Re: test-bitrotate.c missing test cases

[Bruno Haible <bruno@clisp.org>]

Jeffrey,

> Forgive my ignorance... No'oping 0 leaks timing information

There are only few algorithms where leaking timing information is an
issue. For most of the code we deal with, the developer wants to get
optimal performance.

> I also don't think developers are going to write a rotate like:
> 
>     if (n != 0)
>         x = rotr32(x, n);

Sure they will. Here's an example from lib/vasnprintf.c, where a shift
count of 0 is treated specially:


      /* Copy a, shifting it left by s bits, yields r.
         Memory layout:
         At the beginning: r = roomptr[0..a_len],
         at the end: r = roomptr[0..b_len-1], q = roomptr[b_len..a_len]  */
      r_ptr = roomptr;
      if (s == 0)
        {
          memcpy (r_ptr, a_ptr, a_len * sizeof (mp_limb_t));
          r_ptr[a_len] = 0;
        }
      else
        {
          const mp_limb_t *sourceptr = a_ptr;
          mp_limb_t *destptr = r_ptr;
          mp_twolimb_t accu = 0;
          size_t count;
          for (count = a_len; count > 0; count--)
            {
              accu += (mp_twolimb_t) *sourceptr++ << s;
              *destptr++ = (mp_limb_t) accu;
              accu = accu >> GMP_LIMB_BITS;
            }
          *destptr++ = (mp_limb_t) accu;
        }


Bruno