index.html 51 KB
Newer Older
1
<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.0 Transitional//EN" "http://www.w3.org/TR/REC-html40/loose.dtd">
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
2 3 4 5 6
<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" />
7
<link rel="Shortcut icon" href="http://www.inria.fr/extension/site_inria/design/site_inria/images/favicon.ico" type="image/x-icon" />
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
8 9 10 11
</HEAD>

<body>

12
<div class="title">
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
13
<h1><a href="./">StarPU</a></h1>
14 15
<h2>A Unified Runtime System for Heterogeneous Multicore Architectures</h2>
</div>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
16

17
<div class="menu">
18
<a href="https://team.inria.fr/storm/">STORM TEAM</a> |
19
&nbsp; &nbsp; &nbsp;
20
|
21
<a href="#overview">Overview</a> |
Nathalie Furmento's avatar
Nathalie Furmento committed
22
<a href="#news">News</a> |
23
<a href="#contact">Contact</a> |
Nathalie Furmento's avatar
Nathalie Furmento committed
24
<a href="people/">People</a> |
25
<a href="#features">Features</a> |
26
<a href="#software">Software</a> |
THIBAULT Samuel's avatar
THIBAULT Samuel committed
27
<a href="#tryit">Try it!</a> |
28
<a href="#publications">Publications</a> |
29
<a href="internships/">Jobs/Interns</a> |
30
<a href="files/">Download</a> |
Nathalie Furmento's avatar
Nathalie Furmento committed
31
<a href="tutorials/">Tutorials</a> |
32
<a href="https://gforge.inria.fr/plugins/mediawiki/wiki/starpu/index.php/Main_Page">Intranet</a>
Nathalie Furmento's avatar
Nathalie Furmento committed
33
</div>
34

35 36
<div class="section" id="overview">
<h3>Overview</h3>
37 38 39 40 41 42
  <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>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
43
    <li>A graph of tasks, using either the StarPU's high level <b>GCC plugin</b> pragmas, StarPU's rich <b>C API</b>, or <b>OpenMP pragmas</b>.</li>
44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61
    </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>
THIBAULT Samuel's avatar
fix URL  
THIBAULT Samuel committed
62
<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.
63 64 65 66
</p>
</div>

<div class="section emphasize newslist" id="news">
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
67 68
<h3>News</h3>
<p>
69 70 71 72 73 74 75
November 2017 <b>&raquo;&nbsp;</b><a href="http://gforge.inria.fr/frs/?group_id=1570"><b>The
      1.2.3 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>
76 77 78 79 80 81 82
May 2017 <b>&raquo;&nbsp;</b> StarPU is part of the HPCLib project
  that aims at performing static analysis of the Inria Bordeaux
  Sud-Ouest solver stack. See
  the <a href="https://hpclib-sed.bordeaux.inria.fr/sonarqube/overview?id=starpu">StarPU
  page</a> on the sonarqube server of INRIA.
</p>
<p>
83 84 85 86 87 88 89
May 2017 <b>&raquo;&nbsp;</b><a href="http://gforge.inria.fr/frs/?group_id=1570"><b>The
      1.2.2 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>
90 91 92 93 94
April 2017 <b>&raquo;&nbsp;</b>A <a href="https://events.prace-ri.eu/event/618/">tutorial</a>
      "Runtime systems for heterogeneous platform programming" will be
      given at the Maison de la Simulation in May 2017.
</p>
<p>
95 96 97 98 99 100 101
March 2017 <b>&raquo;&nbsp;</b><a href="http://gforge.inria.fr/frs/?group_id=1570"><b>The
      1.2.1 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>
102 103 104 105 106 107
March 2017 <b>&raquo;&nbsp;</b><a href="http://gforge.inria.fr/frs/?group_id=1570"><b>The
      v1.1.7 release of StarPU is now available!</b></a>. This release notably brings the concept of
      scheduling contexts which allows to separate computation
      resources. This is intented to be the last release for the
      branch 1.1.
</p>
108
</div>
109 110

<div class="section emphasizebot" style="text-align: right; font-style: italic;">
111
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>.
112
See also the full <a href="news/">news</a>.
113 114
</div>

115 116 117 118 119 120 121 122 123 124 125
<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>

THIBAULT Samuel's avatar
THIBAULT Samuel committed
126
<div class="section" id="tutorial">
127 128 129 130 131 132 133 134 135 136
<h3>Tutorial material</h3>
<p>
The latest tutorial material for StarPU is composed of two parts:
<ul>
<li><a href="http://starpu.gforge.inria.fr/tutorials/2016-06-PATC/slides/01_introducing_starpu.pdf">Introducing StarPU</a></li>
<li><a href="http://starpu.gforge.inria.fr/tutorials/2016-06-PATC/slides/02_mastering_starpu.pdf">Mastering StarPU</a></li>
</ul>
</p>
</div>

