index.html 59.5 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 86 87 88 89 90
February
2019 <b>&raquo;&nbsp;</b><a href="http://gforge.inria.fr/frs/?group_id=1570"><b>The
      first release candidate for StarPU 1.3.0 is now
      available!</b></a> The 1.3 release brings among other
      functionalities a MPI master-slave support, a tool to replay
      execution through SimGrid, a HDF5 implementation of the
      Out-of-core, a new implementation of StarPU-MPI on top of
      NewMadeleine, implicit support for asynchronous partition
      planning, a resource management module to share processor cores
      and accelerator devices with other parallel runtime systems, ...
</p>
<p>
91 92 93 94 95 96 97
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>
98 99 100 101 102 103 104
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>
105
Avril 2018 <b>&raquo;&nbsp;</b><a href="http://gforge.inria.fr/frs/?group_id=1570"><b>The
106 107
      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
108 109 110 111
      Phi support, an OpenMP runtime support, and a new internal
      communication system for MPI.
</p>
<p>
112 113 114 115 116
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>
117 118 119 120 121 122 123
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>
124 125 126
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
127
  the <a href="https://sonarqube.bordeaux.inria.fr/sonarqube/dashboard?id=storm%3Astarpu%3Arelease%3Av1.2">StarPU
128 129
  page</a> on the sonarqube server of INRIA.
</p>
130
</div>
131 132

<div class="section emphasizebot" style="text-align: right; font-style: italic;">
133
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>.
134
See also the full <a href="news/">news</a>.
135 136
</div>

137 138 139 140 141 142 143 144 145 146 147
<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
148
<div class="section" id="tutorial">
149 150 151 152 153 154 155 156 157 158
<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
159 160 161 162 163 164 165 166
<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>

167 168
<div class="section" id="contact">
<h3>Contact</h3>
169
<p>For any questions regarding StarPU, please contact the StarPU developers mailing list.</p>
170 171 172
<pre>
<a href="mailto:starpu-devel@lists.gforge.inria.fr?subject=StarPU">starpu-devel@lists.gforge.inria.fr</a>
</pre>
173
<p>Details of the <a href="people/">StarPU team people</a> are also available.</p>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
174 175
</div>

176 177
<div class="section" id="features">
<h3>Features</h3>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
178

179
<h4>Portability</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
180
  <p>
181 182 183 184 185 186 187 188
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.
189 190 191 192
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
193 194
  </p>

195 196 197
<h4>Genericity</h4>
  <p>
The StarPU programming interface is very generic. For intance, various data
198
structures are supported mainline (vectors, dense matrices, CSR/BCSR/COO sparse matrices, ...),
199 200 201 202 203 204
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>

205
<h4>Data transfers</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
206
  <p>
207
To relieve programmers from the burden of explicit data transfers, a high-level
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
208
data management library enforces memory coherency over the machine: before a
209 210
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
211 212
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
213
prefetching</b> data, which thus permits to <b>overlap data transfers with computations</b>
THIBAULT Samuel's avatar
update  
THIBAULT Samuel committed
214
(including <b>GPU-GPU direct transfers</b>) to achieve the most of the architecture.
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
215 216
  </p>

217
<h4>Dependencies</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
218
  <p>
219
Dependencies between tasks can be given either of several ways, to provide the
220 221
programmer with best flexibility:
  <ul>
222
    <li><b>implicitly</b> from RAW, WAW, and WAR data dependencies.</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
223
    <li>explicitly through <b>tags</b> which act as rendez-vous points between
224
    tasks (thus including tasks which have not been created yet),</li>
225
    <li><b>explicitly</b> between pairs of tasks,</li>
226
  </ul>
227 228
  </p>
  <p>
229 230 231
  These dependencies are computed in a completely decentralized way.
  </p>
  <p>
232 233 234 235
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).
236 237 238 239 240
  </p>

<h4>Heterogeneous Scheduling</h4>
  <p>
StarPU obtains
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
241
portable performances by efficiently (and easily) using all computing resources
242
at the same time. StarPU also takes advantage of the <b>heterogeneous</b> nature of a
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
243
machine, for instance by using scheduling strategies based on auto-tuned
244 245 246
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>.
247
Various strategies and variants are available. Some of them are centralized, but
248
most of them are <b>completely distributed</b>. <i>dmdas</i> (a data-locality-aware MCT strategy,
249
thus similar to heft but starts executing tasks before the whole task graph is
250 251 252
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), ...
253
The overhead per task is typically around the order of
254 255 256
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
257 258
  </p>

