index.html 50.2 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">
Nathalie Furmento's avatar
Nathalie Furmento committed
18
<a href="https://www.inria.fr/en/teams/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
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>
76 77 78 79 80 81 82
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>
83 84 85 86 87
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>
88 89 90 91 92 93 94
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>
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
      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>
<p>
102 103
August 2016 <b>&raquo;&nbsp;</b><a href="http://gforge.inria.fr/frs/?group_id=1570"><b>The
      1.2.0 release of StarPU is now available!</b></a>.
104 105 106 107
      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>
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>

126 127 128 129 130 131 132 133 134 135 136
<div class="section" id="video">
<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>

137 138
<div class="section" id="contact">
<h3>Contact</h3>
139
<p>For any questions regarding StarPU, please contact the StarPU developers mailing list.</p>
140 141 142
<pre>
<a href="mailto:starpu-devel@lists.gforge.inria.fr?subject=StarPU">starpu-devel@lists.gforge.inria.fr</a>
</pre>
143
<p>Details of the <a href="people/">StarPU team people</a> are also available.</p>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
144 145
</div>

146 147
<div class="section" id="features">
<h3>Features</h3>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
148

149
<h4>Portability</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
150
  <p>
151 152 153 154 155 156 157 158
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.
159 160 161 162
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
163 164
  </p>

165
<h4>Data transfers</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
166
  <p>
167
To relieve programmers from the burden of explicit data transfers, a high-level
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
168
data management library enforces memory coherency over the machine: before a
169 170
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
171 172
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
173
prefetching</b> data, which thus permits to <b>overlap data transfers with computations</b>
THIBAULT Samuel's avatar
update  
THIBAULT Samuel committed
174
(including <b>GPU-GPU direct transfers</b>) to achieve the most of the architecture.
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
175 176
  </p>

177
<h4>Dependencies</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
178
  <p>
179 180 181
Dependencies between tasks can be given several ways, to provide the
programmer with best flexibility:
  <ul>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
182 183
    <li><b>explicitly</b> between pairs of tasks,</li>
    <li>explicitly through <b>tags</b> which act as rendez-vous points between
184
    tasks (thus including tasks which have not been created yet),</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
185
    <li><b>implicitly</b> from RAW, WAW, and WAR data dependencies.</li>
186
  </ul>
187 188
  </p>
  <p>
189 190 191
  These dependencies are computed in a completely decentralized way.
  </p>
  <p>
192 193 194 195
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).
196 197 198 199 200
  </p>

<h4>Heterogeneous Scheduling</h4>
  <p>
StarPU obtains
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
201
portable performances by efficiently (and easily) using all computing resources
202
at the same time. StarPU also takes advantage of the <b>heterogeneous</b> nature of a
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
203
machine, for instance by using scheduling strategies based on auto-tuned
204 205 206
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>.
207
Various strategies and variants are available. Some of them are centralized, but
THIBAULT Samuel's avatar
THIBAULT Samuel committed
208
most of them are <b>completely distributed</b>. dmda (a data-locality-aware MCT strategy,
209
thus similar to heft but starts executing tasks before the whole task graph is
210 211 212
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
213 214 215
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
216 217
  </p>

218 219
<h4>Clusters</h4>
  <p>
220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241
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>
242 243 244 245 246
  </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
247
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
248 249
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.
250 251
  </p>

252 253 254
<h4>Extensions to the C Language</h4>
<p>
  StarPU comes with a GCC plug-in
255
  that <a href="doc/html/cExtensions.html">extends the C programming
256 257
  language</a> with pragmas and attributes that make it easy
  to <b>annotate a sequential C program to turn it into a parallel
258 259 260
  StarPU program</b> (new in v1.0).
</p>

THIBAULT Samuel's avatar
THIBAULT Samuel committed
261 262
<h4>OpenMP 4 -compatible interface</h4>
<p>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
263
  <a href=http://kstar.gforge.inria.fr/>K'Star</a> provides an OpenMP
THIBAULT Samuel's avatar
THIBAULT Samuel committed
264 265 266 267 268 269 270 271 272
  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>

273 274 275 276
<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).
277 278
</p>

