Site Updates

Hello all, its been a few years since I’ve updated and I see that there still is traffic coming to this site every day.  I won’t be continuing with my Fortran adventures, but if anyone else is interested in taking over posting here then please send me a message.

Fortran, Clojure, Haskell and Julia are not at war

A blog response to the Ars Technica article I posted about a while back that discussed the future of scientific computing has recently been posted here:


Once you’re finished reading the post, please direct your attention to it’s comments section where you’ll find the author of the original Ars article responding to defend what he wrote, which is then followed up by a response from the author of this new blog post.


GCC 4.9.1 and NetCDF 4.4.0 Releases

GCC v4.9.1 was released yesterday, bringing OpenMP 4.0 to the gfortran compiler.

Also, v4.4.0 of the netCDF-Fortran library was released a few days ago by Unidata, adding some… stuff (and fixing some things!).

Update to the GUI post below

So when I put together everything for the Creating a GUI using GTK3 and gtk-fortran post about a month ago, I completely screwed up the and apparently included an older version that didn’t have everything named properly.  In any case, I’ve fixed the issue, tested it (I should have done this more thoroughly the first time!), and everything should be working okay.

Here’s the location of the new project template with the fixed .glade file:

nanomsg sockets library for Fortran

A little background:  nanomsg is a sockets library that provides several communication patterns that are designed to be the building blocks for distributed systems.  Built in C, it was created by the lead developer of ZeroMQ, Martin Sustrik.  It is light, powerful, and succinct.

Up until now, sockets-based network communication in Fortran was limited (as far as I know) to using the winsock2 library put together by Intel(?).  Succinctness was not its strong point.  But with the advent of nanomsg and the use of the iso_c_binding intrinsic module standardized by Fortran 2003, I have developed Fortran bindings for nanomsg that will allow for very easy (and succinct!) sockets-based communication:

Included in the nanomsg.f90 file are examples for most of the functions.  I’m confident that there shouldn’t be any problems, but if there are, please let me know by opening an issue on github or leaving a comment here.

One final note: Not only is Fortran 2003 is required for this, but you will also need to compile the nanomsg library and include it in your projects to make this work.


Intel MKL matrix multiplication wrapper

I’ve put together a small wrapper for calling Intel MKL matrix multiplication xGEMM routines.  You’ll need to allocate everything outside the subroutine, but this should make it less confusing since all you’ll be passing in are the matrices and whether or not they’re to be transposed.

subroutine matrix_multiply(first_matrix, first_transposed, second_matrix, second_transposed, result_matrix)
!This subroutine multiplies two matrices using the Intel MKL xGEMM multiplication routines
!The routines can be a little tricky, so I've made it a little easier to use them consistently
!first_matrix and second_matrix are the input matrices
!first_transposed and second_transposed are integers indicating (with a 1 or 0) whether or not
!the respective matrix is transposed
!The resulting matrix is stored in result_matrix, but you'll have to allocate that outside the routine
!The sGEMM below can be changed to dGEMM for double precision without any other changes to the
!subroutine call. You'll then have to change the 'real' variables to 'double precision'.
real, intent(in) :: first_matrix(:,:), second_matrix(:,:)
real, intent(inout) :: result_matrix(:,:)
integer, intent(in) :: first_transposed, second_transposed
integer :: a1,a2,b1,b2
a1 = size(first_matrix,1)
a2 = size(first_matrix,2)
b1 = size(second_matrix,1)
b2 = size(second_matrix,2)
if (first_transposed .AND. second_transposed) then
call sGEMM('T','T',a2,b1,b2,1.0,first_matrix,a1,second_matrix,b1,0.0,result_matrix,a2)
else if (.NOT. first_transposed .AND. second_transposed) then
call sGEMM('N','T',a1,b1,b2,1.0,first_matrix,a1,second_matrix,b1,0.0,result_matrix,a1)
else if (first_transposed .AND. .NOT. second_transposed) then
call sGEMM('T','N',a2,b2,b1,1.0,first_matrix,a1,second_matrix,b1,0.0,result_matrix,a2)
else if (.NOT. first_transposed .AND. .NOT. second_transposed) then
call sGEMM('N','N',a1,b2,b1,1.0,first_matrix,a1,second_matrix,b1,0.0,result_matrix,a1)
end if
end subroutine matrix_multiply

fmincg ported to Fortran

If you’ve had a chance to take Andrew Ng’s wonderful Coursera class on Machine Learning, you’ll no doubt be familiar with the fmincg function that is used in the neural network algorithms.  For his class, he uses Octave/Matlab implementations, but there may be some who wish to continue their work using Fortran.  To facilitate my own continuation, I’ve ported the function to Fortran and made it available to everyone below.  Please let me know what you think (and if there are any bugs!).  Also, I may soon post a version optimized for use with the Intel MKL libraries that will no doubt run much faster!