259 260
<h4>Clusters</h4>
  <p>
261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282
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>
283 284 285 286 287
  </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
288
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
289 290
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.
291 292
  </p>

293 294 295
<h4>Extensions to the C Language</h4>
<p>
  StarPU comes with a GCC plug-in
296
  that <a href="doc/html/cExtensions.html">extends the C programming
297 298
  language</a> with pragmas and attributes that make it easy
  to <b>annotate a sequential C program to turn it into a parallel
299 300 301
  StarPU program</b> (new in v1.0).
</p>

THIBAULT Samuel's avatar
THIBAULT Samuel committed
302 303
<h4>OpenMP 4 -compatible interface</h4>
<p>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
304
  <a href=http://kstar.gforge.inria.fr/>K'Star</a> provides an OpenMP
THIBAULT Samuel's avatar
THIBAULT Samuel committed
305 306 307 308 309 310 311 312 313
  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>

314 315 316 317
<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).
318 319
</p>

320 321 322 323
<h4>Simulation support</h4>
<p>
  StarPU can very accurately simulate an application execution
  and measure the resulting performance thanks to using the
324
  <a href="http://simgrid.gforge.inria.fr">SimGrid simulator</a> (new in v1.1).  This allows
325 326 327 328 329
  to quickly experiment with various scheduling heuristics, various application
  algorithms, and even various platforms (available GPUs and CPUs, available
  bandwidth)!
</p>

330 331
<h4>All in all</h4>
  <p>
332
All that means that, with the help
333
of <a href="doc/html/cExtensions.html">StarPU's extensions to the C
334 335
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
336
StarPU code, possibly running on all the CPUs and GPUs, and given a data
Nathalie Furmento's avatar
Nathalie Furmento committed
337
distribution over MPI nodes, it is even a distributed version!
338 339 340 341 342 343 344 345 346 347 348 349 350
  </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
351

352
<h4>Supported Architectures</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
353
<ul>
354
<li>SMP/Multicore Processors (x86, PPC, ARM, ... all Debian architecture have been tested) </li>
Nathalie Furmento's avatar
Nathalie Furmento committed
355
<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
356 357
<li>OpenCL devices</li>
<li>Cell Processors (experimental)</li>
Nathalie Furmento's avatar
Nathalie Furmento committed
358 359
<li>Intel SCC (experimental, new in v1.2)</li>
<li>Intel MIC / Xeon Phi (new in v1.2)</li>
360
</ul>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
361

362
<h4>Supported Operating Systems</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
363
<ul>
Ludovic Courtès's avatar
Ludovic Courtès committed
364 365
<li>GNU/Linux</li>
<li>Mac OS X</li>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
366 367 368
<li>Windows</li>
</ul>

THIBAULT Samuel's avatar
THIBAULT Samuel committed
369 370 371 372 373 374 375 376 377 378 379
<h4>Stability</h4>
<p>
StarPU is checked every night with
<ul>
<li>Valgrind / Helgrind</li>
<li>gcc' Address/Leak/Thread/Undefined Sanitizers</li>
<li>cppcheck</li>
<li>Coverity</li>
</ul>
</p>

380
<h4>Performance analysis tools</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
381 382 383 384 385 386 387 388 389
  <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>

390
<p>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
391 392 393 394 395 396
<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>
397
<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>
398 399
</p>

Nathalie Furmento's avatar
website  
Nathalie Furmento committed
400 401 402 403 404 405
<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
406
GFlop/s</b>
407
<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
408 409
</p>

THIBAULT Samuel's avatar
THIBAULT Samuel committed
410 411
<p>
<a href="http://www.hlrs.de/temanejo">Temanejo</a> can be used to debug the task
412
graph, as shown below (new in v1.1).
THIBAULT Samuel's avatar
THIBAULT Samuel committed
413 414 415
</p>

<center>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
416
<a href="images/temanejo.png"><img src="images/temanejo.png" width="50%"/></a>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
417 418
</center>

