Re: ECC logging implementation
- From: Terje Mathisen <terje.mathisen@xxxxxxxxxxxxx>
- Date: Mon, 10 Nov 2008 07:49:31 +0100
Glen Herrmannsfeldt wrote:
Terje Mathisen wrote:AFAIK, the official Intel/x86 memory ordering specification says that it should not happen, but I hope I'll never write code which implicitly depends on it.
How about IA64?
"I really don't care?" :-)
Actually I do, but only as an example of a very different and therefore interesting architecture. I no longer expect to write code for it. :-(
Also, for a high-level language you might have to indicate to the
compiler that you are doing things like this. C has volatile,
and PL/I has the abnormal attribute, though as far as I know most
implementations don't do anything different. Changing the order
of loads and stores, within the aliasing rules, would not be
unusual for an optimizing compiler.
Exactly!
This is why I would never rely on an ordinary inline load or store having any sort of implied global visibility: Any operation which does require some sort of synchronization must be encapsulated in an explicit call to a function with the required functionality.
If an architecture supply sufficiently strong ordering guarantees that those function calls can become inline macros without any special lock/sync operation, that would be nice, but I won't depend on it.
Terje
--
- <Terje.Mathisen@xxxxxxxxxxxxx>
"almost all programming can be viewed as an exercise in caching"
.
- Follow-Ups:
- Re: ECC logging implementation
- From: Andrew Reilly
- Re: ECC logging implementation
- From: Chris M. Thomasson
- Re: ECC logging implementation
- References:
- ECC logging implementation
- From: nmm1
- Re: ECC logging implementation
- From: MitchAlsup
- Re: ECC logging implementation
- From: nmm1
- Re: ECC logging implementation
- From: MitchAlsup
- Re: ECC logging implementation
- From: nmm1
- Re: ECC logging implementation
- From: Terje Mathisen
- Re: ECC logging implementation
- From: Glen Herrmannsfeldt
- ECC logging implementation
- Prev by Date: Re: ECC logging implementation
- Next by Date: Re: ECC logging implementation
- Previous by thread: Re: ECC logging implementation
- Next by thread: Re: ECC logging implementation
- Index(es):
Relevant Pages
|