Subject: Re: [ublas] Overhead of matrix indexing
From: Andrea Sansottera (andrea.sansottera_at_[hidden])
Date: 2012-01-20 08:30:40
I am afraid you will get similar results with VS2010. You might want to try
with the Intel compiler for Windows.
Visual Studio 2010, Core i7 640M:
- 32 bit: 1.50114 s - 0.495624 s
- 64 bit: 1.04218 s - 0.496605 s
Il giorno 19 gennaio 2012 17:28, dariomt <dariomt_at_[hidden]> ha scritto:
> Malcolm Reynolds <malcolm.reynolds <at> gmail.com> writes:
> > On Wed, Jan 18, 2012 at 1:21 PM, Oswin Krause
> > <Oswin.Krause <at> ruhr-uni-bochum.de> wrote:
> > > Hi,
> > >
> > > I can't measure any time differences. both take 0.56s.
> > > GCC 4.6.2 used here with "-O3" as optimisation option.
> > >
> > I think sufficiently smart compilers can convert the code to use a
> > single index. GCC 4.2 (still the default compiler for OS X 10.6) is
> > unable to optimise this properly even with -O3, so I also see roughly
> > a factor of three difference in runtime.. However using LLVM-clang the
> > pointer and indexed versions perform the same.
> It's good to hear that a good optimizer is able to do the right thing.
> I'll try other flags in VS2008 or maybe upgrading to VS2010, but I'm tied
> windows and VS. I might try to send this case to VS support.
> Many thanks!
> ublas mailing list
> Sent to: andrea.sansottera_at_[hidden]