279 280 281 282
<h4>Simulation support</h4>
<p>
  StarPU can very accurately simulate an application execution
  and measure the resulting performance thanks to using the
283
  <a href="http://simgrid.gforge.inria.fr">SimGrid simulator</a> (new in v1.1).  This allows
284 285 286 287 288
  to quickly experiment with various scheduling heuristics, various application
  algorithms, and even various platforms (available GPUs and CPUs, available
  bandwidth)!
</p>

289 290
<h4>All in all</h4>
  <p>
291
All that means that, with the help
292
of <a href="doc/html/cExtensions.html">StarPU's extensions to the C
293 294
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
295
StarPU code, possibly running on all the CPUs and GPUs, and given a data
Nathalie Furmento's avatar
Nathalie Furmento committed
296
distribution over MPI nodes, it is even a distributed version!
297 298 299 300 301 302 303 304 305 306 307 308 309
  </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
310

311
<h4>Supported Architectures</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
312
<ul>
313
<li>SMP/Multicore Processors (x86, PPC, ARM, ... all Debian architecture have been tested) </li>
Nathalie Furmento's avatar
Nathalie Furmento committed
314
<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
315 316
<li>OpenCL devices</li>
<li>Cell Processors (experimental)</li>
Nathalie Furmento's avatar
Nathalie Furmento committed
317 318
<li>Intel SCC (experimental, new in v1.2)</li>
<li>Intel MIC / Xeon Phi (new in v1.2)</li>
319
</ul>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
320

321
<h4>Supported Operating Systems</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
322
<ul>
Ludovic Courtès's avatar
Ludovic Courtès committed
323 324
<li>GNU/Linux</li>
<li>Mac OS X</li>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
325 326 327
<li>Windows</li>
</ul>

328
<h4>Performance analysis tools</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
329 330 331 332 333 334 335 336 337
  <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>

338
<p>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
339 340 341 342 343 344
<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>
345
<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>
346 347
</p>

Nathalie Furmento's avatar
website  
Nathalie Furmento committed
348 349 350 351 352 353
<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
354
GFlop/s</b>
355
<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
356 357
</p>

THIBAULT Samuel's avatar
THIBAULT Samuel committed
358 359
<p>
<a href="http://www.hlrs.de/temanejo">Temanejo</a> can be used to debug the task
360
graph, as shown below (new in v1.1).
THIBAULT Samuel's avatar
THIBAULT Samuel committed
361 362 363
</p>

<center>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
364
<a href="images/temanejo.png"><img src="images/temanejo.png" width="50%"/></a>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
365 366
</center>

Nathalie Furmento's avatar
website  
Nathalie Furmento committed
367 368
</div>

369 370 371 372 373
<div class="section" id="software">
<h3>Software using StarPU</h3>

<p>
Some software is known for being able to use StarPU to tackle heterogeneous
374 375
architectures, here is a non-exhaustive list (feel free to ask to be in the
list!):
376 377 378 379
</p>

<ul>
	<li><a href="http://icl.cs.utk.edu/magma/">MAGMA</a>, dense linear algebra library, starting from version 1.1</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
380
	<li><a href="https://project.inria.fr/chameleon/">Chameleon</a>, dense linear algebra library</li>
381 382
	<li><a href="http://www.ida.liu.se/~chrke/skepu/">SkePU</a>, a skeleton programming framework.</li>
	<li><a href="http://pastix.gforge.inria.fr/">PaStiX</a>, sparse linear algebra library, starting from version 5.2.1</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
383
	<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
384
	<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
385
	<li><a href="https://gitlab.inria.fr/solverstack/maphys">MaPHyS</a>, Massively Parallel Hybrid Solver</li>
386
	<li><a href="https://hal.archives-ouvertes.fr/hal-01086246">SignalPU</a>, a Dataflow-Graph-specific programming model. </li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
387
	<li><a href="https://tel.archives-ouvertes.fr/tel-01410049/">SCHNAPS</a>, Solver for Conservative Hypebolic Non-linear systems Applied to PlasmaS. </li>
388
        <li><a href=http://kstar.gforge.inria.fr/>K'Star</a>, OpenMP 4 - compatible interface on top of StarPU.</li>
389 390
</ul>

391
<p>
392
You can find <a href="#PublicationsOnApplications">below</a> the list of publications related to applications using StarPU.
393 394
</p>

