index.html 53.6 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>
62 63 64 65 66 67
<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>
<p>
68 69
The latest documentation in <a href="./testing/master/doc/starpu.pdf">PDF</a>
and <a href="./testing/master/doc/html">HTML</a> is updated everyday, but covers
70
the latest developments which may not be available in the latest release.
71 72 73 74
</p>
</div>

<div class="section emphasize newslist" id="news">
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
75 76
<h3>News</h3>
<p>
77 78 79 80 81
March 2018 <b>&raquo;&nbsp;</b>A <a href="https://events.prace-ri.eu/event/681/">tutorial</a>
      "Runtime systems for heterogeneous platform programming" will be
      given at the Maison de la Simulation in June 2018.
</p>
<p>
82 83 84 85 86 87 88
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>
89 90 91
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
92
  the <a href="https://sonarqube.bordeaux.inria.fr/sonarqube/dashboard?id=storm%3Astarpu%3Arelease%3Av1.2">StarPU
93 94 95
  page</a> on the sonarqube server of INRIA.
</p>
<p>
96 97 98 99 100 101 102
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>
103 104 105 106 107
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>
108 109 110 111 112 113
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>
114
</div>
115 116

<div class="section emphasizebot" style="text-align: right; font-style: italic;">
117
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>.
118
See also the full <a href="news/">news</a>.
119 120
</div>

121 122 123 124 125 126 127 128 129 130 131
<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
132
<div class="section" id="tutorial">
133 134 135 136 137 138 139 140 141 142
<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
143 144 145 146 147 148 149 150
<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>

151 152
<div class="section" id="contact">
<h3>Contact</h3>
153
<p>For any questions regarding StarPU, please contact the StarPU developers mailing list.</p>
154 155 156
<pre>
<a href="mailto:starpu-devel@lists.gforge.inria.fr?subject=StarPU">starpu-devel@lists.gforge.inria.fr</a>
</pre>
157
<p>Details of the <a href="people/">StarPU team people</a> are also available.</p>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
158 159
</div>

160 161
<div class="section" id="features">
<h3>Features</h3>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
162

163
<h4>Portability</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
164
  <p>
165 166 167 168 169 170 171 172
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.
173 174 175 176
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
177 178
  </p>

179
<h4>Data transfers</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
180
  <p>
181
To relieve programmers from the burden of explicit data transfers, a high-level
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
182
data management library enforces memory coherency over the machine: before a
183 184
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
185 186
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
187
prefetching</b> data, which thus permits to <b>overlap data transfers with computations</b>
THIBAULT Samuel's avatar
update  
THIBAULT Samuel committed
188
(including <b>GPU-GPU direct transfers</b>) to achieve the most of the architecture.
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
189 190
  </p>

191
<h4>Dependencies</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
192
  <p>
193
Dependencies between tasks can be given either of several ways, to provide the
194 195
programmer with best flexibility:
  <ul>
196
    <li><b>implicitly</b> from RAW, WAW, and WAR data dependencies.</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
197
    <li>explicitly through <b>tags</b> which act as rendez-vous points between
198
    tasks (thus including tasks which have not been created yet),</li>
199
    <li><b>explicitly</b> between pairs of tasks,</li>
200
  </ul>
201 202
  </p>
  <p>
203 204 205
  These dependencies are computed in a completely decentralized way.
  </p>
  <p>
206 207 208 209
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).
210 211 212 213 214
  </p>

<h4>Heterogeneous Scheduling</h4>
  <p>
StarPU obtains
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
215
portable performances by efficiently (and easily) using all computing resources
216
at the same time. StarPU also takes advantage of the <b>heterogeneous</b> nature of a
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
217
machine, for instance by using scheduling strategies based on auto-tuned
218 219 220
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>.
221
Various strategies and variants are available. Some of them are centralized, but
THIBAULT Samuel's avatar
THIBAULT Samuel committed
222
most of them are <b>completely distributed</b>. dmda (a data-locality-aware MCT strategy,
223
thus similar to heft but starts executing tasks before the whole task graph is
224 225 226
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
227 228 229
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
230 231
  </p>

