Re: Help with Atomic_Components and whole array assignment

[Robert Dewar <robert_dewar@my-deja.com>]

In article <94h55t$9a1$1@nnrp1.deja.com>,
  r_c_chapman@my-deja.com wrote:
> Can anyone confirm this is indeed acceptable behaviour?

Yes, of course this is acceptable behavior, can you point to
anything in the RM that would even HINT that it is not?

No, you can't (to save you the trouble :-)

There is NO legitimate way to ensure that the compiler
generates specific instructions other than using machine
language inserts, since the compiler is always free to
substitute different instructions which have the same effect
*as defined by the semantics in the RM* and the RM knows
nothing about byte stores etc.

pragma Atomic merely says that two tasks cannot intefere with
one another, and that the semantics is as though writes are
done for every write. But there is nothing to stop the compiler
from combining four writes into a word write (in fact this is
a very nice optimization -- all compilers should do it :-)



Sent via Deja.com
http://www.deja.com/