THIBAULT Samuel's avatar
THIBAULT Samuel committed
137 138 139 140 141 142 143 144
<div class="section" id="slides">
<h3>Set of slides</h3>
<p>
A <a href="slides.pdf">set of slides</a> is also available to get an overview of
StarPU.
</p>
</div>

145 146
<div class="section" id="contact">
<h3>Contact</h3>
147
<p>For any questions regarding StarPU, please contact the StarPU developers mailing list.</p>
148 149 150
<pre>
<a href="mailto:starpu-devel@lists.gforge.inria.fr?subject=StarPU">starpu-devel@lists.gforge.inria.fr</a>
</pre>
151
<p>Details of the <a href="people/">StarPU team people</a> are also available.</p>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
152 153
</div>

154 155
<div class="section" id="features">
<h3>Features</h3>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
156

157
<h4>Portability</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
158
  <p>
159 160 161 162 163 164 165 166
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.
167 168 169 170
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).
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
171 172
  </p>

173
<h4>Data transfers</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
174
  <p>
175
To relieve programmers from the burden of explicit data transfers, a high-level
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
176
data management library enforces memory coherency over the machine: before a
177 178
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
THIBAULT Samuel's avatar
THIBAULT Samuel committed
179 180
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
181
prefetching</b> data, which thus permits to <b>overlap data transfers with computations</b>
THIBAULT Samuel's avatar
update  
THIBAULT Samuel committed
182
(including <b>GPU-GPU direct transfers</b>) to achieve the most of the architecture.
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
183 184
  </p>

185
<h4>Dependencies</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
186
  <p>
187
Dependencies between tasks can be given either of several ways, to provide the
188 189
programmer with best flexibility:
  <ul>
190
    <li><b>implicitly</b> from RAW, WAW, and WAR data dependencies.</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
191
    <li>explicitly through <b>tags</b> which act as rendez-vous points between
192
    tasks (thus including tasks which have not been created yet),</li>
193
    <li><b>explicitly</b> between pairs of tasks,</li>
194
  </ul>
195 196
  </p>
  <p>
197 198 199
  These dependencies are computed in a completely decentralized way.
  </p>
  <p>
200 201 202 203
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).
204 205 206 207 208
  </p>

<h4>Heterogeneous Scheduling</h4>
  <p>
StarPU obtains
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
209
portable performances by efficiently (and easily) using all computing resources
210
at the same time. StarPU also takes advantage of the <b>heterogeneous</b> nature of a
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
211
machine, for instance by using scheduling strategies based on auto-tuned
212 213 214
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>.
215
Various strategies and variants are available. Some of them are centralized, but
THIBAULT Samuel's avatar
THIBAULT Samuel committed
216
most of them are <b>completely distributed</b>. dmda (a data-locality-aware MCT strategy,
217
thus similar to heft but starts executing tasks before the whole task graph is
218 219 220
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
221 222 223
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.
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
224 225
  </p>

226 227
<h4>Clusters</h4>
  <p>
228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249
To deal with clusters, StarPU can nicely integrate with <a
	href="doc/html/MPISupport.html">MPI</a>, through explicit or implicit
support, according to the application's preference.

    <ul>
        <li>Explicit network communication requests can be emitted, which will
then be <b>automatically combined and overlapped</b> with the intra-node data
transfers and computation,
        <li>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>automatically generate all required
MPI communications</b> accordingly (new in v0.9). We have gotten excellent
scaling on a 256-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>.
        <li>Starting with v1.3, the application can also just provide the
whole task graph, and let StarPU decide the data distribution and task
distribution, thanks to a master-slave mechanism. This will however by nature
have a more limited scalability than the fully distributed paradigm mentioned
above.
    </ul>
250 251 252 253 254
  </p>

<h4>Out of core</h4>
  <p>
When memory is not big enough for the working set, one may have to resort to
Nathalie Furmento's avatar
Nathalie Furmento committed
255
using disks. StarPU makes this seamless thanks to its <a href="doc/html/OutOfCore.html">out of core support</a> (new in v1.2).
THIBAULT Samuel's avatar
update  
THIBAULT Samuel committed
256 257
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.
258 259
  </p>

260 261 262
<h4>Extensions to the C Language</h4>
<p>
  StarPU comes with a GCC plug-in
263
  that <a href="doc/html/cExtensions.html">extends the C programming
264 265
  language</a> with pragmas and attributes that make it easy
  to <b>annotate a sequential C program to turn it into a parallel
266 267 268
  StarPU program</b> (new in v1.0).
</p>

