This is the mail archive of the gcc-bugs@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

[Bug tree-optimization/80492] [5/6/7 Regression] Wrong code when unrolling a loop with inline asm and local regs


https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80492

Uroš Bizjak <ubizjak at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2017-04-24
                 CC|                            |rguenth at gcc dot gnu.org
          Component|other                       |tree-optimization
     Ever confirmed|0                           |1

--- Comment #7 from Uroš Bizjak <ubizjak at gmail dot com> ---
(In reply to Georg-Johann Lay from comment #6)
> > > The 2 last inline asms are missing the set of ESI.
> > 
> > You should mark that "Some Code" asm clobbers %esi.
> 
> Why? The code consumes ESI and is not set for that purpose.
> 
> It's a condensed test case for a different target (avr) and demonstrates
> that it also occurs on x86.
> 
> In general, we are striving for simpler test cases, nor more complicated
> ones, no?

Ah, I see the problem. The initialization value changes in each iteration.

Looks like a tree-optimization problem, _.optimized dump shows:

do_syscalls ()
{
  register int reg __asm__ (*4);
  register int reg __asm__ (*4);

  <bb 2> [33.33%]:
  reg = 0;
  __asm__ __volatile__("/* Some Code %0 */" :  : "r" reg);
  reg = 1;
  __asm__ __volatile__("/* Some Code %0 */" :  : "r" reg);
  __asm__ __volatile__("/* Some Code %0 */" :  : "r" reg);
  __asm__ __volatile__("/* Some Code %0 */" :  : "r" reg);
  return;

}

Confirmed, CC added.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]