232 233
<h4>Clusters</h4>
  <p>
234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255
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>
256 257 258 259 260
  </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
261
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
262 263
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.
264 265
  </p>

266 267 268
<h4>Extensions to the C Language</h4>
<p>
  StarPU comes with a GCC plug-in
269
  that <a href="doc/html/cExtensions.html">extends the C programming
270 271
  language</a> with pragmas and attributes that make it easy
  to <b>annotate a sequential C program to turn it into a parallel
272 273 274
  StarPU program</b> (new in v1.0).
</p>

THIBAULT Samuel's avatar
THIBAULT Samuel committed
275 276
<h4>OpenMP 4 -compatible interface</h4>
<p>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
277
  <a href=http://kstar.gforge.inria.fr/>K'Star</a> provides an OpenMP
THIBAULT Samuel's avatar
THIBAULT Samuel committed
278 279 280 281 282 283 284 285 286
  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>

287 288 289 290
<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).
291 292
</p>

293 294 295 296
<h4>Simulation support</h4>
<p>
  StarPU can very accurately simulate an application execution
  and measure the resulting performance thanks to using the
297
  <a href="http://simgrid.gforge.inria.fr">SimGrid simulator</a> (new in v1.1).  This allows
298 299 300 301 302
  to quickly experiment with various scheduling heuristics, various application
  algorithms, and even various platforms (available GPUs and CPUs, available
  bandwidth)!
</p>

303 304
<h4>All in all</h4>
  <p>
305
All that means that, with the help
306
of <a href="doc/html/cExtensions.html">StarPU's extensions to the C
307 308
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
309
StarPU code, possibly running on all the CPUs and GPUs, and given a data
Nathalie Furmento's avatar
Nathalie Furmento committed
310
distribution over MPI nodes, it is even a distributed version!
311 312 313 314 315 316 317 318 319 320 321 322 323
  </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
324

325
<h4>Supported Architectures</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
326
<ul>
327
<li>SMP/Multicore Processors (x86, PPC, ARM, ... all Debian architecture have been tested) </li>
Nathalie Furmento's avatar
Nathalie Furmento committed
328
<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
329 330
<li>OpenCL devices</li>
<li>Cell Processors (experimental)</li>
Nathalie Furmento's avatar
Nathalie Furmento committed
331 332
<li>Intel SCC (experimental, new in v1.2)</li>
<li>Intel MIC / Xeon Phi (new in v1.2)</li>
333
</ul>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
334

335
<h4>Supported Operating Systems</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
336
<ul>
Ludovic Courtès's avatar
Ludovic Courtès committed
337 338
<li>GNU/Linux</li>
<li>Mac OS X</li>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
339 340 341
<li>Windows</li>
</ul>

342
<h4>Performance analysis tools</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
343 344 345 346 347 348 349 350 351
  <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>

352
<p>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
353 354 355 356 357 358
<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>
359
<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>
360 361
</p>

Nathalie Furmento's avatar
website  
Nathalie Furmento committed
362 363 364 365 366 367
<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
368
GFlop/s</b>
369
<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
370 371
</p>

THIBAULT Samuel's avatar
THIBAULT Samuel committed
372 373
<p>
<a href="http://www.hlrs.de/temanejo">Temanejo</a> can be used to debug the task
374
graph, as shown below (new in v1.1).
THIBAULT Samuel's avatar
THIBAULT Samuel committed
375 376 377
</p>

<center>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
378
<a href="images/temanejo.png"><img src="images/temanejo.png" width="50%"/></a>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
379 380
</center>

Nathalie Furmento's avatar
website  
Nathalie Furmento committed
381 382
</div>

383 384 385 386 387
<div class="section" id="software">
<h3>Software using StarPU</h3>

<p>
Some software is known for being able to use StarPU to tackle heterogeneous
388
architectures, here is a non-exhaustive list (feel free to ask to be added to the
389
list!):
390 391 392
</p>

