Thanks to all who have responded so far.
The articles of Charles seem to contain the
only "hard" data so far, but maybe there is
more out there. Impressions are interesting
also, of course.
I think what we are trying to do is determine
if there the language itself prevents F90/95
compilers from generating efficient code.
Obviously, this is a hard thing to determine,
but we are making some progress! It depends
on the skill of the vendor, how the programmer
writes code, etc.
[log in to unmask] wrote:
>
> Walt,
>
> Two things come to mind: are we talking efficiency of compilers or efficiency of
> programmers?
>
> Another thought, are we comparing the efficiency of some vendors versus others?
>
> I feel that the vectorised code and the new intrinsics give scope for vendors to
> "make hay while the sun shines".
>
> An intrinsic like MATMUL should save us many lines of code (maintenance) and allow
> the vendor to really optimise this function. A level like LAPACK/ATLAS. Also the
> new intrinsics like DOT_PRODUCT and TRANSPOSE should be able to be heavily
> optimised by the vendor.
>
> Several of the other features, starting from IMPLICIT NONE are really geared to
> increase the programmer's efficiency. Many of the features of this ilk I do not
> find increase my efficiency. But then I am "in disagreement" with several of my
> betters here and on comp.lang.fortran.
>
> In summary, I believe that the vectoring features have potential to increase the
> runtime efficiency of F90 code. Depends on the ability of the vendor. I believe
> mine to be good -- F95 on an Alpha VMS system.
>
> Some other features which, by me, could be termed "baggage" should only affect
> compile time but not runtime.
>
> I believe that the F95 standard adds baggage to compilers not to runtime. I
> suspect that the F2K standard will add even more, but again hopefully not to
> runtime for those using Fortran for what it was originally intended for.
>
> For all my bias, I hope that some comments are useful.
>
> Regards, Paddy
>
> Paddy O'Brien,
> Transmission Development,
> TransGrid,
> PO Box A1000, Sydney South,
> NSW 2000, Australia
> (Street address, 201 Elizabeth Street)
>
> Tel: +61 2 9284-3063
> Fax: +61 2 9284-3050
> Email: [log in to unmask]
>
> Either "\'" or "\s" (to escape the apostrophe) seems to work for most people,
> but that little whizz-bang apostrophe gives me little spam.
--
Walt Brainerd [log in to unmask]
The Fortran Company +1-520-256-1455
11155 E. Mountain Gate +1-520-760-1397 (fax)
Tucson, AZ 85749 USA http://www.fortran.com
|