David Mazziotti

David Mazziotti

100 Reputation

4 Badges

9 years, 296 days
I am a professor in the Department of Chemistry and the James Franck Institute at The University of Chicago. I am also the founder and principal scientific adviser at RDMChem LLC.

MaplePrimes Activity


These are replies submitted by David Mazziotti

@Dave L 

Dear Dave,

       Thanks for your interesting and insightful comments into the machinery of MKL within Maple.  I received a patch for Maple 18.01 today which adds MKL support for AVX instructions which appear in the Sandy Bridge (2012) generation of Intel Core 2 chips.   I just finished testing the patch with my program, and my program now works well in Maple 18.01.  Before today, I also ran a few tests that I think you will find interesting.  Here is the summary:

Before the Patch:

(1) My code works in Maple 18.01 if I set the following environmental variable before running Maple

export MKL_CBWR="SSE4_2"

This environmental variable forces the Intel chip to use SSE4.2 instructions (rather than AVX).  The fact that my code works with this command proves that only the MKL library is responsible for the difference between Maple 17 and Maple 18.01 --- not any of the differences introduced in Maple 18.

(2) My code does not work in Maple 18.01 if I try to insert the correct shared library from MKL with the AVX instructions.  My code runs without a fatal error, but it suffers from line search errors from faulty floating-point operations from MKL AVX.

After the Patch:

My code works just fine --- very nice.  The MKL library should be a significant improvement over the ATLAS library distributed in earlier Linux versions of Maple including Maple 18.00.  Clearly, the patch does something more sophisticated than just dropping the shared library for AVX from MKL into the LINUX bin directory as in point (2) above. 

Dave, thank you and the rest  of the Maple technical team for resolving this issue with the MKL library in a timely fashion.  Your assistance is greatly appreciated.

Best regards,

David

-------------------------------------------------------------

David A. Mazziotti

Professor

Department of Chemistry

James Franck Institute GCIS 105

929 E. 57th St.

The University of Chicago

Chicago, IL 60615 USA

-------------------------------------------------------------

 

       

@acer Thanks for the insightful comments and questions. 

(1) My computer is equipped with two 16-core 2.9 GHz Sandy Bridge Intel Xeon processors which do support the AVX instruction set.  Maple is not finding the mkl_blas_avx_* procedures which indicates that the some or all avx routines are missing.

(2) I have only tried 18.01 since I did not download Maple 18 right away even though I have the Elite Maintenance Plan.  Perhaps it would be worth asking Maple to allow users to download older versions of Maple. 

(3) Does Maple 18.00 use ATLAS like the earlier Linux versions of Maple 16 and 17?

(4) I was able to make Maple 18.01 use a version of MKL that I installed with Intel Composer 2013.  I simply added the locations of the MKL dircetories to the path before launching Maple.  For this project I am using the command-line version of Maple.

(5) Within my program I am performing an optimization with line searches.  While the optimizations work quite correctly for several steps, the later steps give line-search errors indicating that there might be too much round-off error.  The optimization does not converge.  In earlier versions of Maple these errors do not occur.  Perhaps Maple is not compatible with the version of MKL in my path statement.  Or perhaps something else is different in Maple 18.01.  Unfotunately, my code with the line search is not in a form that can be distibuted by worksheet.

(6) Do we know the version of MKL in Maple 18.01 for Linux64?

(7) A great feature of Maple is that typically programs work across versions effortlessly.  I have run my program successfully in Maple 16 and Maple 17.  I can continue to run it in these versions, but it would be good to know that I can also run it with Maple 18.01.  (Curiously, Maple 16 is 5% faster than Maple 17.) If MKL works correctly, it should be faster than the previous ATLAS BLAS-driven versions of Maple.

Acer, I would greatly appreciate any further insights that you can provide.

