Hi Paule,
My examples I referred to were both quick ports from very antiquated
code. The computationally intensive case simply needed some
large 2-dim matrices of real and I could create them up front.
In Fortran (RM) I squeezed some more performance by holding arrays of
starting addresses per row but now I just use normal 2d arrays.
Others can help you more with your questions about best OO practice
for run-time memory allocation. I would suggest using normal
instantiation and using one of the recommended alternative memory
managers if you are instantiating large numbers of objects, but
I don't have personal experience with performance with large numbers
of them.
BTW, if using large static arrays, use the $M directive e.g.
{$M 16384,10000000} { (the default is $M 16384,1048576) }
Glen
-----Original Message-----
From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]
Behalf Of Paule Ecimovic
Sent: Monday, 5 September 2005 6:22 PM
To: Borland's Delphi Discussion List
Subject: Re: Delphi as a Performance optimum between Fortran
andJavaforInteractive Numerical Analysis
Hi, Glen
Could you give me a pointer on how to solve memory allocation problems
by object-oriented means instead of using pointers and allocating "chunks"
of memory in multiples of the size of a given primitive data type? As you
said, Fortran pre-allocation is wonderful, only if you can map out a static
memory space at design time. Adaptive-stepsize integration routines are
limiting cases of this: pre-allocate for the worst-case scenario and check
if the resources for this are available. I guess all allocation schemes do
the latter, while some, especially the late binders, avoid the former. How
does one apply OO-late binding to runtime memory allocation problems. I
certainly missed this somewhere.
Cheers,
Paule
----- Original Message -----
From: "Glen Thompson" <[EMAIL PROTECTED]>
To: "Borland's Delphi Discussion List" <[email protected]>
Sent: Saturday, September 03, 2005 6:23 AM
Subject: RE: Delphi as a Performance optimum between Fortran and
JavaforInteractive Numerical Analysis
> Hi Paule,
>
> I use Delphi for simulating motor vehicle assembly line operations. This
> is
> not computationally intensive and performance is no problem. Compared to
> Fortran the developpment time is much quicker.
>
> I also have a small but computationally very intensive optimization that
> was
> originally in Fortran. One issue here is Delphi's memory management. It
> is
> slow to dynamically allocate and free memory, so I calculate what I'll
> need
> up front. I have heard about 3rd party improved memory managers but
> haven't
> tried them. They would be more important to me if I could not allocate my
> memory up front.
>
> Glen
>
> -----Original Message-----
> From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]
> Behalf Of Paule Ecimovic
> Sent: Saturday, 3 September 2005 12:39 AM
> To: Borland's Delphi Discussion List
> Subject: Delphi as a Performance optimum between Fortran and Java
> forInteractive Numerical Analysis
>
>
> Hi all,
>
> Can someone give me some pointers on the feasibility of using Delphi
> for
> development and testing of numerically-intensive codes, e.g., simulations
> of
> dynamical systems with large numbers of components such as occur in some
> image processing applications driven by partial differential equation
> systems or cellular automata. I would like to be able to implement
> numerical
> integration routines and test them on a standard problem set for various
> sets of parameters and input data. I would then like to represent the
> results, typically image transformations, graphically. I can paint and
> test
> GUI's rapidly in Delphi, although connecting them to numerical code is
> sometimes tricky. Is there some way of making three "plug boards": a GUI
> layer and a test problem layer and a numerical routines layer such that
> they
> can be interconnected much like a telephone exchange routining calls where
> they need to go given the number that was dialed?
> In a way this question harkens back to Robert's response to the
> business
> logic vs gui layer question. Can anyone elaborate that in this case?
>
> Any thoughts on this matter would be very useful.
>
> Cheers,
>
> Paule
> _______________________________________________
> Delphi mailing list -> [email protected]
> http://www.elists.org/mailman/listinfo/delphi
>
>
> _______________________________________________
> Delphi mailing list -> [email protected]
> http://www.elists.org/mailman/listinfo/delphi
_______________________________________________
Delphi mailing list -> [email protected]
http://www.elists.org/mailman/listinfo/delphi
_______________________________________________
Delphi mailing list -> [email protected]
http://www.elists.org/mailman/listinfo/delphi
