AOCL on SeaWulf

Using AMD AOCL on SeaWulf

Overview

AMD Optimizing CPU Libraries (AOCL) is a set of performance-optimized math libraries tailored for scientific and engineering applications on AMD CPUs. AOCL includes routines for linear algebra, FFTs, random number generation, and more.

Available Modules on SeaWulf

Haswell Nodes

  • aocl/4.2.0
  • aocl/4.0.0
  • aocl/3.1.0

Loading a Module

To use AOCL, load the desired version using the module load command. For example:

module load aocl/4.2.0

Choose the version based on compatibility with your code and performance goals.

Using AMD AOCL

Once loaded, AOCL libraries can be linked into your application to accelerate computations. For example:

gcc -o my_program my_program.c -laocl

Make sure to consult the AOCL documentation for detailed linking instructions based on the specific library (e.g., BLIS, LibFLAME, FFTW).

Important Notes

AOCL libraries are optimized for AMD EPYC processors and are tuned to deliver high performance on supported hardware. Use them in place of generic libraries where possible.

Incorrect linking or using an incompatible library version may result in build errors or suboptimal performance. Double-check AOCL compatibility with your application and compiler.

Additional Resources

For help using AOCL or questions about optimization, refer to the official AMD AOCL documentation or contact HPC support for assistance.