On 12/12/2014 09:31 AM, Bastian Koppelmann wrote:
> +uint32_t helper_parity(target_ulong r1)
> +{
> + uint32_t ret;
> + uint32_t nOnes, i;
> +
> + ret = 0;
> + nOnes = 0;
> + for (i = 0; i < 8; i++) {
> + ret ^= (r1 & 1);
> + r1 = r1 >> 1;
> + }
> + /* second byte */
> + nOnes = 0;
> + for (i = 0; i < 8; i++) {
> + nOnes ^= (r1 & 1);
> + r1 = r1 >> 1;
> + }
> + ret |= nOnes << 8;
> + /* third byte */
> + nOnes = 0;
> + for (i = 0; i < 8; i++) {
> + nOnes ^= (r1 & 1);
> + r1 = r1 >> 1;
> + }
> + ret |= nOnes << 16;
> + /* fourth byte */
> + nOnes = 0;
> + for (i = 0; i < 8; i++) {
> + nOnes ^= (r1 & 1);
> + r1 = r1 >> 1;
> + }
> + ret |= nOnes << 24;
> +
> + return ret;
> +}
> +
Probably doesn't matter much, but
ret = (ctpop8(r1) & 1)
| ((ctpop8(r1 >> 8) & 1) << 8)
| ((ctpop8(r1 >> 16) & 1) << 16)
| ((ctpop8(r1 >> 24) & 1) << 24);
One could also make a case for adding new helpers that
use __builtin_parity rather than __builtin_popcount.
I usually like to look at things like this and see how
the general infrastructure can be improved...
Otherwise,
Reviewed-by: Richard Henderson <[email protected]>
r~
