gomp-discuss
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Gomp-discuss] fortran parser :-) ..


From: Lars Segerlund
Subject: Re: [Gomp-discuss] fortran parser :-) ..
Date: Tue, 22 Mar 2005 15:24:24 +0100

 Replying to my self ...

  I have tested a small and stupid 'fix' for gfortrans backtracking, ( basicly 
I remember
 if we already have parsed a comment on this line :-) ... this is likely to 
fail for
 a lot of cases but for now it will do until the real fix ), this gives us some 
kind
 of starting point for inserting the fortran open mp parser there.

  Now what are we going to do :

        1. adapt out C parser ?

        2. write a new parser ?

        3. I have no idéa ...

  As I said earlier, perhaps we can add some kind of instrumentation to the 
parser
 we got, even split it into a front and backend ?

 / regards, Lars Segerlund.


On Tue, 22 Mar 2005 14:00:41 +0100
Lars Segerlund <address@hidden> wrote:

> 
>  Hi all,
> 
>   I have found the problem in the gcc 4.0 fortran parser :-) .... it does 
> some really
>  nasty backtracking in fixed form and f77 compatible form.
>   Now that the problem is identified I have come up with a bugfix, ( ugly 
> hack really ),
>  which should make us able to parse the OMP directives :-), mainly we 
> remember if we
>  have been on this line before.
> 
>   I am going to make a small proof of concept, and post the code to the list, 
> but in the
>  long run the fortran parser will have to be fixed and I will try to work i 
> on this.
> 
>   Now an ugly hack might sound very bad, but it's ok for us, if we 'hijack' 
> the
>  scanner/parser until were done, this way the only thing that goes away with a
>  working scanner/parser is our ugly hack, the code doing gomp stuff stays the 
> same.
> 
>   I hope this sounds reasonable to everyone.
> 
>  Did anybody have a good look at the c/c++ parser and figure out if we have 
> to make
>  a new one or if we can use it for fortran also ?
>  If the statements are exactly the same we might use different backends, ( 
> ie. something
>  table driven ?? ).
> 
>   This is exiting, I really feel like omitting something nasty for the 
> gimplifier :-) .
> 
>  / regards, Lars Segerlund.
> 
> 
> _______________________________________________
> Gomp-discuss mailing list
> address@hidden
> http://lists.nongnu.org/mailman/listinfo/gomp-discuss




reply via email to

[Prev in Thread] Current Thread [Next in Thread]