On Sat, 2014-12-27 at 13:53 +0000, via Digitalmars-d-learn wrote: […] > > I wonder how TSX would work with GIL. I suppose most GIL locks > are short lived enough to be covered by TSX before it fails and > takes a lock.
For Intel chips this is good stuff (stolen from Sun's Rock processor). Hardware supported transactional memory easily beats software transactional memory, but the latter is portable. […] > > To be fair, you also have to compete against commercial solutions > such as SPSS, SAS and others. It is relatively easy to compete against these generally. Small organizations (which actually make up the bulk of users) prefer not to pay the extortionate fees. Anecdotal evidence clearly show a mass move from Matlab to Python+NumPy+… – the anecdotes being my Python Workshops last year where 40%+ of people were in this position. > Then you have OpenMP for C++ and Fortran, which it will be > difficult for D to compete with in terms of performance vs effort. If you said MPI, then yes, it is the de facto standard native code clustering system: on JVM there is Netty and a few other systems. OpenMP is really just a way of hacking sequential code to create parallel code on a multicore single address space; and a very good hack it is too. But it remains a hack and not a good way of transitioning from fundamentally sequential code to fundamentally parallel code. OpenMP exists exactly because Fortran, C and C++ codes had to be made data parallel without being rewritten. D should not be in this boat. -- Russel. ============================================================================= Dr Russel Winder t: +44 20 7585 2200 voip: sip:[email protected] 41 Buckmaster Road m: +44 7770 465 077 xmpp: [email protected] London SW11 1EN, UK w: www.russel.org.uk skype: russel_winder
signature.asc
Description: This is a digitally signed message part
