Groups | Search | Server Info | Login | Register

Groups > comp.arch > #5917

Re: Lock Elision going to Product

Show all headers | View raw

And Paul A. Clayton writes:
> Unfortunately, there is a lot of existing hardware that has
> enough parallelism that coarse grained locks (which would
> benefit the most from HLE) are not practical; but one has to
> start somewhere.

I suspect many people use omp critical regions rather than
allocating many locks and managing them by hand...

I know HLE will save me a large amount of memory if I switch *to*
omp critical regions.  But it brings the possibility of utterly
horrific performance if I happen to hit implementation-defined
or specification-defined (PAUSE) corner cases.  That in itself
may reduce adoption by "advanced" users, but I still suspect that
many naive users will see improved performance.
-- 
Jason

Back to comp.arch | Previous | Next — Previous in thread | Next in thread | Find similar

Thread

Lock Elision going to Product nedbrek <nedbrek@yahoo.com> - 2012-02-09 08:34 -0500
  Re: Lock Elision going to Product "Paul A. Clayton" <paaronclayton@gmail.com> - 2012-02-09 07:52 -0800
    Re: Lock Elision going to Product "Andy (Super) Glew" <andy@SPAM.comp-arch.net> - 2012-02-09 12:05 -0800
    Re: Lock Elision going to Product Jason Riedy <jason@lovesgoodfood.com> - 2012-02-13 16:09 -0500
      Re: Lock Elision going to Product "Paul A. Clayton" <paaronclayton@gmail.com> - 2012-02-13 17:24 -0800
  Re: Lock Elision going to Product Jim <jgoo052@gmail.com> - 2012-03-02 16:59 -0800

csiph-web