News from Dr. Fortran - Intel's Steve Lionel Discusses Newly Released Fortran Studio XE 2011 - Parallel Programming Talk #128

Dr. Fortran (aka Intel's Steve Lionel), who is a friend of this show, dropped in on Clay and Kathy to discuss what we think is good news - the release of Fortran Studio XE 2011. Below the video are some notes and links that I think you will find useful.


http://software.intel.com/en-us/videos/channel/parallel-programming/parallel-programming-talk-128-steve-lionel/1247156004001

It’s time for Parallel Programming Talk – this is show #128 We’ll be talking with friend of the show, Steve Lionel – Otherwise known as Dr. Fortran in a few minutes.
2. SC11 is only a few weeks away. I guess that might be why there isn’t much new these past few weeks. I guess everyone is saving all the cool hardware and software announcements for the conference. Clay, I know the TOP500 list will be released at SC11. Any bets on what will be in the #1 slot?
CLAY: I’d go with the current #1 system, the K Computer from Japan. It’s got too much of a performance lead to be overtaken in 6 months. Plus, it wasn’t completed when the last reported measurements were taken.

Kathy: We’d like to hear from you – do you have a show idea, a listener question or do you have a prediction about who will be on the TOP500 list? Clay, what’s the best way for our viewers to let us know what they’re thinking?

Clay: They can send us email at parallelprogrammingtalk@intel.com

And now it’s time for our guest Steve Lionel – It’s good to have you back again. Before we get to our subject Fortran – can you tell us a little bit about yourself for our new viewers.

Thanks Steve I understand there is some very good news for Fortran Developers.

• What is Fortran Studio XE 2011? What has there been in the past?
• Are there any new features that were not available before?
• Are there any differences between Fortran Studio XE 2011 and the Parallel Studio XE 2011 products?
• What version of Fortran is supported?
• Where can Fortran programmers get more information about Fortran Studio XE 2011?

If you have comments, questions, suggestions for guests or show topics that you think would be of interest, we’d love to hear from you. Send us an email with your thoughts to parallelprogrammingtalk@intel.com

Remember, programming genius is one percent inspiration, ninety-nine percent parallelism.

About or Guest

Steve Lionel -






My life in an online community began on the PLATO IV system in 1974, continued on ARPANET in the late 1970s, CompuServe and then this newfangled Internet tube-thingy. I joined Intel in 2001 coming from DEC/Compaq where I was a Fortran compiler developer. Nowadays I focus more on the support and "evangelizing" side of things, including my "Doctor Fortran" column which I started in 1998 and which I suppose could be considered a blog. I've been in the same office in Nashua, New Hampshire, since 1988, but I'm on my third company badge. Outside of work I like to read science fiction, build web sites and travel.
For more complete information about compiler optimizations, see our Optimization Notice.

Comments

's picture

Dear Dr. Lionel ("Dr. Fortran"),
Dear INTELligences (meant friendly),

1) The Dr. Fortran - "read more in the Dr. Fortran - gets explicit" - does not work, possibly (polite smile) for the reason of a profanity (incompatability with IE, e.g. some IE 8 or so).

2) A hint for more speed, factor 8 (measured with climate model ccm3.6 but should apply to CESM as well: When contents from large structures are explicitely (I get explicite too) transferred by assignement from "array-elements of strcutures" to "normal" arrays, like in F77, then the "downstream computation" (all the other subr.) is a factor of 8 (eight) faster than if arguments are passed from structures into these arrays in the parameter-list. Of course: Data-copying takes time. I think: IVF possibly "computes with any access to a structure-element an adress", comparable to segmented adresses in "the old days" (16 Bit vs 32 Bit flat mem model). The switch from 16 Bit to 32 Bt flat memory (with the
Watcom compiler in the late 1980s / early 1990s) was exactly the same factor 8 (DOS with Rationals 32 Bit DOS-Extender).
I would like to have a F90 to F77 translater (or I write one myself) to get CESM really fast.
F77: Meant: F77 plus allocate.

I want it, for reasons of safety and speed, as F77-like as possible, with "extra wishes" for those who wish (e.g. graphical APIs, whether you
would regard graphical as something like explicity, or, as I do, as the core of profanities).

And: I would like to have a "target OS definable" during linking, such as (yes) Freedos or some other "single process mini-OS", e.g. development
in Windows (command-line) or Redhat but target OS as one likes - for speed. Wonder whether FreeDOS can adress 4 GB and more!

A profanity that comes: High res FEM, some 0.1 mm resolution at - meant serious - 20 by 4 by 4 km(!).

It will be a ultra-large market, also for operational applications with, right now, really excellent FEM packages (COMSOL Multiphysics, possibly
Fastflo from CSIRO, which is really, really fast) need ultra-fast computers.

And things that you feature in your F90/F95 comment ("explicite interfaces"): Such things contribute to unsafe situations, errors difficult to find
and - possibly (sometimes) - less speed at execution-time.

I want it such: "One throws "blindly" the code into the compiler: With /check:all it stops at any error and once one is through, it runs fast. With WATCOM we had been really fast in the 1980s/1990s, just throwing some 10000 lines of code into it in one go (386) and quite soon it was fine.

For this approach, formed at the time of famous WATFIV (Waterloo Fortran IV, they had a really excellent /check:all), one should make it easy for the compiler.

Possibly the IVF might come with an option: Somehow F77-like: Ultra-safe, ultra fast but can use APIs from elsewhere also with MPI, Perl4 as front-end, scripts as you have etc..

And: The stuff I really hate (what many others most likely want), e.g. some F90/F95 things.

I would like even a translater from C to F77 (yes, meant serious).

C As assembler-replacement: OK but not for larger programs.