Nathalie Furmento's avatar
website  
Nathalie Furmento committed
419 420
</div>

421 422 423 424 425
<div class="section" id="software">
<h3>Software using StarPU</h3>

<p>
Some software is known for being able to use StarPU to tackle heterogeneous
426
architectures, here is a non-exhaustive list (feel free to ask to be added to the
427
list!):
428 429 430
</p>

<ul>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
431
	<li>AL4SAN, dense linear algebra library</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
432
	<li><a href="https://project.inria.fr/chameleon/">Chameleon</a>, dense linear algebra library</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
433
	<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
434
	<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
435
	<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
436
	<li><a href="http://github.com/ecrc/hicma">HiCMA</a>, Low-rank general linear algebra library</li>
437
        <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
438
	<li><a href="http://github.com/ecrc/ksvd">KSVD</a>, dense SVD on distributed-memory manycore systems</li>
439 440
	<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
441
	<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>
442
	<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
443
	<li><a href="http://github.com/ecrc/qdwh">QDWH</a>, QR-based Dynamically Weighted Halley</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
444
	<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
445
	<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
446
	<li><a href="https://tel.archives-ouvertes.fr/tel-01410049/">SCHNAPS</a>, Solver for Conservative Hypebolic Non-linear systems Applied to PlasmaS. </li>
447 448
	<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
449
	<li><a href="http://github.com/ecrc/stars-h">STARS-H</a>, HPC low-rank matrix market</li>
450 451
</ul>

452
<p>
453
You can find <a href="#PublicationsOnApplications">below</a> the list of publications related to applications using StarPU.
454 455
</p>

456 457
</div>

THIBAULT Samuel's avatar
THIBAULT Samuel committed
458 459 460 461
<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
462 463 464
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
465 466 467
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
468 469

<tt><pre>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
470 471 472 473 474 475
$ 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
476
$ 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
477 478 479 480 481

<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
482 483 484 485 486 487 488 489 490 491 492 493 494 495
<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
496 497 498 499 500 501 502 503 504 505
<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
506
38400	9915	1903.7</pre></tt>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
507 508
<p>(MPI simulation is not supported yet)</p>

509 510 511 512
<div class="section" id="publications">
<h3>Publications</h3>
<p>
All StarPU related publications are also
513
listed <a href="./publications">here</a>
514 515 516
with the corresponding Bibtex entries.
</p>

517 518
<p>
A good overview is available in
519 520 521
the following <a href="http://hal.archives-ouvertes.fr/inria-00467677">Research Report</a>.
</p>

522 523
<p>
If you need to cite StarPU, please
524
reference <a href="publications/Year/2011.html#AugThiNamWac11CCPE">[StarPU: A Unified Platform
525 526 527 528 529
    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>

