index.html 57.4 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> |
Nathalie Furmento's avatar
Nathalie Furmento committed
28
<a href="help/">Help</a> |
29
<a href="#publications">Publications</a> |
30
<a href="internships/">Jobs/Interns</a> |
31
<a href="files/">Download</a> |
THIBAULT Samuel's avatar
THIBAULT Samuel committed
32
<a href="market/">Market</a> |
Nathalie Furmento's avatar
Nathalie Furmento committed
33
<a href="tutorials/">Tutorials</a> |
34
<a href="https://gforge.inria.fr/plugins/mediawiki/wiki/starpu/index.php/Main_Page">Intranet</a>
Nathalie Furmento's avatar
Nathalie Furmento committed
35
</div>
36

37 38
<div class="section" id="overview">
<h3>Overview</h3>
39 40 41 42 43 44
  <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>
45
    <li>A graph of tasks, using either the StarPU's high level <b>GCC plugin</b> pragmas, StarPU's rich <b>C/C++ API</b>, or <b>OpenMP pragmas</b>.</li>
46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63
    </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>
64 65 66 67 68 69
<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>
70 71
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
72
the latest developments which may not be available in the latest release.
73 74 75 76
</p>
</div>

<div class="section emphasize newslist" id="news">
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
77 78
<h3>News</h3>
<p>
79 80 81 82 83 84 85
September 2018 <b>&raquo;&nbsp;</b><a href="http://gforge.inria.fr/frs/?group_id=1570"><b>The
      1.2.6 release of StarPU is now available!</b></a>.
      The 1.2 release serie 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>
86 87 88 89 90 91 92
August 2018 <b>&raquo;&nbsp;</b><a href="http://gforge.inria.fr/frs/?group_id=1570"><b>The
      1.2.5 release of StarPU is now available!</b></a>.
      The 1.2 release serie 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>
93
Avril 2018 <b>&raquo;&nbsp;</b><a href="http://gforge.inria.fr/frs/?group_id=1570"><b>The
94 95
      1.2.4 release of StarPU is now available!</b></a>.
      The 1.2 release serie notably brings an out-of-core support, a MIC Xeon
96 97 98 99
      Phi support, an OpenMP runtime support, and a new internal
      communication system for MPI.
</p>
<p>
100 101 102 103 104
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>
105 106 107 108 109 110 111
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>
112 113 114
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
115
  the <a href="https://sonarqube.bordeaux.inria.fr/sonarqube/dashboard?id=storm%3Astarpu%3Arelease%3Av1.2">StarPU
116 117 118
  page</a> on the sonarqube server of INRIA.
</p>
<p>
119 120 121 122 123 124
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>
125
</div>
126 127

<div class="section emphasizebot" style="text-align: right; font-style: italic;">
128
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>.
129
See also the full <a href="news/">news</a>.
130 131
</div>

132 133 134 135 136 137 138 139 140 141 142
<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
143
<div class="section" id="tutorial">
144 145 146 147 148 149 150 151 152 153
<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
154 155 156 157 158 159 160 161
<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>

162 163
<div class="section" id="contact">
<h3>Contact</h3>
164
<p>For any questions regarding StarPU, please contact the StarPU developers mailing list.</p>
165 166 167
<pre>
<a href="mailto:starpu-devel@lists.gforge.inria.fr?subject=StarPU">starpu-devel@lists.gforge.inria.fr</a>
</pre>
168
<p>Details of the <a href="people/">StarPU team people</a> are also available.</p>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
169 170
</div>

171 172
<div class="section" id="features">
<h3>Features</h3>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
173

174
<h4>Portability</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
175
  <p>
176 177 178 179 180 181 182 183
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.
184 185 186 187
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
188 189
  </p>

190 191 192
<h4>Genericity</h4>
  <p>
The StarPU programming interface is very generic. For intance, various data
193
structures are supported mainline (vectors, dense matrices, CSR/BCSR/COO sparse matrices, ...),
194 195 196 197 198 199
but application-specific data structures can also be supported, provided that
the application describes how data is to be transfered (e.g. a series of
contiguous blocks). That was for instance used for hierarchically-compressed
matrices (h-matrices).
  </p>

