[Bug tree-optimization/70848] [6/7 Regression] g++ arm-none-eabi ignores volatile qualifier

["prathamesh3492 at gcc dot gnu.org" <gcc-bugzilla at gcc dot gnu dot org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70848

prathamesh3492 at gcc dot gnu.org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |prathamesh3492 at gcc dot gnu.org

--- Comment #11 from prathamesh3492 at gcc dot gnu.org ---
(In reply to ktkachov from comment #10)
> (In reply to Richard Biener from comment #8)
> > With an old cross to arm I can indeed see (.018t.ssa):
> > 
> > void test() ()
> > {
> >   volatile int * _1;
> >   volatile int * _4;
> >   volatile int * _6;
> > 
> >   <bb 2>:
> >   _1 = 327221280B;
> >   *_1 = 97;
> >   _4 = 327221280B;
> >   *_4 = 98;
> >   _6 = 327221280B;
> >   *_6 = 99;
> >   return;
> > 
> > 
> > note the missing {v} at the assignments.
> 
> I wonder why this doesn't happen on x86 or aarch64...
> Is there some target hook called that erroneously strips the 'volatility'
> from the type somewhere?

Hi,
I could reproduce the issue on x86_64 with g++ as of r235554.
Dump of 018t.ssa compiled with g++ -O2:
(also same for aarch64)

;; Function void test() (_Z4testv, funcdef_no=0, decl_uid=2253, cgraph_uid=0,
symbol_order=1)

void test() ()
{
  volatile int * p.0_2;
  volatile int * p.0_4;
  volatile int * p.0_6;

  <bb 2>:
  p.0_2 = p;
  *p.0_2 = 97;
  p.0_4 = p;
  *p.0_4 = 98;
  p.0_6 = p;
  *p.0_6 = 99;
  return;

}

and generated asm for test:
_Z4testv:
.LFB0:
        .cfi_startproc
        movq    p(%rip), %rax
        movl    $99, (%rax)
        ret
        .cfi_endproc

which I suppose is wrong since it avoids stores of 97 and 98 to *p.

With C FE, the ssa dump is as follows:

;; Function test (test, funcdef_no=0, decl_uid=1756, cgraph_uid=0,
symbol_order=1)

test ()
{
  volatile int * p.0_2;
  volatile int * p.0_4;
  volatile int * p.0_6;

  <bb 2>:
  p.0_2 = p;
  *p.0_2 ={v} 97;
  p.0_4 = p;
  *p.0_4 ={v} 98;
  p.0_6 = p;
  *p.0_6 ={v} 99;
  return;

}

generated asm with C FE:
test:
.LFB0:
        .cfi_startproc
        movq    p(%rip), %rax
        movl    $97, (%rax)
        movl    $98, (%rax)
        movl    $99, (%rax)
        ret
        .cfi_endproc

So it perhaps appears to be a C++FE issue with gimplification ?

Thanks,
Prathamesh