Tim Prince wrote:
>
> They are often even bug-for-bug consistent (e.g. with the 0-sized array bug
> in DGEMV). Results of vendor versions are likely to be more accurate.
> There's no way they can be expected to produce identical results.
>
sure... I meant "consistent apart from numerical uncertainties due to
machine precision, floating point optimisations, etc"
> What is this about "without a vendor version?" Do you not consider the
> versions supplied by compiler vendors?
If precompiled, yes. More generally a vendor version is for me anything
beyond the reference implementation sources.
--
-----------------------------------------------------------------------
Pierre Hugonnet R&D Data Processing
COMPAGNIE GENERALE DE GEOPHYSIQUE - Paris Processing Centre
1, rue Leon Migaux / 91341 MASSY cedex / FRANCE
phone: (33) 164 47 45 59 fax:(33) 164 47 32 49
email: phugonnet(at)cgg(dot)com
-----------------------------------------------------------------------
The opinions expressed in this message are not necessarily
those of CGG or of Jacques Chirac
-----------------------------------------------------------------------
|