Using LAPACK libraries with Intel Fortran and Abaqus user subroutines

What is Linear Algebra PACKage (LAPACK)

Linear Algebra PACKage (LAPACK) is a Fortran-based efficient library for linear algebraic operations shared under a 3-Clause BSD License. Original LAPACK was written in FORTRAN 77 but later moved to Fortran 90. Many programming languages and their packages have interfaces or wrappers for the original LAPACK library. The original LAPACK is hosted on Netlib, but many software companies have their own implementations of LAPACK such as Intel has its own version of LAPACK included in the Math Kernel Library (MKL) as a part of the Intel oneAPI package and Apple has its LAPACK library within the Accelerate framework. There are also simplified Fortran 95 interface libraries to the original LAPACK however it is not standardized and interfaces vary among different implementations. With LAPACK, you do not need to write standard linear algebraic procedures. Check the LAPACK documentation on Netlib to see what type of subroutines are available in LAPACK. The primary objective of this blog is to test the LAPACK library from Intel MKL with Abaqus user subroutines.

Installing Visual Studio and Intel oneAPI

To use LAPACK subroutines from the Intel MKL in your Fortran codes, first, you need to have Microsoft Visual Studio and Intel oneAPI Toolkits (Base and HPC) installed.

The community edition of Microsoft Visual Studio is free and the Intel oneAPI has been free for the last few years.

  • Make sure to install compatible versions of Microsoft Visual Studio and Intel oneAPI packages. Use Google to find information about compatible versions. Hyperlinks to the Intel website for the developers’ guide never work.
  • Install Visual Studio first. The installation procedure is simple, and stick to the default option, but make sure to select the Desktop development with C++ option during installation.
  • Now proceed to Install oneAPI packages; first the Base Toolkit and then the HPC Toolkit. Stick to the default installation procedure. Intel oneAPI packages should be installed in the C:\Program Files (x86)\Intel\oneAPI directory.

Once the installation is complete, you can access Intel oneAPI Command Prompt from the Program Menu on Windows.

Compiling and linking LAPACK subroutines from the command line

Save the following Fortran program as mkl_test.f90 in your working directory. This code is seemingly simple; it calls two subroutines lapack_test() and lapack95_test() from the main program which solves for a simple 3x3 linear system using the dgesv subroutine from LAPACK and the gesv interface from Intel MKL’s LAPACK95 library. In both cases, it prints the results on the screen.

Click here to see the Fortran code
program main

  call lapack_test()

  call lapack95_test()

end program main


subroutine lapack_test()
! subroutine for testing original lapack routine

  implicit none

  integer, parameter  :: wp = selected_real_kind(15,307)
  integer, parameter  :: n = 3
  real(wp)            :: A(n,n), b(n), x(n)

  ! required for original lapack
  integer, parameter  :: nrhs = 1
  integer             :: lda, ipiv(n), ldb, info, i

  do i = 1, n
    A(i,1:i-1) = -1.0_wp
    A(i,i+1:n) = -1.0_wp
    A(i,i) = 5.0_wp
    b(i)   = i
  end do

  x = b
  lda = size(A,1)   ! row count of A
  ldb = size(b)     ! row count of b

  ! A returns as A = P*L*U factorization, x is the solution
  call dgesv(n, nrhs, A, lda, ipiv, x, ldb, info)
  print *, 'results from original lapack subroutine: '
  print *, x

end subroutine lapack_test

subroutine lapack95_test()
! subroutine to test lapack95 routine

  use lapack95, only: gesv

  implicit none

  integer, parameter  :: wp = selected_real_kind(15,307)
  integer, parameter  :: n = 3
  real(wp)            :: A(n,n), b(n), x(n)
  integer             :: i

  ! rebuilding A and b since it was changed
  do i = 1, n
    A(i,1:i-1) = -1.0_wp
    A(i,i+1:n) = -1.0_wp
    A(i,i) = 5.0_wp
    b(i)   = i
  end do

  x = b
  ! A returns as A=P*L*U factorization, x is the solution
  call gesv(A,x)    
  print *, 'results from lapack95 module: '
  print *,x

end subroutine lapack95_test

If you are already familiar with the Microsoft Visual Studio environment, you can create a project there and compile and link the code to Intel MKL libraries to create the executables. You can find other tutorials or blogs to learn how to do that. Here we will use the command line option to compile and link the code to generate the executable. Open the Intel one API Command Prompt for Intel 64 for Visual Studio 2019 terminal from the Windows program menu. A simple Fortran program (without MKL library or external library) can be compiled and linked using:

ifort -o test test.f90

Now to compile and link a Fortran code that includes MKL components, we need to add the /Qmkl flag before the -o. Since the code also uses the LAPACK95 module from Intel MKL, we need to add mkl_lapack95_ilp64.lib for further linking.

ifort /Qmkl -o mkl_test mkl_test.f90 mkl_lapack95_ilp64.lib

ifort is the Intel Fortran compiler, the /Qmkl option tells the compiler to link standard .lib files related to MKL (it’s a shortcut), and -o specifies the object file. The last option mkl_lapack95_ilp64.lib is to link the LAPACK95 interface library to the object file (this is not a part of the core MKL).