200
<h4>Data transfers</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
201
  <p>
202
To relieve programmers from the burden of explicit data transfers, a high-level
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
203
data management library enforces memory coherency over the machine: before a
204 205
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
206 207
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
208
prefetching</b> data, which thus permits to <b>overlap data transfers with computations</b>
THIBAULT Samuel's avatar
update  
THIBAULT Samuel committed
209
(including <b>GPU-GPU direct transfers</b>) to achieve the most of the architecture.
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
210 211
  </p>

212
<h4>Dependencies</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
213
  <p>
214
Dependencies between tasks can be given either of several ways, to provide the
215 216
programmer with best flexibility:
  <ul>
217
    <li><b>implicitly</b> from RAW, WAW, and WAR data dependencies.</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
218
    <li>explicitly through <b>tags</b> which act as rendez-vous points between
219
    tasks (thus including tasks which have not been created yet),</li>
220
    <li><b>explicitly</b> between pairs of tasks,</li>
221
  </ul>
222 223
  </p>
  <p>
224 225 226
  These dependencies are computed in a completely decentralized way.
  </p>
  <p>
227 228 229 230
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).
231 232 233 234 235
  </p>

<h4>Heterogeneous Scheduling</h4>
  <p>
StarPU obtains
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
236
portable performances by efficiently (and easily) using all computing resources
237
at the same time. StarPU also takes advantage of the <b>heterogeneous</b> nature of a
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
238
machine, for instance by using scheduling strategies based on auto-tuned
239 240 241
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>.
242
Various strategies and variants are available. Some of them are centralized, but
243
most of them are <b>completely distributed</b>. <i>dmdas</i> (a data-locality-aware MCT strategy,
244
thus similar to heft but starts executing tasks before the whole task graph is
245 246 247
submitted, thus allowing dynamic task submission and a decentralized scheduler,
as well as an energy optimizing extension), <i>eager</i> (dumb centralized
queue), <i>lws</i> (decentralized locality-aware work-stealing), ...
248
The overhead per task is typically around the order of
249 250 251
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
252 253
  </p>

254 255
<h4>Clusters</h4>
  <p>
256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277
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>
278 279 280 281 282
  </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
283
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
284 285
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.
286 287
  </p>

288 289 290
<h4>Extensions to the C Language</h4>
<p>
  StarPU comes with a GCC plug-in
291
  that <a href="doc/html/cExtensions.html">extends the C programming
292 293
  language</a> with pragmas and attributes that make it easy
  to <b>annotate a sequential C program to turn it into a parallel
294 295 296
  StarPU program</b> (new in v1.0).
</p>

THIBAULT Samuel's avatar
THIBAULT Samuel committed
297 298
<h4>OpenMP 4 -compatible interface</h4>
<p>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
299
  <a href=http://kstar.gforge.inria.fr/>K'Star</a> provides an OpenMP
THIBAULT Samuel's avatar
THIBAULT Samuel committed
300 301 302 303 304 305 306 307 308
  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>

309 310 311 312
<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).
313 314
</p>

315 316 317 318
<h4>Simulation support</h4>
<p>
  StarPU can very accurately simulate an application execution
  and measure the resulting performance thanks to using the
319
  <a href="http://simgrid.gforge.inria.fr">SimGrid simulator</a> (new in v1.1).  This allows
320 321 322 323 324
  to quickly experiment with various scheduling heuristics, various application
  algorithms, and even various platforms (available GPUs and CPUs, available
  bandwidth)!
</p>

325 326
<h4>All in all</h4>
  <p>
327
All that means that, with the help
328
of <a href="doc/html/cExtensions.html">StarPU's extensions to the C
329 330
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
331
StarPU code, possibly running on all the CPUs and GPUs, and given a data
Nathalie Furmento's avatar
Nathalie Furmento committed
332
distribution over MPI nodes, it is even a distributed version!
333 334 335 336 337 338 339 340 341 342 343 344 345
  </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
346

