 
> Blech.  It's a bit confusing.
> 
> I don't honestly know.
> I'll do some research.

They can come from the stack, or elsewhere.  You might be able to
track based-on (similar to how prologue-value.h does it) to identify
CFA-relative accesses.  DW_OP_deref has to clear it, though.

-- 
Daniel Jacobowitz
CodeSourcery