Re: [LF] arrays versus pointers

[Lahey Support]

=0D
>Wrote: =0D
>> I have often heard that arithmetic expressions involving pointers can =
=0D
>> not be optimized by compilers and it is better to use arrays, =0D
>> especially on vector machines. =0D
> =0D
=0D
=0D
>A F90 pointer is a struct where the first member is usually the "C" poin=
ter =0D
>to data, followed by bounds, strides, ... =0D
=0D
Does any one know exactally what the struct is?=0D
=0D
>What I understand for optimization is that Alexey is close to the answer=
when =0D
>he said "arithmetic expressions involving pointers can not be optimized =
by =0D
>compilers and it is better to use arrays, especially on vector machines"=
. =0D
It sounds like F90 pointers take up more memory than Fortran 77 array ele=
ments, C pointers or =0D
C array elements. The adjustable arrays of Fortran 90/95 are slightly la=
rger too. For that reason =0D
and the fact that they are newer (and there has been less time to write o=
ptimizing code) I'd suspect =0D
that the Fortran 77/C/C++ array would be a better option.=0D
=0D
Does anyone know what the structure of an adjustable array is?=0D
=0D
> =0D
>As far as paralell processesing is concerned Fortran 90 has the PURE key=
=0D
>word to help with that. It gives you a way of telling the compiler that=
a =0D
>particular section of code can be executed independently of another. =0D
=0D
The PURE keyword helps the compiler to predict that a subprogram has no s=
ide =0D
effect. If the subprogram is called from a loop, the compoiler may then u=
nroll =0D
the loop, paralelize it, ... =0D
=0D
> =0D
>Once again, I didn't write the compiler. =0D
> =0D
>David T. Croft =0D
>www.AdsorptionProcessModeling.com =0D
=0D
Sylvain Bergeron=0D
Conseiller principal=0D
Groupe-conseil Aon=0D
T=E9l=E9phone : (514) 288-7575 ext: 4323=0D
T=E9l=E9copieur : (514) 845-0678