index.html 61.3 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
June
2019 <b>&raquo;&nbsp;</b><a href="http://gforge.inria.fr/frs/?group_id=1570"><b>The
      release 1.3.2 of StarPU 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
May 2019 <b>&raquo;&nbsp;</b><a href="http://gforge.inria.fr/frs/?group_id=1570"><b>The
      v1.1.8 release of StarPU is now available!</b></a>. This release notably brings the concept of
      scheduling contexts which allows to separate computation
      resources. This is really intented to be the last release for the
      branch 1.1.
</p>
<p>
Nathalie Furmento's avatar
Nathalie Furmento committed
98 99 100 101 102 103 104 105 106 107 108 109
April
2019 <b>&raquo;&nbsp;</b><a href="http://gforge.inria.fr/frs/?group_id=1570"><b>The
      release 1.3.1 of StarPU 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>
110
March
111
2019 <b>&raquo;&nbsp;</b><a href="http://gforge.inria.fr/frs/?group_id=1570"><b>The
112
      release 1.3.0 of StarPU is now
113 114 115 116 117 118 119 120 121
      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>
122 123 124 125 126 127 128 129
February 2019 <b>&raquo;&nbsp;</b><a href="http://gforge.inria.fr/frs/?group_id=1570"><b>The
      1.2.8 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.
	(The release 1.2.7 is broken and should not be used)
</p>
<p>
130 131 132 133 134 135 136
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>
137 138 139 140 141 142 143
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>
144
Avril 2018 <b>&raquo;&nbsp;</b><a href="http://gforge.inria.fr/frs/?group_id=1570"><b>The
145 146
      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
147 148 149 150
      Phi support, an OpenMP runtime support, and a new internal
      communication system for MPI.
</p>
<p>
151 152 153 154
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>
155
</div>
156 157

<div class="section emphasizebot" style="text-align: right; font-style: italic;">
158
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>.
159
See also the full <a href="news/">news</a>.
160 161
</div>

162 163 164 165 166 167 168 169 170 171 172
<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
173
<div class="section" id="tutorial">
174 175 176 177 178 179 180 181 182 183
<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
184 185 186 187 188 189 190 191
<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>

192 193
<div class="section" id="contact">
<h3>Contact</h3>
194
<p>For any questions regarding StarPU, please contact the StarPU developers mailing list.</p>
195 196 197
<pre>
<a href="mailto:starpu-devel@lists.gforge.inria.fr?subject=StarPU">starpu-devel@lists.gforge.inria.fr</a>
</pre>
198
<p>Details of the <a href="people/">StarPU team people</a> are also available.</p>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
199 200
</div>

201 202
<div class="section" id="features">
<h3>Features</h3>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
203

204
<h4>Portability</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
205
  <p>
206 207 208 209 210 211 212 213
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.
214 215 216 217
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).
218 219 220
StarPU can execute them concurrently, e.g. one per socket, provided that the
task implementations support it (which is the case for MKL, but unfortunately
most often not for OpenMP).
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
221 222
  </p>

223 224 225
<h4>Genericity</h4>
  <p>
The StarPU programming interface is very generic. For intance, various data
226
structures are supported mainline (vectors, dense matrices, CSR/BCSR/COO sparse matrices, ...),
227 228 229 230 231 232
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>

233
<h4>Data transfers</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
234
  <p>
235
To relieve programmers from the burden of explicit data transfers, a high-level
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
236
data management library enforces memory coherency over the machine: before a
237 238
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
239 240
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
241
prefetching</b> data, which thus permits to <b>overlap data transfers with computations</b>
THIBAULT Samuel's avatar
update  
THIBAULT Samuel committed
242
(including <b>GPU-GPU direct transfers</b>) to achieve the most of the architecture.
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
243 244
  </p>

245
<h4>Dependencies</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
246
  <p>
247
Dependencies between tasks can be given either of several ways, to provide the
248 249
programmer with best flexibility:
  <ul>
250
    <li><b>implicitly</b> from RAW, WAW, and WAR data dependencies.</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
251
    <li>explicitly through <b>tags</b> which act as rendez-vous points between
252
    tasks (thus including tasks which have not been created yet),</li>
253
    <li><b>explicitly</b> between pairs of tasks,</li>
254
  </ul>
255 256
  </p>
  <p>
257 258 259
  These dependencies are computed in a completely decentralized way.
  </p>
  <p>
260 261 262 263
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).
264 265
  </p>

266 267 268 269 270
  <p>
It also supports transparent dependencies tracking between hierarchical subpieces of data
through asynchronous partitioning (new in v1.3).
  </p>