THIBAULT Samuel's avatar
THIBAULT Samuel committed
269 270
<h4>OpenMP 4 -compatible interface</h4>
<p>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
271
  <a href=http://kstar.gforge.inria.fr/>K'Star</a> provides an OpenMP
THIBAULT Samuel's avatar
THIBAULT Samuel committed
272 273 274 275 276 277 278 279 280
  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>

281 282 283 284
<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).
285 286
</p>

287 288 289 290
<h4>Simulation support</h4>
<p>
  StarPU can very accurately simulate an application execution
  and measure the resulting performance thanks to using the
291
  <a href="http://simgrid.gforge.inria.fr">SimGrid simulator</a> (new in v1.1).  This allows
292 293 294 295 296
  to quickly experiment with various scheduling heuristics, various application
  algorithms, and even various platforms (available GPUs and CPUs, available
  bandwidth)!
</p>

297 298
<h4>All in all</h4>
  <p>
299
All that means that, with the help
300
of <a href="doc/html/cExtensions.html">StarPU's extensions to the C
301 302
language</a>, the following sequential source code of a tiled version of
the classical Cholesky factorization algorithm using BLAS is also valid
THIBAULT Samuel's avatar
THIBAULT Samuel committed
303
StarPU code, possibly running on all the CPUs and GPUs, and given a data
Nathalie Furmento's avatar
Nathalie Furmento committed
304
distribution over MPI nodes, it is even a distributed version!
305 306 307 308 309 310 311 312 313 314 315 316 317
  </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>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
318

319
<h4>Supported Architectures</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
320
<ul>
321
<li>SMP/Multicore Processors (x86, PPC, ARM, ... all Debian architecture have been tested) </li>
Nathalie Furmento's avatar
Nathalie Furmento committed
322
<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 v1.1)</li>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
323 324
<li>OpenCL devices</li>
<li>Cell Processors (experimental)</li>
Nathalie Furmento's avatar
Nathalie Furmento committed
325 326
<li>Intel SCC (experimental, new in v1.2)</li>
<li>Intel MIC / Xeon Phi (new in v1.2)</li>
327
</ul>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
328

329
<h4>Supported Operating Systems</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
330
<ul>
Ludovic Courtès's avatar
Ludovic Courtès committed
331 332
<li>GNU/Linux</li>
<li>Mac OS X</li>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
333 334 335
<li>Windows</li>
</ul>

336
<h4>Performance analysis tools</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
337 338 339 340 341 342 343 344 345
  <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>

346
<p>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
347 348 349 350 351 352
<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>
353
<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>
354 355
</p>

Nathalie Furmento's avatar
website  
Nathalie Furmento committed
356 357 358 359 360 361
<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
362
GFlop/s</b>
363
<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>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
364 365
</p>

THIBAULT Samuel's avatar
THIBAULT Samuel committed
366 367
<p>
<a href="http://www.hlrs.de/temanejo">Temanejo</a> can be used to debug the task
368
graph, as shown below (new in v1.1).
THIBAULT Samuel's avatar
THIBAULT Samuel committed
369 370 371
</p>

<center>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
372
<a href="images/temanejo.png"><img src="images/temanejo.png" width="50%"/></a>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
373 374
</center>

Nathalie Furmento's avatar
website  
Nathalie Furmento committed
375 376
</div>

377 378 379 380 381
<div class="section" id="software">
<h3>Software using StarPU</h3>

<p>
Some software is known for being able to use StarPU to tackle heterogeneous
382
architectures, here is a non-exhaustive list (feel free to ask to be added to the
383
list!):
384 385 386
</p>

<ul>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
387
	<li><a href="https://project.inria.fr/chameleon/">Chameleon</a>, dense linear algebra library</li>
388 389 390
        <li><a href=http://kstar.gforge.inria.fr/>K'Star</a>, OpenMP 4 - compatible interface on top of StarPU.</li>
	<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://gitlab.inria.fr/solverstack/maphys">MaPHyS</a>, Massively Parallel Hybrid Solver</li>
391
	<li><a href="http://pastix.gforge.inria.fr/">PaStiX</a>, sparse linear algebra library, starting from version 5.2.1</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
392
	<li><a href="http://buttari.perso.enseeiht.fr/qr_mumps/">qr_mumps</a>, sparse linear algebra library</li>
THIBAULT Samuel's avatar
update  
THIBAULT Samuel committed
393
	<li><a href="http://scalfmm-public.gforge.inria.fr/doc/">ScalFMM</a>, N-body interaction simulation using the Fast Multipole Method. </li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
394
	<li><a href="https://tel.archives-ouvertes.fr/tel-01410049/">SCHNAPS</a>, Solver for Conservative Hypebolic Non-linear systems Applied to PlasmaS. </li>
395 396
	<li><a href="https://hal.archives-ouvertes.fr/hal-01086246">SignalPU</a>, a Dataflow-Graph-specific programming model. </li>
	<li><a href="http://www.ida.liu.se/~chrke/skepu/">SkePU</a>, a skeleton programming framework.</li>
397 398
</ul>

