[Bug other/67893] [5 Regression] memory hog building a C++ file (on x86_64-linux-gnu)

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67893

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2015-10-08
            Version|6.0                         |5.2.1
   Target Milestone|---                         |5.3
     Ever confirmed|0                           |1

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
Can't yet confirm the regression (need to look for a bigger box...) but I can
confirm it uses >2GB even with plain -O2.  Note that -O2 has quite some
quadratic algorithms.

As even plain -O1 goes up to 2GB my usual suspect would be DF.

Note that with -O1 frontend processing peaks at 600MB and after gimplification
and lowering we are at 1GB - so this is a huge CU in the first place.  Early
opts plus IPA get us to 2GB already (same at -O2).

What's the second preprocessed source from?

Confirmed as memory-hog though.

Unfortunately as the testcase is so large it's hard to work with it but it
looks
like we gradually leak memory during late optimization/expansion, there are no
"big" functions in the end it seems.

Btw, the gcc 5 branch at r228565 peaks at 2.3GB with -O2, -g adds that other
1.2GB (hello var-tracking and VTA).