Hello friends,
At last I manage my institute to buy a fortran compiler. Now it is my
tusk to work on which one we should buy.
We actually a scintific research institute, most of the computational
tusk is on condensed matter physics(ab initio molecular dynamics,
Recursion  etc.)
Giving this in mind, may I request you to suggest me a good compiler,
with their pros and cons? I am familiar with ifort, sun, gfortran. I
am zooming in NAG,Lahey and intel again. Any suggestion?

rudra wrote:
> Hello friends,
> At last I manage my institute to buy a fortran compiler. Now it is my
> tusk to work on which one we should buy.
> We actually a scintific research institute, most of the computational
> tusk is on condensed matter physics(ab initio molecular dynamics,
> Recursion  etc.)
> Giving this in mind, may I request you to suggest me a good compiler,
> with their pros and cons? I am familiar with ifort, sun, gfortran. I
> am zooming in NAG,Lahey and intel again. Any suggestion?

  http://www.polyhedron.com/compare0html

On 2009-09-07 10:29:04 -0300, rudra <bnrj.rudra@gmail.com> said:

> Hello friends,
> At last I manage my institute to buy a fortran compiler. Now it is my
> tusk to work on which one we should buy.
> We actually a scintific research institute, most of the computational
> tusk is on condensed matter physics(ab initio molecular dynamics,
> Recursion  etc.)
> Giving this in mind, may I request you to suggest me a good compiler,
> with their pros and cons? I am familiar with ifort, sun, gfortran. I
> am zooming in NAG,Lahey and intel again. Any suggestion?

"Best" implies some crtierion for identifying which one to choose.
What are the other restrictions? For example, which platform? Do
you qualify for the free licenses?

Do you want good debugging and OK speed? Do you want fast executiom for
programs that are already debugged? The answers are different.

If you want both then the answer may be to have one for each attribute.
A fast program may be glamourous but a wrong program is useless, or even
harmful.

I use two. One for debugging and one for fast production. Using two also
means that I am less likely to have trouble with new versions of other
changes.

On Mon, 07 Sep 2009 14:58:59 GMT, Gordon Sande
<g.sande@worldnet.att.net> wrote:

>
>"Best" implies some crtierion for identifying which one to choose.
>What are the other restrictions? For example, which platform? Do
>you qualify for the free licenses?
>
>Do you want good debugging and OK speed? Do you want fast executiom for
>programs that are already debugged? The answers are different.
>
>If you want both then the answer may be to have one for each attribute.
>A fast program may be glamourous but a wrong program is useless, or even
>harmful.
>
>I use two. One for debugging and one for fast production. Using two also
>means that I am less likely to have trouble with new versions of other
>changes.
>
>

Also, if he has to continue using code written on some old compilers,
it would be good to check for vendor specific extensions (for example,
if he used MS PWS and CVF, Intel's would be the most "natural" course
to take).

pp, Luka

Thanks people.
firstly, soory for using the word *Best*, what i am lookin for is a
good one for scientific computing.

let me give the criterions:
1.we are in linux system(both as desktop and in clusters)
>Do you want good debugging and OK speed? Do you want fast executiom for
>programs that are already debugged?
2. I am happy with gdb debugger; and so, I want good optimization and
speed. Also, if there is some auto parallalization, that will be
excellent.(specially i am a fan of sun's omp default(__auto) syntax,
but this is a add on only). But then, there may be better
(and may be easier)  debugger then gdb. So, plz suggest me each of
them.

3. good library.

In article 
<39d24cce-6ad1-47ad-860e-b0ec2751f729@x5g2000prf.googlegroups.com>,
 rudra <bnrj.rudra@gmail.com> wrote:

> I am familiar with ifort, sun, gfortran. I
> am zooming in NAG,Lahey and intel again. Any suggestion?

Regardless of which compiler you choose to buy, I would recommend 
that you install and use the free gfortran in addition.  There are 
several advantages to using multiple compilers when you develop 
codes, particularly scientific codes that will be distributed in 
source form to other scientists.  For example, not only will the 
fortran itself be better (better style, better organization, more 
portable file names, more portable use of the preprocessor, etc.), 
but the associated files such as makefiles and shell scripts and 
perhaps even the I/O data files will be written in a better way as a 
result.

$.02 -Ron Shepard

On 2009-09-07 13:34:49 -0300, rudra <bnrj.rudra@gmail.com> said:

> Thanks people.
> firstly, soory for using the word *Best*, what i am lookin for is a
> good one for scientific computing.
> 
> let me give the criterions:
> 1.we are in linux system(both as desktop and in clusters)
>> Do you want good debugging and OK speed? Do you want fast executiom for
>> programs that are already debugged?
> 2. I am happy with gdb debugger;

gdb is for debugging assembley level programs. (Fortran when it is
throwing wild subscripts is just another buggy assembley program!)
You should try using something like Siverfrost, Lahey/Fujitsu or
NAG with their undefined variable options operating at the Fortran
source level. Much better to find bugs while they are still just
curious values and long before they turn into bad subscripts. As
they say "Once they have seen Paris they don't want to stay on the
farm".

NAG supports Linux, has superb debugging and code quality of whatever
C backend they are using (i.e. usually GCC).

> and so, I want good optimization and
> speed. Also, if there is some auto parallalization, that will be
> excellent.(specially i am a fan of sun's omp default(__auto) syntax,
> but this is a add on only). But then, there may be better
> (and may be easier)  debugger then gdb. So, plz suggest me each of
> them.
> 
> 3. good library.