Best wishes,

David

Actually,

LinearAlgebra:-Eigenvectors(A);

 

David

Hi,

      A bug in Maple 16's classic interface is still present in Maple 17's classic interface.  The interface in Windows will frequently crash with a fault when cutting and pasting. scrolling, or even saving the worksheet.  The save error is the worst because it kills the application during the save, leaving only a small fraction of the worksheet saved and usually overwriting a previous save with more information.   Currently, I am still using the Classic Interface with Maple 13 for most work to avoid crashes.  The only difference between Maple 16/17 and Maple 13 in terms of GUI appears to be the way that Help pages are accessed.  It would be very helpful if Maple coud update Maple 17 with a new upgrade of the Classic Interface in Maple 13 to 17.  As I have said elsewhere, I think that Maple should consider a new interface to Maple that combines elements of the Calssic Interface with the Newer Interface to produce a more modern streamlined interface to Maple with the flexibility of the original Classic Interface.

David 

 

Professor

Department of Chemistry

University of Chicago

Hi Eithne,

    While studying at Princeton, I chose Maple over Mathematica for several reasons:

(1) My father bought me a copy of Maple for my laptop --- having Maple right there made a huge difference --- I originally learned about Maple from using Mathcad.

(2) Maple (in the early '90s) had better official and unofficial documentation than Mathematica.

(3) The Maple language is similar to Fortran --- actually quite similar to modern Fortran 2003/2008 --- I would like to see Maple build a better bridge between Maple and Fortran --- i.e. code generation for Fortran 2003/2008 and a Compile command for compiling a Maple procedure with Fortran.  Intrinsically, there is a close connection between the evolution of Fortran and Maple in the past 15 years even if not intentional.

(4) The open source aspect of Maple including infolevel made Maple much more useful --- plus it was easy to see many examples of advanced Maple programming.  Maple could do more to make others understand the excitement of Maple programming drawing upon its language and its data types.  Maple should consider updating and adopting A. Heck's book as the standard Maple book --- it was one of the very best --- much better than what Mathematica ever had.  Also, First Leaves was such a fun read --- the current documentation lacks the excitement of the early days of computer algebra and the computer algebra wars.

(5) I beta tested Maple in 1994 for Maple V R4, and hence, I helped select the features of the Classic Worksheet.  Currently, I still use the Classic Interface because it has editting features that are missing in Standard.  I would like to see the Classic Interface added as an option in the Standard Interface.  It could be marketed like Coke Classic --- that was a big hit :). 

(6) Maple has typically covered a broader range of mathematics than Mathematica with an eye towards conforming to to traditional mathematical notation. 

(7) Shift-Enter: One of the reasons (strange as it may sound) that I chose Maple over Mathematica was that you could use Enter rather than Shift-Enter to enter expressions and run programs.  In the Standard Interface Maple adopted the Mathematica convention.  Why I ask?  Maple had it right the first time.  Maple should re-exmine and treamline the keystrokes required to perform computations in the Standard Worksheet.  They should be competitive with the Classic Interface or even better --- and certainly they should be better than Mathematica.  Maple V R2 and R4 had simple interfaces like modern Mathematica and modern web browsers like Chrome.  Just as browsers have simplied their buttons, Maple should move to a more modern streamlined interface --- at least such an updated Classic interface should be an option within the modern Java Standard Interface.  Making an interface for advanced Maple would (I think) show even entry users the potential power of Maple in comparison with showy but less substantial alternatives.

(8) Site License: The site license at Princeton also helped me learn Maple because I could run some hefty jobs on the server.  I wish Maple would press harder to put Maple in every major University.  I would like to see a site license at the University of Chicago where I am a professor.  Some of my students select Mathematica because there is a site license --- I try to encourage them to try Maple.

Best wishes,

David

 

Professor

Depertment of ChemistryUniversity of Chicago

I haven't had a chance to play with the new Maple yet, but here is a list of features that excite me:

     (1) new data structure for polynomials --- salute to Michael Monagan

     (2) faster complex arithmetic

     (3) faster sparse Vector and Matrix operations

     (4) the new options for studying convex polytopes

     (5) new group theory package

     (6) new code editor --- something that had been needed to encourage more advanced Maple programming

One of the best updates --- thanks!

David

Page 1 of 1