399
<p>
400
You can find <a href="#PublicationsOnApplications">below</a> the list of publications related to applications using StarPU.
401 402
</p>

403 404
</div>

THIBAULT Samuel's avatar
THIBAULT Samuel committed
405 406 407 408
<div class="section" id="tryit">
<h3>Give it a try!</h3>
<p>
You can easily try the performance on the Cholesky factorization for
THIBAULT Samuel's avatar
THIBAULT Samuel committed
409 410 411
instance. Make sure to have the pkg-config and
<a href="http://www.open-mpi.org/projects/hwloc/">hwloc</a>
software installed for
THIBAULT Samuel's avatar
THIBAULT Samuel committed
412 413 414
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>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
415 416

<tt><pre>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
417 418 419 420 421 422
$ 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
THIBAULT Samuel's avatar
THIBAULT Samuel committed
423
$ STARPU_SCHED=dmdas mpirun -np 4 -machinefile mymachines ./mpi/examples/matrix_decomposition/mpi_cholesky_distributed -size $((960*40*4)) -nblocks $((40*4))</pre></tt>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
424 425 426 427 428

<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>

THIBAULT Samuel's avatar
THIBAULT Samuel committed
429 430 431 432 433 434 435 436 437 438 439 440 441 442
<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>

THIBAULT Samuel's avatar
THIBAULT Samuel committed
443 444 445 446 447 448 449 450 451 452
<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
THIBAULT Samuel's avatar
THIBAULT Samuel committed
453
38400	9915	1903.7</pre></tt>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
454 455
<p>(MPI simulation is not supported yet)</p>

456 457 458 459
<div class="section" id="publications">
<h3>Publications</h3>
<p>
All StarPU related publications are also
460
listed <a href="./publications">here</a>
461 462 463
with the corresponding Bibtex entries.
</p>

464 465
<p>
A good overview is available in
466 467 468
the following <a href="http://hal.archives-ouvertes.fr/inria-00467677">Research Report</a>.
</p>

469 470
<p>
If you need to cite StarPU, please
471
reference <a href="publications/Year/2011.html#AugThiNamWac11CCPE">[StarPU: A Unified Platform
472 473 474 475 476
    for Task Scheduling on Heterogeneous Multicore Architectures]</a>
for a general presentation. Other sub-sections below will give you
references for more specific aspects of StarPU.
</p>

