[2/2] Reducing the overhead of dwarf2 location tracking

[Jakub Jelinek]

On Fri, Mar 04, 2011 at 03:30:34PM +0100, Richard Guenther wrote:
> On Fri, Mar 4, 2011 at 2:56 PM, Richard Sandiford
> > Jakub confirms that this indeed the intended behaviour, and I haven't
> > seen any problem with the output.
> 
> Hm, but isn't it incorrect debug info?  I would have expected this
> non-localized var to be the abstract origin of a copy for location
> list purposes.

I thought we've discussed that on IRC.  For -fcompare-debug reasons
the decision on what decl is localized and what goes into
BLOCK_NONLOCALIZED_VARS can't take into account debug stmts (otherwise
there would be -fcompare-debug failures).
Thus, we sometimes decide to have some BLOCK_NONLOCALIZED_VARS vars
referenced in debug stmts.
The way this works is we do normal var-tracking against those, and then
at dwarf2out.c time just pick their locations up.
See e.g.
http://gcc.gnu.org/ml/gcc-patches/2010-03/msg00445.html
for some of the details.

	Jakub