347
<h4>Supported Architectures</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
348
<ul>
349
<li>SMP/Multicore Processors (x86, PPC, ARM, ... all Debian architecture have been tested) </li>
Nathalie Furmento's avatar
Nathalie Furmento committed
350
<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
351 352
<li>OpenCL devices</li>
<li>Cell Processors (experimental)</li>
Nathalie Furmento's avatar
Nathalie Furmento committed
353 354
<li>Intel SCC (experimental, new in v1.2)</li>
<li>Intel MIC / Xeon Phi (new in v1.2)</li>
355
</ul>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
356

357
<h4>Supported Operating Systems</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
358
<ul>
Ludovic Courtès's avatar
Ludovic Courtès committed
359 360
<li>GNU/Linux</li>
<li>Mac OS X</li>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
361 362 363
<li>Windows</li>
</ul>

364
<h4>Performance analysis tools</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
365 366 367 368 369 370 371 372 373
  <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>

374
<p>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
375 376 377 378 379 380
<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>
381
<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>
382 383
</p>

Nathalie Furmento's avatar
website  
Nathalie Furmento committed
384 385 386 387 388 389
<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
390
GFlop/s</b>
391
<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
392 393
</p>

THIBAULT Samuel's avatar
THIBAULT Samuel committed
394 395
<p>
<a href="http://www.hlrs.de/temanejo">Temanejo</a> can be used to debug the task
396
graph, as shown below (new in v1.1).
THIBAULT Samuel's avatar
THIBAULT Samuel committed
397 398 399
</p>

<center>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
400
<a href="images/temanejo.png"><img src="images/temanejo.png" width="50%"/></a>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
401 402
</center>

Nathalie Furmento's avatar
website  
Nathalie Furmento committed
403 404
</div>

405 406 407 408 409
<div class="section" id="software">
<h3>Software using StarPU</h3>

<p>
Some software is known for being able to use StarPU to tackle heterogeneous
410
architectures, here is a non-exhaustive list (feel free to ask to be added to the
411
list!):
412 413 414
</p>

<ul>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
415
	<li>AL4SAN, dense linear algebra library</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
416
	<li><a href="https://project.inria.fr/chameleon/">Chameleon</a>, dense linear algebra library</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
417
	<li><a href="http://exa2pro.eu">Exa2pro</a>, Enhancing Programmability and boosting Performance Portability for Exascale Computing Systems</li>
THIBAULT Samuel's avatar
fix URL  
THIBAULT Samuel committed
418
	<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
419
	<li><a href="https://hal.inria.fr/hal-01507613">FLUSEPA</a>, Navier-Stokes Solver for Unsteady Problems with Bodies in Relative Motion</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
420
	<li><a href="http://github.com/ecrc/hicma">HiCMA</a>, Low-rank general linear algebra library</li>
421
        <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
422
	<li><a href="http://github.com/ecrc/ksvd">KSVD</a>, dense SVD on distributed-memory manycore systems</li>
423 424
	<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
425
	<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>
426
	<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
427
	<li><a href="http://github.com/ecrc/qdwh">QDWH</a>, QR-based Dynamically Weighted Halley</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
428
	<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
429
	<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
430
	<li><a href="https://tel.archives-ouvertes.fr/tel-01410049/">SCHNAPS</a>, Solver for Conservative Hypebolic Non-linear systems Applied to PlasmaS. </li>
431 432
	<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
433
	<li><a href="http://github.com/ecrc/stars-h">STARS-H</a>, HPC low-rank matrix market</li>
434 435
</ul>

436
<p>
437
You can find <a href="#PublicationsOnApplications">below</a> the list of publications related to applications using StarPU.
438 439
</p>

440 441
</div>

THIBAULT Samuel's avatar
THIBAULT Samuel committed
442 443 444 445
<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
446 447 448
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
449 450 451
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
452 453

