THE UNIVERSITY OF BRITISH COLUMBIA
 
Physics 210 Assignment # 7:
 
FORTRAN

SOLUTIONS:
 
Tue. 26 Oct. 2010 - finish by Tue. 02 Nov.

You now know how to "program" in various "programming environments" - bash, muview, gnuplot, extrema, matlab and octave - and in a full-featured "scripted language", python. There are many, many more of both, and by now you should sense that a given task can probably be implemented in any one of them, as long as it is "Turing-complete",¹ but each has its own strengths and weaknesses.

One weakness of scripted languages is that the computer's CPU has to interpret the instructions before it can execute them. This takes time, so if you want your algorithm to run faster you should get the interpretation out of the way ahead of time and make a binary executable version of your program in machine code that your CPU can execute directly. This is called compiling, and all applications that involve massive, time-consuming calculations are written in compilable languages. The most common of these are C and C++, but the oldest and in many respects the simplest is FORTRAN, which is still the most widely used for certain types of computational physics. So this week we will introduce ourselves to FORTRAN.

On hyper, the FORTRAN compiler is gfortran, which stands for the GNU implementation of Fortran 95. (There is a lot of idiosyncratic history here - big surprise . . . . )

As usual, create your /home2/phys210/$USER/a07/ directory and put all your finished results in it.

1. TELL A fib: Recall the Fibonacci numbers? The file ~phys210/fib.f contains the FORTRAN source code to generate a sequence of Fibonacci numbers, starting from one specified by the user and continuing for an additional number of Fibonacci numbers also specified by the user.²

   • Go through the file fib.f and add comments that explain what each step is doing. Be verbose! Remember, in FORTRAN any line that starts with a C or a c is a comment, and is ignored by the compiler.

     ANSWER: See fib_comments.f

   • Compile your edited source code with all the comments using the command

     gfortran fib.f -o fib

     which creates an executable image in the file fib (if you left out the "-o fib" part, it would put the executable image in the default file a.out).
   • Try it out using the command "./fib" 
     If it works, you are done with this part. If not, go back to the first step. (I may have made a mistake.  :-)

2. LINFIT PROGRAM: Using what you have learned about FORTRAN from fib.f (plus lots of additional Google-ing and reading, no doubt), "port" the python code you wrote last week into a FORTRAN program linfit.f (source code) and compile the linfit executable image. Your program should ask a question like, "Where is the file with the data you want read in?" It should read and interpret your answer, read in the data from the specified file, calculate the best linear fit, print out the results, and exit.

   ANSWER: See linfit.f

3. LINFIT SUBROUTINE: You may notice that there are two types of things going on here: a bunch of input/output ("I/O") and a calculation that could easily be generalized and used in a lot of different situations. When this happens, it is tempting to put the generalizable part into a subroutine which is "called" from the main program when the data is all read in and ready to fit. To do this we separate our source code into two files, the first (linfit_main.f) containing the main program with its I/O specifics and a line like

   call linfit (npts, x, y, w, p0, p1)

   and the second (linfit_sub.f) containing the subroutine "source" code, starting with the lines

   subroutine linfit (npts, x, y, w, p0, p1)
   integer npts
   real x(*), y(*), w(*), p0, p1

   . . . and ending with . . .

   return
   end

   ANSWER: See linfit_main.f and linfit_sub.f

   Generating the executable image is only slightly more complicated now that it has to be assembled from two components. See if you can guess how to do this. (Try the most obvious guess first!)

   ANSWER: The command is "gfortran linfit_main.f linfit_sub.f -o linfit"

4. GO TO THE LIBRARY: When we get a lot of subroutines and functions³ it becomes useful to store the components as object modules (*.o) in a library like libmy.a (a "static" library) or /usr/lib/libmyutils.so (a "dynamic" or "shared" library). Here you will only need to manage the first type, which are used like a grocery store: you select the ingredients you need from the inventory, take them home and make a tasty meal from them. The latter type are more like pizza delivery: when you want to run a program that uses modules from a shared library, the required modules are installed at run time and can be (as the name suggests) shared by many other tasks running simultaneously. If you want to learn more about the latter, Google shared dynamic libraries.

   To learn about static libraries, just say "man ar" - because the program that manages static libraries is called ar (for "archiver"). All you really need to know are these three steps:

   • Generate an object module linfit_sub.o from the source code linfit_sub.f using the gfortran compiler:

        gfortran -c linfit_sub.f

   • Insert (or replace) the linfit_sub.o in the libmy.a library:

        ar rcs libmy.a linfit_sub.o

   • Build your program (again with gfortran, retrieving linfit_sub.o from libmy.a:

        gfortran linfit_main.f libmy.a -o linfit⁴

   Do it. Save all your results in the usual place, along with all your source code, object modules and executable files.

   ANSWER: there is nothing left to explain!

5. make IT SO: All this coding and compiling and archiving and building and installing can get very complicated, especially when you are building an application that uses object modules from many libraries, and particularly when you are making changes to required subroutines and functions in different libraries and trying to keep the assorted libraries (and the final product) all up to date! Although it would be wonderful if one "spoke Geek" well enough to give all these instructions to the computer in the right order every time without any goofs, any reasonable person would want to have (at least as a "backup") some sort of stored script that remembers it all for you - a program in its own right, in other words, but this time a program for building an application.

   In Linux this script is called a Makefile, because the program that interprets the commands in the Makefile is called make and, as usual, it is invoked by the command "make". See (e.g.) http://en.wikipedia.org/wiki/Make_(software) or Google "Makefile" for details.

   The syntax of a Makefile is extremely arcane. Some consider it to be the ultimate in "Geek lingo". Nevertheless it is possible to adapt a working Makefile to make your own project, by just changing the names of the programs, libraries and sources. If you copy the files fib_main.f, fib_sub.f and Makefile from ~phys210/ to your own /home2/phys210/$USER/a07/ directory, cd to that directory and enter the terse command

       make

   you should get a new version of the fib executable and a new file called libmy.a containing the fib_sub.o object module. Then, without necessarily understanding exactly what the Makefile is doing,⁵ edit and adapt your copy of the Makefile to add linfit_sub.o to the libmy.a static library and, from it and linfit_main.f, compile and build your linfit executable.

   ANSWER: See the adapted Makefile

   Once this works, you can make any modifications you like to either linfit_main.f or linfit_sub.f and generate a new, up-to-date, executable linfit just by entering "make".