530
<h4>General Presentations</h4> 
531
<a name="PublicationsGeneralPresentations"></a>
532 533
<ol>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
534 535 536 537 538 539 540
<a name="thibault:tel-01959127"></a>Samuel Thibault<br/>
<strong>On Runtime Systems for Task-based Programming on Heterogeneous Platforms</strong><br/>
Habilitation à diriger des recherches, Université de Bordeaux, December 2018<br/>
[<a href="https://hal.inria.fr/tel-01959127">WWW</a>]
[<a href="https://hal.inria.fr/tel-01959127/file/hdr.pdf">PDF</a>]
</li>
<li>
541 542 543
<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/>
544
[<a href="http://tel.archives-ouvertes.fr/tel-00777154">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
545
[<a href="http://tel.archives-ouvertes.fr/tel-00777154/document">PDF</a>]
546 547
</li>
<li>
548 549
<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/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
550
<em>CCPE - Concurrency and Computation: Practice and Experience, Special Issue: Euro-Par 2009</em>, 23:187-198, February 2011<br/>
551
[<a href="http://hal.inria.fr/inria-00550877">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
552
[<a href="http://hal.inria.fr/inria-00550877/document">PDF</a>]
553
[doi:<a href="http://dx.doi.org/10.1002/cpe.1631">10.1002/cpe.1631</a>]
554
</li>
555
<li>
556 557
<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/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
558
Research Report RR-7240, INRIA, March 2010<br/>
559
[<a href="http://hal.inria.fr/inria-00467677">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
560
[<a href="http://hal.inria.fr/inria-00467677/document">PDF</a>]
561 562
</li>
<li>
563 564 565 566
<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/>
567
[<a href="http://hal.inria.fr/inria-00411581">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
568
[<a href="http://hal.inria.fr/inria-00411581/document">PDF</a>]
569 570
</li>
<li>
571 572
<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/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
573
In <em>Euro-Par - 15th International Conference on Parallel Processing</em>, volume 5704 of <em>Lecture Notes in Computer Science</em>, Delft, The Netherlands, pages 863-874, August 2009<br/>
574
Springer<br/>
575
[<a href="http://hal.inria.fr/inria-00384363">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
576
[<a href="http://hal.inria.fr/inria-00384363/document">PDF</a>]
577 578 579 580 581 582 583 584
[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/>
585
[<a href="http://hal.inria.fr/inria-00326917">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
586
[<a href="http://hal.inria.fr/inria-00326917/document">PDF</a>]
587 588 589 590 591 592
[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/>
593
[<a href="http://hal.inria.fr/inria-00289361">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
594
[<a href="http://hal.inria.fr/inria-00289361/document">PDF</a>]
595 596
</li>
</ol>
597
<h4>On Composability</h4> 
598
<a name="PublicationsOnComposability"></a>
Nathalie Furmento's avatar
Nathalie Furmento committed
599 600
<ol>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
601 602 603 604 605 606 607
<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>
608 609 610
<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/>
611
[<a href="http://hal.inria.fr/hal-00773610">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
612
[<a href="http://hal.inria.fr/hal-00773610/document">PDF</a>]
Nathalie Furmento's avatar
Nathalie Furmento committed
613 614
</li>
<li>
615 616 617
<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/>
618
[<a href="http://hal.inria.fr/hal-00824514">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
619
[<a href="http://hal.inria.fr/hal-00824514/document">PDF</a>]
620 621 622 623 624
</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/>
625
[<a href="http://hal.inria.fr/inria-00619654/en/">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
626
[<a href="http://hal.inria.fr/inria-00619654/document">PDF</a>]
Nathalie Furmento's avatar
Nathalie Furmento committed
627 628
</li>
</ol>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
629 630 631 632 633 634 635 636 637 638 639
<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>
640
<h4>On Scheduling</h4> 
641
<a name="PublicationsOnScheduling"></a>
642
<ol>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
643
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
644 645
<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/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
646
<em>CCPE - Concurrency and Computation: Practice and Experience</em>, 30, April 2018<br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
647 648 649 650 651
[<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>
652
<a name="pinto:hal-01842038"></a>Vinicius Garcia Pinto, Lucas Mello Schnorr, Arnaud Legrand, Samuel Thibault, Luka Stanisic,  and Vincent Danjean<br/>
653 654
<strong>Detecç ao de Anomalias de Desempenho em Aplicaç oes de Alto Desempenho baseadas em Tarefas em Clusters Hìbridos</strong><br/>
In <em>WPerformance - 17º Workshop em Desempenho de Sistemas Computacionais e de Comunicaç ao</em>, Natal, Brazil, July 2018<br/>
655 656 657 658
[<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
659 660
<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
661
PhD thesis, Université de Bordeaux, April 2017<br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