395 396
</div>

THIBAULT Samuel's avatar
THIBAULT Samuel committed
397 398 399 400
<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
401 402 403
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
404 405 406
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
407 408

<tt><pre>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
409 410 411 412 413 414
$ 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
415
$ 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
416 417 418 419 420

<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
421 422 423 424 425 426 427 428 429 430 431 432 433 434
<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
435 436 437 438 439 440 441 442 443 444
<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
445
38400	9915	1903.7</pre></tt>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
446 447
<p>(MPI simulation is not supported yet)</p>

448 449 450 451
<div class="section" id="publications">
<h3>Publications</h3>
<p>
All StarPU related publications are also
452
listed <a href="./publications">here</a>
453 454 455
with the corresponding Bibtex entries.
</p>

456 457
<p>
A good overview is available in
458 459 460
the following <a href="http://hal.archives-ouvertes.fr/inria-00467677">Research Report</a>.
</p>

461 462
<p>
If you need to cite StarPU, please
463
reference <a href="publications/Year/2011.html#AugThiNamWac11CCPE">[StarPU: A Unified Platform
464 465 466 467 468
    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>

469
<h4>General Presentations</h4> 
470
<a name="PublicationsGeneralPresentations"></a>
471 472
<ol>
<li>
473 474 475
<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/>
476
[<a href="http://tel.archives-ouvertes.fr/tel-00777154">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
477
[<a href="http://tel.archives-ouvertes.fr/tel-00777154/document">PDF</a>]
478 479
</li>
<li>
480 481 482
<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/>
483
[<a href="http://hal.inria.fr/inria-00550877">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
484
[<a href="http://hal.inria.fr/inria-00550877/document">PDF</a>]
485
[doi:<a href="http://dx.doi.org/10.1002/cpe.1631">10.1002/cpe.1631</a>]
486
</li>
487
<li>
488 489 490 491
<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
492
[<a href="http://hal.inria.fr/inria-00467677/document">PDF</a>]
493 494
</li>
<li>
495 496 497 498
<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/>
499
[<a href="http://hal.inria.fr/inria-00411581">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
500
[<a href="http://hal.inria.fr/inria-00411581/document">PDF</a>]
501 502
</li>
<li>
503 504 505 506
<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/>
507
[<a href="http://hal.inria.fr/inria-00384363">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
508
[<a href="http://hal.inria.fr/inria-00384363/document">PDF</a>]
509 510 511 512 513 514 515 516
[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/>
517
[<a href="http://hal.inria.fr/inria-00326917">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
518
[<a href="http://hal.inria.fr/inria-00326917/document">PDF</a>]
519 520 521 522 523 524
[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/>
525
[<a href="http://hal.inria.fr/inria-00289361">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
526
[<a href="http://hal.inria.fr/inria-00289361/document">PDF</a>]
527 528
</li>
</ol>
529
<h4>On Composability</h4> 
530
<a name="PublicationsOnComposability"></a>
Nathalie Furmento's avatar
Nathalie Furmento committed
531 532
<ol>
<li>
533 534 535
<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/>
536
[<a href="http://hal.inria.fr/hal-00773610">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
537
[<a href="http://hal.inria.fr/hal-00773610/document">PDF</a>]
Nathalie Furmento's avatar
Nathalie Furmento committed
538 539
</li>
<li>
540 541 542
<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/>
543
[<a href="http://hal.inria.fr/hal-00824514">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
544
[<a href="http://hal.inria.fr/hal-00824514/document">PDF</a>]
545 546 547 548 549
</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/>
550
[<a href="http://hal.inria.fr/inria-00619654/en/">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
551
[<a href="http://hal.inria.fr/inria-00619654/document">PDF</a>]
Nathalie Furmento's avatar
Nathalie Furmento committed
552 553
</li>
</ol>
554
<h4>On Scheduling</h4> 
555
<a name="PublicationsOnScheduling"></a>
