mixGamma should fix results with low popcount

[Lysxia]

The mixGamma function should "fix" its result if the popcount is too low, but in its current version it is doing the opposite.

pringo/stubs.c

Lines 77 to 78 in fa170bc

	if (popcount64(z ^ (z >> 1)) >= 24) z ^= 0xaaaaaaaaaaaaaaaaULL;
	return z;

This is a bug inherited from a typo in the code of the original SplitMix paper. The text gives the correct interpretation:

Therefore we require that the number of such pairs, as computed by Long.bitCount(z ^ (z >>> 1)), exceed 24; if it does not, then the candidate z is replaced by the XOR of z and 0xaaaaaaaaaaaaaaaaL

I found that in this blogpost: http://www.pcg-random.org/posts/bugs-in-splitmix.html

[xavierleroy]

Thanks a lot for reporting this error.