Compilers comparison

[jhp]

I wanted to know how CVF compares with other fortran 77/90/95 compilers as LAHEYs or ABSOFT in terms of quality of the compiler itself, support and updates. I have read polihedron comparisons but they do not get to a conclusion, just make interesting comparisons but without a final opinion and feedback from users. My idea is laheys compiler (LF95) is best than competitors in terms of diagnosis and support, and CVF could be best for developing windows programs but is that true?, and what about other features?. Compaq Visual Fortran at least is much more friendly and comfortable to use. I wanted to ask users for a recomendation for which compiler to choose and WHY when working under windows 98 and use code that is also developed in unix platforms as alpha or sun, and creating windows would not be a primary task.

[Intel_C_Intel]

There is a lot of useful information located at http://www.polyhedron.co.uk/. They do a lot of compiler comparisons. I have found that each compiler is better and worse than another depending upon what you are doing. CVF runtimes are consistently the best, but their compile times are terrible compared to Lahey and Salford.

So, if you are doing a lot of development work and thus concerned primarily with compile times, the Salford and then Lahey compilers would be preferable to CVF. However, if you are doing computationally intensive runs, then CVF and then Lahey would be your compiler(s) of choice. I do primarily development of hydrodynamic and water quality codes, so I rely mainly on Salford because of the short compile times during develpment work, but use CVF for producing distributable executables. The Salford and Lahey compilesr also have an incredibly powerful capability for finding unintialized variables at runtime.

Bottom line - I have 5 different compilers that work in the Windows environment and use them all depending upon what I am trying to develop/test.

Tom

[jhp]

Tom: From your experience with different compilers how will you compare quality of diagnosis, included documentation and support from LAHEYs and CVF? Those are the most important for me. And also, which versions are you dealing with(that could be important for diagnosis).

[Intel_C_Intel]

Inigo - That is a tough question. The documentation is more extensive for CVF, but the documentation is also quite extensive for Lahey. The diagnostic capabilities are slightly stronger for Lahey, but CVF is no slouch in this arena either. Each one will catch different coding problems that the other won't catch. The debugger and development environment are much better in CVF, but Lahey is suitable. Bottom line - I would recommend getting them both as each has strong capabilities that the other doesn't have and that are very useful when developing code. Anytime a compiler saves me a days worth of work tracking down a coding problem, it has paid for itself. Both of these will probably save you many days worth of work thus justifying their initial cost.

Tom

[njuffa]

If you can afford it, buy two compilers. You can cross check code across these
to platforms. I have fours compilers myself. Which two should you get? I nominate
CVF and Lahey LF95.

If you want just a single compiler, go with CVF. It has the best and most extensive
documentation. It integrates seamlessly into Microsoft Developer Studio so you can
use its powerful debugger. It has best (most responsive) support of any Fortran
compiler. On many real world problems it also produces the fastest code, although
Intel Fortran is catching up fast and leads in things like SPECfp benchmarks, LFK,
etc.

-- Norbert

[Intel_C_Intel]

We found that Compaq Visual Fortran was the only Fortran compiler that met our requirements:

1) Compiling very large source files (some other compilers used far too much memory, others failed because the space allocated for symbol tables was too small).

2) Producing fairly efficient code (see below).

3) Having excellent support.

4) Being compatible with "legacy" code.

5) Compling quickly.

Many other compilers satisfied most requirements, but only CVF had them all. Our stringent compilation timing requirements meant that we had to switch off most of the Compaq Fortran optimisations, but this only reduced execution times (in our case) by about 15%. The Compaq support team helped us achieve this.

In some situations, the new Fortran 90/95 features allowed us to improve our algorithms, and thus dramatically increase functionality.

Eddie

[njuffa]

One issue I failed to address in my original reply (and that one rarely thinks about
when first buying a compiler) is the question of version upgrades. I first started
with CVF version 5, and have found the upgrade process extremely reasonable.
Bug fix updates are free for download, and in addition Compaq made CD-ROM versions available at cost ($15) for those who are not on a broadband connection
and have difficulties downloading 20-40 MB version upgrades. The major version
upgrades so far also have been very reasonably priced. By contrast, Intel Fortran
has had no upgrade path (at least none I am aware of) and required to one to buy
the complete product for every version at about $400, which so annoyed me that I
stopped upgrading after version 4.5 (they are now at 5 and 6 is imminent).

-- Norbert

[Steven_L_Intel1]

I appreciate Norbert's comments on CVF - we do believe that CVF has the lowest "total cost of ownership" of the complete Fortran products out there. However, I want to correct a misimpression - I'm not sure what Intel's policies were in the past, but Intel does offer a discounted version upgrade for their Fortran - $295 vs. $499 for new installations. Intel also now offers one year of Premiere Support which includes any new versions that happen to be released during that time at no extra charge. It is likely that Intel Visual Fortran, when it makes its debut, will follow a similar model.

Steve