This will generate the mkl_test.exe executable and mkl_test.obj object file in the working directory. You can then run the code from any Windows terminal using:

./mkl_test

This will now execute and print out results from two different subroutines used in the code. Now LAPACK95 interfaces may appeal to you more than the original LAPACK subroutines because of the simplicity. But there is a caveat, LAPACK95 implementations are not as universal as the original LAPACK. LAPACK95 interfaces included in the Intel MKL and provided by Netlib are different. To avoid compatibility issues, it is probably better to use the keywords when calling a LAPACK95 subroutine.

TIP

If you are using a Linux or Unix-based operating system (macOS), you can follow this simple tutorial on fortran-lang.org to learn how to compile source codes and link libraries.

Using LAPACK libraries with Abaqus user subroutines

If you are looking to use LAPACK with Abaqus user subroutines, I am assuming, you already have configured Abaqus to compile and link user subroutines. If you have not done it yet, follow the steps described in a previous blog post. At the end of the post, it is shown how to configure the Abaqus environment file to use the MKL library.

You need to have admin access to make changes described in that blog post. If you are working on a shared computer cluster, ask the system administrator to make changes for you.

In case you are unable to make changes to the environment file in the installation directory for compatibility reasons, you can add a abaqus_v6.env file to your home directory or the project’s working directory with customized compiling options as shown below for Windows.

# job-specific abaqus_v6 environment file (Windows)
# this should be located in the working directory

compile_fortran += ['/Qmkl:sequential']

If you are using Abaqus with Intel Fortran compiler on Linux, the compiler option needs to be -mkl=sequential. This command will extend the default compiler options provided by Abaqus and let you compile user subroutines that uses functions and routines from the Intel oneMKL library.

TIP

It is possible to add compile_fortran += ['/free'] (Windows) or compile_fortran += ['-free'] (Linux) to the abaqus_v6.env file to execute free form Fortran codes as Abaqus subroutine. You will also need to change the Abaqus-provided fixed-form subroutine interface to a free-form one. Multiple compile options can be added to this way by seperating each option by a comma (,). If your subroutine requires additional compiling or linking options, it is perhaps better to share the customized abaqus_v6.env file with other users to execute the code.

Writing and testing Abaqus user subroutine with LAPACK

Now we will include the lapack_test() subroutine to the Abaqus UEL subroutine interface. For the simple (and dummy) model we built, Abaqus will call the UEL subroutine for that. Within the UEL subroutine, we call the lapack_test() followed by calling the xit function from Abaqus to terminate the program. Copy and save the following code as uel_mkl.for in your working directory. For simplicity, we will use the fixed-form UEL interface provided by Abaqus in its documentation.

Click here to see the Abaqus UEL subroutine code
      SUBROUTINE UEL(RHS,AMATRX,SVARS,ENERGY,NDOFEL,NRHS,NSVARS,
     1 PROPS,NPROPS,COORDS,MCRD,NNODE,U,DU,V,A,JTYPE,TIME,DTIME,
     2 KSTEP,KINC,JELEM,PARAMS,NDLOAD,JDLTYP,ADLMAG,PREDEF,NPREDF,
     3 LFLAGS,MLVARX,DDLMAG,MDLOAD,PNEWDT,JPROPS,NJPROP,PERIOD)

      INCLUDE 'ABA_PARAM.INC'

      DIMENSION RHS(MLVARX,*),AMATRX(NDOFEL,NDOFEL),PROPS(*),
     1 SVARS(*),ENERGY(8),COORDS(MCRD,NNODE),U(NDOFEL),
     2 DU(MLVARX,*),V(NDOFEL),A(NDOFEL),TIME(2),PARAMS(*),
     3 JDLTYP(MDLOAD,*),ADLMAG(MDLOAD,*),DDLMAG(MDLOAD,*),
     4 PREDEF(2,NPREDF,NNODE),LFLAGS(*),JPROPS(*)

      
        call lapack_test()

        call xit

      END SUBROUTINE UEL


      subroutine lapack_test()
      
        implicit none

        integer, parameter  :: wp = selected_real_kind(15,307)
        integer, parameter  :: n = 3
        real(wp)            :: A(n,n), b(n), x(n)

        integer, parameter  :: nrhs = 1
        integer             :: lda, ipiv(n), ldb, info, i

        do i = 1, n
          A(i,1:i-1) = -1.0_wp
          A(i,i+1:n) = -1.0_wp
          A(i,i) = 5.0_wp
          b(i)   = i
        end do

        x = b
        lda = size(A,1)   ! row count of A
        ldb = size(b)     ! row count of b