662 663 664 665
[<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
666 667 668 669 670 671 672 673
<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>
674 675 676 677 678 679 680 681
<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
682 683 684 685
<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
686
[<a href="https://hal.inria.fr/hal-01361992v2/document">PDF</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
687 688
</li>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
689 690
<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/>
691
In <em>HeteroPar'2016 workshop of Euro-Par</em>, Grenoble, France, August 2016<br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
692 693
[<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
694 695
</li>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
696 697
<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/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
698
In <em>VPA - 3rd Workshop on Visual Performance Analysis</em>, Salt Lake City, United States, November 2016<br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
699 700
Note: Held in conjunction with SC16<br/>
[<a href="https://hal.inria.fr/hal-01353962">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
701 702
[<a href="https://hal.inria.fr/hal-01353962v2/document">PDF</a>]
[doi:<a href="http://dx.doi.org/10.1109/VPA.2016.008">10.1109/VPA.2016.008</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
703
</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
704
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
705 706 707
<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
708
[<a href="http://ieeexplore.ieee.org/abstract/document/7789322/">WWW</a>]
709 710 711
[doi:<a href="http://dx.doi.org/10.1109/SBAC-PAD.2016.14">10.1109/SBAC-PAD.2016.14</a>]
</li>
<li>
712 713 714 715 716 717 718
<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>
719 720
<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/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
721
In <em>HCW'2015 - Heterogeneity in Computing Workshop of IPDPS</em>, Hyderabad, India, May 2015<br/>
722
[<a href="https://hal.inria.fr/hal-01120507">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
723
[<a href="https://hal.inria.fr/hal-01120507/document">PDF</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
724
[doi:<a href="http://dx.doi.org/10.1109/IPDPSW.2015.35">10.1109/IPDPSW.2015.35</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
725
</li>
726
<li>
727 728 729
<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/>
730 731
[<a href="http://hal.inria.fr/hal-00978364">WWW</a>]
[<a href="http://hal.inria.fr/hal-00978364/PDF/ordonnanceurs_modulaires.pdf">PDF</a>]
732 733
</li>
</ol>
734
<h4>On The C Extensions</h4> 
735
<a name="PublicationsOnTheCExtensions"></a>
736
<ol>
737 738 739 740
<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/>
741 742
[<a href="http://hal.inria.fr/hal-00807033">WWW</a>]
[<a href="http://hal.inria.fr/hal-00807033/PDF/RR-8278.pdf">PDF</a>]
743 744
</li>
</ol>
745
<h4>On OpenMP Support on top of StarPU</h4> 
746
<a name="PublicationsOnOpenMPSupportontopofStarPU"></a>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
747
<ol>
748
<li>
Nathalie Furmento's avatar
Nathalie Furmento committed
749 750 751 752 753 754 755 756
<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>
757
<a name="agullo:hal-01372022"></a>Emmanuel Agullo, Olivier Aumage, Berenger Bramas, Olivier Coulaud,  and Samuel Pitoiset<br/>
758
<strong>Bridging the gap between OpenMP 4.0 and native runtime systems for the fast multipole method</strong><br/>
759 760 761 762 763
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>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
764
<a name="virouleau:hal-01081974"></a>Philippe Virouleau, Pierrick Brunet, François Broquedis, Nathalie Furmento, Samuel Thibault, Olivier Aumage,  and Thierry Gautier<br/>
765
<strong>Evaluation of OpenMP Dependent Tasks with the KASTORS Benchmark Suite</strong><br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
766
In <em>IWOMP2014 - 10th International Workshop on OpenMP</em>, 10th International Workshop on OpenMP, IWOMP2014, Salvador, Brazil, France, pages 16 - 29, September 2014<br/>
767
Springer<br/>
768
[<a href="https://hal.inria.fr/hal-01081974">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
769
[<a href="https://hal.inria.fr/hal-01081974/document">PDF</a>]
770
[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
771 772
</li>
</ol>
773
<h4>On MPI Support</h4> 
774
<a name="PublicationsOnMPISupport"></a>
775 776
<ol>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
777
<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
778
<strong>Achieving High Performance on Supercomputers with a Sequential Task-based Programming Model</strong><br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
779
<em>TPDS - IEEE Transactions on Parallel and Distributed Systems</em>, December 2017<br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
780 781
[<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
782
[doi:<a href="http://dx.doi.org/10.1109/TPDS.2017.2766064">10.1109/TPDS.2017.2766064</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
783 784 785 786
</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
787
PhD thesis, Université de Bordeaux, December 2016<br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
788 789
[<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
