RFR 8020292 <was> :Re: class SplittableRandom
dl at cs.oswego.edu
Mon Jul 15 23:51:11 UTC 2013
On 07/15/13 09:02, Aleksey Shipilev wrote:
> OK, early performance data follows.
> - 64 bit: on moderately saturated machine (1, 16 threads), SR double is
> marginally faster than TLR, and SR int/long is significantly faster than TLR
> - 32 bit: TLR is vastly faster in int/long standalone scenarios
> The conclusion is: SR is a reasonable alternative with no obvious
> performance show-stoppers. That said, TLR is also reasonably fast at
> this point, ...
Thanks. Given the apparent increasing agreement that
both APIs are needed and useful, it would be great if
people could choose among them based solely on usage context,
and not contaminated with speed/quality tradeoffs
of the underlying algorithms. To do this, it looks like
we'd end up with TLR faster than it is now on 64bit, but
slower on 32bit platforms.
Without yet committing to details (especially because I don't
have any yet!) does anyone disagree that this is an OK tradeoff?
More information about the core-libs-dev