index.html

<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.0 Transitional//EN" "http://www.w3.org/TR/REC-html40/loose.dtd">
<html xmlns="http://www.w3.org/1999/xhtml" xml:lang="en">
<HEAD>
<meta http-equiv="content-type" content="text/html; charset=UTF-8" />
<TITLE>StarPU</TITLE>
<link rel="stylesheet" type="text/css" href="style.css" />
<link rel="Shortcut icon" href="http://www.inria.fr/extension/site_inria/design/site_inria/images/favicon.ico" type="image/x-icon" />
</HEAD>

<body>

<div class="title">
<h1><a href="./">StarPU</a></h1>
<h2>A Unified Runtime System for Heterogeneous Multicore Architectures</h2>
</div>

<div class="menu">
<a href="http://runtime.bordeaux.inria.fr/">RUNTIME TEAM</a> |
&nbsp; &nbsp; &nbsp;
|
<a href="#overview">Overview</a> |
<a href="#news">News</a> |
<a href="#contact">Contact</a> |
<a href="#features">Features</a> |
<a href="#software">Software</a> |
<a href="#tryit">Try it!</a> |
<a href="#publications">Publications</a> |
<a href="internships/">Jobs/Interns</a> |
<a href="files/">Download</a> |
<a href="tutorials">Tutorials</a> |
<a href="https://wiki.bordeaux.inria.fr/runtime/doku.php?id=starpu">Intranet</a>
</div>

<div class="section" id="overview">
<h3>Overview</h3>
  <p>
<span class="important">StarPU is a task programming library for hybrid architectures</span>
<ol>
<li><b>The application provides algorithms and constraints</b>
    <ul>
    <li>CPU/GPU implementations of tasks</li>
    <li>A graph of tasks, using either the StarPU's high level <b>GCC plugin</b> pragmas or StarPU's rich <b>C API</b></li>
    </ul>
<br>
</li>
<li><b>StarPU handles run-time concerns</b>
    <ul>
    <li>Task dependencies</li>
    <li>Optimized heterogeneous scheduling</li>
    <li>Optimized data transfers and replication between main memory and discrete memories</li>
    <li>Optimized cluster communications</li>
    </ul>
</li>
</ol>
</p>
<p>
<span class="important">Rather than handling low-level issues, <b>programmers can concentrate on algorithmic concerns!</b></span>
</p>

<p>
<span class="note">The StarPU documentation is available in <a href="./doc/starpu.pdf">PDF</a> and in <a href="./doc/html/">HTML</a>.</span> Please note that these documents are up-to-date with the latest release of StarPU.
</p>
</div>

<div class="section emphasize newslist" id="news">
<h3>News</h3>
<p>
August 2016 <b>&raquo;&nbsp;</b><a href="http://gforge.inria.fr/frs/?group_id=1570"><b>The
      v1.1.6 release of StarPU is now available!</b></a>. This release notably brings the concept of
      scheduling contexts which allows to separate computation
      resources.
</p>
<p>
August 2016 <b>&raquo;&nbsp;</b><a href="http://gforge.inria.fr/frs/?group_id=1570"><b>The
      1.2.0 release of StarPU is now available!</b></a>.
      This release notably brings an out-of-core support, a MIC Xeon
      Phi support, an OpenMP runtime support, and a new internal
      communication system for MPI.
</p>
<p>
August 2016 <b>&raquo;&nbsp;</b><a href="http://gforge.inria.fr/frs/?group_id=1570"><b>The
      sixth (and really hopefully last) release candidate of the v1.2.0 release of StarPU is now
      available!</b></a>.
      This release notably brings an out-of-core support, a MIC Xeon
      Phi support, an OpenMP runtime support, and a new internal
      communication system for MPI.
</p>
<p>
March 2016 <b>&raquo;&nbsp;</b> <b>Engineer job offer</b> at Inria: more
details on the job and on how to apply are available <a href="internships/hibox.html">here</a>
</p>
<p>
December 2015 <b>&raquo;&nbsp;</b><a href="http://gforge.inria.fr/frs/?group_id=1570"><b>The
      fifth (and hopefully last) release candidate of the v1.2.0 release of StarPU is now
      available!</b></a>.
      This release notably brings an out-of-core support, a MIC Xeon
      Phi support, an OpenMP runtime support, and a new internal
      communication system for MPI.
</p>
<p>
September 2015 <b>&raquo;&nbsp;</b><a href="http://gforge.inria.fr/frs/?group_id=1570"><b>The
      v1.1.5 release of StarPU is now available!</b></a>. This release notably brings the concept of
      scheduling contexts which allows to separate computation
      resources.
</p>
<p>
August 2015 <b>&raquo;&nbsp;</b><a href="http://gforge.inria.fr/frs/?group_id=1570"><b>The
      fourth release candidate of the v1.2.0 release of StarPU is now
      available!</b></a>.
      This release notably brings an out-of-core support, a MIC Xeon
      Phi support, an OpenMP runtime support, and a new internal
      communication system for MPI.
</p>
<p>
July 2015 <b>&raquo;&nbsp;</b><a href="http://gforge.inria.fr/frs/?group_id=1570"><b>The
      third release candidate of the v1.2.0 release of StarPU is now
      available!</b></a>.
      This release notably brings an out-of-core support, a MIC Xeon
      Phi support, an OpenMP runtime support, and a new internal
      communication system for MPI.
</p>
<p>
May 2015 <b>&raquo;&nbsp;</b><a href="http://gforge.inria.fr/frs/?group_id=1570"><b>The
      second release candidate of the v1.2.0 release of StarPU is now
      available!</b></a>.
      This release notably brings an out-of-core support, a MIC Xeon
      Phi support, an OpenMP runtime support, and a new internal
      communication system for MPI.
</p>
<p>
April 2015 <b>&raquo;&nbsp;</b>A <a href="https://events.prace-ri.eu/event/339/">tutorial</a> on runtime systems including
StarPU will be given at INRIA Bordeaux in June 2015.
</p>
<p>
March 2015 <b>&raquo;&nbsp;</b><a href="http://gforge.inria.fr/frs/?group_id=1570"><b>The
      first release candidate of the v1.2.0 release of StarPU is now
      available!</b></a>.
      This release notably brings an out-of-core support, a MIC Xeon
      Phi support, an OpenMP runtime support, and a new internal
      communication system for MPI.
</p>
<p>
March 2015 <b>&raquo;&nbsp;</b><a href="http://gforge.inria.fr/frs/?group_id=1570"><b>The
      v1.1.4 release of StarPU is now available!</b></a>. This release notably brings the concept of
      scheduling contexts which allows to separate computation
      resources.
</p>
</div>

