Re: [boost] Re: Matrix

From: David Abrahams (abrahams_at_[hidden])
Date: 2001-03-13 21:58:35

Is the idea of building in the ability to dispatch to BLAS based on any real
measurements, or just the assumption that BLAS will be faster than anything
we can reasonably do in C++?
Jeremy's paper on MTL seems to indicate that involving BLAS is unneccessary.
Overgeneralizing is one sure way to kill a software project.

-Dave

----- Original Message -----
From: "Dave Steffen" <steffend_at_[hidden]>
To: <boost_at_[hidden]>
Sent: Tuesday, March 13, 2001 8:41 PM
Subject: Re: [boost] Re: Matrix

> Toon Knapen writes:
>  > lums_at_[hidden] wrote:
> [...]
>  > > Now, vendor-tuned libraries can give you good performance (which is
>  > > where BLAS get their reputation).  The problem with this is that a)
>  > > the performance is not portable, b) you have to pay for them, c) they
>  > > are highly optimized for only the subset of BLAS necessary to market
>  > > their machines well, and d) the BLAS do not really cover everything
>  > > that you would want to do in an efficient way.
>  >
>  > I'm also definitly in favor of an C++ only approach. Nevertheless I
>  > think using in some cases BLAS internally would be necessary as a
>  > migration path.
>
>  Maybe we should (try to) design our hypothetical library ("Boost
>  BLASt"? ;-) so that it's reasonably easy to "plug in" calls to an
>  available BLAS library if it's available.  I.E. have
>
> template <class Matrix, class Vector>
> Matrix operator* (const Matrix& M, const Vector& v)
> {
>   // ... implement M * v
> }
>
>  ... and then, if BLAS happens to be available (or desirable), somehow
>  we also get a specialization for something that BLAS can handle:
>
> template <>
> Matrix<double> operator* <Matrix<double>, Vector<double> >
> (const Matrix<double>& M, const Vector<double>& v)
> {
> // ... calls DGEMV with appropriate stuff
> }
>
>
>  Now, I'm not claiming that I know how to build the machinery to do
>  this; but it'd be a _really_ nice thing to have. ;-)
>
>  I'm also thinking that this would make it _really_ easy to benchmark
>  straight C++ solutions against local (tuned) BLAS routines.
>
> --------------------------------------------------------------------------
> Dave Steffen                      Wave after wave will flow with the tide
> Dept. of Physics                    And bury the world as it does
> Colorado State University         Tide after tide will flow and recede
> steffend_at_[hidden]        Leaving life to go on as it was...
> - Peart / RUSH
> "The reason that our people suffer in this way....
> is that our ancestors failed to rule wisely".   -General Choi, Hong Hi
>
>
>
>
> List-Unsubscribe: <mailto:boost-unsubscribe_at_[hidden]>
>
>
> Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/
>
>