[Bug tree-optimization/123140] New: extracting element from a vec_duplicate SVE type should be a no-op.

[tnfchris at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=123140

            Bug ID: 123140
           Summary: extracting element from a vec_duplicate SVE type
                    should be a no-op.
           Product: gcc
           Version: 16.0
            Status: UNCONFIRMED
          Keywords: missed-optimization
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: tnfchris at gcc dot gnu.org
  Target Milestone: ---
            Target: aarch64*

The following example

#include <stdint.h>
#include <arm_sve.h>

uint32_t g;

svuint32_t foo(uint32_t *p) {
  svuint32_t v = svdup_u32(*p);
  g = v[0];
  return v;
}

at -O3 -march=armv9-a generates

foo(unsigned int*):
        adrp    x1, .LANCHOR0
        add     x1, x1, :lo12:.LANCHOR0
        ptrue   p3.b, all
        ld1rw   z0.s, p3/z, [x0]
        st1     {v0.s}[0], [x1]
        ret

which is quite silly and it's due to in GIMPLE we don't see that

  <bb 2> [local count: 1073741824]:
  _1 = *p_5(D);
  _2 = [vec_duplicate_expr] _1;
  _3 = BIT_FIELD_REF <_2, 32, 0>;
  g = _3;
  return _2;

the BIT_FIELD_REF of the vec_duplicate_expr is a no-op.
This is likely due to the VLA type.