Autotuning under Tight Budget Constraints: a Transparent Design of Experiments Approach

This repository contains the scripts to generate the pdf and the figures in our CCGRID’19 submission. The data produced by our the experiments is stored in two other git repositories:

• GPU Laplacian Kernel
• SPAPT benchmark suite

The Grid5000 scripts we used are stored in this repository:

• Grid5000 configuration & run scripts

Our implementation of DLMT is publicly hosted in GitHub in the following files:

• Implementation in Python2 (Orio)
• Implementation in R

This paper was written and exported to pdf using Emacs Org mode, so you can clone the repositories, run our data analyses, and generate all images from Emacs.

Org Mode LaTex: Exporting IEEEtran

After installing the IEEEtran template, add the following to your Emacs configuration to be able to export using the org-ieeetran class:

(with-eval-after-load 'ox-latex
  (add-to-list 'org-latex-classes
                '("org-ieeetran"
                  "\\documentclass{IEEEtran}"
                  ("\\section{%s}" . "\\section*{%s}")
                  ("\\subsection{%s}" . "\\subsection*{%s}")
                  ("\\subsubsection{%s}" . "\\subsubsection*{%s}")
                  ("\\paragraph{%s}" . "\\paragraph*{%s}")
                  ("\\subparagraph{%s}" . "\\subparagraph*{%s}"))))

You must use the pdflatex compiler to generate the correct template in pdf. You can use latexmk in your Emacs configuration to properly export references:

(setq org-latex-pdf-process (list "latexmk -pdflatex='pdflatex' -pdf -f %f"))