<div class="section emphasizebot" style="text-align: right; font-style: italic;">
Get the latest StarPU news by subscribing to the <a href="http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/starpu-announce">starpu-announce mailing list</a>.
See also the full <a href="news/">news</a>.
</div>

<div class="section" id="video">
<h3>Video Conference</h3>
<p>
A video recording (26') of a <a href=http://www.x.org/wiki/Events/XDC2014/XDC2014ThibaultStarPU/>presentation at the XDC2014 conference</a> gives an overview of StarPU
(<a href=http://www.x.org/wiki/Events/XDC2014/XDC2014ThibaultStarPU/xdc_starpu.pdf>slides</a>):
</p>
<center>
<iframe width="420" height="315" src="https://www.youtube.com/embed/frsWSqb8UJU" frameborder="0" allowfullscreen></iframe>
</center>
</div>

<div class="section" id="contact">
<h3>Contact</h3>
<p>For any questions regarding StarPU, please contact the StarPU developers mailing list.</p>
<pre>
<a href="mailto:starpu-devel@lists.gforge.inria.fr?subject=StarPU">starpu-devel@lists.gforge.inria.fr</a>
</pre>
</div>

<div class="section" id="features">
<h3>Features</h3>

<h4>Portability</h4>
  <p>
Portability is obtained by the means of a unified abstraction of the machine.
StarPU offers a unified offloadable task abstraction named <em>codelet</em>. Rather
than rewriting the entire code, programmers can encapsulate existing functions
within codelets. In case a codelet can run on heterogeneous architectures, <b>it
is possible to specify one function for each architectures</b> (e.g. one function
for CUDA and one function for CPUs). StarPU takes care of scheduling and
executing those codelets as efficiently as possible over the entire machine, include
multiple GPUs.
One can even specify <b>several functions for each architecture</b> (new in
v1.0) as well as
<b>parallel implementations</b> (e.g. in OpenMP), and StarPU will
automatically determine which version is best for each input size (new in v0.9).
  </p>

<h4>Data transfers</h4>
  <p>
To relieve programmers from the burden of explicit data transfers, a high-level
data management library enforces memory coherency over the machine: before a
codelet starts (e.g. on an accelerator), all its <b>data are automatically made
available on the compute resource</b>. Data are also kept on e.g. GPUs as long as
they are needed for further tasks. When a device runs out of memory, StarPU uses
an LRU strategy to <b>evict unused data</b>. StarPU also takes care of <b>automatically
prefetching</b> data, which thus permits to <b>overlap data transfers with computations</b>
(including <b>GPU-GPU direct transfers</b>) to achieve the most of the architecture.
  </p>

<h4>Dependencies</h4>
  <p>
Dependencies between tasks can be given several ways, to provide the
programmer with best flexibility:
  <ul>
    <li><b>explicitly</b> between pairs of tasks,</li>
    <li>explicitly through <b>tags</b> which act as rendez-vous points between
    tasks (thus including tasks which have not been created yet),</li>
    <li><b>implicitly</b> from RAW, WAW, and WAR data dependencies.</li>
  </ul>
  </p>
  <p>
  These dependencies are computed in a completely decentralized way.
  </p>
  <p>
StarPU also supports an OpenMP-like <a href="doc/html/DataManagement.html#DataReduction">reduction</a> access mode (new in v0.9).
  </p>
  <p>
It also supports a <a href="doc/html/DataManagement.html#DataCommute">commute</a> access mode to allow data access commutativity (new in v1.2).
  </p>

<h4>Heterogeneous Scheduling</h4>
  <p>
StarPU obtains
portable performances by efficiently (and easily) using all computing resources
at the same time. StarPU also takes advantage of the <b>heterogeneous</b> nature of a
machine, for instance by using scheduling strategies based on auto-tuned
performance models. These determine the relative performance achieved
by the different processing units for the various kinds of task, and thus
permits to <b>automatically let processing units execute the tasks they are the best for</b>.
Various strategies and variants are available. Some of them are centralized, but
most of them are <b>completely distributed</b>. dmda (a data-aware MCT strategy,
thus similar to heft but starts executing tasks before the whole task graph is
submitted, thus allowing dynamic task submission and a decentralized scheduler),
eager (dumb centralized queue), decentralized locality-aware work-stealing, ...
The overhead per task is typically around the order of
magnitude of a microsecond. Tasks should thus be a few orders of magnitude
bigger, such as 100 microseconds or 1 millisecond, to make the overhead
negligible.
  </p>

<h4>Clusters</h4>
  <p>
To deal with clusters, StarPU can nicely integrate with <a href="doc/html/MPISupport.html">MPI</a> through
explicit network communications, which will then be <b>automatically combined and
overlapped</b> with the intra-node data transfers and computation. The application
can also just provide the whole task graph, a data distribution over MPI nodes, and StarPU
will automatically determine which MPI node should execute which task, and
<b>generate all required MPI communications</b> accordingly (new in v0.9). We
have gotten excellent scaling on a 144-node cluster with GPUs, we have not yet
had the opportunity to test on a yet larger cluster. We have however measured
that with naive task submission, it should scale to a thousand nodes, and with
pruning-tuned task submission, it should scale to about a <b>million nodes</b>.
  </p>

<h4>Out of core</h4>
  <p>
When memory is not big enough for the working set, one may have to resort to
using disks. StarPU makes this seamless thanks to its <a href="doc/html/OutOfCore.html">out of core support</a> (new in 1.2).
StarPU will <b>automatically evict</b> data from the main memory in advance, and
<b>prefetch back</b> required data before it is needed for tasks.
  </p>

<h4>Extensions to the C Language</h4>
<p>
  StarPU comes with a GCC plug-in
  that <a href="doc/html/cExtensions.html">extends the C programming
  language</a> with pragmas and attributes that make it easy
  to <b>annotate a sequential C program to turn it into a parallel
  StarPU program</b> (new in v1.0).
</p>

<h4>OpenMP 4 -compatible interface</h4>
<p>
  <a href=http://kstar.gforge.inria.fr/>K'Star</a> provides an OpenMP
  4 -compatible interface on top of StarPU. This allows to just rebuild OpenMP
  applications with the K'Star source-to-source compiler, then build it with the
  usual compiler, and the result will use the StarPU runtime.
</p>
<p>
  K'Star also provides some extensions to the OpenMP 4 standard, to let the
  StarPU runtime perform online optimizations.
</p>