Oh and by the way, you will need to use the ieee_arithmetic module for this.

Read more of this post

Creating a GUI using GTK3 and gtk-fortran

Update (6/26/14): If you downloaded the old project template .zip I first had up here, you probably couldn’t get the window to load.  That’s because the file in the template had the signals (and the objects) named incorrectly, so the window would never launch.  I’ve since made corrections and uploaded the fixes to here:

I’ve also completely rewritten that section of the post so there won’t be anymore confusion.  If you have no idea what I’m talking about then don’t worry about it…  Everything is great 🙂

Update (6/27/14): I’ve gone ahead and re-written a major of the second half of the post to include some basic concepts for what you’re doing and also to make the steps a bit more clear.


Original Post:  

If there’s one thing that I’ve struggled with in the past year, its been finding an easy way to create a GUI using Fortran.  While I haven’t actually spent an entire year trying to figure it out, I certainly have stumbled during those times when I did try.  I even attempted to make my own library based on gtk-fortran, but then I realized that I was just making things more complicated than they needed to be.  Rather than try and reinvent the wheel, I abandoned the project.

Recently though, I have come across a relatively pain-free way of doing it.  First, a little bit of an introduction:

There are several libraries available to make a GUI, but most don’t have a GUI to make a GUI (i.e. a RAD tool).  You can use DISLIN to code a GUI, and it certainly can turn out nice.  But because you have to code it by hand and then compile/run to see if you did it right, it can be a lot of work.  And unfortunately, I’m really lazy (in public I call it being “pragmatic”).  Not to mention for anything other than your own little home project it costs money to license.  There are other GUI libraries around (Winteracter, GINO), but they cost money, too.  What we need is a free one that you can build a fully-functional program with most of the bells and whistles.

Enter GTK and Glade.

GTK, if you didn’t know already, is an open-source, cross-platform API written in C that’s for creating graphical user interfaces.  In order to interface that into Fortran, gtk-fortran was created.  It is a “partial GTK+ / Fortran binding 100% written in Fortran.”  To do such binding, the ISO_C_BINDING module is used.  The current version of GTK+ is v3.x, so we’ll be concentrating on that rather than the previous v2.x.

Using GTK and gtk-fortran alone, you can create the entire GUI.  But why do that when there’s Glade?  Glade is a RAD tool to make GUIs for GTK, and it works really well.  You can read more about it on its own page, but if you’re going to download it, you should look at glade-main-page-thumbgetting v3.18.x, which is for GTK 3.  There’s another version, 3.8.x, but that’s for GTK 2.   Now, I have mixed up Glade files and used ones that were designed for GTK 2, but I wouldn’t recommend it.

So now you’ve got a library and a GUI for making your GUI.  What next?  How do you put it all together?  That can be a lot harder than it sounds, even for someone that’s experienced.  I struggled for a while until I found this webpage put together by Darius Markauskas:

He goes through the steps of putting everything together and even graciously provides links for a couple of Code::Blocks Fortran projects to demonstrate.  He really has everything there.  So why am I writing this?  Well, for coverage of the topic.  And also because I want to try to simplify the process as much possible.  So, I’m going to cut the steps down and also include a bare-bones project template for Code::Blocks.  That way you’ll be able to create a new project directly from the template with the absolute essentials (and a tiny little demo from the original project by Markauskas).

New stuff:  

Some basic concepts before we begin:

Think of your program as having several different levels of things you need work with.  The bottom level is all the GTK code that draws stuff on screen.  You won’t be messing with any of that.  In fact, you can’t see any of it since its stuffed away in a compiled library in the \GTK3\bin folder.  The next level up is the gtk-fortran code, which calls the GTK code.  Its written in Fortran and visible for you to work with and modify.  However, I would highly recommend against this.  What you’re going to do instead is call the subroutines and functions of gtk-fortran (and provide specific arguments) in your own code, which is the next level up.  Within your own code, you’ll be using things like call gtk_init() and call gtk_widget_show_all(window) to do the heavy-lifting for you.  But, you’ll also need to actually arrange the layout of the GUI and give the various pieces names.  In theory, you can do this in code, but its very cumbersome.  What we want to do is to create a separate file, outside any of the code, with the layout and names and use the Glade editor to work with it.  That way, we can visually see what it will look like and generally have a much easier time when putting it together.  That file will essentially be the highest level up.  It’s really just an XML file with the names of things, their locations, and signals that are associated with “stuff.”  What are signals?  Well, if you’ve never worked with a GUI before, signals are basically specifically-named “signals” that are broadcast by the program when something is done.  So let’s say you click a button.  A “clicked” signal is sent out.  In Glade, you’ve assigned this particular “clicked” signal a unique name that you can use later on.  That unique name is called the “handler name” (yes, its a bit confusing but you’ll get used to it).  You assign the handler the name “button_hello_clicked.”  Every time you click that button, that “clicked” signal is broadcast through the program with the handler name “button_hello_clicked.”  But you, of course, want something to happen when you click a button.  So there are gtk-fortran functions that look for these signals.  You can set up the program using those functions so that whenever it sees a specific signal, it will call a function you specifically designate.  For instance, take a look at the code below:

select case (h_name)
case (“button_hello_clicked”)
call g_signal_connect (object, signal_name, c_funloc(button_hello_clicked), c_this)

case (“button_quit_clicked”)
call g_signal_connect (object, signal_name, c_funloc(button_quit_clicked), c_this)

end case

The above is code from the demo project that I’ve made available below.  The h_name is the handler name of signal.  So the select case is setting up for all the possible signals that could be broadcast.  The call g_signal_connect is a subroutine that then connects that signal to a function of your choice.  Within the arguments for g_signal_connect is c_funloc(button_hello_clicked).  The c_funloc is always required.  You don’t need to know all the details of it, but its a C thing that you need for GTK to know stuff.  What’s more important is the button_hello_clicked.  That’s the actual name of the function in your program that you want to associate with the button being clicked.

And that’s it.  That’s all you need to do.

Well, for the most part.

There’s a lot more you can do, sure, but the above is the very most basic concept you need to get started.


Edited Stuff:  

Steps to get your GUI up and running:

  1. Make sure you have a Fortran compiler and the Code::Blocks IDE
  2. Download the GTK+ 3.x All-In-One bundles (for example, choose the Windows 32-bit platform and then download this All-In-One bundle) and install it to same base directory that you keep your project folders in.  Well, you don’t really have to, but it’ll make things easier.  For example, using the base directory of C:\dev, we’ll eventually have C:\dev\GTK3 (for GTK) and C:\dev\fortran-gui-demo (for your project)
  3. This is very important: Add the bin folder of the above GTK path (e.g. C:\dev\GTK3\bin) to your PATH variable or equivalent.  When you run the program, its going to look for a number of libraries in the GTK3\bin folder.
  4. Download gtk-fortran from the gtk-fortran GitHub page (using the “Download ZIP” button on the right side of the page).  You don’t need this for the demo project I provide, but you’ll need it for the future.
  5. Download Glade so you can edit the GUI’s .glade file later if you want.  Remember, this is the program that you use to visually arrange the GUI’s layout and assign the handler names to the signals.  You may want to download the source for Glade and compile it yourself, so you can go here for that (remember, grab v3.18 or higher).  If you don’t want to compile it, you can download a precompiled binary (even though its only v3.14, it should work fine).
  6. Download this demo project and load it with CodeBlocks:

You’ll notice that in the demo project, there is a directory called gtk_src.  This is the gtk-fortran files that I mentioned you wouldn’t have to worry about for now.  Its simply a convenience that they’re there.  Also notice the file.  Once you load the project in CodeBlocks, you’ll even see it listed as one of the files.  That’s another little thing I did to make things a bit easier.  That’s not normally going to happen.  You’ll specify in the code itself what .glade file you’re using (see call create_window(demo_win,””//c_null_char) )

Keep in mind, too, that I really just changed a bunch of things around in the code from the original say_hello project by Markauskas to make things a little easier to understand.  If you’re still having trouble, go to his page and try working through his directions.  They might work better for you.

Stanford JOLTS Fortran Video Tutorials

Stanford Jolts Modern Fortran

Just found a series of video tutorials from Stanford JOLTS on some of the best features of Fortran 2003 and 2008, such as object-oriented and parallel programming.

More Press for Ars Technica Article on Fortran

Slashdot is now covering the Ars Technica article previously mentioned.  Check out the discussion and leave some input if you can!