<div dir="ltr">Hi,<div><br></div><div>I think that the story that Simon is telling in that abstract is really well explained and would certainly grab my attention and make me want to see the talk :)</div><div><br></div><div>
I do agree with James that we might want to be careful about how general we make the 2x claim. In your abstact you are careful to say there are some 'examples' of this, but the natural question arises about the generality of this improvement. It would be good to include some average improvement figures in the talk to ward this problem off, even if the abstract remains unchanged.</div>
<div><br></div><div>Simon</div></div><div class="gmail_extra"><br><br><div class="gmail_quote">On 29 August 2014 10:30, James Pallister <span dir="ltr"><<a href="mailto:James.Pallister@bristol.ac.uk" target="_blank">James.Pallister@bristol.ac.uk</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<div bgcolor="#FFFFFF" text="#000000">
<div>Hi Simon,<div class=""><br>
<br>
<blockquote type="cite">We will present the latest performance
results, which show<br>
that optimizing for energy is a double win, because such
programs are almost<br>
always much faster, and will include examples where energy usage
is halved and<br>
speed doubled compared to -O3.</blockquote></div>
Do we know if MAGEEC can find these sets of optimizations? I can
see a few complaints with saying we can double speed (assuming
that this is coming from the genetic algorithm stuff I did a while
ago). It was actually a genetic algorithm, made heavy use of
optimization ordering and it was with an 'old' version of LLVM
(3.1 or 3.2). Also it was on a Cortex-A8, rather than the embedded
platforms we're targeting.<br>
<br>
It might be better to say that we've found cases where the
potential is almost 3 times better than -O3, and we expect MAGEEC
to get closer to that?<br>
<br>
I've attached the graph.<br>
<br>
Cheers,<br>
James<div><div class="h5"><br>
<br>
On 27/08/14 08:26, Simon Cook wrote:<br>
</div></div></div>
<blockquote type="cite"><div><div class="h5">
<div dir="ltr">
<div>
<div>
<div>Hi all,<br>
<br>
</div>
Jeremy and I have been working on an abstract for a talk at
the upcoming LLVM conference. Here is what we have so far.<br>
<br>
"Title: Machine Guided Compilation and Compiling to Minimize
Energy Usage<br>
<br>
Compilers currently optimize for execution time and code
size. But in a world<br>
of diminishing battery life, ever larger data centers and
nanoscale energy<br>
scavenging devices, what if the compiler could instead
optimize for energy<br>
efficiency of the compiled code? Imagine having "-Oe" as an
option to LLVM.<br>
<br>
Since June 2013 Embecosm and the University of Bristol,
supported by the UK<br>
Technology Strategy Board have developed the Machine Guided
Energy Efficient<br>
Compliation system. MAGEEC (<a href="http://www.mageec.org" target="_blank">www.mageec.org</a>)
uses machine learning to train a<br>
compiler on the combinations of compiler optimizations which
will minimize the<br>
energy consumed by the compiled program. The original target
of the project was<br>
to reduce the energy of compiled code by 40% compared to
optimized code today.<br>
<br>
Unlike previous projects, MAGEEC is completely generic,
capable of working with<br>
any compiler and target. It can optimize for any criterion,
not just energy<br>
efficiency as we have used, and can be integrated with any
machine learning<br>
approach. The project is entirely free and open source, with
the code and all<br>
project design materials freely available online.<br>
<br>
In July 2014, the first complete version of the MAGEEC
infrastructure was<br>
released, supporting both GCC and LLVM, and is now
undergoing refinement and<br>
evaluation. To support energy efficiency optimization, the
project has<br>
developed its own low-cost (approx $50) energy measurement
board, capable of<br>
sampling voltage and current up to 6 million times per
second to an accuracy of<br>
better than 1%. Training also requires a large pool of
benchmark programs, and<br>
as part of this project we have also developed a new
standard free and open<br>
source benchmark suite, BEEBS (<a href="http://www.beebs.eu" target="_blank">www.beebs.eu</a>),
suitable for use with low energy<br>
deeply embedded systems.<br>
<br>
In this talk we will explain the detailed design of MAGEEC,
comparing the<br>
challenges in developing for LLVM and GCC. This will
include the details of<br>
the integration with the machine learning system, allowing
it to control the<br>
pass manager, and which allows us to select the best
optimizations on a per<br>
function basis. We will present the latest performance
results, which show<br>
that optimizing for energy is a double win, because such
programs are almost<br>
always much faster, and will include examples where energy
usage is halved and<br>
speed doubled compared to -O3.<br>
<br>
We will conclude with a demonstration of MAGEEC in action on
ARM processors,<br>
using the energy measurement board to measure the energy
savings achieved."<br>
<br>
</div>
<br>
We're attempting to cover the entire history of the work we've
done so far, as we don't have the benefit of having done a
couple of talks previously for background information. If
anyone has any thoughts before we submit this, that would be
great. In particular, James, the graphs you had a few months
ago with data from trying different passes with opt, have we
quoted have the correct decrease in energy/execution time?<br>
<br>
</div>
Many thanks,<br>
Simon<br>
</div>
<br>
<fieldset></fieldset>
<br>
</div></div><pre>_______________________________________________
mageec mailing list
<a href="mailto:mageec@mageec.org" target="_blank">mageec@mageec.org</a>
<a href="http://mageec.org/cgi-bin/mailman/listinfo/mageec" target="_blank">http://mageec.org/cgi-bin/mailman/listinfo/mageec</a>
</pre>
</blockquote>
<br>
</div>
<br>_______________________________________________<br>
mageec mailing list<br>
<a href="mailto:mageec@mageec.org">mageec@mageec.org</a><br>
<a href="http://mageec.org/cgi-bin/mailman/listinfo/mageec" target="_blank">http://mageec.org/cgi-bin/mailman/listinfo/mageec</a><br>
<br></blockquote></div><br></div>