271 272 273
<h4>Heterogeneous Scheduling</h4>
  <p>
StarPU obtains
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
274
portable performances by efficiently (and easily) using all computing resources
275
at the same time. StarPU also takes advantage of the <b>heterogeneous</b> nature of a
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
276
machine, for instance by using scheduling strategies based on auto-tuned
277 278 279
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>.
280
Various strategies and variants are available. Some of them are centralized, but
281
most of them are <b>completely distributed</b>. <i>dmdas</i> (a data-locality-aware MCT strategy,
282
thus similar to heft but starts executing tasks before the whole task graph is
283 284 285
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), ...
286
The overhead per task is typically around the order of
287 288 289
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
290 291
  </p>

292 293
<h4>Clusters</h4>
  <p>
294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315
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>
316 317 318 319 320
  </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
321
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
322 323
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.
324 325
  </p>

326 327 328
<h4>Extensions to the C Language</h4>
<p>
  StarPU comes with a GCC plug-in
329
  that <a href="doc/html/cExtensions.html">extends the C programming
330 331
  language</a> with pragmas and attributes that make it easy
  to <b>annotate a sequential C program to turn it into a parallel
332 333 334
  StarPU program</b> (new in v1.0).
</p>

THIBAULT Samuel's avatar
THIBAULT Samuel committed
335 336
<h4>OpenMP 4 -compatible interface</h4>
<p>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
337
  <a href=http://kstar.gforge.inria.fr/>K'Star</a> provides an OpenMP
THIBAULT Samuel's avatar
THIBAULT Samuel committed
338 339 340 341 342 343 344 345 346
  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>

347 348 349 350
<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).
351 352
</p>

353 354 355 356
<h4>Simulation support</h4>
<p>
  StarPU can very accurately simulate an application execution
  and measure the resulting performance thanks to using the
357
  <a href="http://simgrid.gforge.inria.fr">SimGrid simulator</a> (new in v1.1).  This allows
358 359 360 361 362
  to quickly experiment with various scheduling heuristics, various application
  algorithms, and even various platforms (available GPUs and CPUs, available
  bandwidth)!
</p>

363 364
<h4>All in all</h4>
  <p>
365
All that means that, with the help
366
of <a href="doc/html/cExtensions.html">StarPU's extensions to the C
367 368
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
369
StarPU code, possibly running on all the CPUs and GPUs, and given a data
Nathalie Furmento's avatar
Nathalie Furmento committed
370
distribution over MPI nodes, it is even a distributed version!
371 372 373 374 375 376 377 378 379 380 381 382 383
  </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
384

385
<h4>Supported Architectures</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
386
<ul>
387
<li>SMP/Multicore Processors (x86, PPC, ARM, ... all Debian architecture have been tested) </li>
Nathalie Furmento's avatar
Nathalie Furmento committed
388
<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
389 390
<li>OpenCL devices</li>
<li>Cell Processors (experimental)</li>
Nathalie Furmento's avatar
Nathalie Furmento committed
391 392
<li>Intel SCC (experimental, new in v1.2)</li>
<li>Intel MIC / Xeon Phi (new in v1.2)</li>
393
</ul>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
394

395
<h4>Supported Operating Systems</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
396
<ul>
Ludovic Courtès's avatar
Ludovic Courtès committed
397 398
<li>GNU/Linux</li>
<li>Mac OS X</li>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
399 400 401
<li>Windows</li>
</ul>

THIBAULT Samuel's avatar
THIBAULT Samuel committed
402 403 404 405 406 407 408 409 410 411 412
<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>

413
<h4>Performance analysis tools</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
414 415 416 417 418 419 420 421 422
  <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>

423
<p>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
424 425 426 427 428 429
<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>
430
<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>
431 432
</p>

Nathalie Furmento's avatar
website  
Nathalie Furmento committed
433 434 435 436 437 438
<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
439
GFlop/s</b>
440
<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
441 442
</p>

THIBAULT Samuel's avatar
THIBAULT Samuel committed
443 444
<p>
<a href="http://www.hlrs.de/temanejo">Temanejo</a> can be used to debug the task
445
graph, as shown below (new in v1.1).
THIBAULT Samuel's avatar
THIBAULT Samuel committed
446 447 448
</p>

<center>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
449
<a href="images/temanejo.png"><img src="images/temanejo.png" width="50%"/></a>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
450 451
</center>

Nathalie Furmento's avatar
website  
Nathalie Furmento committed
452 453
</div>

454 455 456 457 458
<div class="section" id="software">
<h3>Software using StarPU</h3>

