14 Dec
2009
14 Dec
'09
5:59 p.m.
On Monday 14 December 2009, at 21.21.55, Paul Davis
<paul(a)linuxaudiosystems.com> wrote:
On Mon, Dec 14, 2009 at 3:04 PM, Nick Copeland
<nickycopeland(a)hotmail.com> wrote:
the kernel implementation of atomic integer writes suggested
otherwise, if my memory holds correct.
since the kernel doesn't have atomic float ops, and i did see kernel
code for this that noted
the issue, i think my memory is correct. but its been wrong so many
times before.
I have a faint memory of this having to do with the memory subsystem of SPARC
SMP systems... I'm not so sure this has much to do with atomic reads and
writes, though. A quick search gave me this:
http://lwn.net/Articles/71732/
--
//David Olofson - Developer, Artist, Open Source Advocate
.--- Games, examples, libraries, scripting, sound, music, graphics ---.
| http://olofson.net http://kobodeluxe.com http://audiality.org |
| http://eel.olofson.net http://zeespace.net http://reologica.se |
'---------------------------------------------------------------------'
guaranteed for 24 bits of a 32 bit value. funny old world, eh? :)
Wasn't that just for 32 bit floats though? They were implemented as a
structure
that implied multiple write operations, they only garanteed the mantissa,
not the exponent. I am pretty sure integer memory IO was guaranteed as
that is where the 24 bits came from.