Why oneMKL? Accelerate Math Computation on the Latest Hardware

With 20 years of maturity under its belt, Intel® Math Kernel Library remains the fastest and most-used math library for Intel-based systems and continues to hold this distinction based on continual optimizations that result in best-in-class performance.

This session focuses on its most recent iteration: Intel® oneAPI Math Kernel Library (oneMKL), optimized for implementing fast math-processing routines targeting heterogeneous, multiarchitecture compute.

The session includes:

• How to use oneMKL to take the best advantage of the latest built-in hardware acceleration engines such as Intel® Advanced Vector Extensions 512, Intel® Advanced Matrix Extensions, and the new bfloat16 data type commonly used for machine learning.
• An illustration—with syntax specifics—of how function domains (Basic Linear Algebra Subprograms [BLAS], Linear Algebra Package [LAPACK], fast Fourier transform [FFT], Rnd, PARDISO) take advantage of the 4th gen Intel® Xeon® Scalable processor and Intel® Max Series product family.
• How oneMKL supports the latest OpenMP* standard, expansion into SYCL*, and open-standards-based C++ cross-architectural compute framework.
• Instruction and a demo of how to map CUDA* math library calls (for example, cuBLAS, cuFFT, and cuRAND libraries) to oneMKL

Skill level: Intermediate and expert

Download the stand-alone version of oneMKL or as part of the Intel® oneAPI Base Toolkit.

Code samples (GitHub*):

• Fourier Correlation
• cuBLAS Migration
• Matrix Multiplication
• All oneMKL Samples
• [Bookmark] All oneAPI Code Samples