<ul>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
393
	<li><a href="https://project.inria.fr/chameleon/">Chameleon</a>, dense linear algebra library</li>
THIBAULT Samuel's avatar
fix URL  
THIBAULT Samuel committed
394
	<li><a href="http://github.com/ecrc/exageostat">ExaGeoStat</a>, Machine learning framework for Climate/Weather prediction applications</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
395
	<li><a href="http://github.com/ecrc/hicma">HiCMA</a>, Low-rank general linear algebra library</li>
396
        <li><a href=http://kstar.gforge.inria.fr/>K'Star</a>, OpenMP 4 - compatible interface on top of StarPU.</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
397
	<li><a href="http://github.com/ecrc/ksvd">KSVD</a>, dense SVD on distributed-memory manycore systems</li>
398 399
	<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>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
400
	<li><a href="http://github.com/ecrc/moao">MOAO</a>, HPC framework for computational astronomy, servicing the European Extremely Large Telescope and the Japanese Subaru Telescope</li>
401
	<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
402
	<li><a href="http://github.com/ecrc/qdwh">QDWH</a>, QR-based Dynamically Weighted Halley</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
403
	<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
404
	<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
405
	<li><a href="https://tel.archives-ouvertes.fr/tel-01410049/">SCHNAPS</a>, Solver for Conservative Hypebolic Non-linear systems Applied to PlasmaS. </li>
406 407
	<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>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
408
	<li><a href="http://github.com/ecrc/stars-h">STARS-H</a>, HPC low-rank matrix market</li>
409 410
</ul>

411
<p>
412
You can find <a href="#PublicationsOnApplications">below</a> the list of publications related to applications using StarPU.
413 414
</p>

415 416
</div>

THIBAULT Samuel's avatar
THIBAULT Samuel committed
417 418 419 420
<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
421 422 423
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
424 425 426
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
427 428

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

<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
441 442 443 444 445 446 447 448 449 450 451 452 453 454
<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
455 456 457 458 459 460 461 462 463 464
<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
465
38400	9915	1903.7</pre></tt>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
466 467
<p>(MPI simulation is not supported yet)</p>

468 469 470 471
<div class="section" id="publications">
<h3>Publications</h3>
<p>
All StarPU related publications are also
472
listed <a href="./publications">here</a>
473 474 475
with the corresponding Bibtex entries.
</p>

476 477
<p>
A good overview is available in
478 479 480
the following <a href="http://hal.archives-ouvertes.fr/inria-00467677">Research Report</a>.
</p>

481 482
<p>
If you need to cite StarPU, please
483
reference <a href="publications/Year/2011.html#AugThiNamWac11CCPE">[StarPU: A Unified Platform
484 485 486 487 488
    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>

