[Bug target/126450] [16/17 Regression] wrong code at -O2 -mavx2 since r16-8339

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126450

--- Comment #4 from GCC Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-16 branch has been updated by H.J. Lu <[email protected]>:

https://gcc.gnu.org/g:ad1212765889961594a678912c21c55d1e447679

commit r16-9435-gad1212765889961594a678912c21c55d1e447679
Author: H.J. Lu <[email protected]>
Date:   Wed Jul 29 06:23:57 2026 +0800

    x86: Check argument-linked memory used for local variable

    Since in

    typedef int V [[gnu::vector_size (16)]];
    long long a, b, c, d, e;
    short f;
    [[gnu::vector_size (8 * sizeof (int))]] int g;
    V h;
    _Bool i;

    void
    foo (V x)
    {
      _Bool j = 0;
      short l = 0;
    l1:
      b = l;
      x = h;
      ...
    }

    the register argument, x, is used as local variable, also return true if
    spilling an SSA_NAME into an argument-linked memory.

    gcc/

            PR target/126450
            * config/i386/i386.cc (ix86_spill_register_argument_p): Check
            the argument-linked memory used to store local variable.

    gcc/testsuite/

            PR target/126450
            * gcc.target/i386/pr126450-1.c: New test.
            * gcc.target/i386/pr126450-2.c: Likewise.

    Signed-off-by: H.J. Lu <[email protected]>
    (cherry picked from commit ea27cb48a346633d6bfabcde75f88affb9463e40)