RFR (XS): CR 8006176: Switch to optimal identity hash code generator

Aleksey Shipilev aleksey.shipilev at oracle.com
Mon Jan 14 06:50:31 PST 2013

On 01/14/2013 06:41 PM, Doug Lea wrote:
> My take is that the only possible concern is statistical
> independence of the thread-local versions? But this is pretty
> easy to address if it hasn't already been done.

Seems already done, see thread.cpp, init of _hashStateX [1];
thread-local PRNGs seem to be pre-seeded with HotSpot's internal linear
PRNG. That should be one of the reasons why my simple spectral tests do
not detect any particular patterns in mode-5 sequences (they do on
address-based generators).



More information about the hotspot-runtime-dev mailing list