<tt><pre>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
454 455 456 457 458 459
$ 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
460
$ 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
461 462 463 464 465

<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
466 467 468 469 470 471 472 473 474 475 476 477 478 479
<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
480 481 482 483 484 485 486 487 488 489
<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
490
38400	9915	1903.7</pre></tt>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
491 492
<p>(MPI simulation is not supported yet)</p>

493 494 495 496
<div class="section" id="publications">
<h3>Publications</h3>
<p>
All StarPU related publications are also
497
listed <a href="./publications">here</a>
498 499 500
with the corresponding Bibtex entries.
</p>

501 502
<p>
A good overview is available in
503 504 505
the following <a href="http://hal.archives-ouvertes.fr/inria-00467677">Research Report</a>.
</p>

506 507
<p>
If you need to cite StarPU, please
508
reference <a href="publications/Year/2011.html#AugThiNamWac11CCPE">[StarPU: A Unified Platform
509 510 511 512 513
    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>

514
<h4>General Presentations</h4> 
515
<a name="PublicationsGeneralPresentations"></a>
516 517
<ol>
<li>
518 519 520
<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/>
521
[<a href="http://tel.archives-ouvertes.fr/tel-00777154">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
522
[<a href="http://tel.archives-ouvertes.fr/tel-00777154/document">PDF</a>]
523 524
</li>
<li>
525 526 527
<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/>
528
[<a href="http://hal.inria.fr/inria-00550877">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
529
[<a href="http://hal.inria.fr/inria-00550877/document">PDF</a>]
530
[doi:<a href="http://dx.doi.org/10.1002/cpe.1631">10.1002/cpe.1631</a>]
531
</li>
532
<li>
533 534 535 536
<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
537
[<a href="http://hal.inria.fr/inria-00467677/document">PDF</a>]
538 539
</li>
<li>
540 541 542 543
<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/>
544
[<a href="http://hal.inria.fr/inria-00411581">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
545
[<a href="http://hal.inria.fr/inria-00411581/document">PDF</a>]
546 547
</li>
<li>
548 549 550 551
<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/>
552
[<a href="http://hal.inria.fr/inria-00384363">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
553
[<a href="http://hal.inria.fr/inria-00384363/document">PDF</a>]
554 555 556 557 558 559 560 561
[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/>
562
[<a href="http://hal.inria.fr/inria-00326917">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
563
[<a href="http://hal.inria.fr/inria-00326917/document">PDF</a>]
564 565 566 567 568 569
[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/>
570
[<a href="http://hal.inria.fr/inria-00289361">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
571
[<a href="http://hal.inria.fr/inria-00289361/document">PDF</a>]
572 573
</li>
</ol>
574
<h4>On Composability</h4> 
575
<a name="PublicationsOnComposability"></a>
Nathalie Furmento's avatar
Nathalie Furmento committed
576 577
<ol>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
578 579 580 581 582 583 584
<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>
585 586 587
<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/>
588
[<a href="http://hal.inria.fr/hal-00773610">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
589
[<a href="http://hal.inria.fr/hal-00773610/document">PDF</a>]
Nathalie Furmento's avatar
Nathalie Furmento committed
590 591
</li>
<li>
592 593 594
<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/>
595
[<a href="http://hal.inria.fr/hal-00824514">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
596
[<a href="http://hal.inria.fr/hal-00824514/document">PDF</a>]
597 598 599 600 601
</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/>
602
[<a href="http://hal.inria.fr/inria-00619654/en/">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
603
[<a href="http://hal.inria.fr/inria-00619654/document">PDF</a>]
Nathalie Furmento's avatar
Nathalie Furmento committed
604 605
</li>
