[patch] Backport fixes for PR54146 to GCC 4.7

[Steven Bosscher]

On Wed, Aug 22, 2012 at 3:44 PM, Jakub Jelinek <jakub@redhat.com> wrote:
> To me the patch looks way too long, changing too many things, to be suitable
> for the release branch.

Most of the changes are only to allocate bitmaps on dedicated
obstacks. It makes the patch long, but the changes are not really
significant except for bringing down the memory footprint (e.g. for
c-common.c this patch halves the peak memory usage in top).

Only the tree-ssa-loop-manip.c changes are the more complex bit,
actual algorithm changes. The rewrite of add_exit_phis() makes an
O(b_basic_blocks^2) algorithm linear, but I can see why this may be
considered unsuitable for a release branch. I could simplyleave these
changes out and only apply the bitmap_obstack bits .

I think it's important that GCC 4.7 gets less memory-hungry. This is
the first GCC with a usable LTO, but it currently consumes way too
much memory. My patches, and those from Richard and you to plug leaks
in various places (which I plan to test and propose for back-porting
also), make GCC 4.7 a better compiler with almost no risk.

Would it be acceptable for you if I apply the patches one at a time
(after testing them individually first)?

Ciao!
Steven