On Sep 7, 9:29=A0am, rudra <bnrj.ru...@gmail.com> wrote:
> Hello friends,
> At last I manage my institute to buy a fortran compiler. Now it is my
> tusk to work on which one we should buy.
> We actually a scintific research institute, most of the computational
> tusk is on condensed matter physics(ab initio molecular dynamics,
> Recursion =A0etc.)
> Giving this in mind, may I request you to suggest me a good compiler,
> with their pros and cons? I am familiar with ifort, sun, gfortran. I
> am zooming in NAG,Lahey and intel again. Any suggestion?
Afaik Sun compilers are free alongwith the debugger, performance
analyzer, math libs etc.

rusi_pathan wrote:

[snip]

> Afaik Sun compilers are free alongwith the debugger, performance
> analyzer, math libs etc.


Let's hope it stays that way if/when Oracle takes control of Sun.

I'm a Physicist; so the obvious question for Rundra is "is speed of
execution going to be important?"

There are many kinds of calculations in Physics, especially in the
quantum and astrophysical areas that require horrendously long times
to execute.

Such a requirement means you look for a highly-optimisising compiler.
However, a simpler bug-finding compiler is very useful for getting the
program code correct before re-compiling with a good commercial
compiler for the execution phase.

rudra wrote:
> Hello friends,
> At last I manage my institute to buy a fortran compiler. Now it is my
> tusk to work on which one we should buy.
> We actually a scintific research institute, most of the computational
> tusk is on condensed matter physics(ab initio molecular dynamics,
> Recursion  etc.)
> Giving this in mind, may I request you to suggest me a good compiler,
> with their pros and cons? I am familiar with ifort, sun, gfortran. I
> am zooming in NAG,Lahey and intel again. Any suggestion?

I don't think anyone can say what's best - for a start you'd need to be 
very familiar with all the available compilers.  I use Intel Visual 
Fortran for scientific computing and I'm very happy with it.  Execution 
speed is my main concern.

as i told earlier, i am in linux.  so IVF is out of my choice.

"rudra" <bnrj.rudra@gmail.com> wrote in message 
news:25e7f224-d1ab-42fc-9cfe-89b38b1bb416@y10g2000prf.googlegroups.com...
> as i told earlier, i am in linux.  so IVF is out of my choice.
>

Intel also do a Fortran compiler for Linux

Les 

Gordon Sande wrote:
> On 2009-09-07 10:29:04 -0300, rudra <bnrj.rudra@gmail.com> said:
> 
>> Hello friends,
>> At last I manage my institute to buy a fortran compiler. Now it is my
>> tusk to work on which one we should buy.
>> We actually a scintific research institute, most of the computational
>> tusk is on condensed matter physics(ab initio molecular dynamics,
>> Recursion  etc.)
>> Giving this in mind, may I request you to suggest me a good compiler,
>> with their pros and cons? I am familiar with ifort, sun, gfortran. I
>> am zooming in NAG,Lahey and intel again. Any suggestion?
> 
> "Best" implies some crtierion for identifying which one to choose.
> What are the other restrictions? For example, which platform? Do
> you qualify for the free licenses?
> 
> Do you want good debugging and OK speed? Do you want fast executiom for
> programs that are already debugged? The answers are different.
> 
> If you want both then the answer may be to have one for each attribute.
> A fast program may be glamourous but a wrong program is useless, or even
> harmful.
> 
> I use two. One for debugging and one for fast production. Using two also
> means that I am less likely to have trouble with new versions of other
> changes.

I see your two and raise you at least four:  I set up my "Makefile"s
so that  objects and executables go into distinct compiler/option
specific directories, and then "make bins" builds the project using
multiple compilers (currently gnu, intel, pathscale ,and sun (in
asciibetical order), with debug and optimized versions of each.
Building with at least three compilers gives me more info to work with...

-- Carlie Coats

On Sep 8, 4:02=A0pm, "Les Neilson" <l.neil...@nospam.co.uk> wrote:
> "rudra" <bnrj.ru...@gmail.com> wrote in message
>
> news:25e7f224-d1ab-42fc-9cfe-89b38b1bb416@y10g2000prf.googlegroups.com...
>
> > as i told earlier, i am in linux. =A0so IVF is out of my choice.
>
> Intel also do a Fortran compiler for Linux
>
> Les

right, but not "Visual" one!

Terence wrote:

> I'm a Physicist; so the obvious question for Rundra is "is speed of
> execution going to be important?"
> 
> There are many kinds of calculations in Physics, especially in the
> quantum and astrophysical areas that require horrendously long times
> to execute.
> 
> Such a requirement means you look for a highly-optimisising compiler.
> However, a simpler bug-finding compiler is very useful for getting the
> program code correct before re-compiling with a good commercial
> compiler for the execution phase.

Get at least 2 different compilers: if one gives you error messages you
can't understand or don't believe, another may help sort out the problem. 
If both compile and run but give different outputs, investigate whether the
problem is in your program or in at least one of the compilers. (What use
is fast execution if the results are wrong?) Beware: it can take a lot of
your time hunting and reporting compiler bugs, though I suspect not as much
time as the compiler developers have to spend checking whether you are
right, and fixing them if you are :-)

-- John Harper

rudra wrote:
> On Sep 8, 4:02 pm, "Les Neilson" <l.neil...@nospam.co.uk> wrote:
>> "rudra" <bnrj.ru...@gmail.com> wrote in message
>>
>> news:25e7f224-d1ab-42fc-9cfe-89b38b1bb416@y10g2000prf.googlegroups.com...
>>
>>> as i told earlier, i am in linux.  so IVF is out of my choice.
>> Intel also do a Fortran compiler for Linux
>>
>> Les
> 
> right, but not "Visual" one!

It's basically the same compiler.