</ol>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
606 607 608 609 610 611 612 613 614 615 616
<h4>On Parallel Tasks</h4> 
<a name="PublicationsOnParallelTasks"></a>
<ol>
<li>
<a name="cojean:tel-01816341"></a>Terry Cojean<br/>
<strong>Programmation of heterogeneous architectures using moldable tasks</strong><br/>
Theses, Université de Bordeaux, March 2018<br/>
[<a href="https://tel.archives-ouvertes.fr/tel-01816341">WWW</a>]
[<a href="https://tel.archives-ouvertes.fr/tel-01816341/file/COJEAN_TERRY_2018.pdf">PDF</a>]
</li>
</ol>
617
<h4>On Scheduling</h4> 
618
<a name="PublicationsOnScheduling"></a>
619
<ol>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
620
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
621 622 623 624 625 626 627 628
<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/>
<em>Concurrency and Computation: Practice and Experience</em>, April 2018<br/>
[<a href="https://hal.inria.fr/hal-01616632">WWW</a>]
[<a href="https://hal.inria.fr/hal-01616632/file/CCPE_article_submitted_2018_02_06.pdf">PDF</a>]
[doi:<a href="http://dx.doi.org/10.1002/cpe.4472">10.1002/cpe.4472</a>]
</li>
<li>
629 630 631 632 633 634 635
<a name="pinto:hal-01842038"></a>Vinicius Garcia Pinto, Lucas Mello Schnorr, Arnaud Legrand, Samuel Thibault, Luka Stanisic,  and Vincent Danjean<br/>
<strong>Detecção de Anomalias de Desempenho em Aplicações de Alto Desempenho baseadas em Tarefas em Clusters Hìbridos</strong><br/>
In <em>17º Workshop em Desempenho de Sistemas Computacionais e de Comunicação (WPerformance)</em>, Natal, Brazil, July 2018<br/>
[<a href="https://hal.inria.fr/hal-01842038">WWW</a>]
[<a href="https://hal.inria.fr/hal-01842038/file/181587_1.pdf">PDF</a>]
</li>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
636 637
<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
638
PhD thesis, Université de Bordeaux, April 2017<br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
639 640 641 642
[<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
643 644 645 646 647 648 649 650
<a name="beaumont:hal-01386174"></a>O. Beaumont, L. Eyraud-Dubois,  and S. Kumar<br/>
<strong>Approximation Proofs of a Fast and Efficient List Scheduling Algorithm for Task-Based Runtime Systems on Multicores and GPUs</strong><br/>
In <em>2017 IEEE International Parallel and Distributed Processing Symposium (IPDPS)</em>, pages 768-777, May 2017<br/>
[<a href="https://hal.inria.fr/hal-01386174">WWW</a>]
[<a href="https://hal.inria.fr/hal-01386174/file/heteroPrioApproxProofsRR.pdf">PDF</a>]
[doi:<a href="http://dx.doi.org/10.1109/IPDPS.2017.71">10.1109/IPDPS.2017.71</a>]
</li>
<li>
651 652 653 654 655 656 657 658
<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
659 660 661 662
<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
663
[<a href="https://hal.inria.fr/hal-01361992v2/document">PDF</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
664 665
</li>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
666 667
<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/>
668
In <em>HeteroPar'2016 workshop of Euro-Par</em>, Grenoble, France, August 2016<br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
669 670
[<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
671 672
</li>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
673 674 675 676 677 678 679
<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
680
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
681 682 683
<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
684
[<a href="http://ieeexplore.ieee.org/abstract/document/7789322/">WWW</a>]
685 686 687
[doi:<a href="http://dx.doi.org/10.1109/SBAC-PAD.2016.14">10.1109/SBAC-PAD.2016.14</a>]
</li>
<li>
688 689 690 691 692 693 694
<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>
695 696 697
<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/>
698
[<a href="https://hal.inria.fr/hal-01120507">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
699
[<a href="https://hal.inria.fr/hal-01120507/document">PDF</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
700
</li>
701
<li>
702 703 704
<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/>
705 706
[<a href="http://hal.inria.fr/hal-00978364">WWW</a>]
[<a href="http://hal.inria.fr/hal-00978364/PDF/ordonnanceurs_modulaires.pdf">PDF</a>]
