Community
cancel
Showing results for 
Search instead for 
Did you mean: 
Highlighted
New Contributor I

Old Fortran

The interesting issue in considering TABS or ETABS or ULARC or AXISHELL is why waste time on these old programs.  I spent a fair amount of time and so did people on this board helping with the '73 Harrison program.  Putting this interesting little program with PARDISO and FEAST gives on access to a very fast simple program that uses txt files, which makes you think about the modelling and the modelling methods - as we move deeper into full dynamic methods and the problems grow in size it is interesting to look back at some of the techniques used in the OLDEN days.

TABS takes a huge problem and reduces it to a simple set of equations, is this a suitable assumption, the only way to find out is to try it and compare the results from real data.

Sometimes it helps and sometimes it does not.  But TABS for instance has a great step forward in the topological representation of a model as data.  Playing with the FORTRAN code to understand that method is worth all of the effort. His data method is more LISP like and as we attempt to create models for millions of bridges around the world we need forward thinking to bridge the data gap.

But there is absolutely minimal error checking so even slight mistakes in data entry make a huge difference with the original TABS - so it is an excellent research tool - which is what I do.

The TABS manual is worth reading as he puts in a lot of stuff related to the modelling issues. 

So one has to be very lucky to have access to such code and the people to help make it work.

Where else can you ask some of the best Fortran programmers in the world a question - but here.

And on the other side of the world I have a engineer asking me to explain why we get a burst of energy in the bridge vibration at 150 Hz or higher. That requires some thought and FEAST.

Steve:  Quick question what is roofline? 

 

0 Kudos
4 Replies
Highlighted

Steve:  Quick question what

Steve:  Quick question what is roofline? 

You're asking me? It is not a term I am familiar with (assuming it is software-related and you're not asking about asphalt shingles.)

Retired 12/31/2016
0 Kudos
Highlighted
New Contributor I

Does not really explain what

Does not really explain what it is for?

Dear Beta participant,

 

Intel® Advisor 2017 provides unique capabilities to simplify efficient vectorization.  It also introduces support for 2nd generation Intel® Xeon Phi™ processors (codename Knights Landing). 

 

We have a limited time offer to join a subscription-only preview of the new vectorization advisor “Roofline” feature which has generated a lot of excitement.  Join now to get more information and subscribe to the “Roofline” preview program.

 

Sincerely,

 

The Intel Parallel Studio XE Beta Program Team

Developer Products Division

Intel Corporation

0 Kudos
Highlighted

Oh - I had not heard that

Oh - I had not heard that name (Roofline) before.  Vectorization Advisor I know and it is great. It's sort of a cross between VTune and the old GAP (Guided Auto Parallelization) feature. You run the program under control of Vectorization Advisor and it gives you clear and direct advice as to what you can do to enable or enhance vectorization in your program. Every customer I have seen it demonstrated to loves it. It's being offered as a component of Advisor XE and is all new. (The "subscription" aspect is unfamiliar to me, though.)

Retired 12/31/2016
0 Kudos
Highlighted
Black Belt

I looked at that and couldn't

I looked at that and couldn't see that the subscription idea is distinguished from participating in the Parallel Studio beta test, which has already been open for some time.  I'm guessing it may be a beta update version on the way, as I couldn't find the feature in the current beta Advisor.

The Roofline idea seems based on determining where performance limits lie based on memory bandwidth and computational rate and checking how close an application comes to its limits.  People began talking about it for Intel cluster applications about 3 years ago.  Presumably it might begin to fulfill the promise which was made about Advisor being a tool for estimating potential for performance improvement.

I suppose it's late in the beta program for changes to be incorporated.  The ifort update already corrected a couple of problem reports I had submitted.

I've had several questions about random changes in performance associated with changes in unrolling and alignment which became prominent when running under Advisor.  The new Advisor works fairly well with the 2016 release as well as beta compilers.

0 Kudos