[gclist] Re: PPC & GC, or GC and threads
Mon, 26 Jan 1998 13:20:57 -0500
At 09:39 AM 1/26/98 -0800, Hans Boehm wrote:
>There is every reason to expect that with an implementation that's integrated
>into the compiler even the 2% number could be significantly lower. After all,
>empirically nearly all optimized C code is GC-safe, even on PPC.
>You can combine all of this with a safe-point notion and reduce the cost
>further. But I suspect that by the time you have an integrated implementation
>you're talking about reducing the slowdown from 1% to 0.5%.
The difficulty is in convincing someone to take the time to
design and implement such a compiler. One odd problem with
conservative collection is that its robustness makes it
difficult to test for compiler cooperation; if an alleged
cooperative compiler failed to cooperate in certain instances,
how much would you need to test it to be sure of this?
Given a choice between the blunt club described in the 1996
PLDI paper, and a compiler claimed to be cooperative-by-
design, I might well give up a bit of performance to get
the cooperation guaranteed by the blunt club.
This somewhat different for compacting collectors, especially
if you unmap the old space after it has been evacuated.