<p>
Some software is known for being able to use StarPU to tackle heterogeneous
459
architectures, here is a non-exhaustive list (feel free to ask to be added to the
460
list!):
461 462 463
</p>

<ul>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
464
	<li>AL4SAN, dense linear algebra library</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
465
	<li><a href="https://project.inria.fr/chameleon/">Chameleon</a>, dense linear algebra library</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
466
	<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
467
	<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
468
	<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
469
	<li><a href="http://github.com/ecrc/hicma">HiCMA</a>, Low-rank general linear algebra library</li>
470
	<li>hmat, hierarchical matrix C/C++ library</li>
471
        <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
472
	<li><a href="http://github.com/ecrc/ksvd">KSVD</a>, dense SVD on distributed-memory manycore systems</li>
473 474
	<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
475
	<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>
476
	<li><a href="http://pastix.gforge.inria.fr/">PaStiX</a>, sparse linear algebra library, starting from version 5.2.1</li>
477
	<li>PEPPHER, Performance Portability and Programmability for Heterogeneous Many-core Architectures</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
478
	<li><a href="http://github.com/ecrc/qdwh">QDWH</a>, QR-based Dynamically Weighted Halley</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
479
	<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
480
	<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
481
	<li><a href="https://tel.archives-ouvertes.fr/tel-01410049/">SCHNAPS</a>, Solver for Conservative Hypebolic Non-linear systems Applied to PlasmaS. </li>
482 483
	<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
484
	<li><a href="http://github.com/ecrc/stars-h">STARS-H</a>, HPC low-rank matrix market</li>
485
	<li><a href="http://www.xcalablemp.org/">XcalableMP</a>, Directive-based language eXtension for Scalable and performance-aware Parallel Programming</li>
486 487
</ul>

488
<p>
489
You can find <a href="#PublicationsOnApplications">below</a> the list of publications related to applications using StarPU.
490 491
</p>

492 493
</div>

THIBAULT Samuel's avatar
THIBAULT Samuel committed
494 495 496 497
<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
498 499 500
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
501 502 503
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
504 505

<tt><pre>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
506 507 508 509 510 511
$ 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
512
$ 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
513 514 515 516 517

<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
518 519 520 521 522 523 524 525 526 527 528 529 530 531
<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
532 533 534 535 536 537 538 539 540 541
<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
542
38400	9915	1903.7</pre></tt>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
543 544
<p>(MPI simulation is not supported yet)</p>

545 546 547 548
<div class="section" id="publications">
<h3>Publications</h3>
<p>
All StarPU related publications are also
549
listed <a href="./publications">here</a>
550 551 552
with the corresponding Bibtex entries.
</p>

553 554
<p>
A good overview is available in
555 556 557
the following <a href="http://hal.archives-ouvertes.fr/inria-00467677">Research Report</a>.
</p>

558 559
<p>
If you need to cite StarPU, please
560
reference <a href="publications/Year/2011.html#AugThiNamWac11CCPE">[StarPU: A Unified Platform
561 562 563 564 565
    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>