556
<ol>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
557
<li>
558 559 560 561 562 563 564 565
<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
566 567 568 569
<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
570
[<a href="https://hal.inria.fr/hal-01361992v2/document">PDF</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
571 572
</li>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
573 574 575 576 577 578 579
<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
580
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
581 582 583
<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
584
[<a href="http://ieeexplore.ieee.org/abstract/document/7789322/">WWW</a>]
585 586 587
[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
588 589 590 591 592 593
<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
594
<li>
595 596 597
<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/>
598
[<a href="https://hal.inria.fr/hal-01120507">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
599
[<a href="https://hal.inria.fr/hal-01120507/document">PDF</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
600
</li>
601
<li>
602 603 604
<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/>
605 606
[<a href="http://hal.inria.fr/hal-00978364">WWW</a>]
[<a href="http://hal.inria.fr/hal-00978364/PDF/ordonnanceurs_modulaires.pdf">PDF</a>]
607 608
</li>
</ol>
609
<h4>On The C Extensions</h4> 
610
<a name="PublicationsOnTheCExtensions"></a>
611
<ol>
612 613 614 615
<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/>
616 617
[<a href="http://hal.inria.fr/hal-00807033">WWW</a>]
[<a href="http://hal.inria.fr/hal-00807033/PDF/RR-8278.pdf">PDF</a>]
618 619
</li>
</ol>
620
<h4>On OpenMP Support on top of StarPU</h4> 
621
<a name="PublicationsOnOpenMPSupportontopofStarPU"></a>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
622
<ol>
623
<li>
Nathalie Furmento's avatar
Nathalie Furmento committed
624 625 626 627 628 629 630 631
<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>
632
<a name="agullo:hal-01372022"></a>Emmanuel Agullo, Olivier Aumage, Berenger Bramas, Olivier Coulaud,  and Samuel Pitoiset<br/>
633
<strong>Bridging the gap between OpenMP 4.0 and native runtime systems for the fast multipole method</strong><br/>
634 635 636 637 638
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>
639 640 641 642
<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/>
643
[<a href="https://hal.inria.fr/hal-01081974">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
644
[<a href="https://hal.inria.fr/hal-01081974/document">PDF</a>]
645
[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
646 647
</li>
</ol>
648
<h4>On MPI Support</h4> 
649
<a name="PublicationsOnMPISupport"></a>
650 651
<ol>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
652 653 654 655 656 657 658
<a name="agullo:hal-01332774"></a>Emmanuel Agullo, Olivier Aumage, Mathieu Faverge, Nathalie Furmento, Florent Pruvost, Marc Sergent,  and Samuel Thibault<br/>
<strong>Achieving High Performance on Supercomputers with a Sequential Task-based Programming Model</strong><br/>
Research Report RR-8927, Inria Bordeaux Sud-Ouest ; Bordeaux INP ; CNRS ; Université de Bordeaux ; CEA, June 2016<br/>
[<a href="https://hal.inria.fr/hal-01332774">WWW</a>]
[<a href="https://hal.inria.fr/hal-01332774/file/RR-8927.pdf">PDF</a>]
</li>
<li>
659 660 661 662 663 664 665
<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>
666 667 668 669 670 671 672
<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>
673 674 675 676 677
<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/>
678
[<a href="http://hal.inria.fr/hal-00725477">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
679
[<a href="http://hal.inria.fr/hal-00725477/document">PDF</a>]
680
</li>
681
</ol>
Nathalie Furmento's avatar
Nathalie Furmento committed
682
<h4>On Memory Control</h4> 
683
<a name="PublicationsOnMemoryControl"></a>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
684 685
<ol>
<li>
Nathalie Furmento's avatar
Nathalie Furmento committed
686
<a name="sergent:hal-01284004"></a>Marc Sergent, David Goudin, Samuel Thibault,  and Olivier Aumage<br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
687
<strong>Controlling the Memory Subscription of Distributed Applications with a Task-Based Runtime System</strong><br/>
Nathalie Furmento's avatar
Nathalie Furmento committed
688
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
689 690 691
[<a href="https://hal.inria.fr/hal-01284004">WWW</a>]