707 708
</li>
</ol>
709
<h4>On The C Extensions</h4> 
710
<a name="PublicationsOnTheCExtensions"></a>
711
<ol>
712 713 714 715
<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/>
716 717
[<a href="http://hal.inria.fr/hal-00807033">WWW</a>]
[<a href="http://hal.inria.fr/hal-00807033/PDF/RR-8278.pdf">PDF</a>]
718 719
</li>
</ol>
720
<h4>On OpenMP Support on top of StarPU</h4> 
721
<a name="PublicationsOnOpenMPSupportontopofStarPU"></a>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
722
<ol>
723
<li>
Nathalie Furmento's avatar
Nathalie Furmento committed
724 725 726 727 728 729 730 731
<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>
732
<a name="agullo:hal-01372022"></a>Emmanuel Agullo, Olivier Aumage, Berenger Bramas, Olivier Coulaud,  and Samuel Pitoiset<br/>
733
<strong>Bridging the gap between OpenMP 4.0 and native runtime systems for the fast multipole method</strong><br/>
734 735 736 737 738
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>
739 740 741 742
<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/>
743
[<a href="https://hal.inria.fr/hal-01081974">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
744
[<a href="https://hal.inria.fr/hal-01081974/document">PDF</a>]
745
[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
746 747
</li>
</ol>
748
<h4>On MPI Support</h4> 
749
<a name="PublicationsOnMPISupport"></a>
750 751
<ol>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
752
<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
753
<strong>Achieving High Performance on Supercomputers with a Sequential Task-based Programming Model</strong><br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
754 755 756
<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
757 758 759 760
</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
761
PhD thesis, Université de Bordeaux, December 2016<br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
762 763
[<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
764 765
</li>
<li>
766 767 768 769 770 771 772
<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>
773 774 775 776 777 778 779
<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>
780 781 782 783 784
<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/>
785
[<a href="http://hal.inria.fr/hal-00725477">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
786
[<a href="http://hal.inria.fr/hal-00725477/document">PDF</a>]
787
</li>
788
</ol>
Nathalie Furmento's avatar
Nathalie Furmento committed
789
<h4>On Memory Control</h4> 
790
<a name="PublicationsOnMemoryControl"></a>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
791 792
<ol>
<li>
793 794 795 796 797 798 799
<a name="chevalier:hal-01718280"></a>Arthur Chevalier<br/>
<strong>Critical resources management and scheduling under StarPU</strong><br/>
Master's thesis, Université de Bordeaux, September 2017<br/>
[<a href="https://hal.inria.fr/hal-01718280">WWW</a>]
[<a href="https://hal.inria.fr/hal-01718280/file/Memoire.pdf">PDF</a>]
</li>
<li>
Nathalie Furmento's avatar
Nathalie Furmento committed
800
<a name="sergent:hal-01284004"></a>Marc Sergent, David Goudin, Samuel Thibault,  and Olivier Aumage<br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
801
<strong>Controlling the Memory Subscription of Distributed Applications with a Task-Based Runtime System</strong><br/>
Nathalie Furmento's avatar
Nathalie Furmento committed
802
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
803 804 805 806
[<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>
807
<h4>On Data Transfer Management</h4> 
808
<a name="PublicationsOnDataTransferManagement"></a>
809 810
<ol>
<li>
811 812 813
<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/>
814
[<a href="http://hal.inria.fr/inria-00523937">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
815
[<a href="http://hal.inria.fr/inria-00523937/document">PDF</a>]
816
[doi:<a href="http://dx.doi.org/10.1109/ICPADS.2010.129">10.1109/ICPADS.2010.129</a>]
817 818
</li>
</ol>
819
<h4>On Performance Model Tuning</h4> 
820
<a name="PublicationsOnPerformanceModelTuning"></a>
821 822
<ol>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
823 824 825 826 827 828 829
<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>
830 831 832 833
<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/>
