Developer Guide

Developer Guide for Intel® oneAPI Math Kernel Library Linux*

Download PDF
ID 766690
Date 6/24/2024
Public

A newer version of this document is available. Customers should click here to go to the newest version.

Document Table of Contents
Document Table of Contents x
Developer Guide for Intel® oneAPI Math Kernel Library for Linux*
Developer Guide for Intel® oneAPI Math Kernel Library for Linux* x
Getting Help and Support What's New Notational Conventions Related Information Getting Started Structure of the Intel® oneAPI Math Kernel Library Linking Your Application with the Intel® oneAPI Math Kernel Library Managing Performance and Memory Language-specific Usage Options Obtaining Numerically Reproducible Results Coding Tips Managing Output Working with the Intel® oneAPI Math Kernel Library Cluster Software Managing Behavior of the Intel® oneAPI Math Kernel Library with Environment Variables Configuring Your Integrated Development Environment to Link with Intel® oneAPI Math Kernel Library Intel® oneAPI Math Kernel Library Benchmarks Appendix A: Intel® oneAPI Math Kernel Library Language Interfaces Support Appendix B: Support for Third-Party Interfaces Appendix C: Directory Structure in Detail Notices and Disclaimers
Getting Started x
Shared Library Versioning CMake Config for oneMKL Checking Your Installation Setting Environment Variables Compiler Support Using Code Examples What You Need to Know Before You Begin Using the Intel® oneAPI Math Kernel Library
Setting Environment Variables x
Scripts to Set Environment Variables Modulefiles to Set Environment Variables Automating the Process of Setting Environment Variables Using the CMake Config File
Structure of the Intel® oneAPI Math Kernel Library x
Architecture Support High-Level Directory Structure Layered Model Concept
Linking Your Application with the Intel® oneAPI Math Kernel Library x
Linking Quick Start Linking Examples Linking in Detail Building Custom Shared Objects
Linking Quick Start x
Using the -qmkl Compiler Option Using the -qmkl-ilp64 Compiler Option Using the Single Dynamic Library Selecting Libraries to Link with Using the Link-line Advisor Using the Command-Line Link Tool
Linking Examples x
Linking on IA-32 Architecture Systems Linking on Intel(R) 64 Architecture Systems
Linking in Detail x
Listing Libraries on a Link Line Dynamically Selecting the Interface and Threading Layer Linking with Interface Libraries Linking with Threading Libraries Linking with Computational Libraries Linking with Compiler Run-time Libraries Linking with System Libraries
Linking with Interface Libraries x
Using the ILP64 Interface vs. LP64 Interface Linking with Fortran 95 Interface Libraries
Building Custom Shared Objects x
Using the Custom Shared Object Builder Composing a List of Functions Specifying Function Names Distributing Your Custom Shared Object
Managing Performance and Memory x
Improving Performance with Threading Improving Performance for Small Size Problems Other Tips and Techniques to Improve Performance Using Memory Functions
Improving Performance with Threading x
OpenMP* Threaded Functions and Problems Functions Threaded with Intel® Threading Building Blocks Avoiding Conflicts in the Execution Environment Techniques to Set the Number of Threads Setting the Number of Threads Using an OpenMP* Environment Variable Changing the Number of OpenMP* Threads at Run Time Using Additional Threading Control Calling oneMKL Functions from Multi-threaded Applications Using Intel® Hyper-Threading Technology Managing Multi-core Performance Managing Performance with Heterogeneous Cores
Using Additional Threading Control x
oneMKL-specific Environment Variables for OpenMP Threading Control MKL_DYNAMIC MKL_DOMAIN_NUM_THREADS MKL_NUM_STRIPES Setting the Environment Variables for Threading Control
Improving Performance for Small Size Problems x
Using MKL_DIRECT_CALL in C Applications Using MKL_DIRECT_CALL in Fortran Applications Limitations of the Direct Call
Other Tips and Techniques to Improve Performance x
Coding Techniques Improving oneMKL Performance on Specific Processors Operating on Denormals
Using Memory Functions x
Avoiding Memory Leaks in oneMKL Redefining Memory Functions
Language-specific Usage Options x
Using Language-Specific Interfaces with Intel® oneAPI Math Kernel Library Mixed-language Programming with the Intel Math Kernel Library
