On Fri, 19 Jun 2026, Andrea Pinski wrote:

> On Fri, Jun 19, 2026 at 4:12 AM Richard Biener <[email protected]> wrote:
> >
> > The following avoids diagnosing bit sets/clears that involve RMW
> > cycles to uninitialized memory.  This is for example exposed by
> > __builtin_clear_padding lowering which suppresses diagnostics but
> > those can re-surface with vectorization.
> >
> > Bootstrapped and tested on x86_64-unknown-linux-gnu.
> >
> > OK?
> >
> > Thanks,
> > Richard.
> >
> >         PR tree-optimization/110743
> >         * tree-ssa-uninit.cc (maybe_warn_operand): Suppress diagnostics
> >         on RMW cycles setting/clearing bits.
> >
> >         * gcc.dg/uninit-pr110743.c: New testcase.
> > ---
> >  gcc/testsuite/gcc.dg/uninit-pr110743.c | 22 ++++++++++++++++++++++
> >  gcc/tree-ssa-uninit.cc                 | 24 ++++++++++++++++++++++++
> >  2 files changed, 46 insertions(+)
> >  create mode 100644 gcc/testsuite/gcc.dg/uninit-pr110743.c
> >
> > diff --git a/gcc/testsuite/gcc.dg/uninit-pr110743.c 
> > b/gcc/testsuite/gcc.dg/uninit-pr110743.c
> > new file mode 100644
> > index 00000000000..5a95f60cf9b
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.dg/uninit-pr110743.c
> > @@ -0,0 +1,22 @@
> > +/* { dg-do compile } */
> > +/* { dg-options "-O2 -Wall" } */
> > +
> > +void foo (int *);
> > +
> > +void bar()
> > +{
> > +  int x, tem;
> > +  tem = x; /* { dg-bogus "used uninitialized" } */
> > +  tem = tem | 5;
> > +  x = tem;
> > +  foo (&x);
> > +}
> > +
> > +void baz()
> > +{
> > +  int x, tem;
> > +  tem = x; /* { dg-bogus "used uninitialized" } */
> > +  tem = tem & 5;
> > +  x = tem;
> > +  foo (&x);
> > +}
> > diff --git a/gcc/tree-ssa-uninit.cc b/gcc/tree-ssa-uninit.cc
> > index dbd187e21d7..107ed21e10b 100644
> > --- a/gcc/tree-ssa-uninit.cc
> > +++ b/gcc/tree-ssa-uninit.cc
> > @@ -758,6 +758,30 @@ maybe_warn_operand (ao_ref &ref, gimple *stmt, tree 
> > lhs, tree rhs,
> >    if (is_empty_type (rhstype))
> >      return NULL_TREE;
> >
> > +  /* Avoid diagnosing read-modify-write cycles that in the end only
> > +     sets a subset of bits to zero or one.
> > +     ???  Note that further reads will then appear (partly) initialized
> > +     and will not be diagnosed.  */
> > +  gimple *use_stmt;
> > +  if (gimple_assign_load_p (stmt)
> > +      && TREE_CODE (lhs) == SSA_NAME
> > +      && single_imm_use (lhs, &luse_p, &use_stmt))
> > +    {
> > +      gassign *use_ass = dyn_cast <gassign *> (use_stmt);
> > +      for (int i = 0; i < 2; ++i)
> > +       if (use_ass
> > +           && (gimple_assign_rhs_code (use_ass) == BIT_AND_EXPR
> > +               || gimple_assign_rhs_code (use_ass) == BIT_IOR_EXPR)
> > +           && single_imm_use (gimple_assign_lhs (use_ass), &luse_p,
> > +                              &use_stmt))
> > +         use_ass = dyn_cast <gassign *> (use_stmt);
> > +      if (use_ass
> > +         && gimple_vdef (use_ass)
> > +         && operand_equal_p (gimple_assign_rhs1 (stmt),
> > +                             gimple_assign_lhs (use_ass)))
> > +       return NULL_TREE;
> > +    }
> 
> Hmm, is there a way to combine this with the BIT_INSERT_EXPR loop
> since both are basically trying to prevent the same thing? It is
> handled closer to the top of the function.

I considered that but while it is conceptually similar
(RMW) it's structured differently enough to make the result
less readable.

Richard.

> Thanks,
> Andrea
> 
> > +
> >    bool warned = false;
> >    /* We didn't find any may-defs so on all paths either
> >       reached function entry or a killing clobber.  */
> > --
> > 2.51.0
> 

-- 
Richard Biener <[email protected]>
SUSE Software Solutions Germany GmbH,
Frankenstrasse 146, 90461 Nuernberg, Germany;
GF: Jochen Jaser, Andrew McDonald, Werner Knoblich; (HRB 36809, AG Nuernberg)

Reply via email to