[PATCH] Fix ICE in match_asm_constraints_1 (PR inline-asm/84941)

[Michael Matz]

Hi,

On Tue, 20 Mar 2018, Jakub Jelinek wrote:

> It is very common that input is one of the above cases, during x86_64-linux
> and i686-linux bootstraps+regtests I got:
> 13201x CONST_INT, 1959x MEM, 114x SUBREG, 110x SYMBOL_REF,
> 2x PLUS (the new testcase only)
> and most of those were actually from input-output constraints, like:
>   var = 1;
>   asm ("" : "+g" (var));
>   var2 = &static_var3;
>   asm ("" : "+g" (var2));
> etc.  I believe the mini-pass does a useful transformation for these that
> ought to make it easier for reload to actually handle the matching constraints.

Well, then, so your handling of them makes somewhat sense.

> > -      if (end == constraint)
> > +      if (end == constraint || *end)
> >         continue;
> 
> That wouldn't handle e.g.
>   asm volatile ("" : "=m,m" (b), "=r,r" (b) : "1,1" (b));
> case.

No, it wouldn't, which was my intention.  I'm fine with either way, but 
wouldn't have bothered with these cases.  Human-written asms don't use 
alternatives very often (the asm template can't easily make use of them), 
nor do they use funny matching-or-something-else constraints.  But if you 
want to handle them: more power to you :)


Ciao,
Michael.