Compiling the following code with "gcc-11 -fanalyzer -c"

void f (unsigned long *p, int r, int i)
{
  int b = 64, n = r % 64;

  while (i >= 0 && b >= 0)
    {
      if (b <= n)
        p[i--] = 1UL << b;
      b -= n;
    }
}

gives:

warn-shiftcount.c: In function ‘f’:
warn-shiftcount.c:8:22: warning: shift by count (‘64’) >= precision of type (‘6’) [-Wanalyzer-shift-count-overflow]
    8 |         p[i--] = 1UL << b;
      |                  ~~~~^~~~
  ‘f’: events 1-5
    |
    |    5 |   while (i >= 0 && b >= 0)
    |      |          ~~~~~~~^~~~~~~~~
    |      |                 |
    |      |                 (1) following ‘true’ branch...
    |    6 |     {
    |    7 |       if (b <= n)
    |      |          ~
    |      |          |
    |      |          (2) ...to here
    |      |          (3) following ‘true’ branch (when ‘b <= n’)...
    |    8 |         p[i--] = 1UL << b;
    |      |           ~~~    ~~~~~~~~
    |      |            |         |
    |      |            |         (5) shift by count ‘64’ here
    |      |            (4) ...to here

This warning is incorrect, since 1UL << b is in the "if" branch,
where b <= n and n <= 63 (thus b <= 63).

No such issue with gcc-10 and before.

This makes testing of GNU MPFR fail (this testcase was derived from
GNU MPFR's random2.c).