This is the mail archive of the gcc@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: Compile time speed


 > But the kind of patches mentioned above isn't something I
 > would consider. 
 > -- Gaby

I agree.  The physmem heuristics patch is not safe enough to backport
for 3.2.x.  Not only are we unsure of potential problems, but another
issue is that the patch relies on the --param interface to allow
tweeking the values.  Backporting physmem + --param would be even more
destablizing.  Without the --param ability, one couldn't manually
override the heuristics to ensure a given bug isn't related to GC
parameters.


We might want to consider bumping up the default hardcoded GC
values in 3.2.x.  As in here (but not quite so big):
http://gcc.gnu.org/ml/gcc/2003-02/msg02002.html

E.g. we could bump up heapsize (last allocated) to 8M or 16M and
expand could be 1.4 or 1.5.

Thoughts?

		--Kaveh
--
Kaveh R. Ghazi			ghazi at caip dot rutgers dot edu


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