477
<h4>General Presentations</h4> 
478
<a name="PublicationsGeneralPresentations"></a>
479 480
<ol>
<li>
481 482 483
<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/>
484
[<a href="http://tel.archives-ouvertes.fr/tel-00777154">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
485
[<a href="http://tel.archives-ouvertes.fr/tel-00777154/document">PDF</a>]
486 487
</li>
<li>
488 489 490
<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/>
491
[<a href="http://hal.inria.fr/inria-00550877">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
492
[<a href="http://hal.inria.fr/inria-00550877/document">PDF</a>]
493
[doi:<a href="http://dx.doi.org/10.1002/cpe.1631">10.1002/cpe.1631</a>]
494
</li>
495
<li>
496 497 498 499
<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>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
500
[<a href="http://hal.inria.fr/inria-00467677/document">PDF</a>]
501 502
</li>
<li>
503 504 505 506
<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/>
507
[<a href="http://hal.inria.fr/inria-00411581">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
508
[<a href="http://hal.inria.fr/inria-00411581/document">PDF</a>]
509 510
</li>
<li>
511 512 513 514
<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/>
515
[<a href="http://hal.inria.fr/inria-00384363">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
516
[<a href="http://hal.inria.fr/inria-00384363/document">PDF</a>]
517 518 519 520 521 522 523 524
[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/>
525
[<a href="http://hal.inria.fr/inria-00326917">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
526
[<a href="http://hal.inria.fr/inria-00326917/document">PDF</a>]
527 528 529 530 531 532
[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/>
533
[<a href="http://hal.inria.fr/inria-00289361">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
534
[<a href="http://hal.inria.fr/inria-00289361/document">PDF</a>]
535 536
</li>
</ol>
537
<h4>On Composability</h4> 
538
<a name="PublicationsOnComposability"></a>
Nathalie Furmento's avatar
Nathalie Furmento committed
539 540
<ol>
<li>
541 542 543
<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/>
544
[<a href="http://hal.inria.fr/hal-00773610">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
545
[<a href="http://hal.inria.fr/hal-00773610/document">PDF</a>]
Nathalie Furmento's avatar
Nathalie Furmento committed
546 547
</li>
<li>
548 549 550
<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/>
551
[<a href="http://hal.inria.fr/hal-00824514">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
552
[<a href="http://hal.inria.fr/hal-00824514/document">PDF</a>]
553 554 555 556 557
</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/>
558
[<a href="http://hal.inria.fr/inria-00619654/en/">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
559
[<a href="http://hal.inria.fr/inria-00619654/document">PDF</a>]
Nathalie Furmento's avatar
Nathalie Furmento committed
560 561
</li>
</ol>
562
<h4>On Scheduling</h4> 
563
<a name="PublicationsOnScheduling"></a>
564
<ol>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
565
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
566 567
<a name="kumar:tel-01538516"></a>Suraj Kumar<br/>
<strong>Scheduling of Dense Linear Algebra Kernels on Heterogeneous Resources</strong><br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
568
PhD thesis, Université de Bordeaux, April 2017<br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
569 570 571 572
[<a href="https://tel.archives-ouvertes.fr/tel-01538516">WWW</a>]
[<a href="https://tel.archives-ouvertes.fr/tel-01538516/file/KUMAR_SURAL_2017.pdf">PDF</a>]
</li>
<li>
573 574 575 576 577 578 579 580
<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>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
581 582 583 584
<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>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
585
[<a href="https://hal.inria.fr/hal-01361992v2/document">PDF</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
586 587
</li>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
588 589 590 591 592 593 594
<a name="garciapinto:hal-01353962"></a>Vinicius Garcia Pinto, Luka Stanisic, Arnaud Legrand, Lucas Mello Schnorr, Samuel Thibault,  and Vincent Danjean<br/>
<strong>Analyzing Dynamic Task-Based Applications on Hybrid Platforms: An Agile Scripting Approach</strong><br/>
In <em>3rd Workshop on Visual Performance Analysis (VPA)</em>, Salt Lake City, United States, November 2016<br/>
Note: Held in conjunction with SC16<br/>
[<a href="https://hal.inria.fr/hal-01353962">WWW</a>]
[<a href="https://hal.inria.fr/hal-01353962/file/VPA_2016_paper_3.pdf">PDF</a>]
</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
595
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
596 597 598
<a name="JaBlHU2016a"></a>Johan Janzén, David Black-Schaffer,  and Andra Hugo<br/>
<strong>Partitioning GPUs for Improved Scalability</strong><br/>
In <em>IEEE 28th International Symposium on Computer Architecture and High Performance Computing (SBAC-PAD)</em>, October 2016<br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
599
