[vta] more propagation of information in debug annotations

[Alexandre Oliva]

A long time ago, I noticed we were dropping a lot of useful debug
information early on.  I know I said a few days ago that I was going
to work next on the transformation of debug insns in var location
notes, but I got a personal request to look into a problem that turned
out to be partially caused by this early loss of information.

So I fixed it, and ran into a few other issues, some caused by the
formerly-unexpected presence of addressable memory in debug
annotations, some caused by the resulting complexity of debug
annotations, and a few latent issues exposed by these changes, because
of which -fvar-tracking-assignments caused slightly different code to
be generated.

Today's patchset fixes all of these.

I'm starting with the most complex patch, that addresses the issue of
propagation of non-constant SSA assignments and addressable-memory
expressions into debug annotations, and the immediate fallout.


ATM, I don't add VOPs to debug annotations.  I can't quite see where
they'd be useful, but I haven't thought too hard about this issue.  I
just know that, if we end up having reason to add them, we have to be
very careful to avoid messing with the decl UIDs out of the various
memory tags.  Thoughts?


I'm checking this in the vta branch.  Bootstrap-debugged and
bootstrap4-debug/g0ed on x86_64-linux-gnu, i686-linux-gnu,
ppc64-linux-gnu and ia64-linux-gnu.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: vta-ssa-fwprop-more.patch
Type: text/x-patch
Size: 7997 bytes
Desc: not available
URL: <http://gcc.gnu.org/pipermail/gcc-patches/attachments/20080109/8c977ca1/attachment.bin>
-------------- next part --------------

-- 
Alexandre Oliva         http://www.lsd.ic.unicamp.br/~oliva/
FSF Latin America Board Member         http://www.fsfla.org/
Red Hat Compiler Engineer   aoliva@{redhat.com, gcc.gnu.org}
Free Software Evangelist  oliva@{lsd.ic.unicamp.br, gnu.org}