[<a href="https://hal.inria.fr/hal-01284004/file/PID4127657.pdf">PDF</a>]
</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
692 693 694 695 696 697 698
<li>
<a name="agullo:hal-01332774"></a>Emmanuel Agullo, Olivier Aumage, Mathieu Faverge, Nathalie Furmento, Florent Pruvost, Marc Sergent,  and Samuel Thibault<br/>
<strong>Achieving High Performance on Supercomputers with a Sequential Task-based Programming Model</strong><br/>
Research Report RR-8927, Inria Bordeaux Sud-Ouest ; Bordeaux INP ; CNRS ; Université de Bordeaux ; CEA, June 2016<br/>
[<a href="https://hal.inria.fr/hal-01332774">WWW</a>]
[<a href="https://hal.inria.fr/hal-01332774/file/RR-8927.pdf">PDF</a>]
</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
699
</ol>
700
<h4>On Data Transfer Management</h4> 
701
<a name="PublicationsOnDataTransferManagement"></a>
702 703
<ol>
<li>
704 705 706
<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/>
707
[<a href="http://hal.inria.fr/inria-00523937">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
708
[<a href="http://hal.inria.fr/inria-00523937/document">PDF</a>]
709
[doi:<a href="http://dx.doi.org/10.1109/ICPADS.2010.129">10.1109/ICPADS.2010.129</a>]
710 711
</li>
</ol>
712
<h4>On Performance Model Tuning</h4> 
713
<a name="PublicationsOnPerformanceModelTuning"></a>
714 715
<ol>
<li>
716 717 718 719
<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/>
720
[<a href="http://hal.inria.fr/inria-00421333">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
721
[<a href="http://hal.inria.fr/inria-00421333/document">PDF</a>]
722
[doi:<a href="http://dx.doi.org/10.1007/978-3-642-14122-5_9">10.1007/978-3-642-14122-5_9</a>]
723 724
</li>
</ol>
725
<h4>On The Simulation Support through SimGrid</h4> 
726
<a name="PublicationsOnTheSimulationSupportthroughSimGrid"></a>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
727
<ol>
THIBAULT Samuel's avatar
update  
THIBAULT Samuel committed
728
<li>
729 730 731
<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/>
732 733
[<a href="https://hal.inria.fr/hal-01147997">WWW</a>]
[<a href="https://hal.inria.fr/hal-01147997/file/CCPE14_article.pdf">PDF</a>]
734
[doi:<a href="http://dx.doi.org/10.1002/cpe">10.1002/cpe</a>]
THIBAULT Samuel's avatar
update  
THIBAULT Samuel committed
735
</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
736
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
737 738 739 740 741 742 743
<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>
744 745 746 747
<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/>
748 749
[<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
750 751
</li>
</ol>
752
<h4>On The Cell Support</h4> 
753
<a name="PublicationsOnTheCellSupport"></a>
754 755
<ol>
<li>
756 757 758
<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/>
759
[<a href="http://hal.inria.fr/inria-00378705">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
760
[<a href="http://hal.inria.fr/inria-00378705/document">PDF</a>]
761
[doi:<a href="http://dx.doi.org/10.1007/978-3-642-03138-0_36">10.1007/978-3-642-03138-0_36</a>]
762 763
</li>
</ol>
764
<h4>On Applications</h4> 
765
<a name="PublicationsOnApplications"></a>
766
<ol>
THIBAULT Samuel's avatar
update  
THIBAULT Samuel committed
767
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
768 769 770 771 772 773 774 775
<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
776 777 778 779 780 781 782
<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
783 784 785 786 787 788 789
<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
790 791 792 793 794 795 796
<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>
797 798 799
<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/>
800 801
[<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
802
</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
803
<li>
804 805 806
<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/>
807 808
[<a href="https://hal.inria.fr/hal-01182746">WWW</a>]
[<a href="https://hal.inria.fr/hal-01182746/file/sbac2015_soumission.pdf">PDF</a>]
809 810
</li>
<li>
811 812 813 814
<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/>
815 816
[<a href="http://hal.inria.fr/hal-01005765">WWW</a>]
[<a href="http://hal.inria.fr/hal-01005765/PDF/final.pdf">PDF</a>]
817 818
</li>
<li>
819 820 821 822 823
<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/>
824 825
[<a href="http://hal.inria.fr/hal-00987094">WWW</a>]
[<a href="http://hal.inria.fr/hal-00987094/PDF/sparsegpus.pdf">PDF</a>]
826 827
</li>
