> -----Original Message-----
> From: owner-javamemorymodel@cs.umd.edu
> [mailto:owner-javamemorymodel@cs.umd.edu] On Behalf Of Ben Wint
> Sent: Wednesday, February 19, 2003 1:52 PM
> To: javamemorymodel@cs.umd.edu
> Subject: Re: JavaMemoryModel: Question: Why is it that Memory
> Models are Intuitively Hard?
>
>
>
> I would have thought the hardware guys could use some of the
> googolplexes of gates they are getting, to figure out the
> rare cases when they need to slow down to maintain sequential
> consistency, and when they can go at full speed.
>
But they have! (See Gharchorloo et al. ICPP'91, Pai et al. ASPLOS'96,
Ranganathan et al. SPAA'97, Hill IEEE Computer'98, Gniady et al.
ISCA'99, and others.) The compiler people, however, still have to figure
out how to reorder memory accesses and use registers safely (the theory
is there, but looks like practice is still some ways away). As long as
the compiler violates sequential consistency, sequentially consistent
hardware doesn't do Java programmers any good (and so in turn reduces
the motivation for hardware to be sequentially consistent).
Sarita
-------------------------------
JavaMemoryModel mailing list - http://www.cs.umd.edu/~pugh/java/memoryModel
This archive was generated by hypermail 2b29 : Thu Oct 13 2005 - 07:00:42 EDT