Intel Compiler, MKL, MPI

Last update: 27 Nov. 2024

The Intel compiler is one of the software components of the Intel oneAPI base & HPC toolkit. This page mainly explains how to compile programming languages Fortran, C, and C++, and how to create scripts for batch processing.

Available users

Module

Commands

Commands of Intel Compiler for compilation and linkage are as follows:

*1 Recommended option for creating optimized load modules.
*2 OpenMP options are disabled by default.
*3 Please load the MPI library module for MPI parallel.

Compilation and Linkage

Setup Environment (module)

module command should be executed at the compile time and the execution time.

OneAPI 2024.1

$ module load intel/2024.1

OneAPI 2023.2

$ module load intel

Intel MPI

$ module load intel
$ module load impi

C/C++

Use icc / icpc commands for C/C++ programs without MPI, and mpiicc / mpiicpc commands for C/C++ programs with MPI. (Following examples show commands for C programs. Change the command to icpx and mpiicpx when you compile C++ programs.)

Example 1) Serial C program

$ icc sample.c

Example 2) Thread parallel C program with OpenMP

$ icc -qopenmp sample.c

Example 3) Process parallel C program with MPI

$ mpiicc sample.c

Example 4) Hybrid parallel C program with OpenMP and MPI

$ mpiicc -qopenmp sample.c

Fortran

Use ifort command for compiling and linking Fortran programs. To compile parallel programs with MPI, use mpiifort command.

Example 1) Serial Fortran program

$ ifort ample.f90

Example 2) Thread parallel Fortran program with OpenMP

$ ifort -qopenmp sample.f90

Example 3) Process parallel Fortran program with MPI

$ mpiifort sample.f90

Example 4) Hybrid parallel Fortran program with OpenMP and MPI

$ mpiifort -qopenmp sample.f90

Frequently Used Options

This option should be specified for the programs that chose MPI_THREAD_FUNNELED, MPI_THREAD_SERIALIZED or MPI_THREAD_MULTIPLE as the thread mode.
If -qopenmp or -parallel is specified, thread-safe MPI library will be linked without this options. |
| -ilp64（for Intel MPI） | Assume all integers appear in MPI functions as 64bit integer. (Supports ILP64)
Used in the case -i8 option is specified for Fortran programs. |
| -free or -nofixed（Fortran only） | Notify that the source code is written in free format. |
| -fixed or -nofree（Fortran only） | Notify that the source code is written in fixed format. |
| -extend-source number | Notify that the source code is written in fiexed format with the line length of number. Default value of number is 132. |

Refer to the document of “Intel HPC Toolkit Documentation” for detailed information about options.

Intel Math Kernel Library

Intel Math Kernel Library (MKL) is a library for numerical operations tuned for Intel processors. Specify -qmkl option to link this library.
This option accepts additional sub-options. -qmkl=sequtntial links the non-parallel library. -qmkl=parallel links the thread-parallelized library. -qmkl=cluster links the MPI-parallelized library, such as ScaLAPACK, FFT and direct sparse solvers. -qmkl without sub-option means -qmkl=parallel.

MKL has the following functions:

BLAS, LAPACK, ScaLAPACK, BLACS, PBLAS, Sparse BLAS, sparse matrix arithmetics (includes PARDISO), Fourier Transform, partial differential equations, non-linear optimization solvers, data-fitting functions, GNU multiple precision (GMP) functions, vectorized mathematical library (VML), statistical functions (including pseudo random number generator), etc.

Batch Job

On subsystems A / B of GENKAI, programs should be executed as Batch Jobs.

Refer to the following page for details about batch jobs:
Batch Job

Followings are examples of scripts for executing programs compiled by Intel Compiler as batch jobs.

Example 1) Serial Program

#!/bin/bash

#PJM -L rscgrp=a-batch
#PJM -L vnode-core=1
#PJM -L elapse=0:10:00
#PJM -j

module load intel

./a.out

• 1 node, 1 process, 1 core（1 thread）
• Maximum Execution Time: 10 minutes
• Store both standard output and standard error into the same file

Example 2) Thread Parallel Program (OpenMP)

#!/bin/bash

#PJM -L rscgrp=a-batch
#PJM -L vnode-core=16
#PJM -L elapse=1:00:00
#PJM -j

module load intel
export OMP_NUM_THREADS=16

./a.out

• 1 node, 1 process, 1 core（16 threads）
• Maximum Execution Time: 1 hour

Example 3) Process Parallel Program (MPI)

#!/bin/bash

#PJM -L rscgrp=a-batch
#PJM -L vnode-core=16
#PJM --mpi proc=16
#PJM -L elapse=1:00:00
#PJM -j

module load intel
module load impi

mpiexec ./a.out

• 1node, 16 processes

Example 4) Hybrid Parallel Programs (MPI + OpenMP)

#!/bin/bash

#PJM -L rscgrp=a-batch
#PJM -L vnode-core=32
#PJM --mpi proc=4
#PJM -L elapse=2:00:00
#PJM -j

module load intel
module load impi
export OMP_NUM_THREADS=8

mpiexec ./a.out

• 1 node, 4 processes, 8 threads

Example 5) Process Parallel Program (MPI) with single node (exclusive use)

#!/bin/bash

#PJM -L rscgrp=a-batch
#PJM -L node=1
#PJM --mpi proc=32
#PJM -L elapse=2:00:00
#PJM -j

module load intel
module load impi

mpiexec ./a.out

Example 6) Process Parallel Program (MPI) with multiple nodes (exclusive use)

#!/bin/bash

#PJM -L rscgrp=a-batch
#PJM -L node=2
#PJM --mpi proc=120
#PJM -L elapse=2:00:00
#PJM -j

module load intel
module load impi

mpiexec ./a.out

References

• Intel HPC Toolkit Documentation