Re: [Gomp-discuss] CIL representation ...

[Lars Segerlund]

 Since you seem to be the one with the most experience in parallel 
compiler ( practical ), I thought it would be apropriate to ask your 
opinion on this matter.

 On the list we decided to try to make the internal representation 
independent of the memory or parallel model ( or atleast capable of 
representing different ones ), ( a cople of days ago ).

 And with this in mind, do you think that a representation such as CIL 
is able to cover these cases ? My objection is how to construct the 
necessary graphs since the control parallism is available through 
mechanisms ( locks, mutexes, restrictions ) instead of symbolic form ( 
graph ).

 /Lars

Diego Novillo wrote:
> On Wed, 2003-03-12 at 05:14, Lars Segerlund wrote:
>
> >  I read the CIL doc's, and this sums it up.
> >  CIL doesn't have everything needed to represent a parallell program at 
> > the abstract level, it got locks and mutexes and non of the dependencies 
> > needed for the variables. It's not a graph, it's a bytecode, so I think 
> > we can drop it as a format interesting for gomp.
>
> Wait.  Data dependencies are not an IL feature.  That's something that
> the compiler deals with separately.  The questions you need to answer
> are:
>
> (1) Can we represent parallelism?  If you can represent control
> parallelism (think fork/join) and data parallelism (think parallel
> loops), then that's enough to get started.
>
> (2) Can we represent synchronization?  Here you don't really need
> language constructs.  If you have library calls for mutex locking,
> barriers, events and/or semaphores then you're set.  Language features
> may help (e.g., monitors).
>
>
> Diego.
>
>
>
> _______________________________________________
> Gomp-discuss mailing list
> address@hidden
> http://mail.nongnu.org/mailman/listinfo/gomp-discuss