Re[4]: [Gc] Re: [bigloo] Performance evaluation

Ivan Maidanski ivmai at mail.ru
Wed Aug 11 23:05:29 PDT 2010


Thu, 12 Aug 2010 07:41:58 +0200 Manuel.Serrano at sophia.inria.fr:

> > The better question is: what are the options passed to compiler?
> "-O3  -Wpointer-arith -Wswitch -Wtrigraphs -Iinclude -Ilibatomic_ops-install/include -DFINALIZE_ON_DEMAND -fPIC"

-Ilibatomic_ops-install/include is redundant here (as you are not using threads).

And, you don't use GC_enable_incremental(), right? (or just add -D GC_DISABLE_INCREMENTAL to exclude incremental support).

BTW. For speed (I'm not talking about this experiment), you could also add -fprofile-generate/use -mtune=native -freorder-blocks-and-partition and compile extra/gc.c instead of all *.c files ;).

Note 2: IMHO, it's dangerous to optimize GC without -fno-strict-aliasing.

> 
> > Also, it would be better if you replace gc72a4 with the recent CVS snapshot.
> I can imagine doing that for conducting an experiment but not for 
> shipping the stable Bigloo version.

Yes, purely for the measurement.

> 
> -- 
> Manuel



More information about the Gc mailing list