Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!uunet!tdatirv!sarima
From: sarima@tdatirv.UUCP (Stanley Friesen)
Newsgroups: comp.lang.c++
Subject: Re: 64 bit architectures and C/C++
Message-ID: <226@tdatirv.UUCP>
Date: 30 Apr 91 00:35:20 GMT
References: <168@shasta.Stanford.EDU> <224@tdatirv.UUCP> <295@dumbcat.sf.ca.us>
Reply-To: sarima@tdatirv.UUCP (Stanley Friesen)
Organization: Teradata Corp., Irvine
Lines: 41

In article <295@dumbcat.sf.ca.us> marc@dumbcat.sf.ca.us (Marco S Hyman) writes:
>I hope not -- at least not without some other way of describing a 16-bit
>value.  64-bit architecture machines will still have to communicate with other
>machines that do support 16-bit values.  Swapping bytes between big and little
>endian machines is bad enough. Think about the overhead of converting a pair
>of bytes to a 16-bit value.  

Binary communication between machines should use XDR *not* C structures.

As for how to do it, see below.

>Hmmm.  How would such a processor communicate with hardware devices requiring
>16-bit I/O?  How would a structure that maps an external device's registers be
>coded if the registers are 16-bits wide?  If there is a way to do these things
>then a 16-bit wide data type is probably necessary.

Well, two seperate points.

First, if the machine does not have 16-bit operations in its instruction set,
then 16-bit I/O registers will probably be wired into 32-bit pseudo-words
(with the high-order 16-bits wired into the bit-bucket).  Remember, the lack
of 16-bit instructions would mean that 16-bit bus transfers are not possible.

Secondly, there are two possible implementations of this sort of thing.

First, C has this concept called bit fields, which can be of any size you
want - so if you *must* have a 16-bit quantity, use a 16-bit bit field.
(This also answers the previous objection - if you *must* be machine
dependent, at least make it stand out).


Alternatively, for the I/O register case, just send two sequential bytes.
(Unless the machine does not have byte adressing - which is quite possible
in a 64-bit super-computer).


Remember, binary compatibility between machines is a very shaky proposition
at best, so it is better not to depend on it at all.
-- 
---------------
uunet!tdatirv!sarima				(Stanley Friesen)