Intel® Compiler Base Tutorial
- RIS-SVC-conf-publish
Overview
The purpose of this tutorial is to demonstrate usage of the Intel® Compiler Base image for use on the Scientific Compute Platform. Please refer to this page for more information on the Intel® Compiler Base image. In this tutorial, we will be compiling a sample MPI-enabled Hello World program. The tutorial uses material from this page.
Requirements
-
Docker Desktop (https://www.docker.com/products/docker-desktop)
-
Free Docker Hub Account (https://hub.docker.com)
Multi-Stage Build
This section of the tutorial will demonstrate how to build a multi-stage Docker image for use on the Scientific Compute Platform. An MPI-enabled Hello World program is compiled using the Intel® Compiler Base image.
MPI Hello World Code Example
Below is sample code for a simple MPI-enabled Hello World program.
#include <mpi.h>
#include <stdio.h>
int main(int argc, char** argv) {
// Initialize the MPI environment
MPI_Init(NULL, NULL);
// Get the number of processes
int world_size;
MPI_Comm_size(MPI_COMM_WORLD, &world_size);
// Get the rank of the process
int world_rank;
MPI_Comm_rank(MPI_COMM_WORLD, &world_rank);
// Get the name of the processor
char processor_name[MPI_MAX_PROCESSOR_NAME];
int name_len;
MPI_Get_processor_name(processor_name, &name_len);
// Print off a hello world message
printf("Hello world from processor %s, rank %d out of %d processors\n",
processor_name, world_rank, world_size);
// Finalize the MPI environment.
MPI_Finalize();
}
Dockerfile for a Multi-Stage Build
Using your favorite text editor, save the above code to a file called mpi_hello_world.c. In the same folder as the mpi_hello_world.c file, create a file called Dockerfile. Add the following lines of code to the Dockerfile file:
# Begin Stage 1 with the base compiler image.
FROM us.gcr.io/ris-appeng-shared-dev/compiler-base:oneapi2021.1.1_centos7 as build
# Additional dependencies for the Intel® MPI C compiler
RUN yum install -y gcc-c++
# Change working directory
WORKDIR /opt/mpi_hello_world/src/
# Copy the source code to the working directory
COPY mpi_hello_world.c .
# Copy source code to a new location inside the container.
RUN cd /opt/mpi_hello_world/src/ && \
. /opt/intel/oneapi/setvars.sh && \
mpiicc -o mpi_hello_world mpi_hello_world.c && \
cp -f mpi_hello_world /usr/local/bin
# Begin Stage 2 with a new base image.
FROM centos:7
# Copy only the needed parts of Stage 1.
COPY --from=build /usr/local/bin/mpi_hello_world /usr/local/bin
# Add any additional runtime dependencies here
RUN yum install -y wget perl numactl-libs \
gtk2 atk cairo gcc-gfortran tcsh libnl3 \
libmnl tcl tk
# Set up MLNX_OFED driver.
ENV MOFED_VERSION 5.4-3.1.0.0
ENV OS_VERSION rhel7.7
ENV PLATFORM x86_64
RUN wget -q http://content.mellanox.com/ofed/MLNX_OFED-${MOFED_VERSION}/MLNX_OFED_LINUX-${MOFED_VERSION}-${OS_VERSION}-${PLATFORM}.tgz && \
tar -xvf MLNX_OFED_LINUX-${MOFED_VERSION}-${OS_VERSION}-${PLATFORM}.tgz && \
MLNX_OFED_LINUX-${MOFED_VERSION}-${OS_VERSION}-${PLATFORM}/mlnxofedinstall --user-space-only --without-fw-update -q --distro ${OS_VERSION} && \
cd .. && \
rm -rf ${MOFED_DIR} && \
rm -rf *.tgz
Building and Pushing the Docker Image
Please see this section for more information on building and pushing the Docker image to Docker Hub.
Interactive Command-Line Session
Job Submission
The following command can be used to submit an interactive command-line job to run the MPI-enabled Hello World program. Make sure to replace docker/image with the name of the Docker image you created.
LSF_DOCKER_VOLUMES="/scratch1/fs1/ris/application/intel/oneapi/:/opt/intel/oneapi" LSF_DOCKER_NETWORK=host LSF_DOCKER_IPC=host bsub -n 20 -Is -q general-interactive -R "affinity[core(1):distribute=pack] span[ptile=4]" -a "docker(docker/image)" /bin/bash
The job submission command uses environment variables required for parallel computing on RIS. The job submission is also requesting 20 vCPUs spread across 5 exec nodes. Please see the parallel computing documentation for more information.