<h4>OpenCL-compatible interface</h4>
<p>
  StarPU provides an <a href="doc/html/SOCLOpenclExtensions.html">OpenCL-compatible interface, SOCL</a>
  which allows to simply run OpenCL applications on top of StarPU (new in v1.0).
</p>

<h4>Simulation support</h4>
<p>
  StarPU can very accurately simulate an application execution
  and measure the resulting performance thanks to using the
  <a href="http://simgrid.gforge.inria.fr">SimGrid simulator</a> (new in v1.1).  This allows
  to quickly experiment with various scheduling heuristics, various application
  algorithms, and even various platforms (available GPUs and CPUs, available
  bandwidth)!
</p>

<h4>All in all</h4>
  <p>
All that means that, with the help
of <a href="doc/html/cExtensions.html">StarPU's extensions to the C
language</a>, the following sequential source code of a tiled version of
the classical Cholesky factorization algorithm using BLAS is also valid
StarPU code, possibly running on all the CPUs and GPUs, and given a data
distribution over MPI nodes, it is even a distributed version!
  </p>

  <tt><pre>
for (k = 0; k < tiles; k++) {
  potrf(A[k,k])
  for (m = k+1; m < tiles; m++)
    trsm(A[k,k], A[m,k])
  for (m = k+1; m < tiles; m++)
    syrk(A[m,k], A[m, m])
  for (m = k+1, m < tiles; m++)
    for (n = k+1, n < m; n++)
      gemm(A[m,k], A[n,k], A[m,n])
}</pre></tt>

<h4>Supported Architectures</h4>
<ul>
<li>SMP/Multicore Processors (x86, PPC, ...) </li>
<li>NVIDIA GPUs (e.g. heterogeneous multi-GPU), with pipelined and concurrent kernel execution support (new in v1.2) and GPU-GPU direct transfers (new in 1.1)</li>
<li>OpenCL devices</li>
<li>Cell Processors (experimental)</li>
</ul>
and soon (in v1.2)
<ul>
<li>Intel SCC</li>
<li>Intel MIC / Xeon Phi</li>
</ul>

<h4>Supported Operating Systems</h4>
<ul>
<li>GNU/Linux</li>
<li>Mac OS X</li>
<li>Windows</li>
</ul>

<h4>Performance analysis tools</h4>
  <p>
In order to understand the performance obtained by StarPU, it is helpful to
visualize the actual behaviour of the applications running on complex
heterogeneous multicore architectures.  StarPU therefore makes it possible to
generate Pajé traces that can be visualized thanks to the <a
href="http://vite.gforge.inria.fr/"><b>ViTE</b> (Visual Trace Explorer) open
source tool.</a>
  </p>

<p>
<b>Example:</b> LU decomposition on 3 CPU cores and a GPU using a very simple
greedy scheduling strategy. The green (resp. red) sections indicate when the
corresponding processing unit is busy (resp. idle). The number of ready tasks
is displayed in the curve on top: it appears that with this scheduling policy,
the algorithm suffers a certain lack of parallelism. <b>Measured speed: 175.32
GFlop/s</b>
<center><a href="./images/greedy-lu-16k-fx5800.png"> <img src="./images/greedy-lu-16k-fx5800.png" alt="LU decomposition (greedy)" width="75%"></a></center>
</p>

<p>
This second trace depicts the behaviour of the same application using a
scheduling strategy trying to minimize load imbalance thanks to auto-tuned
performance models and to keep data locality as high as possible. In this
example, the Pajé trace clearly shows that this scheduling strategy outperforms
the previous one in terms of processor usage. <b>Measured speed: 239.60
GFlop/s</b>
<center><a href="./images/dmda-lu-16k-fx5800.png"><img src="./images/dmda-lu-16k-fx5800.png" alt="LU decomposition (dmda)" width="75%"></a></center>
</p>

<p>
<a href="http://www.hlrs.de/temanejo">Temanejo</a> can be used to debug the task
graph, as shown below (new in v1.1).
</p>

<center>
<a href="images/temanejo.png"><img src="images/temanejo.png" width="50%"/></a>
</center>

</div>

<div class="section" id="software">
<h3>Software using StarPU</h3>

<p>
Some software is known for being able to use StarPU to tackle heterogeneous
architectures, here is a non-exhaustive list (feel free to ask to be in the
list!):
</p>

<ul>
	<li><a href="http://icl.cs.utk.edu/magma/">MAGMA</a>, dense linear algebra library, starting from version 1.1</li>
	<li><a href="https://project.inria.fr/chameleon/">Chameleon</a>, dense linear algebra library</li>
	<li><a href="http://www.ida.liu.se/~chrke/skepu/">SkePU</a>, a skeleton programming framework.</li>
	<li><a href="http://pastix.gforge.inria.fr/">PaStiX</a>, sparse linear algebra library, starting from version 5.2.1</li>
	<li><a href="http://buttari.perso.enseeiht.fr/qr_mumps/">qr_mumps</a>, sparse linear algebra library</li>
	<li><a href="http://scalfmm-public.gforge.inria.fr/doc/">ScalFMM</a>, N-body interaction simulation using the Fast Multipole Method. </li>
	<li><a href="https://project.inria.fr/maphys/fr/">MaPHyS</a>, Massively Parallel Hybrid Solver</li>
	<li><a href="https://hal.archives-ouvertes.fr/hal-01086246">SignalPU</a>, a Dataflow-Graph-specific programming model. </li>
</ul>

<p>
You can find below the list of publications related to applications
using StarPU.
</p>

</div>

<div class="section" id="tryit">
<h3>Give it a try!</h3>
<p>
You can easily try the performance on the Cholesky factorization for
instance. Make sure to have the pkg-config and
<a href="http://www.open-mpi.org/projects/hwloc/">hwloc</a>
software installed for
proper CPU control and BLAS kernels for your computation units and configured in
your environment (e.g. MKL for CPUs and CUBLAS for GPUs).
</p>

<tt><pre>
$ wget http://starpu.gforge.inria.fr/files/starpu-someversion.tar.gz
$ tar xf starpu-someversion.tar.gz
$ cd starpu-someversion
$ ./configure
$ make -j 12
$ STARPU_SCHED=dmdas ./examples/cholesky/cholesky_implicit -size $((960*40)) -nblocks 40
$ STARPU_SCHED=dmdas mpirun -np 4 -machinefile mymachines ./mpi/examples/matrix_decomposition/mpi_cholesky_distributed -size $((960*40*4)) -nblocks $((40*4))</pre></tt>

<p>Note that the dmdas scheduler uses performance models, and thus needs
calibration execution before exhibiting optimized performance (until the "model
something is not calibrated enough" messages go away).</p>