834
[<a href="http://hal.inria.fr/inria-00421333">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
835
[<a href="http://hal.inria.fr/inria-00421333/document">PDF</a>]
836
[doi:<a href="http://dx.doi.org/10.1007/978-3-642-14122-5_9">10.1007/978-3-642-14122-5_9</a>]
837 838
</li>
</ol>
839
<h4>On The Simulation Support through SimGrid</h4> 
840
<a name="PublicationsOnTheSimulationSupportthroughSimGrid"></a>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
841
<ol>
THIBAULT Samuel's avatar
update  
THIBAULT Samuel committed
842
<li>
843 844 845
<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/>
846 847
[<a href="https://hal.inria.fr/hal-01147997">WWW</a>]
[<a href="https://hal.inria.fr/hal-01147997/file/CCPE14_article.pdf">PDF</a>]
848
[doi:<a href="http://dx.doi.org/10.1002/cpe">10.1002/cpe</a>]
THIBAULT Samuel's avatar
update  
THIBAULT Samuel committed
849
</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
850
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
851 852 853 854 855 856 857
<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>
858 859 860 861
<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/>
862 863
[<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
864 865
</li>
</ol>
866
<h4>On The Cell Support</h4> 
867
<a name="PublicationsOnTheCellSupport"></a>
868 869
<ol>
<li>
870 871 872
<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/>
873
[<a href="http://hal.inria.fr/inria-00378705">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
874
[<a href="http://hal.inria.fr/inria-00378705/document">PDF</a>]
875
[doi:<a href="http://dx.doi.org/10.1007/978-3-642-03138-0_36">10.1007/978-3-642-03138-0_36</a>]
876 877
</li>
</ol>
878
<h4>On Applications</h4> 
879
<a name="PublicationsOnApplications"></a>
880
<ol>
THIBAULT Samuel's avatar
update  
THIBAULT Samuel committed
881
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
882 883 884 885 886 887 888 889
<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
890 891 892 893 894 895 896
<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
897 898 899 900 901 902 903
<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
904 905 906 907 908 909 910
<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>
911 912 913
<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/>
914 915
[<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
916
</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
917
<li>
918 919 920
<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/>
921 922
[<a href="https://hal.inria.fr/hal-01182746">WWW</a>]
[<a href="https://hal.inria.fr/hal-01182746/file/sbac2015_soumission.pdf">PDF</a>]
923 924
</li>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
925 926 927 928 929 930 931 932
<a name="agullo:hal-00911856"></a>Emmanuel Agullo, Bérenger Bramas, Olivier Coulaud, Eric Darve, Matthias Messner,  and Toru Takahashi<br/>
<strong>Task-Based FMM for Multicore Architectures</strong><br/>
<em>SIAM Journal on Scientific Computing</em>, 36(1):66-93, 2014<br/>
[<a href="https://hal.inria.fr/hal-00911856">WWW</a>]
[<a href="https://hal.inria.fr/hal-00911856/file/sisc-cpu.pdf">PDF</a>]
[doi:<a href="http://dx.doi.org/10.1137/130915662">10.1137/130915662</a>]
</li>
<li>
933 934 935 936
<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/>
937 938
[<a href="http://hal.inria.fr/hal-01005765">WWW</a>]
[<a href="http://hal.inria.fr/hal-01005765/PDF/final.pdf">PDF</a>]
939 940
</li>
<li>
941 942 943 944 945
<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/>
946 947
[<a href="http://hal.inria.fr/hal-00987094">WWW</a>]
[<a href="http://hal.inria.fr/hal-00987094/PDF/sparsegpus.pdf">PDF</a>]
948 949
</li>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
950 951 952 953 954 955 956
<a name="agullo:hal-00974674"></a>Emmanuel Agullo, Berenger Bramas, Olivier Coulaud, Eric Darve, Matthias Messner,  and Toru Takahashi<br/>
<strong>Task-based FMM for heterogeneous architectures</strong><br/>
Research Report RR-8513, Inria Bordeaux - Sud-Ouest, April 2014<br/>
[<a href="https://hal.inria.fr/hal-00974674">WWW</a>]
[<a href="https://hal.inria.fr/hal-00974674/file/RR-8513.pdf">PDF</a>]
</li>
<li>
957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973
<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