790 791
</li>
<li>
792 793 794 795 796 797 798
<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>
799 800
<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/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
801
Research Report RR-8538, INRIA, May 2014<br/>
802 803 804 805
[<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>
806 807 808 809 810
<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/>
811
[<a href="http://hal.inria.fr/hal-00725477">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
812
[<a href="http://hal.inria.fr/hal-00725477/document">PDF</a>]
813
</li>
814
</ol>
Nathalie Furmento's avatar
Nathalie Furmento committed
815
<h4>On Memory Control</h4> 
816
<a name="PublicationsOnMemoryControl"></a>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
817 818
<ol>
<li>
819 820 821 822 823 824 825
<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
826
<a name="sergent:hal-01284004"></a>Marc Sergent, David Goudin, Samuel Thibault,  and Olivier Aumage<br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
827
<strong>Controlling the Memory Subscription of Distributed Applications with a Task-Based Runtime System</strong><br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
828
In <em>HIPS - 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
829 830
[<a href="https://hal.inria.fr/hal-01284004">WWW</a>]
[<a href="https://hal.inria.fr/hal-01284004/file/PID4127657.pdf">PDF</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
831
[doi:<a href="http://dx.doi.org/10.1109/IPDPSW.2016.105">10.1109/IPDPSW.2016.105</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
832 833
</li>
</ol>
834
<h4>On Data Transfer Management</h4> 
835
<a name="PublicationsOnDataTransferManagement"></a>
836 837
<ol>
<li>
838 839 840
<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/>
841
[<a href="http://hal.inria.fr/inria-00523937">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
842
[<a href="http://hal.inria.fr/inria-00523937/document">PDF</a>]
843
[doi:<a href="http://dx.doi.org/10.1109/ICPADS.2010.129">10.1109/ICPADS.2010.129</a>]
844 845
</li>
</ol>
846
<h4>On Performance Model Tuning</h4> 
847
<a name="PublicationsOnPerformanceModelTuning"></a>
848 849
<ol>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
850 851 852 853 854 855 856
<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>
857 858
<a name="AugThiNam09HPPC"></a>Cédric Augonnet, Samuel Thibault,  and Raymond Namyst<br/>
<strong>Automatic Calibration of Performance Models on Heterogeneous Multicore Architectures</strong><br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
859
In <em>HPPC - Proceedings of the International Euro-Par Workshops, Highly Parallel Processing on a Chip</em>, volume 6043 of <em>Lecture Notes in Computer Science</em>, Delft, The Netherlands, pages 56-65, August 2009<br/>
860
Springer<br/>
861
[<a href="http://hal.inria.fr/inria-00421333">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
862
[<a href="http://hal.inria.fr/inria-00421333/document">PDF</a>]
863
[doi:<a href="http://dx.doi.org/10.1007/978-3-642-14122-5_9">10.1007/978-3-642-14122-5_9</a>]
864 865
</li>
</ol>
866
<h4>On The Simulation Support through SimGrid</h4> 
867
<a name="PublicationsOnTheSimulationSupportthroughSimGrid"></a>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
868
<ol>
THIBAULT Samuel's avatar
update  
THIBAULT Samuel committed
869
<li>
870 871
<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/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
872
<em>CCPE - Concurrency and Computation: Practice and Experience</em>, pp 16, May 2015<br/>
873 874
[<a href="https://hal.inria.fr/hal-01147997">WWW</a>]
[<a href="https://hal.inria.fr/hal-01147997/file/CCPE14_article.pdf">PDF</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
875
[doi:<a href="http://dx.doi.org/10.1002/cpe.3555">10.1002/cpe.3555</a>]
THIBAULT Samuel's avatar
update  
THIBAULT Samuel committed
876
</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
877
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
878 879 880 881 882 883 884
<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>
885 886
<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/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
887
In <em>Euro-Par - 20th International Conference on Parallel Processing</em>, Porto, Portugal, August 2014<br/>
888
Springer-Verlag<br/>
889 890
[<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
891
[doi:<a href="http://dx.doi.org/10.1007/978-3-319-09873-9_5">10.1007/978-3-319-09873-9_5</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
892 893
</li>
</ol>
894
<h4>On The Cell Support</h4> 
895
<a name="PublicationsOnTheCellSupport"></a>
896 897
<ol>
<li>
898 899 900
<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/>
901
[<a href="http://hal.inria.fr/inria-00378705">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
902
[<a href="http://hal.inria.fr/inria-00378705/document">PDF</a>]
903
[doi:<a href="http://dx.doi.org/10.1007/978-3-642-03138-0_36">10.1007/978-3-642-03138-0_36</a>]
904 905
</li>
</ol>
906
<h4>On Applications</h4> 
907
<a name="PublicationsOnApplications"></a>
908
<ol>
THIBAULT Samuel's avatar
update  
THIBAULT Samuel committed
909
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
910 911 912 913 914 915 916 917
<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
918 919 920 921 922 923 924
<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
925 926 927 928