489
<h4>General Presentations</h4> 
490
<a name="PublicationsGeneralPresentations"></a>
491 492
<ol>
<li>
493 494 495
<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/>
496
[<a href="http://tel.archives-ouvertes.fr/tel-00777154">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
497
[<a href="http://tel.archives-ouvertes.fr/tel-00777154/document">PDF</a>]
498 499
</li>
<li>
500 501 502
<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/>
503
[<a href="http://hal.inria.fr/inria-00550877">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
504
[<a href="http://hal.inria.fr/inria-00550877/document">PDF</a>]
505
[doi:<a href="http://dx.doi.org/10.1002/cpe.1631">10.1002/cpe.1631</a>]
506
</li>
507
<li>
508 509 510 511
<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
512
[<a href="http://hal.inria.fr/inria-00467677/document">PDF</a>]
513 514
</li>
<li>
515 516 517 518
<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/>
519
[<a href="http://hal.inria.fr/inria-00411581">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
520
[<a href="http://hal.inria.fr/inria-00411581/document">PDF</a>]
521 522
</li>
<li>
523 524 525 526
<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/>
527
[<a href="http://hal.inria.fr/inria-00384363">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
528
[<a href="http://hal.inria.fr/inria-00384363/document">PDF</a>]
529 530 531 532 533 534 535 536
[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/>
537
[<a href="http://hal.inria.fr/inria-00326917">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
538
[<a href="http://hal.inria.fr/inria-00326917/document">PDF</a>]
539 540 541 542 543 544
[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/>
545
[<a href="http://hal.inria.fr/inria-00289361">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
546
[<a href="http://hal.inria.fr/inria-00289361/document">PDF</a>]
547 548
</li>
</ol>
549
<h4>On Composability</h4> 
550
<a name="PublicationsOnComposability"></a>
Nathalie Furmento's avatar
Nathalie Furmento committed
551 552
<ol>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
553 554 555 556 557 558 559
<a name="hugo:tel-01162975"></a>Andra-Ecaterina Hugo<br/>
<strong>Composability of parallel codes on heterogeneous architectures</strong><br/>
Theses, Université de Bordeaux, December 2014<br/>
[<a href="https://tel.archives-ouvertes.fr/tel-01162975">WWW</a>]
[<a href="https://tel.archives-ouvertes.fr/tel-01162975/file/HUGO_ANDRA_2014.pdf">PDF</a>]
</li>
<li>
560 561 562
<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/>
563
[<a href="http://hal.inria.fr/hal-00773610">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
564
[<a href="http://hal.inria.fr/hal-00773610/document">PDF</a>]
Nathalie Furmento's avatar
Nathalie Furmento committed
565 566
</li>
<li>
567 568 569
<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/>
570
[<a href="http://hal.inria.fr/hal-00824514">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
571
[<a href="http://hal.inria.fr/hal-00824514/document">PDF</a>]
572 573 574 575 576
</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/>
577
[<a href="http://hal.inria.fr/inria-00619654/en/">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
578
[<a href="http://hal.inria.fr/inria-00619654/document">PDF</a>]
Nathalie Furmento's avatar
Nathalie Furmento committed
579 580
</li>
</ol>
581
<h4>On Scheduling</h4> 
582
<a name="PublicationsOnScheduling"></a>
583
<ol>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
584
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
585 586
<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
587
PhD thesis, Université de Bordeaux, April 2017<br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
588 589 590 591
[<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>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
592 593 594 595 596 597 598
<a name="garciapinto:hal-01616632"></a>Vinicius Garcia Pinto, Lucas Mello Schnorr, Luka Stanisic, Arnaud Legrand, Samuel Thibault,  and Vincent Danjean<br/>
<strong>A Visual Performance Analysis Framework for Task-based Parallel Applications running on Hybrid Clusters</strong><br/>
Note: Working paper or preprint, October 2017<br/>
[<a href="https://hal.inria.fr/hal-01616632">WWW</a>]
[<a href="https://hal.inria.fr/hal-01616632/file/CCPE_article_submitted_2017_09_29-gz.pdf">PDF</a>]
</li>
<li>
599 600 601 602 603 604 605 606
<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
607 608 609 610
<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
611
[<a href="https://hal.inria.fr/hal-01361992v2/document">PDF</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
612 613
</li>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
614 615
<a name="cojean:hal-01181135"></a>Terry Cojean, Abdou Guermouche, Andra Hugo, Raymond Namyst,  and Pierre-André Wacrenier<br/>
<strong>Resource aggregation for task-based Cholesky Factorization on top of heterogeneous machines</strong><br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
616
In <em>HeteroPar'2016 worshop of Euro-Par</em>, Grenoble, France, August 2016<br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