[<a href="http://ieeexplore.ieee.org/abstract/document/7789322/">WWW</a>]
600 601 602
[doi:<a href="http://dx.doi.org/10.1109/SBAC-PAD.2016.14">10.1109/SBAC-PAD.2016.14</a>]
</li>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
603 604 605 606 607 608
<a name="beaumont:hal-01386174"></a>Olivier Beaumont, Lionel Eyraud-Dubois,  and Suraj Kumar<br/>
<strong>Approximation Proofs of a Fast and Efficient List Scheduling Algorithm for Task-Based Runtime Systems on Multicores and GPUs</strong><br/>
Note: Working paper or preprint, October 2016<br/>
[<a href="https://hal.inria.fr/hal-01386174">WWW</a>]
[<a href="https://hal.inria.fr/hal-01386174/file/heteroPrioApproxProofsRR.pdf">PDF</a>]
</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
609
<li>
610 611 612
<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/>
613
[<a href="https://hal.inria.fr/hal-01120507">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
614
[<a href="https://hal.inria.fr/hal-01120507/document">PDF</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
615
</li>
616
<li>
617 618 619
<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/>
620 621
[<a href="http://hal.inria.fr/hal-00978364">WWW</a>]
[<a href="http://hal.inria.fr/hal-00978364/PDF/ordonnanceurs_modulaires.pdf">PDF</a>]
622 623
</li>
</ol>
624
<h4>On The C Extensions</h4> 
625
<a name="PublicationsOnTheCExtensions"></a>
626
<ol>
627 628 629 630
<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/>
631 632
[<a href="http://hal.inria.fr/hal-00807033">WWW</a>]
[<a href="http://hal.inria.fr/hal-00807033/PDF/RR-8278.pdf">PDF</a>]
633 634
</li>
</ol>
635
<h4>On OpenMP Support on top of StarPU</h4> 
636
<a name="PublicationsOnOpenMPSupportontopofStarPU"></a>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
637
<ol>
638
<li>
Nathalie Furmento's avatar
Nathalie Furmento committed
639 640 641 642 643 644 645 646
<a name="agullo:hal-01517153"></a>Emmanuel Agullo, Olivier Aumage, Berenger Bramas, Olivier Coulaud,  and Samuel Pitoiset<br/>
<strong>Bridging the gap between OpenMP and task-based runtime systems for the fast multipole method</strong><br/>
<em>IEEE Transactions on Parallel and Distributed Systems</em>, April 2017<br/>
[<a href="https://hal.inria.fr/hal-01517153">WWW</a>]
[<a href="https://hal.inria.fr/hal-01517153/file/tpds_kstar_scalfmm_print.pdf">PDF</a>]
[doi:<a href="http://dx.doi.org/10.1109/TPDS.2017.2697857">10.1109/TPDS.2017.2697857</a>]
</li>
<li>
647
<a name="agullo:hal-01372022"></a>Emmanuel Agullo, Olivier Aumage, Berenger Bramas, Olivier Coulaud,  and Samuel Pitoiset<br/>
648
<strong>Bridging the gap between OpenMP 4.0 and native runtime systems for the fast multipole method</strong><br/>
649 650 651 652 653
Research Report RR-8953, Inria, March 2016<br/>
[<a href="https://hal.inria.fr/hal-01372022">WWW</a>]
[<a href="https://hal.inria.fr/hal-01372022/file/RR-8953.pdf">PDF</a>]
</li>
<li>
654 655 656 657
<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/>
658
[<a href="https://hal.inria.fr/hal-01081974">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
659
[<a href="https://hal.inria.fr/hal-01081974/document">PDF</a>]
660
[doi:<a href="http://dx.doi.org/10.1007/978-3-319-11454-5_2">10.1007/978-3-319-11454-5_2</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
661 662
</li>
</ol>
663
<h4>On MPI Support</h4> 
664
<a name="PublicationsOnMPISupport"></a>
665 666
<ol>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
667
<a name="agullo:hal-01618526"></a>Emmanuel Agullo, Olivier Aumage, Mathieu Faverge, Nathalie Furmento, Florent Pruvost, Marc Sergent,  and Samuel Thibault<br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
668
<strong>Achieving High Performance on Supercomputers with a Sequential Task-based Programming Model</strong><br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
669 670 671
<em>IEEE Transactions on Parallel and Distributed Systems</em>, 2017<br/>
[<a href="https://hal.inria.fr/hal-01618526">WWW</a>]
[<a href="https://hal.inria.fr/hal-01618526/file/tpds14.pdf">PDF</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
672 673 674 675
</li>
<li>
<a name="sergent:tel-01483666"></a>Marc Sergent<br/>
<strong>Scalability of a task-based runtime system for dense linear algebra applications</strong><br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
676
PhD thesis, Université de Bordeaux, December 2016<br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
677 678
[<a href="https://tel.archives-ouvertes.fr/tel-01483666">WWW</a>]
[<a href="https://tel.archives-ouvertes.fr/tel-01483666/file/SERGENT_MARC_2016.pdf">PDF</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
679 680
</li>
<li>
681 682 683 684 685 686 687
<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>
688 689 690 691 692 693 694
<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>
695 696 697 698 699
<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/>
700
[<a href="http://hal.inria.fr/hal-00725477">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
701
[<a href="http://hal.inria.fr/hal-00725477/document">PDF</a>]
702
</li>
703
</ol>
Nathalie Furmento's avatar
Nathalie Furmento committed
704
<h4>On Memory Control</h4> 
