kerncraft

Loop Kernel Analysis and Performance Modeling Toolkit

This tool allows automatic analysis of loop kernels using the Execution Cache Memory (ECM) model, the Roofline model and actual benchmarks. kerncraft provides a framework to investigate the data reuse and cache requirements by static code analysis. In combination with the Intel IACA tool kerncraft can give a good overview of both in-core and memory bottlenecks and use that data to apply performance models.

For a detailed documentation see publications in doc/.

Installation

On most systems with python pip and setuputils installed, just run:

pip install --user kerncraft

for the latest release. In order to get the Intel Achitecture Code Analyzer (IACA), required by the ECM, ECMCPU and RooflineASM performance models, read this and run:

iaca_get --I-accept-the-Intel-What-If-Pre-Release-License-Agreement-and-please-take-my-soul

Additional requirements are:

• likwid (used in Benchmark model and by likwid_bench_auto)

Usage

1. Get an example kernel and machine file from the examples directory

wget https://raw.githubusercontent.com/RRZE-HPC/kerncraft/master/examples/machine-files/SandyBridgeEP_E5-2680.yml

wget https://raw.githubusercontent.com/RRZE-HPC/kerncraft/master/examples/kernels/2d-5pt.c

1. Have a look at the machine file and change it to match your targeted machine (above we downloaded a file for a Sandy Bridge EP machine)
2. Run kerncraft

kerncraft -p ECM -m SandyBridgeEP_E5-2680.yml 2d-5pt.c -D N 10000 -D M 10000 add -vv for more information on the kernel and ECM model analysis.

Citations

When using Kerncraft for your work, please consider citing the following publication:

Kerncraft: A Tool for Analytic Performance Modeling of Loop Kernels (preprint)

J. Hammer, J. Eitzinger, G. Hager, and G. Wellein: Kerncraft: A Tool for Analytic Performance Modeling of Loop Kernels. In: Tools for High Performance Computing 2016, ISBN 978-3-319-56702-0, 1-22 (2017). Proceedings of IPTW 2016, the 10th International Parallel Tools Workshop, October 4-5, 2016, Stuttgart, Germany. Springer, Cham. DOI: 10.1007/978-3-319-56702-0_1, Preprint: arXiv:1702.04653``

Credits

Implementation: Julian Hammer;
ECM Model (theory): Georg Hager, Holger Stengel, Jan Treibig;
LC generalization: Julian Hammer

License

AGPLv3