hpac / elaps

Experimental Linear Algebra Performance Studies

ELAPS

Experimental Linear Algebra Performance Studies

Table of Contents generated with DocToc

• What is ELAPS?
• Requirements
• Overview
• Installation

What is ELAPS?

ELAPS is a multi-platform open source environment for fast yet powerful experimentation with dense linear algebra kernels, algorithms, and libraries. It allows to construct experiments to investigate how performance and efficiency vary depending on factors such as caching, algorithmic parameters, problem size, and parallelism. Experiments are designed either through Python scripts or a specialized GUI, and run on the whole spectrum of architectures, ranging from laptops to clusters, accelerators, and supercomputers. The resulting experiment reports provide various metrics and statistics that can be analyzed both numerically and visually.

Requirements

• C/C++ compiler

• Python version 2.7.x

• PyQt4

• Matplotlib

• Kernels to measure (e.g. BLAS/LAPACK libraries)

Overview

The Framework consists for three layers:

• The first, "bottom" layer is written in C/C++ and contains the Sampler, a low-level command line tool responsible for executing and timing individual kernels. The Sampler has to be compiled for each specific combination of hardware and libraries (the only stage in which the user needs to configure the system); ELAPS can interface with any number of Samplers. See docs/Sampler.md.

  The Sampler comes with a set of utility routines that cover basic tasks of experiment setups, such as matrix initializations and file-I/O. See docs/Utility.md.

• The second, "middle" layer is the Python library elaps, which centers around the class Experiment that implements the previously introduced experiments. An Experiment can be executed on different Samplers, both locally or through job submission systems. The outcome is a Report, which provides not only structured access to the individual measurements, but also functionality to analyze different metrics and statistics. See docs/Experiment.md and docs/Report.md.

  This layer also includes the plot module, which is based on the matplotlib library, and is used to easily visualize Reports in graphical form. See docs/plot.md.

• The third, "top" layer adds a graphical user interface, written in PyQt4, to both design Experiments in the PlayMat and study Reports and plots in the Viewer. See docs/PlayMat.md and docs/Viewer.md.

Installation

To use ELAPS:

• clone the GitHub repository, and
• compile one or more Samplers (see docs/Sampler.md).

Now, ELAPS is ready to go. To get started, run bin/PlayMat.

elaps's Issues

Viewer: wrong efficiency calculation when using omp

dgemm with "#omp for j=1:10" and 4 threads give about 25Gflops/s, sequential is about 7.5
displayed efficiency for the omp case is at ~7.8% which should rather be ~78%

PlayMat prints "[<class 'elaps.symbolic.Plus'>]" in matrix size field instead of e.g. "i+j"

error occured on local machine as well as on the RWTH cluster-linux with all required modules loaded (python/2.7.9)

PlayMat converts i-1 to [<class 'elaps.symbolic.Plus'>] for BasicCalls

Expressions that contains '+' or '-' are converted to [<class 'elaps.symbolic.Plus'>] in the *.calls file for BasicCall routines like dorgtr or dsycon.

PlayMat zsyr2k turns into csyr2k

PlayMat job error when there are whitespaces in current path

*.err file shows something like:

bash: Zeile 5: /home/[...]/[dirname]: Datei oder Verzeichnis nicht gefunden
ERROR 127

where [dirname] is only showing the part of the actual directory name up to the first whitespace
removing all whitespaces from the current path and rebuilding the sampler fixed this issue