<p>To get a glimpse at what happened, you can get an execution trace by
installing
<a href="http://savannah.nongnu.org/projects/fkt">FxT</a>
and <a href="http://vite.gforge.inria.fr/">ViTE</a>, and enabling traces:
</p>

<tt><pre>
$ ./configure --with-fxt
$ make -j 12
$ STARPU_SCHED=dmdas ./examples/cholesky/cholesky_implicit -size $((960*40)) -nblocks 40
$ ./tools/starpu_fxt_tool -i /tmp/prof_file_${USER}_0
$ vite paje.trace
</pre></tt>

<p>
Starting with StarPU 1.1, it is also possible to reproduce the performance that
we show in our articles on our machines, by installing simgrid, and then using
the simulation mode of StarPU using the performance models of our machines:
</p>
  <tt><pre>
$ ./configure --enable-simgrid
$ make -j 12
$ STARPU_PERF_MODEL_DIR=$PWD/tools/perfmodels/sampling STARPU_HOSTNAME=mirage STARPU_SCHED=dmdas ./examples/cholesky/cholesky_implicit -size $((960*40)) -nblocks 40
# size	ms	GFlops
38400	10216	1847.6</pre></tt>
<p>(MPI simulation is not supported yet)</p>

<div class="section" id="publications">
<h3>Publications</h3>
<p>
All StarPU related publications are also
listed <a href="./publications">here</a>
with the corresponding Bibtex entries.
</p>

<p>A good overview is available in
the following <a href="http://hal.archives-ouvertes.fr/inria-00467677">Research Report</a>.
</p>