Using Language-Specific Interfaces with Intel® oneAPI Math Kernel Library x
Interface Libraries and Modules Fortran 95 Interfaces to LAPACK and BLAS Compiler-dependent Functions and Fortran 90 Modules
Mixed-language Programming with the Intel Math Kernel Library x
Calling LAPACK, BLAS, and CBLAS Routines from C/C++ Language Environments Using Complex Types in C/C++ Calling BLAS Functions That Return the Complex Values in C/C++ Code
Obtaining Numerically Reproducible Results x
Getting Started with Conditional Numerical Reproducibility Specifying Code Branches Reproducibility Conditions Setting the Environment Variable for Conditional Numerical Reproducibility Code Examples
Coding Tips x
Example of Data Alignment Using Predefined Preprocessor Symbols for Intel® MKL Version-Dependent Compilation
Managing Output x
Using oneMKL Verbose Mode
Using oneMKL Verbose Mode x
Version Information Line Call Description Line
Working with the Intel® oneAPI Math Kernel Library Cluster Software x
Linking with oneMKL Cluster Software Setting the Number of OpenMP* Threads Using Shared Libraries Setting Environment Variables on a Cluster Interaction with the Message-passing Interface Using a Custom Message-Passing Interface Examples of Linking for Clusters
Examples of Linking for Clusters x
Examples for Linking a C Application Examples for Linking a Fortran Application
Managing Behavior of the Intel® oneAPI Math Kernel Library with Environment Variables x
Managing Behavior of Function Domains with Environment Variables Instruction Set–Specific Dispatching on Intel® Architectures
Managing Behavior of Function Domains with Environment Variables x
Setting the Default Mode of Vector Math with an Environment Variable Managing Performance of the Cluster Fourier Transform Functions Managing Invalid Input Checking in LAPACKE Functions
Configuring Your Integrated Development Environment to Link with Intel® oneAPI Math Kernel Library x
Configuring the Eclipse* IDE CDT to Link with oneMKL
Intel® oneAPI Math Kernel Library Benchmarks x
Intel Optimized LINPACK Benchmark for Linux* Intel® Distribution for LINPACK* Benchmark and Intel® Optimized HPL-AI* Benchmark Intel® Optimized High Performance Conjugate Gradient Benchmark
Intel Optimized LINPACK Benchmark for Linux* x
Contents of the Intel® Optimized LINPACK Benchmark Running the Software Known Limitations of the Intel® Optimized LINPACK Benchmark
Intel® Distribution for LINPACK* Benchmark and Intel® Optimized HPL-AI* Benchmark x
Overview of the Intel® Distribution for LINPACK* Benchmark Overview of the Intel® Optimized HPL-AI* Benchmark Contents of the Intel® Distribution for LINPACK* Benchmark and the Intel® Optimized HPL-AI* Benchmark Building the Intel® Distribution for LINPACK* Benchmark and the Intel® Optimized HPL-AI* Benchmark for a Customized MPI Implementation Building the Netlib HPL from Source Code Configuring Parameters Ease-of-use Command-line Parameters Running the Intel® Distribution for LINPACK* Benchmark and the Intel® Optimized HPL-AI* Benchmark Heterogeneous Support in the Intel® Distribution for LINPACK* Benchmark Example Environment Variables Improving Performance of Your Cluster
Intel® Optimized High Performance Conjugate Gradient Benchmark x
Overview of the Intel Optimized HPCG Versions of the Intel® CPU Optimized HPCG Versions of the Intel® GPU Optimized HPCG Getting Started with Intel® CPU Optimized HPCG Getting Started with Intel® GPU Optimized HPCG Choosing the Best Configuration and Problem Sizes for CPUs Choosing the Best HPCG Configuration for GPUs
Appendix A: Intel® oneAPI Math Kernel Library Language Interfaces Support x
Language Interfaces Support, by Function Domain Include Files
Appendix B: Support for Third-Party Interfaces x
FFTW Interface Support
Appendix C: Directory Structure in Detail x
Detailed Structure of the IA-32 Architecture Directories Detailed Structure of the Intel® 64 Architecture Directories
Detailed Structure of the IA-32 Architecture Directories x
Static Libraries in the lib32 Directory Dynamic Libraries in the lib32 Directory
Detailed Structure of the Intel® 64 Architecture Directories x
Static Libraries in the lib Directory Dynamic Libraries in the lib Directory
Developer Guide for Intel® oneAPI Math Kernel Library for Linux*

