Exploiting Dual Core's with Py_NewInterpreter's separated GIL ?
cginboston at hotmail.com
Tue Nov 7 22:25:36 CET 2006
Joe Seigh wrote:
> Martin v. Löwis wrote:
>> You still didn't say what you would suggest to make it thread-safe
>> again; most likely, you proposal would be to add locking. If I
>> understand Joe's approach correctly, he has a solution that does
>> not involve locking (although I don't understand how it works).
> Sun had applied for a patent on it. You can go to the
> uspto search page here http://www.uspto.gov/patft/index.html
> and look for
> 20060218561 Code preparation technique employing lock-free pointer
> 20060037026 Lightweight reference counting using single-target
> Click on the images link on the patent application where the illustrations
> are which show the concepts probably better than the text.
> The first one above is actually a continuation patent on three different
> techniques. One using double wide compare and swap, one using ROP (Repeat
> Offender Problem), a form of PDR, and one using DCAS (compare and swap
> of two separate locations) which only exists on MC68020 and MC68030
Check out the work in the '80s from the NYU Ultra project. They did a
great deal of work on using atomic incr/decr for all sorts of algorithms
to get around locking on parallel processors.
More information about the Python-list