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: Faster compilation speed


On Wed, Aug 14, 2002 at 09:35:07PM +0100, Jamie Lokier wrote:
> Timothy J. Wood wrote:
> >    I'm not sure what you mean.  The external RC table only gets modified 
> > once every 2^N operations already (and never if the internal RC never 
> > overflows the internal RC).  Stated another way, the internal RC holds 
> > the low order bits and the external RC holds the high order bits.
> 
> Ah, right, I misunderstood the code.  Your way is a fine way.
> Presumably if it only gets called every 2^N operations at most, then
> it's not _strictly_ the high order bits in the hash table.  I.e. I'm
> thinking of the reference pattern 127 -> 128 -> 127 -> 128...

A simpler strategy is to just make every object with RC >= 2^n immortal.
Something like:

add: if (rc) rc++;

subtract: if (rc) if (!--rc) delete;

For a program like gcc that doesn't have to absolutely guarantee it
leaks no memory, this can be an acceptable tradeoff.

(for belt and suspenders people, run a infrequent lazy gc to clean up
the scraps)

-Tim


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