<li>
828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844
<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
845 846
[<a href="https://tel.archives-ouvertes.fr/tel-00958494">WWW</a>]
[<a href="https://tel.archives-ouvertes.fr/tel-00958494/document">PDF</a>]
847 848 849 850 851 852
</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
853
[<a href="http://tel.archives-ouvertes.fr/tel-00948309/document">PDF</a>]
854 855
</li>
<li>
856 857 858
<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/>
859 860
[<a href="http://hal.inria.fr/hal-00851122">WWW</a>]
[<a href="http://hal.inria.fr/hal-00851122/PDF/fhpc13.pdf">PDF</a>]
861 862
</li>
<li>
863 864 865
<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/>
866 867
[<a href="http://hal.inria.fr/hal-00920915">WWW</a>]
[<a href="http://hal.inria.fr/hal-00920915/PDF/ADPC2013-117.pdf">PDF</a>]
868 869 870 871 872 873
</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/>
874
[<a href="http://hal.inria.fr/hal-00926144">WWW</a>]
875 876 877 878 879
</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/>
880
[<a href="http://hal.inria.fr/hal-00773571">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
881
[<a href="http://hal.inria.fr/hal-00773571/document">PDF</a>]
882 883
</li>
<li>
884 885 886
<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/>
887 888
[<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>]
889
</li>
890
<li>
891 892 893 894
<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
895
[<a href="http://hal.inria.fr/hal-00772742/document">PDF</a>]
896 897 898 899 900 901 902 903
</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>
904 905 906
<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/>
907 908
[<a href="http://hal.inria.fr/hal-00697020">WWW</a>]
[<a href="http://hal.inria.fr/hal-00697020/PDF/europar2012-submitted.pdf">PDF</a>]
909 910
</li>
<li>
911 912 913
<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/>
914 915
[<a href="http://hal.inria.fr/hal-00776610">WWW</a>]
[<a href="http://hal.inria.fr/hal-00776610/PDF/date12-paper.pdf">PDF</a>]
916 917
</li>
<li>
918 919 920 921 922
<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
923
[<a href="http://hal.inria.fr/hal-00648480/document">PDF</a>]
924 925 926
[doi:<a href="http://dx.doi.org/10.1109/MM.2011.67">10.1109/MM.2011.67</a>]
</li>
<li>
927 928 929
<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/>
930
[<a href="http://hal.inria.fr/hal-00654193">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
931
[<a href="http://hal.inria.fr/hal-00654193/document">PDF</a>]
932 933 934 935 936
</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/>
937
[<a href="http://hal.inria.fr/inria-00547614">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
938
[<a href="http://hal.inria.fr/inria-00547614/document">PDF</a>]
939 940 941 942 943 944
[doi:<a href="http://dx.doi.org/10.1109/IPDPS.2011.90">10.1109/IPDPS.2011.90</a>]
</li>
<li>
<a name="DasKesThi11ParCo"></a>Usman Dastgeer, Christoph Kessler,  and Samuel Thibault<br/>
<strong>Flexible runtime support for efficient skeleton programming on hybrid systems</strong><br/>
In <em>Proceedings of the International Conference on Parallel Computing (ParCo), Applications, Tools and Techniques on the Road to Exascale Computing</em>, volume 22 of <em>Advances of Parallel Computing</em>, Gent, Belgium, pages 159-166, August 2011<br/>
945
[<a href="http://hal.inria.fr/inria-00606200/">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
946
[<a href="http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.699.6216&rep=rep1&type=pdf">PDF</a>]
947 948 949 950 951
</li>
<li>
<a name="Hen11Renpar"></a>Sylvain Henry<br/>
<strong>Programmation multi-accélérateurs unifiée en OpenCL</strong><br/>
In <em>20èmes Rencontres Francophones du Parallélisme (RenPar'20)</em>, Saint Malo, France, May 2011<br/>
952
[<a href="http://hal.archives-ouvertes.fr/hal-00643257">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
953
[<a href="http://hal.archives-ouvertes.fr/hal-00643257/document">PDF</a>]
954 955 956 957 958
</li>
<li>
<a name="MahManAugThi11Renpar20"></a>Sidi Ahmed Mahmoudi, Pierre Manneback, Cédric Augonnet,  and Samuel Thibault<br/>
<strong>Détection optimale des coins et contours dans des bases d'images volumineuses sur architectures multicoeurs hétérogènes</strong><br/>
In <em>20èmes Rencontres Francophones du Parallélisme</em>, Saint-Malo / France, May 2011<br/>
959
[<a href="http://hal.inria.fr/inria-00606195">WWW</a>]
960 961
</li>
<li>
962