Ralf Wildenhues <[EMAIL PROTECTED]> wrote:
> Hi Jim,
>
> sorry for not looking earier.
Hi Ralf,
Thanks for looking.
> * Jim Meyering wrote on Sat, Nov 29, 2008 at 12:00:34PM CET:
>> +/* Return the number of bits by which a d_type value must be shifted
>> + left in order to put it into the S_IFMT bits of stat.st_mode. */
>> +static int
>> +s_ifmt_shift_bits (void)
>> +{
>> + unsigned int v = S_IFMT; /* usually, 0170000 */
>> + static const int MultiplyDeBruijnBitPosition[32] =
>> + {
>> + 0, 1, 28, 2, 29, 14, 24, 3, 30, 22, 20, 15, 25, 17, 4, 8,
>> + 31, 27, 13, 23, 21, 19, 16, 7, 26, 12, 18, 6, 11, 5, 10, 9
>> + };
>> +
>> + /* Find the smallest power of two, P (e.g., 0010000) such that P & V ==
>> P. */
>> + unsigned int p = v ^ (v & (v - 1));
>> +
>> + /* Compute and return r = log2 (p), using code from
>> + http://graphics.stanford.edu/~seander/bithacks.html#IntegerLogDeBruijn
>> */
>> + return MultiplyDeBruijnBitPosition[(uint32_t) (p * 0x077CB531UL) >> 27];
>> +}
>
> This is a constant you could compute at configure time, no?
> See AC_COMPUTE_INT.
Sure, but why bother? Since any decent compiler will optimize that
function to "return 12;", I prefer to leave the function right next to
the code that uses it.
Of course, if there is ever a second place where we need to map d_type
to the type in stat.st_mode, I'll factor it into a new module.