617 618
[<a href="https://hal.inria.fr/hal-01181135">WWW</a>]
[<a href="https://hal.inria.fr/hal-01181135/file/papier%20%281%29.pdf">PDF</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
619 620
</li>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
621 622 623 624 625 626 627
<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
628
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
629 630 631
<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
632
[<a href="http://ieeexplore.ieee.org/abstract/document/7789322/">WWW</a>]
633 634 635
[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
636 637 638 639 640 641
<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
642
<li>
643 644 645 646 647 648 649
<a name="cojean:hal-01409965"></a>Terry Cojean, Abdou Guermouche, Andra Hugo, Raymond Namyst,  and Pierre-André Wacrenier<br/>
<strong>Resource aggregation for task-based Cholesky Factorization on top of modern architectures</strong><br/>
Note: This paper is submitted for review to the Parallel Computing special issue for HCW and HeteroPar 16 workshops, November 2016<br/>
[<a href="https://hal.inria.fr/hal-01409965">WWW</a>]
[<a href="https://hal.inria.fr/hal-01409965/file/submission.pdf">PDF</a>]
</li>
<li>
650 651 652
<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/>
653
[<a href="https://hal.inria.fr/hal-01120507">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
654
[<a href="https://hal.inria.fr/hal-01120507/document">PDF</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
655
</li>
656
<li>
657 658 659
<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/>
660 661
[<a href="http://hal.inria.fr/hal-00978364">WWW</a>]
[<a href="http://hal.inria.fr/hal-00978364/PDF/ordonnanceurs_modulaires.pdf">PDF</a>]
662 663
</li>
</ol>
664
<h4>On The C Extensions</h4> 
665
<a name="PublicationsOnTheCExtensions"></a>
666
<ol>
667 668 669 670
<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/>
671 672
[<a href="http://hal.inria.fr/hal-00807033">WWW</a>]
[<a href="http://hal.inria.fr/hal-00807033/PDF/RR-8278.pdf">PDF</a>]
673 674
</li>
</ol>
675
<h4>On OpenMP Support on top of StarPU</h4> 
676
<a name="PublicationsOnOpenMPSupportontopofStarPU"></a>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
677
<ol>
678
<li>
Nathalie Furmento's avatar
Nathalie Furmento committed
679 680 681 682 683 684 685 686
<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>
687
<a name="agullo:hal-01372022"></a>Emmanuel Agullo, Olivier Aumage, Berenger Bramas, Olivier Coulaud,  and Samuel Pitoiset<br/>
688
<strong>Bridging the gap between OpenMP 4.0 and native runtime systems for the fast multipole method</strong><br/>
689 690 691 692 693
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>
694 695 696 697
<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/>
698
[<a href="https://hal.inria.fr/hal-01081974">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
699
[<a href="https://hal.inria.fr/hal-01081974/document">PDF</a>]
700
[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
701 702
</li>
</ol>
703
<h4>On MPI Support</h4> 
704
<a name="PublicationsOnMPISupport"></a>
705 706
<ol>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
707
<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
708
<strong>Achieving High Performance on Supercomputers with a Sequential Task-based Programming Model</strong><br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
709 710 711
<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
712 713 714 715
</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
716
PhD thesis, Université de Bordeaux, December 2016<br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
717 718
[<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
719 720
</li>
<li>
721 722 723 724 725 726 727
<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>
728 729 730 731 732 733 734
<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>
735 736 737 738 739
<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/>
740
[<a href="http://hal.inria.fr/hal-00725477">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
741
[<a href="http://hal.inria.fr/hal-00725477/document">PDF</a>]
742
</li>
743
</ol>
Nathalie Furmento's avatar
Nathalie Furmento committed
744
<h4>On Memory Control</h4> 
745
<a name="PublicationsOnMemoryControl"></a>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
746 747
<ol>
<li>
Nathalie Furmento's avatar
Nathalie Furmento committed
748
<a name="sergent:hal-01284004"></a>Marc Sergent, David Goudin, Samuel Thibault,  and Olivier Aumage<br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
749
<strong>Controlling the Memory Subscription of Distributed Applications with a Task-Based Runtime System</strong><br/>
Nathalie Furmento's avatar
Nathalie Furmento committed
750
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
751 752 753 754
[<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>
755
<h4>On Data Transfer Management</h4> 
756
<a name="PublicationsOnDataTransferManagement"></a>
757 758
<ol>
<li>
759 760 761
<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/>
762
