Hi John,
1) Did you introduce long double because your
algorithms need a higher accuracy ?
If so,
a. That is extremely worrisome, have a look at
"Accuracy and Stability of Numerical
Algorithms"
by Nicholas Higham
b. Actually do not , there is money to be made
from your algorithms :)
2) Are you seeing differences between Windows and
linux using TMatrix . That should not be the case.
test/stressLinear performs some numerically
intensive/difficult tasks that run ok on both
platforms .
Could you provide an example ?
Eddy
--- "Dr. John Krane" <jkrane@netzero.com> wrote:
> Hi Axel,
>
> > And as Philippe has pointed out, the most probable
> cause is that you
> > don't link against the dictionary objects. The
> exports are only
> > relevant if you create a shared lib containing
> your objects, and you
> > want to link against that shared lib. Looking at
> the output you've
> > attached to your emails this doesn't seem to be
> the case.
>
> I think I am lost in the terminology.
>
> I have a set of analysis code that uses my personal
> TMatrixQ class,
> which is like TMatrixD only with long double instead
> of Double_t
> elements. I need to implement on Windows now - the
> financial world uses
> Windows. I stripped out my personal class and used
> TMatrixD when I moved
> stuff to Windows because I wanted to avoid build
> trouble in .NET... My
> .dll file, which is a plug-in for Excel, worked
> great, but gave me
> different answers from the Linux analysis with the
> same inputs and same
> code (with the TMatrixD substitution). This code I
> describe so far has a
> .def file that "exports" the one function call to
> Excel (or to my text
> executable, or any other .exe that would use my
> .dll), but this might
> not be the "export" you are talking about.
>
> Anyway, I must put TMatrixQ back in. I made a
> separate area called
> root_extend, just like I have in my Linux libraries,
> with separate build
> rules in .NET to reflect what Francois-Xavier's page
> says. TMatrixQ
> depends on ROOT more heavily than the rest of the
> to-be-implemented code
> because TMatrixQ (and TArrayQ, TMatrixQUtils,
> TVectorQ) inherits from
> TObject and needs all that Cint stuff.
>
> So you think I can address my TMatrixQ problems with
> dictionary work
> only, and without making an "export library"? Or do
> you now agree with
> Valeri?
>
> - John
>
> --
>
> Dr. John Krane
> jkrane@netzero.com
>
>
This archive was generated by hypermail 2b29 : Sun Jan 02 2005 - 05:50:07 MET