        ! A returns as LU factorization, x is the solution
        call dgesv(n, nrhs, A, lda, ipiv, x, ldb, info)
        print *, 'results from original lapack subroutine: '
        print *, x

      end subroutine lapack_test

Executing user subroutine with a dummy Abaqus model

Since we chose to test the LAPACK subroutine inside the Abaqus UEL subroutine, we need to build an Abaqus model with a user element. Remember, our Abaqus subroutine does not do anything but perform a simple linear algebra calculation. So we do not need a real Abaqus model, rather we need something minimalistic that can help us to execute the subroutine. Following is a simple example of a user element, where I defined a single 4-node quadrilateral element with one property and static step. We did not apply any boundary or loading condition to this. Copy the following input file and save it as test.inp in the working directory.

Click here to see the Abaqus input file
*User element, type=U1, nodes=4, coordinates=2, properties=1, iproperties=0
1, 2
*Node
1, 0., 0.
2, 1., 0.
3, 0., 1.
4, 1., 1.
*Element, type=U1, elset=all
1, 1, 2, 4, 3
*Uel property, elset=all
1.0
*Step, name=test
*Static
1., 1., 1.E-3, 1.
*End step

Abaqus UEL subroutine needs to be executed from the command line. From the PowerShell terminal, Windows Command Prompt, or Abaqus Command terminal, you can execute the subroutine with the sample input file using the following command:

abaqus interactive job=test user=uel_mkl.for

This will compile and link the user subroutine to the Abaqus executable and run it from the command line. As we already discussed, this is not a real Abaqus model or user subroutine code, we are not hoping to get any real finite element result. Executing this subroutine will just print out the same output as before on the terminal before exiting Abaqus with an error (remember, we used xit function for termination).

Fortran and LAPACK resources

Since Fortran has been revised multiple times to include new features, it may make you a bit confused when you look at the legacy code vs how Fortran codes are written today. It is good to know the features from the older Fortran standard and modern Fortran standard and follow a consistent style in developing subroutines. Some resources that may help in that regard are:

Using LAPACK is not as intuitive or straightforward as using MATLAB or NumPy. Look into the documentation carefully and you can use the following to find what LAPACK subroutines to use and how to use them for specific tasks.

Intel also has LAPACK implementation examples from oneMKL avaiable online, but I am afraid to share the hyperlinks as they often change. It is better to search online when needed. Additionally, the same examples can be found in one of the Intel oneAPI installation subdirectories.

Bonus item: Adding Intel oneAPI Command Prompt to PowerShell and Windows Terminal

You can not run the Intel oneAPI command line from the PowerShell terminal directly, but you can run it from the original Windows Command Prompt terminal. However, Command Prompt, even the default old PowerShell (PowerShell 5) does not have the latest features such as auto-completion. If you have not used the new PowerShell, it’s highly recommended on Windows. Download and install the latest version of PowerShell 7 from here. The new PowerShell executable is named as pwsh.exe.

Windows Terminal App is great at managing different Shell profiles. Terminal does not have to be boring; you can customize your terminal experience a lot using this app. Download and install the Windows Terminal app from here. Once installed, you can manage different Shell profiles there (Cmd, PowerShell, pwsh, WSL, etc.). To add the Intel oneAPI profile there,

  • Open the app, and from the dropdown menu on the top bar, click on Settings.

  • Now scroll down along the left-side menu bar, click Add a new Profile, and select the option Empty New Profil. Name the new profile as Intel oneAPI and add the following following to the Command Line option.

    cmd.exe /k ""C:\Program Files (x86)\Intel\oneAPI\setvars.bat" intel64 vs2019" && pwsh
    

    Make sure you have the correct version of Visual Studio instead of vs2019. Other options to set the new profile are trivial. Now you can save the profile. When you start the profile from the Terminal app, this will execute the Windows command prompt first, then it will set the environment variables for Intel oneAPI and Visual Studio, and then run the new PowerShell 7 (pwsh.exe). You can now compile and build programs using Intel Fortran in a modern terminal environment.




Enjoy Reading This Article?

Here are some more articles you might like to read next:

  • Linking Intel oneAPI and Visual Studio with Abaqus
  • Getting started with FEniCSx
  • Licensing and citing academic software and tutorials
  • LaTeX in MS PowerPoint and VS Code on macOS
  • Relearning mathematics