10 Mar
2011
10 Mar
'11
1:32 p.m.
On 03/10/2011 12:33 PM, SATAN66613 wrote:
apart from some 3-to-20% boost (oh, sorry for a pun) cow saves some heap at least cow _DOES NOT_ make it worse
Was this a multithreaded test where interlocked ops on refcounts can potentially cause ping-ponging of cache lines across multiple cores? I'm not saying that this is an overriding concern, only that the "no free lunch" principle suggests that there is likely to be least some possible benchmark for which even the best optimization performs worse. - Marsh