Re: [RFC] asm goto

[Alex Rosenberg <alexr at leftfield dot org>]

On Jul 29, 2009, at 6:33 AM, Paolo Bonzini wrote:

> > > > and it needs a means to specify the exit block/exceptions.
> > >
> > > Hmm?
> >
> > What if I want to say that the asm might throw an exception? That
> > requires an edge to the exit block.
>
> You can add a label and throw there, like:
>
> #define JO	_JO(__COUNTER__)
>
> #define_JO(N) \
> { \
>  register char *foo asm("%rax"); \
>  if (1) \
>    asm goto ("mov %0, %%rax; jo %l0" : : "a" ("overflow") : : \
>             throw##N); \
>  else throw##N: throw overflow_exception(foo); \
> }

On Jul 29, 2009, at 8:34 AM, Richard Henderson wrote:

> No it doesn't.  Exception handling isn't modeled that way.
> It does require stmt_could_throw_p return true though.

When I said "exception" I was referring to whatever the language  
models as such. As I said before "The functional-language-to-C people  
will certainly abuse this extension." They don't necessarily convert  
their exception model into GCC's model. An edge to the exit block is  
likely more than sufficient for some situations, including cases like  
noreturn.

+------------------------------------------------------------+
| Alexander M. Rosenberg        <mailto:alexr@leftfield.org> |
| Nobody cares what I say, so no disclaimer appears here.    |