Running the MPI-enabled Hello World program
Run the MPI-enabled Hello World program using 20 vCPUs with the following command:
mpirun -np 20 /usr/local/bin/mpi_hello_world
The following output should be displayed after the image has finished downloading on all exec nodes:
Hello world from processor compute1-exec-98.ris.wustl.edu, rank 12 out of 20 processors Hello world from processor compute1-exec-98.ris.wustl.edu, rank 13 out of 20 processors Hello world from processor compute1-exec-98.ris.wustl.edu, rank 14 out of 20 processors Hello world from processor compute1-exec-98.ris.wustl.edu, rank 15 out of 20 processors Hello world from processor compute1-exec-6.ris.wustl.edu, rank 5 out of 20 processors Hello world from processor compute1-exec-6.ris.wustl.edu, rank 6 out of 20 processors Hello world from processor compute1-exec-6.ris.wustl.edu, rank 4 out of 20 processors Hello world from processor compute1-exec-6.ris.wustl.edu, rank 7 out of 20 processors Hello world from processor compute1-exec-5.ris.wustl.edu, rank 1 out of 20 processors Hello world from processor compute1-exec-218.ris.wustl.edu, rank 8 out of 20 processors Hello world from processor compute1-exec-218.ris.wustl.edu, rank 9 out of 20 processors Hello world from processor compute1-exec-5.ris.wustl.edu, rank 3 out of 20 processors Hello world from processor compute1-exec-218.ris.wustl.edu, rank 10 out of 20 processors Hello world from processor compute1-exec-5.ris.wustl.edu, rank 0 out of 20 processors Hello world from processor compute1-exec-218.ris.wustl.edu, rank 11 out of 20 processors Hello world from processor compute1-exec-5.ris.wustl.edu, rank 2 out of 20 processors Hello world from processor compute1-exec-217.ris.wustl.edu, rank 19 out of 20 processors Hello world from processor compute1-exec-217.ris.wustl.edu, rank 16 out of 20 processors Hello world from processor compute1-exec-217.ris.wustl.edu, rank 17 out of 20 processors Hello world from processor compute1-exec-217.ris.wustl.edu, rank 18 out of 20 processors
Single-Stage Build
Should you require a single-stage build, the following Dockerfile can be used. Please be aware that this method results in a larger Docker image which may cause increased computing time/resources/cost. This method also caches the source code in build layers resulting in public exposure, which may be unwanted. When possible, it is advised to use the multi-stage build method.
# Begin with the base compiler image.
FROM us.gcr.io/ris-appeng-shared-dev/compiler-base:oneapi2021.1.1_centos7 as build
# Additional dependencies for the Intel® MPI C compiler and MLNX
RUN yum install -y gcc-c++ wget perl numactl-libs \
gtk2 atk cairo gcc-gfortran tcsh libnl3 \
libmnl tcl tk
# Change working directory
WORKDIR /opt/mpi_hello_world/src/
# Copy the source code to the working directory
COPY mpi_hello_world.c .
# Copy source code to a new location inside the container.
RUN cd /opt/mpi_hello_world/src/ && \
. /opt/intel/oneapi/setvars.sh && \
mpiicc -o mpi_hello_world mpi_hello_world.c && \
cp -f mpi_hello_world /usr/local/bin
# Set up MLNX_OFED driver.
ENV MOFED_VERSION 5.4-3.1.0.0
ENV OS_VERSION rhel7.7
ENV PLATFORM x86_64
RUN wget -q http://content.mellanox.com/ofed/MLNX_OFED-${MOFED_VERSION}/MLNX_OFED_LINUX-${MOFED_VERSION}-${OS_VERSION}-${PLATFORM}.tgz && \
tar -xvf MLNX_OFED_LINUX-${MOFED_VERSION}-${OS_VERSION}-${PLATFORM}.tgz && \
MLNX_OFED_LINUX-${MOFED_VERSION}-${OS_VERSION}-${PLATFORM}/mlnxofedinstall --user-space-only --without-fw-update -q --distro ${OS_VERSION} && \
cd .. && \
rm -rf ${MOFED_DIR} && \
rm -rf *.tgz