[<a href="http://hal.inria.fr/inria-00523937">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
763
[<a href="http://hal.inria.fr/inria-00523937/document">PDF</a>]
764
[doi:<a href="http://dx.doi.org/10.1109/ICPADS.2010.129">10.1109/ICPADS.2010.129</a>]
765 766
</li>
</ol>
767
<h4>On Performance Model Tuning</h4> 
768
<a name="PublicationsOnPerformanceModelTuning"></a>
769 770
<ol>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
771 772 773 774 775 776 777
<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>
778 779 780 781
<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/>
782
[<a href="http://hal.inria.fr/inria-00421333">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
783
[<a href="http://hal.inria.fr/inria-00421333/document">PDF</a>]
784
[doi:<a href="http://dx.doi.org/10.1007/978-3-642-14122-5_9">10.1007/978-3-642-14122-5_9</a>]
785 786
</li>
</ol>
787
<h4>On The Simulation Support through SimGrid</h4> 
788
<a name="PublicationsOnTheSimulationSupportthroughSimGrid"></a>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
789
<ol>
THIBAULT Samuel's avatar
update  
THIBAULT Samuel committed
790
<li>
791 792 793
<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/>
794 795
[<a href="https://hal.inria.fr/hal-01147997">WWW</a>]
[<a href="https://hal.inria.fr/hal-01147997/file/CCPE14_article.pdf">PDF</a>]
796
[doi:<a href="http://dx.doi.org/10.1002/cpe">10.1002/cpe</a>]
THIBAULT Samuel's avatar
update  
THIBAULT Samuel committed
797
</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
798
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
799 800 801 802 803 804 805
<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>
806 807 808 809
<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/>
810 811
[<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
812 813
</li>
</ol>
814
<h4>On The Cell Support</h4> 
815
<a name="PublicationsOnTheCellSupport"></a>
816 817
<ol>
<li>
818 819 820
<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/>
821
[<a href="http://hal.inria.fr/inria-00378705">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
822
[<a href="http://hal.inria.fr/inria-00378705/document">PDF</a>]
823
[doi:<a href="http://dx.doi.org/10.1007/978-3-642-03138-0_36">10.1007/978-3-642-03138-0_36</a>]
824 825
</li>
</ol>
826
<h4>On Applications</h4> 
827
<a name="PublicationsOnApplications"></a>
828
<ol>
THIBAULT Samuel's avatar
update  
THIBAULT Samuel committed
829
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
830 831 832 833 834 835 836 837
<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
838 839 840 841 842 843 844
<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
845 846 847 848 849 850 851
<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
852 853 854 855 856 857 858
<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>
859 860 861
<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/>
862 863
[<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
864
</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
865
<li>
866 867 868
<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/>
869 870
[<a href="https://hal.inria.fr/hal-01182746">WWW</a>]
[<a href="https://hal.inria.fr/hal-01182746/file/sbac2015_soumission.pdf">PDF</a>]
871 872
</li>
<li>
873 874 875 876
<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/>
877 878
[<a href="http://hal.inria.fr/hal-01005765">WWW</a>]
[<a href="http://hal.inria.fr/hal-01005765/PDF/final.pdf">PDF</a>]
879 880
</li>
<li>
881 882 883 884 885
<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/>
886 887
[<a href="http://hal.inria.fr/hal-00987094">WWW</a>]
[<a href="http://hal.inria.fr/hal-00987094/PDF/sparsegpus.pdf">PDF</a>]
888 889
</li>
<li>
890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906
<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
907 908
[<a href="https://tel.archives-ouvertes.fr/tel-00958494">WWW</a>]
[<a href="https://tel.archives-ouvertes.fr/tel-00958494/document">PDF</a>]
909 910 911 912 913 914
</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
915
[<a href="http://tel.archives-ouvertes.fr/tel-00948309/document">PDF</a>]
916 917
</li>
<li>
918 919 920
<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/>
921 922
[<a href="http://hal.inria.fr/hal-00851122">WWW</a>]
[<a href="http://hal.inria.fr/hal-00851122/PDF/fhpc13.pdf">PDF</a>]
923 924
</li>
<li>
925 926 927
<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/>
928 929
[<a href="http://hal.inria.fr/hal-00920915">WWW</a>]
[<a href="http://hal.inria.fr/hal-00920915/PDF/ADPC2013-117.pdf">PDF</a>]
930 931 932 933 934 935
</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/>
936
[<a href="http://hal.inria.fr/hal-00926144">WWW</a>]
937 938 939 940 941
</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/>
942
[<a href="http://hal.inria.fr/hal-00773571">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
943
[<a href="http://hal.inria.fr/hal-00773571/document">PDF</a>]
944 945
</li>
<li>
946 947 948
<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/>
949 950
[<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>]
951
</li>
952
<li>
953 954 955 956
<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