Re: [PATCH] match.pd: popcount(X & -X) -> (X & -X) != 0 [PR102486]

[Andrew Pinski]

On Tue, Dec 30, 2025 at 8:43 AM Daniel Henrique Barboza
<daniel.barb...@oss.qualcomm.com> wrote:
>
>
>
> On 12/29/25 7:22 PM, Andrew Pinski wrote:
> > On Mon, Dec 29, 2025 at 12:14 PM Daniel Barboza
> > <daniel.barb...@oss.qualcomm.com> wrote:
> >>
> >>          PR 102486
> >>
> >> gcc/ChangeLog:
> >>
> >>           * match.pd (`popcount (X & -X) -> (X & -X) != 0`): New pattern.
> >>
> >> gcc/testsuite/ChangeLog:
> >>
> >>          * gcc.dg/torture/pr102486.c: New test.
> >>
> >> Signed-off-by: Daniel Barboza <daniel.barb...@oss.qualcomm.com>
> >> ---
> >>   gcc/match.pd                            |  8 ++++++++
> >>   gcc/testsuite/gcc.dg/torture/pr102486.c | 15 +++++++++++++++
> >>   2 files changed, 23 insertions(+)
> >>   create mode 100644 gcc/testsuite/gcc.dg/torture/pr102486.c
> >>
> >> diff --git a/gcc/match.pd b/gcc/match.pd
> >> index 45d96844673..6b01aaea9b6 100644
> >> --- a/gcc/match.pd
> >> +++ b/gcc/match.pd
> >> @@ -10541,6 +10541,14 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
> >>                (popcount:s @1))
> >>         (popcount (log2 @0 @1)))))
> >>
> >> +/* popcount (X & -X) is (X & -X) != 0.  */
> >> +(simplify
> >> +  (POPCOUNT (convert?@1 (bit_and:c @0 (negate @0))))
> >> +  /* Use an explicit bool conversion to avoid errors when the
> >> +     POPCOUNT result is assigned to a type (e.g. int) that
> >> +     will generate a "bogus comparison result type" error.  */
> >> +  (convert:boolean_type_node (ne @1 { build_zero_cst (TREE_TYPE (@1)); 
> >> })))
> >
> > I think this needs a `convert` at the new pattern.
>
> Do you mean this?
>
>    (convert? (POPCOUNT (convert?@1 (bit_and:c @0 (negate @0)))))
>
> Because if I do that I won't be able to match one of the tests I'm
> adding in the testsuite because there's not explicit convert/cast in
> the popcount call to match.
>
> If I do this:
>
>    (convert (POPCOUNT (convert?@1 (bit_and:c @0 (negate @0)))))
>
>
> Then I got no matches in the testcase. Thanks,


No I meant:
```
(simplify
  (POPCOUNT (convert?@1 (bit_and:c @0 (negate @0))))
  /* Use an explicit bool conversion to avoid errors when the
     POPCOUNT result is assigned to a type (e.g. int) that
     will generate a "bogus comparison result type" error.  */
 (convert:type (convert:boolean_type_node (ne @1 { build_zero_cst
(TREE_TYPE (@1)); }))))
```

Thanks,
Andrew Pinski

>
> Daniel
>
> > Also I think this is wrong for vector types so it needs a check that
> > type is an INTEGRAL_TYPE_P.
> >
> >> +
> >>   #if GIMPLE
> >>   /* popcount(zext(X)) == popcount(X).  */
> >>   (simplify
> >> diff --git a/gcc/testsuite/gcc.dg/torture/pr102486.c 
> >> b/gcc/testsuite/gcc.dg/torture/pr102486.c
> >> new file mode 100644
> >> index 00000000000..d900149e011
> >> --- /dev/null
> >> +++ b/gcc/testsuite/gcc.dg/torture/pr102486.c
> >> @@ -0,0 +1,15 @@
> >> +/* { dg-do compile } */
> >> +/* { dg-options "-fdump-tree-optimized" } */
> >> +/* { dg-skip-if "" { *-*-* } { "-O0" "-fno-fat-lto-objects" } { "" } } */
> >
> > No reason to run this under torture. Just put it as
> > `gcc.dg/tree-ssa/pr102486.c` and `-O1` to the dg-options.
> >
> > Thanks,
> > Andrew Pinski
> >
> >> +
> >> +int f (unsigned y)
> >> +{
> >> +  return __builtin_popcount (y & -y);
> >> +}
> >> +
> >> +int f2 (int y)
> >> +{
> >> +  return __builtin_popcount (y & -y);
> >> +}
> >> +
> >> +/* { dg-final { scan-tree-dump-times "popcount" 0 "optimized" } } */
> >> --
> >> 2.43.0
> >>
>