705
<a name="PublicationsOnMemoryControl"></a>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
706 707
<ol>
<li>
Nathalie Furmento's avatar
Nathalie Furmento committed
708
<a name="sergent:hal-01284004"></a>Marc Sergent, David Goudin, Samuel Thibault,  and Olivier Aumage<br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
709
<strong>Controlling the Memory Subscription of Distributed Applications with a Task-Based Runtime System</strong><br/>
Nathalie Furmento's avatar
Nathalie Furmento committed
710
In <em>21st International Workshop on High-Level Parallel Programming Models and Supportive Environments</em>, Chicago, United States, May 2016<br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
711 712 713 714
[<a href="https://hal.inria.fr/hal-01284004">WWW</a>]
[<a href="https://hal.inria.fr/hal-01284004/file/PID4127657.pdf">PDF</a>]
</li>
</ol>
715
<h4>On Data Transfer Management</h4> 
716
<a name="PublicationsOnDataTransferManagement"></a>
717 718
<ol>
<li>
719 720 721
<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/>
722
[<a href="http://hal.inria.fr/inria-00523937">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
723
[<a href="http://hal.inria.fr/inria-00523937/document">PDF</a>]
724
[doi:<a href="http://dx.doi.org/10.1109/ICPADS.2010.129">10.1109/ICPADS.2010.129</a>]
725 726
</li>
</ol>
727
<h4>On Performance Model Tuning</h4> 
728
<a name="PublicationsOnPerformanceModelTuning"></a>
729 730
<ol>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
731 732 733 734 735 736 737
<a name="agullo:hal-01474556"></a>Emmanuel Agullo, Bérenger Bramas, Olivier Coulaud, Luka Stanisic,  and Samuel Thibault<br/>
<strong>Modeling Irregular Kernels of Task-based codes: Illustration with the Fast Multipole Method</strong><br/>
Research Report RR-9036, INRIA Bordeaux, February 2017<br/>
[<a href="https://hal.inria.fr/hal-01474556">WWW</a>]
[<a href="https://hal.inria.fr/hal-01474556/file/rapport.pdf">PDF</a>]
</li>
<li>
738 739 740 741
<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/>
742
[<a href="http://hal.inria.fr/inria-00421333">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
743
[<a href="http://hal.inria.fr/inria-00421333/document">PDF</a>]
744
[doi:<a href="http://dx.doi.org/10.1007/978-3-642-14122-5_9">10.1007/978-3-642-14122-5_9</a>]
745 746
</li>
</ol>
747
<h4>On The Simulation Support through SimGrid</h4> 
748
<a name="PublicationsOnTheSimulationSupportthroughSimGrid"></a>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
749
<ol>
THIBAULT Samuel's avatar
update  
THIBAULT Samuel committed
750
<li>
751 752 753
<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/>
754 755
[<a href="https://hal.inria.fr/hal-01147997">WWW</a>]
[<a href="https://hal.inria.fr/hal-01147997/file/CCPE14_article.pdf">PDF</a>]
756
[doi:<a href="http://dx.doi.org/10.1002/cpe">10.1002/cpe</a>]
THIBAULT Samuel's avatar
update  
THIBAULT Samuel committed
757
</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
758
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
759 760 761 762 763 764 765
<a name="stanisic:hal-01180272"></a>Luka Stanisic, Emmanuel Agullo, Alfredo Buttari, Abdou Guermouche, Arnaud Legrand, Florent Lopez,  and Brice Videau<br/>
<strong>Fast and Accurate Simulation of Multithreaded Sparse Linear Algebra Solvers</strong><br/>
In <em>The 21st IEEE International Conference on Parallel and Distributed Systems</em>, Melbourne, Australia, December 2015<br/>
[<a href="https://hal.inria.fr/hal-01180272">WWW</a>]
[<a href="https://hal.inria.fr/hal-01180272/file/QRMSTARSG_article.pdf">PDF</a>]
</li>
<li>
766 767 768 769
<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/>
770 771
[<a href="http://hal.inria.fr/hal-01011633">WWW</a>]
[<a href="http://hal.inria.fr/hal-01011633/PDF/StarPUSG_article.pdf">PDF</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
772 773
</li>
</ol>
774
<h4>On The Cell Support</h4> 
775
<a name="PublicationsOnTheCellSupport"></a>
776 777
<ol>
<li>
778 779 780
<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/>
781
[<a href="http://hal.inria.fr/inria-00378705">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
782
[<a href="http://hal.inria.fr/inria-00378705/document">PDF</a>]
783
[doi:<a href="http://dx.doi.org/10.1007/978-3-642-03138-0_36">10.1007/978-3-642-03138-0_36</a>]
784 785
</li>
</ol>
786
<h4>On Applications</h4> 
787
<a name="PublicationsOnApplications"></a>
788
<ol>
THIBAULT Samuel's avatar
update  
THIBAULT Samuel committed
789
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
790 791 792 793 794 795 796 797
<a name="couteyencarpaye:hal-01507613"></a>Jean Marie Couteyen Carpaye, Jean Roman,  and Pierre Brenner<br/>
<strong>Design and Analysis of a Task-based Parallelization over a Runtime System of an Explicit Finite-Volume CFD Code with Adaptive Time Stepping</strong><br/>
<em>International Journal of Computational Science and Engineering</em>, pp 1 - 22, 2017<br/>
[<a href="https://hal.inria.fr/hal-01507613">WWW</a>]
[<a href="https://hal.inria.fr/hal-01507613/file/flusepa-task-hal-inria-preprint.pdf">PDF</a>]
[doi:<a href="http://dx.doi.org/10.1016/j.jocs.2017.03.008">10.1016/j.jocs.2017.03.008</a>]
