Subject: Re: amd64 stable for production ?
To: Johnny Billquist <bqt@softjar.se>
From: Andrew Reilly <andrew-netbsd@areilly.bpc-users.org>
List: netbsd-users
Date: 12/12/2006 12:05:55
On Mon, 11 Dec 2006 19:50:08 +0100
Johnny Billquist <bqt@softjar.se> wrote:
> Now, bytes on the other hand could lead to a very funny discussion that
> many today don't appreciate fully.
There are still quite a few strictly word-addressed machines in
the embedded/DSP arena. That is,
sizeof(char)==sizeof(int)==sizeof(long)==sizeof(void*)==1
[and they all occupy 16,18,20,24,32,40, or 48 bits: take your
pick; mostly it's obvious for the processor architecture;
sometimes it depends on the address range...], and you need
software support and special APIs to deal with packed
byte-strings... This is also how many older mainframes of the
Cray, Burroughs or PDP-10 vintage operated. That's the "true"
origin and meaning of the term "word". Any of this newfangled
PDP-11/Microsoft-mandated fixed size rubbish is just
unnecessarily confusing. Or, rather, it's confusing because the
significance of the term isn't quite as obvious on modern,
byte-addressed machines that have a whole slew of plausible
"word" sizes.
Personally, I use "word" in an algorithm-specific sense to mean
any atomically readable/writeable quantity larger than a byte
that could plausibly be ported to a word-addressed machine and
still work correctly.
Cheers,
--
Andrew