Mailinglist Archive: radeonhd (312 mails)

[Matthias Hopf <mhopf@xxxxxxx>]

On Mar 26, 09 10:02:00 +1300, Michael Cree wrote:
> On 26/03/2009, at 3:17 AM, Matthias Hopf wrote:
> > I don't know why this cpu reorders volatile memory accesses, though.
> Volatile is a direction to the C compiler that the memory datum may change 
> at any time (whether by task switch or interrupt driver, etc.).  It turns 
> off optimisation strategies that assume the memory datum is only acted upon 
> by the present code under compilation.  It does _not_ tell the compiler 
> that the datum at the memory location is an I/O port.
[...]

I know all that, just wanted to indicate that due to being volatile the
compiler won't reorder them (in contrast to regular memory accesses that
can be reordered in between sequence points. AFAIR a volatile access is
a sequence point, function calls are sequence points as well.

Until now I was pretty sure that only few architectures (especially
including alpha) did memory reordering in hardware. I learned different
today.

CU

Matthias

-- 
Matthias Hopf <mhopf@xxxxxxx>      __        __   __
Maxfeldstr. 5 / 90409 Nuernberg   (_   | |  (_   |__          mat@xxxxxxxxx
Phone +49-911-74053-715           __)  |_|  __)  |__  R & D   www.mshopf.de
-- 
To unsubscribe, e-mail: radeonhd+unsubscribe@xxxxxxxxxxxx
For additional commands, e-mail: radeonhd+help@xxxxxxxxxxxx