Re: [eigen] Parallelizable operations

[ Thread Index | Date Index | More Archives ]

Hi Joël ;)

It may also depend on the processor you are addressing, and this is
difficult without outside support (i.e. defining macros to specific
values retrieved or infered from the processor description).

Interesting link, I'll to check if I can have access to it.


2010/7/29 joel falcou <joel.falcou@xxxxxx>:
> On 29/07/10 14:22, Matthieu Brucher wrote:
>> Hi,
>> I don't think it would be a huge gain, if there is a gain at all. If
>> we consider single processor multiple cores, the memory bandwidth is
>> shared accross cores. Addition and substraction are meory bandwidth
>> limited, so there would only be additional contention and thus less
>> performance.Parallelization works best with far more computations than
>> memory accesses.
>> Matthieu
> Well, there is a simple model that can be evaluated at compile-time w/r to
> operations cycles count knwown before hand
> to decide if parallelization is useful or not.
> <shamelessplug>
> </shamelessplug>
> in NT2 it works OK but then again, we don't have the same compiletime
> constraints

Information System Engineer, Ph.D.

Mail converted by MHonArc 2.6.19+