<h4>General Presentations</h4> 
<ol>
<li>
<a name="agullo:hal-01283949"></a>Emmanuel Agullo, Olivier Aumage, Mathieu Faverge, Nathalie Furmento, Florent Pruvost, Marc Sergent,  and Samuel Thibault<br/>
<strong>Harnessing clusters of hybrid nodes with a sequential task-based programming model</strong><br/>
In <em>8th International Workshop on Parallel Matrix Algorithms and Applications</em>, July 2014<br/>
[<a href="https://hal.inria.fr/hal-01283949">WWW</a>]
[<a href="https://hal.inria.fr/hal-01283949/file/pmaa14.pdf">PDF</a>]
</li>
<li>
<a name="Aug11Thesis"></a>Cédric Augonnet<br/>
<strong>Scheduling Tasks over Multicore machines enhanced with Accelerators: a Runtime System's Perspective</strong><br/>
PhD thesis, Université Bordeaux 1, 351 cours de la Libération --- 33405 TALENCE cedex, December 2011<br/>
[<a href="http://tel.archives-ouvertes.fr/tel-00777154">WWW</a>]
</li>
<li>
<a name="AugThiNamWac11CCPE"></a>Cédric Augonnet, Samuel Thibault, Raymond Namyst,  and Pierre-André Wacrenier<br/>
<strong>StarPU: A Unified Platform for Task Scheduling on Heterogeneous Multicore Architectures</strong><br/>
<em>Concurrency and Computation: Practice and Experience, Special Issue: Euro-Par 2009</em>, 23:187-198, February 2011<br/>
[<a href="http://hal.inria.fr/inria-00550877">WWW</a>]
[doi:<a href="http://dx.doi.org/10.1002/cpe.1631">10.1002/cpe.1631</a>]
</li>
<li>
<a name="AugThiNamWac10RR7240"></a>Cédric Augonnet, Samuel Thibault,  and Raymond Namyst<br/>
<strong>StarPU: a Runtime System for Scheduling Tasks over Accelerator-Based Multicore Machines</strong><br/>
Technical Report 7240, INRIA, March 2010<br/>
[<a href="http://hal.inria.fr/inria-00467677">WWW</a>]
</li>
<li>
<a name="Aug09Renpar19"></a>Cédric Augonnet<br/>
<strong>StarPU: un support exécutif unifié pour les architectures multicoeurs hétérogènes</strong><br/>
In <em>19èmes Rencontres Francophones du Parallélisme</em>, Toulouse / France, September 2009<br/>
Note: Best Paper Award<br/>
[<a href="http://hal.inria.fr/inria-00411581">WWW</a>]
</li>
<li>
<a name="AugThiNamWac09Europar"></a>Cédric Augonnet, Samuel Thibault, Raymond Namyst,  and Pierre-André Wacrenier<br/>
<strong>StarPU: A Unified Platform for Task Scheduling on Heterogeneous Multicore Architectures</strong><br/>
In <em>Proceedings of the 15th International Euro-Par Conference</em>, volume 5704 of <em>Lecture Notes in Computer Science</em>, Delft, The Netherlands, pages 863-874, August 2009<br/>
Springer<br/>
[<a href="http://hal.inria.fr/inria-00384363">WWW</a>]
[doi:<a href="http://dx.doi.org/10.1007/978-3-642-03869-3_80">10.1007/978-3-642-03869-3_80</a>]
</li>
<li>
<a name="AugNam08HPPC"></a>Cédric Augonnet and Raymond Namyst<br/>
<strong>A unified runtime system for heterogeneous multicore architectures</strong><br/>
In <em>Proceedings of the International Euro-Par Workshops 2008, HPPC'08</em>, volume 5415 of <em>Lecture Notes in Computer Science</em>, Las Palmas de Gran Canaria, Spain, pages 174-183, August 2008<br/>
Springer<br/>
<strong>ISBN:</strong> 978-3-642-00954-9<br/>
[<a href="http://hal.inria.fr/inria-00326917">WWW</a>]
[doi:<a href="http://dx.doi.org/10.1007/978-3-642-00955-6_22">10.1007/978-3-642-00955-6_22</a>]
</li>
<li>
<a name="Aug08Master"></a>Cédric Augonnet<br/>
<strong>Vers des supports d'exécution capables d'exploiter les machines multicoeurs hétérogènes</strong><br/>
Mémoire de DEA, Université Bordeaux 1, June 2008<br/>
[<a href="http://hal.inria.fr/inria-00289361">WWW</a>]
</li>
</ol>
<h4>On Composability</h4> 
<ol>
<li>
<a name="AH13Renpar"></a>Andra Hugo<br/>
<strong>Le problème de la composition parallèle : une approche supervisée</strong><br/>
In <em>21èmes Rencontres Francophones du Parallélisme (RenPar'21)</em>, Grenoble, France, January 2013<br/>
[<a href="http://hal.inria.fr/hal-00773610">WWW</a>]
</li>
<li>
<a name="hugo:hal-00824514"></a>Andra Hugo, Abdou Guermouche, Raymond Namyst,  and Pierre-André Wacrenier<br/>
<strong>Composing multiple StarPU applications over heterogeneous machines: a supervised approach</strong><br/>
In <em>Third International Workshop on Accelerators and Hybrid Exascale Systems</em>, Boston, USA, May 2013<br/>
[<a href="http://hal.inria.fr/hal-00824514">WWW</a>]
</li>
<li>
<a name="AH11Master"></a>Andra Hugo<br/>
<strong>Composabilité de codes parallèles sur architectures hétérogènes</strong><br/>
Mémoire de Master, Université Bordeaux 1, June 2011<br/>
[<a href="http://hal.inria.fr/inria-00619654/en/">WWW</a>]
</li>
</ol>
<h4>On Scheduling</h4> 
<ol>
<li>
<a name="agullo:hal-01223573"></a>Emmanuel Agullo, Olivier Beaumont, Lionel Eyraud-Dubois,  and Suraj Kumar<br/>
<strong>Are Static Schedules so Bad ? A Case Study on Cholesky Factorization</strong><br/>
In <em>IPDPS'16</em>, Proceedings of the 30th IEEE International Parallel & Distributed Processing Symposium, IPDPS'16, Chicago, IL, United States, May 2016<br/>
IEEE<br/>
[<a href="https://hal.inria.fr/hal-01223573">WWW</a>]
[<a href="https://hal.inria.fr/hal-01223573/file/heteroprioCameraReady-ieeeCompatiable.pdf">PDF</a>]
</li>
<li>
<a name="beaumont:hal-01361992"></a>Olivier Beaumont, Terry Cojean, Lionel Eyraud-Dubois, Abdou Guermouche,  and Suraj Kumar<br/>
<strong>Scheduling of Linear Algebra Kernels on Multiple Heterogeneous Resources</strong><br/>
In <em>International Conference on High Performance Computing, Data, and Analytics (HiPC)</em>, Hyderabad, India, December 2016<br/>
[<a href="https://hal.inria.fr/hal-01361992">WWW</a>]
[<a href="https://hal.inria.fr/hal-01361992/file/HeteroPrioMultiResource.pdf">PDF</a>]
</li>
<li>
<a name="agullo:hal-01120507"></a>Emmanuel Agullo, Olivier Beaumont, Lionel Eyraud-Dubois, Julien Herrmann, Suraj Kumar, Loris Marchal,  and Samuel Thibault<br/>
<strong>Bridging the Gap between Performance and Bounds of Cholesky Factorization on Heterogeneous Platforms</strong><br/>
In <em>Heterogeneity in Computing Workshop 2015</em>, Hyderabad, India, May 2015<br/>
[<a href="https://hal.inria.fr/hal-01120507">WWW</a>]
