On Fri, Nov 30, 2018 at 9:33 AM Alexander Monakov <amona...@ispras.ru> wrote:
>
> On Fri, 30 Nov 2018, Richard Biener wrote:
> > > Ping - as I think this approach addresses the root of the problem, I 
> > > wouldn't
> > > like it to be forgotten.
> >
> > I agree this is also useful but it addresses another issue (that may appear 
> > to
> > be related).  asm inline is really a hint to the inliner estimates (with no 
> > way
> > to get semantics botched) while marking off-section parts is making the
> > asm text more precise also affecting code generation and thus has the
> > possibility to cause correctness issues (if you say mark everything as
> > off-section just to make it inline better).
>
> I don't think that's true: if the user marks too much of the template as
> off-section and makes GCC under-estimate branch ranges, they may receive an
> error from the assembler. But that's a build failure, not a correctness
> issue. Surely build error is a reasonable outcome from misuse of inline asm.

Yes, but I say they can't mark all of the asm test off-section to make it more
likely be inlined.  Because that might fire back (only on some weird targets
in weird circumstances).  But they can use asm inline for that.

> > I'm sympathtetic to both patches but clearly the kernel folks have shown
> > need for the inline hint (arguably the kernel folks are the ones we could
> > expect to get usages of %` correct ...).  I haven't seen any comments
> > from possible users of %`
>
> FWIW, when I raised the idea in the kernel thread, the response from Borislav
> was:
>
> >> I don't mind it but I see you guys are still discussing what would be
> >> the better solution here, on the gcc ML. And we, as one user, are a
> >> happy camper as long as it does what it is meant to do. But how the
> >> feature looks like syntactically is something for gcc folk to decide as
> >> they're going to support it for the foreseeable future and I'm very well
> >> aware of how important it is for a supportable feature to be designed
> >> properly.
>
> Alexander

Reply via email to