[cpp-threads] A question about N2153

[Chris Thomasson]

> On Wed, Jan 17, 2007 at 11:56:29AM -0800, Chris Thomasson wrote:
>> ----- Original Message -----
>> From: "Paul E. McKenney" <paulmck at linux.vnet.ibm.com>
>> To: "C++ threads standardisation" <cpp-threads at decadentplace.org.uk>
>> Sent: Wednesday, January 17, 2007 11:20 AM
>> Subject: Re: [cpp-threads] A question about N2153
>>
>>
>> >On Wed, Jan 17, 2007 at 08:44:00PM +0200, Peter Dimov wrote:
>> >>What is the reference implementation of acquire_fence for x86, SPARC 
>> >>RMO,
>> >>PowerPC, IA-64? I'm guessing (no op), #LoadLoad | #LoadStore, lwsync, 
>> >>mf.
>> >
>> >Stupid side question...  Why acquire_fence and release_fence?  This is
>> >the only group I have come across that uses these.  Where I come from,
>> >we use the following types:
>>
>> Well, an acquire_"fence" is like this:
>>
>> // do your atomic operation here
>> #StoreLoad | #StoreStore
>>
>> And a release_"fence" is like this:
>>
>> #LoadStore | #StoreStore
>> // do your atomic operation here
>
> OK.  But I already understood what you guys meant by these two primitives.
> My question is why choose these particular definitions?  From my
> viewpoint, the definitions are quite strange, and they don't do what I
> want in many cases.  I often need a StoreStore without the StoreLoad,

Yeah... Same here. The acquire/release thin is closely related to the 
barriers that you need wrt implementing a POSIX mutex.


> and
> I often need to enforce data dependencies.  This latter especially is not
> supported well by the cpp-threads definitions, as most CPUs don't need any
> barriers in that case (just disable compiler code-motion optimizations).
>
> Thoughts?

Something like this might be better:

load_acquire == #LoadStore | #LoadLoad
load_depends == *#LoadDepends | #LoadLoad
load_ordered == #LoadStore
load_relaxed == #LoadLoad
store_acquire == #StoreLoad | #StoreStore
store_release == #LoadStore | #StoreStore
store_ordered == #StoreLoad
store_relaxed == #StoreStore
full_prolog == store_release
full_epilog == store_ordered


*- I would like to see this barrier added to the SPARC membar instruction 
operands..



Na! Perhaps something like this:

Well, the following language
additions would be nice:

Okay, a normal load from a shared variable in C:

static void *x = 0;

int main(void) {
  void *l = x; /* naked load */
  return 0;
}

Well, how about adding support for the following logic:

static void *x = 0;

int main(void) {
  void *l:(#LoadStore | #LoadLoad) = x; /* acquire load */

  l:(#LoadDepends | #LoadLoad) = x; /* dependant load */

  /* here is what a store could look like: */
  x:(#LoadStore | #StoreStore) = l; /* store release */
  return 0;
}


Humm...