</li>
<li>
<a name="sergent:hal-00978364"></a>Marc Sergent and Simon Archipoff<br/>
<strong>Modulariser les ordonnanceurs de tâches : une approche structurelle</strong><br/>
In <em>Compas'2014</em>, Neuchâtel, Suisse, April 2014<br/>
[<a href="http://hal.inria.fr/hal-00978364">WWW</a>]
[<a href="http://hal.inria.fr/hal-00978364/PDF/ordonnanceurs_modulaires.pdf">PDF</a>]
</li>
</ol>
<h4>On The C Extensions</h4> 
<ol>
<li>
<a name="LC13Report"></a>Ludovic Courtès<br/>
<strong>C Language Extensions for Hybrid CPU/GPU Programming with StarPU</strong><br/>
Research Report RR-8278, INRIA, April 2013<br/>
[<a href="http://hal.inria.fr/hal-00807033">WWW</a>]
[<a href="http://hal.inria.fr/hal-00807033/PDF/RR-8278.pdf">PDF</a>]
</li>
</ol>
<h4>On OpenMP Support on top of StarPU</h4> 
<ol>
<li>
<a name="virouleau:hal-01081974"></a>Philippe Virouleau, Pierrick BRUNET, François Broquedis, Nathalie Furmento, Samuel Thibault, Olivier Aumage,  and Thierry Gautier<br/>
<strong>Evaluation of OpenMP Dependent Tasks with the KASTORS Benchmark Suite</strong><br/>
In <em>10th International Workshop on OpenMP, IWOMP2014</em>, 10th International Workshop on OpenMP, IWOMP2014, Salvador, Brazil, France, pages 16 - 29, September 2014<br/>
Springer<br/>
[<a href="https://hal.inria.fr/hal-01081974">WWW</a>]
[doi:<a href="http://dx.doi.org/10.1007/978-3-319-11454-5_2">10.1007/978-3-319-11454-5_2</a>]
</li>
</ol>
<h4>On MPI Support</h4> 
<ol>
<li>
<a name="augonnet:hal-00992208"></a>Cédric Augonnet, Olivier Aumage, Nathalie Furmento, Samuel Thibault,  and Raymond Namyst<br/>
<strong>StarPU-MPI: Task Programming over Clusters of Machines Enhanced with Accelerators</strong><br/>
Rapport de recherche RR-8538, INRIA, May 2014<br/>
[<a href="http://hal.inria.fr/hal-00992208">WWW</a>]
[<a href="http://hal.inria.fr/hal-00992208/PDF/RR-8538.pdf">PDF</a>]
</li>
<li>
<a name="AugAumFurNamThi2012EuroMPI"></a>Cédric Augonnet, Olivier Aumage, Nathalie Furmento, Raymond Namyst,  and Samuel Thibault<br/>
<strong>StarPU-MPI: Task Programming over Clusters of Machines Enhanced with Accelerators</strong><br/>
In Siegfried Benkner Jesper Larsson Träff and Jack Dongarra, editors, <em>EuroMPI 2012</em>, volume 7490 of <em>LNCS</em>, September 2012<br/>
Springer<br/>
Note: Poster Session<br/>
[<a href="http://hal.inria.fr/hal-00725477">WWW</a>]
</li>
</ol>
<h4>On Memory Control</h4> 
<ol>
<li>
<a name="sergent:hal-01284004"></a>Marc Sergent, David Goudin, Samuel Thibault,  and Olivier Aumage<br/>
<strong>Controlling the Memory Subscription of Distributed Applications with a Task-Based Runtime System</strong><br/>
In <em>21st International Workshop on High-Level Parallel Programming Models and Supportive Environments</em>, Chicago, United States, May 2016<br/>
[<a href="https://hal.inria.fr/hal-01284004">WWW</a>]
[<a href="https://hal.inria.fr/hal-01284004/file/PID4127657.pdf">PDF</a>]
</li>
<li>
<a name="agullo:hal-01332774"></a>Emmanuel Agullo, Olivier Aumage, Mathieu Faverge, Nathalie Furmento, Florent Pruvost, Marc Sergent,  and Samuel Thibault<br/>
<strong>Achieving High Performance on Supercomputers with a Sequential Task-based Programming Model</strong><br/>
Research Report RR-8927, Inria Bordeaux Sud-Ouest ; Bordeaux INP ; CNRS ; Université de Bordeaux ; CEA, June 2016<br/>
[<a href="https://hal.inria.fr/hal-01332774">WWW</a>]
[<a href="https://hal.inria.fr/hal-01332774/file/RR-8927.pdf">PDF</a>]
</li>
</ol>
<h4>On Data Transfer Management</h4> 
<ol>
<li>
<a name="AugCleThiNam10ICPADS"></a>Cédric Augonnet, Jérôme Clet-Ortega, Samuel Thibault,  and Raymond Namyst<br/>
<strong>Data-Aware Task Scheduling on Multi-Accelerator based Platforms</strong><br/>
In <em>The 16th International Conference on Parallel and Distributed Systems (ICPADS)</em>, Shanghai, China, December 2010<br/>
[<a href="http://hal.inria.fr/inria-00523937">WWW</a>]
[doi:<a href="http://dx.doi.org/10.1109/ICPADS.2010.129">10.1109/ICPADS.2010.129</a>]
</li>
</ol>
<h4>On Performance Model Tuning</h4> 
<ol>
<li>
<a name="AugThiNam09HPPC"></a>Cédric Augonnet, Samuel Thibault,  and Raymond Namyst<br/>
<strong>Automatic Calibration of Performance Models on Heterogeneous Multicore Architectures</strong><br/>
In <em>Proceedings of the International Euro-Par Workshops 2009, HPPC'09</em>, volume 6043 of <em>Lecture Notes in Computer Science</em>, Delft, The Netherlands, pages 56-65, August 2009<br/>
Springer<br/>
[<a href="http://hal.inria.fr/inria-00421333">WWW</a>]
[doi:<a href="http://dx.doi.org/10.1007/978-3-642-14122-5_9">10.1007/978-3-642-14122-5_9</a>]
</li>
</ol>
<h4>On The Simulation Support through SimGrid</h4> 
<ol>
<li>
<a name="stanisic:hal-01147997"></a>Luka Stanisic, Samuel Thibault, Arnaud Legrand, Brice Videau,  and Jean-François Méhaut<br/>
<strong>Faithful Performance Prediction of a Dynamic Task-Based Runtime System for Heterogeneous Multi-Core Architectures</strong><br/>
<em>Concurrency and Computation: Practice and Experience</em>, pp 16, May 2015<br/>
[<a href="https://hal.inria.fr/hal-01147997">WWW</a>]
[<a href="https://hal.inria.fr/hal-01147997/file/CCPE14_article.pdf">PDF</a>]
[doi:<a href="http://dx.doi.org/10.1002/cpe">10.1002/cpe</a>]
</li>
<li>
<a name="stanisic:hal-01011633"></a>Luka Stanisic, Samuel Thibault, Arnaud Legrand, Brice Videau,  and Jean-François Méhaut<br/>
<strong>Modeling and Simulation of a Dynamic Task-Based Runtime System for Heterogeneous Multi-Core Architectures</strong><br/>
In <em>Euro-par - 20th International Conference on Parallel Processing</em>, Porto, Portugal, August 2014<br/>
Springer-Verlag<br/>
[<a href="http://hal.inria.fr/hal-01011633">WWW</a>]
[<a href="http://hal.inria.fr/hal-01011633/PDF/StarPUSG_article.pdf">PDF</a>]
</li>
</ol>
<h4>On The Cell Support</h4> 
<ol>
<li>
<a name="AugThiNamNij09Samos"></a>Cédric Augonnet, Samuel Thibault, Raymond Namyst,  and Maik Nijhuis<br/>
<strong>Exploiting the Cell/BE architecture with the StarPU unified runtime system</strong><br/>
