This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [RFC][4.3 regression] Fix massive compile-time and memory consumption for large numbers of SFTs


On Fri, Jul 18, 2008 at 01:46, Richard Guenther <rguenther@suse.de> wrote:

> IMHO this is a way too "simplistic" heuristic for the branch.  The
> SFTs itself are not too expensive - where is the memory gone?

Yes, it is simplistic by design.  I very explicitly wanted to
implement a straightforward heuristic.  If there are too many symbols
that can be partitioned, it really doesn't make sense to do so.

The problem is the usual massive increase in the number of symbols and
virtual operands.  Operand scanning time goes through the roof, memory
consumption also explodes, etc.

> In the end at least some people will want to use GCC 4.3 for
> compiling performance critical applications, so - can you
> 1) share your testcase, 2) provide SPEC 2k6 numbers before/after
> your patch?

I can do SPEC2k, I don't have SPEC2k6 setup.  I don't know if I can
share the test case, I'll find out.


Diego.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]