Heterogeneous Support in the Intel® Distribution for LINPACK* Benchmark

Intel® Distribution for LINPACK Benchmark achieves heterogeneous support by distributing the matrix data unequally between the nodes. The heterogeneous factor command-line parameter f controls the amount of work to be assigned to the more powerful nodes, while the command-line parameter c controls the number of process columns for the faster nodes:

./xhpl –n <problem size> –b <block size> –p <grid row dimn> –q <grid column dimn> –f <heterogeneous factor> –c <number of faster processor columns>

If the heterogeneous factor is 2.5, roughly 2.5 times the work will be put on the more powerful nodes. The more work you put on the more powerful nodes, the more memory you might be wasting on the other nodes if all nodes have equal amount of memory. If your cluster includes many different types of nodes, you may need multiple heterogeneous factors.

Let P be the number of rows and Q the number of columns in your processor grid (PxQ). The work must be homogeneous within each processor column because vertical operations, such as pivoting or panel factorization, are synchronizing operations. When there are two different types of nodes, use MPI to process all the faster nodes first and make sure the "PMAP process mapping" (line 9) of HPL.dat is set to 1 for Column-major mapping. Because all the nodes must be the same within a process column, the number of faster nodes must always be a multiple of P, and you can specify the faster nodes by setting the number of process columns C for the faster nodes with the c command-line parameter. The -f 1.0 -c 0 setting corresponds to the default homogeneous behavior.

To understand how to choose the problem size N for a heterogeneous run, first consider a homogeneous system, where you might choose N as follows:

N ~= sqrt(Memory Utilization * P * Q * Memory Size in Bytes / 8)

Memory Utilization is usually around 0.8 for homogeneous Intel® Xeon® processor systems. On a heterogeneous system, you may apply a different formula for N for each set of nodes that are the same and then choose the minimum N over all sets. Suppose you have a cluster with only one heterogeneous factor F and the number of processor columns (out of the total Q) in the group with that heterogeneous factor equal to C. That group contains P*C nodes. First compute the sum of the parts: S =F*P*C + P*(Q-C). Note that on a homogeneous system S=P*Q,F=1, and C=Q. Take N as

N ~= sqrt(Memory Utilization * P * Q * ((F*P*C)/S) * Memory Size in Bytes / 8)

or simply scale down the value of N for the homogeneous system by sqrt(F*P*C/S).

Example

Suppose the cluster has 100 nodes each having 64 GB of memory, and 20 of the nodes are 2.7 times as powerful as the other 80. Run one MPI process per node for a total of 100 MPI processes. Assume a square processor grid P=Q=10, which conveniently divides up the faster nodes evenly. Normally, the HPL documentation recommends choosing a matrix size that consumes 80 percent of available memory. If N is the size of the matrix, the matrix consumes 8N^2/(P*Q) bytes. So a homogeneous run might look like:

./xhpl –n 820000 –b 256 –p 10 –q 10

If you redistribute the matrix and run the heterogeneous Intel® Distribution for LINPACK Benchmark, you can take advantage of the faster nodes. But because some of the nodes will contain 2.7 times as much data as the other nodes, you must shrink the problem size (unless the faster nodes also happen to have 2.7 times as much memory). Instead of 0.8*64GB*100 total memory size, we have only 0.8*64GB*20 + 0.8*64GB/2.7*80 total memory size, which is less than half the original space. So the problem size in this case would be 526000. Because P=10 and there are 20 faster nodes, two processor columns are faster. If you arrange MPI to send these nodes first to the application, the command line looks like:

./xhpl –n 526000 –b 1024 –p 10 –q 10 –f 2.7 –c 2

The m parameter may be misleading for heterogeneous calculations because it calculates the problem size assuming all the nodes have the same amount of data.

WARNING:

The number of faster nodes must be C*P. If the number of faster nodes is not divisible by P, you might not be able to take advantage of the extra performance potential by giving the faster nodes extra work.

While it suffices to simply provide f and c command-line parameters if you need only one heterogeneous factor, you must add lines to the HPL.dat input to support multiple heterogeneous factors. For the above example (two processor columns have nodes that are 2.7 times faster), instead of passing f and c command-line parameters you can modify the HPL.dat input file by adding these two lines to the end: 

1         number of heterogeneous factors
0 1 2.7   [start_column, stop_column, heterogeneous factor for that range]
NOTE:

Numbering of processor columns starts at 0. The start and stopping numbers must be between 0 and Q-1 (inclusive).

If instead there are three different types of nodes in a cluster and you need at least two heterogeneous factors, change the number in the first row above from 1 to 2 and follow that line with two lines specifying the start column, stopping column, and heterogeneous factor.

When choosing parameters for heterogeneous support in HPL.dat, primarily focus on the most powerful nodes. The larger the heterogeneous factor, the more balanced the cluster may be from a performance viewpoint, but the more imbalanced from a memory viewpoint. At some point, further performance balancing might affect the memory too much. If this is the case, try to reduce any changes done for the faster nodes (such as in block sizes). Experiment with values in HPL.dat carefully because wrong values may greatly hinder performance.

When tuning on a heterogeneous cluster, do not immediately attempt a heterogeneous run, but do the following:

  1. Break the cluster down into multiple homogeneous clusters.

  2. Make heterogeneous adjustments for performance balancing. For instance, if you have two different sets of nodes where one is three times as powerful as the other, it must do three times the work.

  3. Figure out the approximate size of the problem (per node) that you can run on each piece.

  4. Do some homogeneous runs with those problem sizes per node and the final block size needed for the heterogeneous run and find the best parameters.

  5. Use these parameters for an initial heterogeneous run.

Product and Performance Information

Performance varies by use, configuration and other factors. Learn more at www.Intel.com/PerformanceIndex.

Notice revision #20201201

Parent topic: Intel® Distribution for LINPACK* Benchmark and Intel® Optimized HPL-AI* Benchmark
Level Two Title