In <em>SAMOS Workshop - International Workshop on Systems, Architectures, Modeling, and Simulation</em>, volume 5657 of <em>Lecture Notes in Computer Science</em>, Samos, Greece, July 2009<br/>
[<a href="http://hal.inria.fr/inria-00378705">WWW</a>]
[doi:<a href="http://dx.doi.org/10.1007/978-3-642-03138-0_36">10.1007/978-3-642-03138-0_36</a>]
</li>
</ol>
<h4>On Applications</h4> 
<ol>
<li>
<a name="agullo:hal-01316982"></a>E Agullo, L Giraud, A Guermouche, S Nakov,  and Jean Roman<br/>
<strong>Task-based Conjugate Gradient: from multi-GPU towards heterogeneous architectures</strong><br/>
Research Report 8912, Inria Bordeaux Sud-Ouest, May 2016<br/>
[<a href="https://hal.inria.fr/hal-01316982">WWW</a>]
[<a href="https://hal.inria.fr/hal-01316982/file/RR-8912.pdf">PDF</a>]
</li>
<li>
<a name="rossignon:tel-01230876"></a>Corentin Rossignon<br/>
<strong>A fine grain model programming for parallelization of sparse linear solver</strong><br/>
PhD thesis, Université de Bordeaux, July 2015<br/>
[<a href="https://tel.archives-ouvertes.fr/tel-01230876">WWW</a>]
[<a href="https://tel.archives-ouvertes.fr/tel-01230876/file/ROSSIGNON_CORENTIN_2015.pdf">PDF</a>]
</li>
<li>
<a name="MaMiDuAuThiAoNa15"></a>Vìctor Martìnez, David Michéa, Fabrice Dupros, Olivier Aumage, Samuel Thibault, Hideo Aochi,  and Philippe Olivier Alexandre Navaux<br/>
<strong>Towards seismic wave modeling on heterogeneous many-core architectures using task-based runtime system</strong><br/>
In <em>27th International Symposium on Computer Architecture and High Performance Computing (SBAC-PAD)</em>, Florianopolis, Brazil, October 2015<br/>
[<a href="https://hal.inria.fr/hal-01182746">WWW</a>]
[<a href="https://hal.inria.fr/hal-01182746/file/sbac2015_soumission.pdf">PDF</a>]
</li>
<li>
<a name="sylvain:hal-01005765"></a>Sylvain Henry, Alexandre Denis, Denis Barthou, Marie-Christine Counilh,  and Raymond Namyst<br/>
<strong>Toward OpenCL Automatic Multi-Device Support</strong><br/>
In Fernando Silva, Ines Dutra,  and Vitor Santos Costa, editors, <em>Euro-Par 2014</em>, Porto, Portugal, August 2014<br/>
Springer<br/>
[<a href="http://hal.inria.fr/hal-01005765">WWW</a>]
[<a href="http://hal.inria.fr/hal-01005765/PDF/final.pdf">PDF</a>]
</li>
<li>
<a name="lacoste:hal-00987094"></a>Xavier Lacoste, Mathieu Faverge, Pierre Ramet, Samuel Thibault,  and George Bosilca<br/>
<strong>Taking advantage of hybrid systems for sparse direct solvers via task-based runtimes</strong><br/>
In <em>HCW'2014 workshop of IPDPS</em>, Phoenix, États-Unis, May 2014<br/>
IEEE<br/>
Note: RR-8446 RR-8446<br/>
[<a href="http://hal.inria.fr/hal-00987094">WWW</a>]
[<a href="http://hal.inria.fr/hal-00987094/PDF/sparsegpus.pdf">PDF</a>]
</li>
<li>
<a name="lacoste:hal-00925017"></a>Xavier Lacoste, Mathieu Faverge, Pierre Ramet, Samuel Thibault,  and George Bosilca<br/>
<strong>Taking advantage of hybrid systems for sparse direct solvers via task-based runtimes</strong><br/>
Rapport de recherche RR-8446, INRIA, January 2014<br/>
[<a href="http://hal.inria.fr/hal-00925017">WWW</a>]
[<a href="http://hal.inria.fr/hal-00925017/PDF/RR-8446.pdf">PDF</a>]
</li>
<li>
<a name="sergent:hal-00978602"></a>Emmanuel Agullo, Olivier Aumage, Mathieu Faverge, Nathalie Furmento, Florent Pruvost, Marc Sergent,  and Samuel Thibault<br/>
<strong>Overview of Distributed Linear Algebra on Hybrid Nodes over the StarPU Runtime</strong><br/>
SIAM Conference on Parallel Processing for Scientific Computing, February 2014<br/>
[<a href="http://hal.inria.fr/hal-00978602">WWW</a>]
[<a href="http://hal.inria.fr/hal-00978602/PDF/siampp14.pdf">PDF</a>]
</li>
<li>
<a name="Bor13Thesis"></a>Cyril Bordage<br/>
<strong>Ordonnancement dynamique, adapté aux architectures hétérogènes, de la méthode multipôle pour les équations de Maxwell, en électromagnétisme</strong><br/>
PhD thesis, Université Bordeaux 1, 351 cours de la Libération --- 33405 TALENCE cedex, December 2013<br/>
</li>
<li>
<a name="Hen13Thesis"></a>Sylvain Henry<br/>
<strong>Modèles de programmation et supports exécutifs pour architectures hétérogènes</strong><br/>
PhD thesis, Université Bordeaux 1, 351 cours de la Libération --- 33405 TALENCE cedex, November 2013<br/>
[<a href="http://tel.archives-ouvertes.fr/tel-00948309">WWW</a>]
</li>
<li>
<a name="hen13fhpc"></a>Sylvain Henry<br/>
<strong>ViperVM: a Runtime System for Parallel Functional High-Performance Computing on Heterogeneous Architectures</strong><br/>
In <em>2nd Workshop on Functional High-Performance Computing (FHPC'13)</em>, Boston, États-Unis, September 2013<br/>
[<a href="http://hal.inria.fr/hal-00851122">WWW</a>]
[<a href="http://hal.inria.fr/hal-00851122/PDF/fhpc13.pdf">PDF</a>]
</li>
<li>
<a name="odajima:hal-00920915"></a>Tetsuya Odajima, Taisuke Boku, Mitsuhisa Sato, Toshihiro Hanawa, Yuetsu Kodama, Raymond Namyst, Samuel Thibault,  and Olivier Aumage<br/>
<strong>Adaptive Task Size Control on High Level Programming for GPU/CPU Work Sharing</strong><br/>
In <em>The 2013 International Symposium on Advances of Distributed and Parallel Computing (ADPC 2013)</em>, Vietri sul Mare, Italie, December 2013<br/>
[<a href="http://hal.inria.fr/hal-00920915">WWW</a>]
[<a href="http://hal.inria.fr/hal-00920915/PDF/ADPC2013-117.pdf">PDF</a>]
</li>
<li>
<a name="ohshima:hal-00926144"></a>Satoshi Ohshima, Satoshi Katagiri, Kengo Nakajima, Samuel Thibault,  and Raymond Namyst<br/>
<strong>Implementation of FEM Application on GPU with StarPU</strong><br/>
In <em>SIAM CSE13 - SIAM Conference on Computational Science and Engineering 2013</em>, Boston, États-Unis, February 2013<br/>
SIAM<br/>
[<a href="http://hal.inria.fr/hal-00926144">WWW</a>]
</li>
<li>
<a name="Ros13Renpar"></a>Corentin Rossignon<br/>
<strong>Optimisation du produit matrice-vecteur creux sur architecture GPU pour un simulateur de reservoir</strong><br/>
In <em>21èmes Rencontres Francophones du Parallélisme (RenPar'21)</em>, Grenoble, France, January 2013<br/>
[<a href="http://hal.inria.fr/hal-00773571">WWW</a>]
</li>
<li>