</li>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
798 799 800 801 802 803 804
<a name="agullo:hal-01473475"></a>Emmanuel Agullo, Alfredo Buttari, Mikko Byckling, Abdou Guermouche,  and Ian Masliah<br/>
<strong>Achieving high-performance with a sparse direct solver on Intel KNL</strong><br/>
Research Report RR-9035, Inria Bordeaux Sud-Ouest ; CNRS-IRIT ; Intel corporation ; Université Bordeaux, February 2017<br/>
[<a href="https://hal.inria.fr/hal-01473475">WWW</a>]
[<a href="https://hal.inria.fr/hal-01473475/file/RR-9035.pdf">PDF</a>]
</li>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
805 806 807 808 809 810 811
<a name="agullo:hal-01387482"></a>Emmanuel Agullo, Bérenger Bramas, Olivier Coulaud, Martin Khannouz,  and Luka Stanisic<br/>
<strong>Task-based fast multipole method for clusters of multicore processors</strong><br/>
Research Report RR-8970, Inria Bordeaux Sud-Ouest, October 2016<br/>
[<a href="https://hal.inria.fr/hal-01387482">WWW</a>]
[<a href="https://hal.inria.fr/hal-01387482/file/report-8970.pdf">PDF</a>]
</li>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
812 813 814 815 816 817 818
<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>
819 820 821
<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/>
822 823
[<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>]
THIBAULT Samuel's avatar
update  
THIBAULT Samuel committed
824
</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
825
<li>
826 827 828
<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/>
829 830
[<a href="https://hal.inria.fr/hal-01182746">WWW</a>]
[<a href="https://hal.inria.fr/hal-01182746/file/sbac2015_soumission.pdf">PDF</a>]
831 832
</li>
<li>
833 834 835 836
<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/>
837 838
[<a href="http://hal.inria.fr/hal-01005765">WWW</a>]
[<a href="http://hal.inria.fr/hal-01005765/PDF/final.pdf">PDF</a>]
839 840
</li>
<li>
841 842 843 844 845
<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/>
846 847
[<a href="http://hal.inria.fr/hal-00987094">WWW</a>]
[<a href="http://hal.inria.fr/hal-00987094/PDF/sparsegpus.pdf">PDF</a>]
848 849
</li>
<li>
850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866
<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/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
867 868
[<a href="https://tel.archives-ouvertes.fr/tel-00958494">WWW</a>]
[<a href="https://tel.archives-ouvertes.fr/tel-00958494/document">PDF</a>]
869 870 871 872 873 874
</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>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
875
[<a href="http://tel.archives-ouvertes.fr/tel-00948309/document">PDF</a>]
876 877
</li>
<li>
878 879 880
<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/>
881 882
[<a href="http://hal.inria.fr/hal-00851122">WWW</a>]
[<a href="http://hal.inria.fr/hal-00851122/PDF/fhpc13.pdf">PDF</a>]
883 884
</li>
<li>
885 886 887
<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/>
888 889
[<a href="http://hal.inria.fr/hal-00920915">WWW</a>]
[<a href="http://hal.inria.fr/hal-00920915/PDF/ADPC2013-117.pdf">PDF</a>]
890 891 892 893 894 895
</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/>
896
[<a href="http://hal.inria.fr/hal-00926144">WWW</a>]
897 898 899 900 901
</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/>
902
[<a href="http://hal.inria.fr/hal-00773571">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
903
[<a href="http://hal.inria.fr/hal-00773571/document">PDF</a>]
904 905
</li>
<li>
906 907 908
<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/>
909 910
[<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>]
911
</li>
912
<li>
913 914 915 916
<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>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
917
[<a href="http://hal.inria.fr/hal-00772742/document">PDF</a>]
918 919 920 921 922 923 924 925
</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>
926 927 928
<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/>
929 930
[<a href="http://hal.inria.fr/hal-00697020">WWW</a>]
[<a href="http://hal.inria.fr/hal-00697020/PDF/europar2012-submitted.pdf">PDF</a>]
931 932
</li>
<li>
933 934 935
<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/>
936 937
[<a href="http://hal.inria.fr/hal-00776610">WWW</a>]
[<a href="http://hal.inria.fr/hal-00776610/PDF/date12-paper.pdf">PDF</a>]
938 939
</li>
<li>
940 941 942 943 944
<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>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
945
[<a href="http://hal.inria.fr/hal-00648480/document">PDF</a>]
946 947 948
[doi:<a href="http://dx.doi.org/10.1109/MM.2011.67">10.1109/MM.2011.67</a>]
</li>
<li>
949 950 951
<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/>
952
[<a href="http://hal.inria.fr/hal-00654193">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
953
[<a href="http://hal.inria.fr/hal-00654193/document">PDF</a>]
954 955 956 957 958
</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/>
959
[<a href="http://hal.inria.fr/inria-00547614">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
960
[<a href="http://hal.inria.fr/inria-00547614/document">PDF</a>]