Re:STOKE – a stochastic optimizer for x86_64 assembly
by barsonme on Tuesday January 19, 2016 @05:12AM
(#10929706)
> I thought was a little underwhelming.
I agree, but...
> so it added some overhead to special case 0 even though 0 only takes one iteration through the loop anyway, and the loop still required up to 64 iterations if you have the high bit set.
Didn't it simply replace the body of `size_t popcnt(uint64_t x)` with the `popcnt` instruction? Or am I looking at the wrong test?
I agree, but...
> so it added some overhead to special case 0 even though 0 only takes one iteration through the loop anyway, and the loop still required up to 64 iterations if you have the high bit set.
Didn't it simply replace the body of `size_t popcnt(uint64_t x)` with the `popcnt` instruction? Or am I looking at the wrong test?