<a name="rossignon:hal-00858350"></a>Corentin Rossignon, Pascal Hénon, Olivier Aumage,  and Samuel Thibault<br/>
<strong>A NUMA-aware fine grain parallelization framework for multi-core architecture</strong><br/>
In <em>PDSEC - 14th IEEE International Workshop on Parallel and Distributed Scientific and Engineering Computing - 2013</em>, Boston, États-Unis, May 2013<br/>
[<a href="http://hal.inria.fr/hal-00858350">WWW</a>]
[<a href="http://hal.inria.fr/hal-00858350/PDF/taggre_pdsec_2013.pdf">PDF</a>]
</li>
<li>
<a name="HenDenBar2012TSI"></a>Sylvain Henry, Alexandre Denis,  and Denis Barthou<br/>
<strong>Programmation unifiée multi-accélérateur OpenCL</strong><br/>
<em>Techniques et Sciences Informatiques</em>, (8-9-10):1233-1249, 2012<br/>
[<a href="http://hal.inria.fr/hal-00772742">WWW</a>]
</li>
<li>
<a name="MahManAugThi12TSI"></a>Sidi Ahmed Mahmoudi, Pierre Manneback, Cédric Augonnet,  and Samuel Thibault<br/>
<strong>Traitements d'Images sur Architectures Parallèles et Hétérogènes</strong><br/>
<em>Technique et Science Informatiques</em>, 2012<br/>
[<a href="http://hal.inria.fr/hal-00714858/">WWW</a>]
</li>
<li>
<a name="BenkBajMarSanNamThiEuroPar2012"></a>Siegfried Benkner, Enes Bajrovic, Erich Marth, Martin Sandrieser, Raymond Namyst,  and Samuel Thibault<br/>
<strong>High-Level Support for Pipeline Parallelism on Many-Core Architectures</strong><br/>
In <em>Europar - International European Conference on Parallel and Distributed Computing - 2012</em>, Rhodes Island, Grèce, August 2012<br/>
[<a href="http://hal.inria.fr/hal-00697020">WWW</a>]
[<a href="http://hal.inria.fr/hal-00697020/PDF/europar2012-submitted.pdf">PDF</a>]
</li>
<li>
<a name="kessler:hal-00776610"></a>Christoph Kessler, Usman Dastgeer, Samuel Thibault, Raymond Namyst, Andrew Richards, Uwe Dolinsky, Siegfried Benkner, Jesper Larsson Träff,  and Sabri Pllana<br/>
<strong>Programmability and Performance Portability Aspects of Heterogeneous Multi-/Manycore Systems</strong><br/>
In <em>Design, Automation and Test in Europe (DATE)</em>, Dresden, Allemagne, March 2012<br/>
[<a href="http://hal.inria.fr/hal-00776610">WWW</a>]
[<a href="http://hal.inria.fr/hal-00776610/PDF/date12-paper.pdf">PDF</a>]
</li>
<li>
<a name="BenPllTraTsiDolAugBacKesMolOsi11IEEEMicro"></a>Siegfried Benkner, Sabri Pllana, Jesper Larsson Träff, Philippas Tsigas, Uwe Dolinsky, Cédric Augonnet, Beverly Bachmayer, Christoph Kessler, David Moloney,  and Vitaly Osipov<br/>
<strong>PEPPHER: Efficient and Productive Usage of Hybrid Computing Systems</strong><br/>
<em>IEEE Micro</em>, 31(5):28-41, September 2011<br/>
<strong>ISSN:</strong> 0272-1732<br/>
[<a href="http://hal.inria.fr/hal-00648480">WWW</a>]
[doi:<a href="http://dx.doi.org/10.1109/MM.2011.67">10.1109/MM.2011.67</a>]
</li>
<li>
<a name="AguAugDonFavLanLtaTomAICCSA11"></a>Emmanuel Agullo, Cédric Augonnet, Jack Dongarra, Mathieu Faverge, Julien Langou, Hatem Ltaief,  and Stanimire Tomov<br/>
<strong>LU factorization for accelerator-based systems</strong><br/>
In <em>9th ACS/IEEE International Conference on Computer Systems and Applications (AICCSA 11)</em>, Sharm El-Sheikh, Egypt, June 2011<br/>
[<a href="http://hal.inria.fr/hal-00654193">WWW</a>]
</li>
<li>
<a name="AguAugDonFavLtaThiTom11IPDPS"></a>Emmanuel Agullo, Cédric Augonnet, Jack Dongarra, Mathieu Faverge, Hatem Ltaief, Samuel Thibault,  and Stanimire Tomov<br/>
<strong>QR Factorization on a Multicore Node Enhanced with Multiple GPU Accelerators</strong><br/>
In <em>25th IEEE International Parallel & Distributed Processing Symposium (IEEE IPDPS 2011)</em>, Anchorage, Alaska, USA, May 2011<br/>
[<a href="http://hal.inria.fr/inria-00547614">WWW</a>]
[doi:<a href="http://dx.doi.org/10.1109/IPDPS.2011.90">10.1109/IPDPS.2011.90</a>]
</li>
<li>
<a name="DasKesThi11ParCo"></a>Usman Dastgeer, Christoph Kessler,  and Samuel Thibault<br/>
<strong>Flexible runtime support for efficient skeleton programming on hybrid systems</strong><br/>
In <em>Proceedings of the International Conference on Parallel Computing (ParCo), Applications, Tools and Techniques on the Road to Exascale Computing</em>, volume 22 of <em>Advances of Parallel Computing</em>, Gent, Belgium, pages 159-166, August 2011<br/>
[<a href="http://hal.inria.fr/inria-00606200/">WWW</a>]
</li>
<li>
<a name="Hen11Renpar"></a>Sylvain Henry<br/>
<strong>Programmation multi-accélérateurs unifiée en OpenCL</strong><br/>
In <em>20èmes Rencontres Francophones du Parallélisme (RenPar'20)</em>, Saint Malo, France, May 2011<br/>
[<a href="http://hal.archives-ouvertes.fr/hal-00643257">WWW</a>]
</li>
<li>
<a name="MahManAugThi11Renpar20"></a>Sidi Ahmed Mahmoudi, Pierre Manneback, Cédric Augonnet,  and Samuel Thibault<br/>
<strong>Détection optimale des coins et contours dans des bases d'images volumineuses sur architectures multicoeurs hétérogènes</strong><br/>
In <em>20èmes Rencontres Francophones du Parallélisme</em>, Saint-Malo / France, May 2011<br/>
[<a href="http://hal.inria.fr/inria-00606195">WWW</a>]
</li>
<li>
<a name="AguAugDonLtaNamThiTomGPUgems"></a>Emmanuel Agullo, Cédric Augonnet, Jack Dongarra, Hatem Ltaief, Raymond Namyst, Samuel Thibault,  and Stanimire Tomov<br/>
<strong>A Hybridization Methodology for High-Performance Linear Algebra Software for GPUs</strong><br/>
In Wen-mei W<br/> Hwu, editor, <em>GPU Computing Gems</em>, volume 2<br/>
Morgan Kaufmann, September 2010<br/>
[<a href="http://hal.inria.fr/inria-00547847">WWW</a>]
</li>
<li>
<a name="AguAugDonLtaNamRomThiTom10SAAHPC"></a>Emmanuel Agullo, Cédric Augonnet, Jack Dongarra, Hatem Ltaief, Raymond Namyst, Jean Roman, Samuel Thibault,  and Stanimire Tomov<br/>
<strong>Dynamically scheduled Cholesky factorization on multicore architectures with GPU accelerators</strong><br/>
In <em>Symposium on Application Accelerators in High Performance Computing (SAAHPC)</em>, Knoxville, USA, July 2010<br/>
[<a href="http://hal.inria.fr/inria-00547616">WWW</a>]
</li>
</ol>

<div class="section bot">
<p class="updated">
  Last updated on 2016/04/13.
</p>
</div>

</body>
</html>