566
<h4>General Presentations</h4> 
567
<a name="PublicationsGeneralPresentations"></a>
568 569
<ol>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
570 571 572 573 574 575 576
<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>
577 578 579
<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/>
580
[<a href="http://tel.archives-ouvertes.fr/tel-00777154">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
581
[<a href="http://tel.archives-ouvertes.fr/tel-00777154/document">PDF</a>]
582 583
</li>
<li>
584 585
<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
586
<em>CCPE - Concurrency and Computation: Practice and Experience, Special Issue: Euro-Par 2009</em>, 23:187-198, February 2011<br/>
587
[<a href="http://hal.inria.fr/inria-00550877">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
588
[<a href="http://hal.inria.fr/inria-00550877/document">PDF</a>]
589
[doi:<a href="http://dx.doi.org/10.1002/cpe.1631">10.1002/cpe.1631</a>]
590
</li>
591
<li>
592 593
<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
594
Research Report RR-7240, INRIA, March 2010<br/>
595
[<a href="http://hal.inria.fr/inria-00467677">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
596
[<a href="http://hal.inria.fr/inria-00467677/document">PDF</a>]
597 598
</li>
<li>
599 600 601 602
<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/>
603
[<a href="http://hal.inria.fr/inria-00411581">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
604
[<a href="http://hal.inria.fr/inria-00411581/document">PDF</a>]
605 606
</li>
<li>
607 608
<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
609
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/>
610
Springer<br/>
611
[<a href="http://hal.inria.fr/inria-00384363">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
612
[<a href="http://hal.inria.fr/inria-00384363/document">PDF</a>]
613 614 615 616 617 618 619 620
[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/>
621
[<a href="http://hal.inria.fr/inria-00326917">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
622
[<a href="http://hal.inria.fr/inria-00326917/document">PDF</a>]
623 624 625 626 627 628
[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/>
629
[<a href="http://hal.inria.fr/inria-00289361">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
630
[<a href="http://hal.inria.fr/inria-00289361/document">PDF</a>]
631 632
</li>
</ol>
633
<h4>On Composability</h4> 
634
<a name="PublicationsOnComposability"></a>
Nathalie Furmento's avatar
Nathalie Furmento committed
635 636
<ol>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
637 638 639 640 641 642 643
<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>
644 645 646
<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/>
647
[<a href="http://hal.inria.fr/hal-00773610">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
648
[<a href="http://hal.inria.fr/hal-00773610/document">PDF</a>]
Nathalie Furmento's avatar
Nathalie Furmento committed
649 650
</li>
<li>
651 652 653
<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/>
654
[<a href="http://hal.inria.fr/hal-00824514">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
655
[<a href="http://hal.inria.fr/hal-00824514/document">PDF</a>]
656 657 658 659 660
</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/>
661
[<a href="http://hal.inria.fr/inria-00619654/en/">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
662
[<a href="http://hal.inria.fr/inria-00619654/document">PDF</a>]
Nathalie Furmento's avatar
Nathalie Furmento committed
663 664
</li>
</ol>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
665 666 667 668 669 670 671 672 673 674 675
<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>
676
<h4>On Scheduling</h4> 
677
<a name="PublicationsOnScheduling"></a>
678
<ol>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
679
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
680 681
<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
682
<em>CCPE - Concurrency and Computation: Practice and Experience</em>, 30, April 2018<br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
683 684 685 686 687
[<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>
688
<a name="pinto:hal-01842038"></a>Vinicius Garcia Pinto, Lucas Mello Schnorr, Arnaud Legrand, Samuel Thibault, Luka Stanisic,  and Vincent Danjean<br/>
689 690
<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/>
691 692 693 694
[<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
695 696
<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
697
PhD thesis, Université de Bordeaux, April 2017<br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
698 699 700 701
[<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
702 703 704 705 706 707 708 709
<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>
710 711 712 713 714 715 716 717
<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
718 719 720 721
<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
722
[<a href="https://hal.inria.fr/hal-01361992v2/document">PDF</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
723 724
</li>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
725 726
<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/>
727
In <em>HeteroPar'2016 workshop of Euro-Par</em>, Grenoble, France, August 2016<br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
728 729
[<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
730 731
</li>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
732 733
<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
734
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
735 736
Note: Held in conjunction with SC16<br/>
[<a href="https://hal.inria.fr/hal-01353962">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
737 738
[<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
739
</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
740
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
741 742 743
<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
744
[<a href="http://ieeexplore.ieee.org/abstract/document/7789322/">WWW</a>]
745 746 747
[doi:<a href="http://dx.doi.org/10.1109/SBAC-PAD.2016.14">10.1109/SBAC-PAD.2016.14</a>]
</li>
<li>
748 749 750 751 752 753 754
<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>
755 756
<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
757
In <em>HCW'2015 - Heterogeneity in Computing Workshop of IPDPS</em>, Hyderabad, India, May 2015<br/>
758
[<a href="https://hal.inria.fr/hal-01120507">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
759
[<a href="https://hal.inria.fr/hal-01120507/document">PDF</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
760
[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
761
</li>
762
<li>
763 764 765
<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/>
766 767
[<a href="http://hal.inria.fr/hal-00978364">WWW</a>]
[<a href="http://hal.inria.fr/hal-00978364/PDF/ordonnanceurs_modulaires.pdf">PDF</a>]
768
</li>
769 770 771 772 773 774 775 776
<li>
<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/>
[<a href="http://hal.inria.fr/inria-00523937">WWW</a>]
[<a href="http://hal.inria.fr/inria-00523937/document">PDF</a>]
[doi:<a href="http://dx.doi.org/10.1109/ICPADS.2010.129">10.1109/ICPADS.2010.129</a>]
</li>
777
</ol>
778
<h4>On The C Extensions</h4> 
779
<a name="PublicationsOnTheCExtensions"></a>
780
<ol>
781 782 783 784
<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/>
785 786
[<a href="http://hal.inria.fr/hal-00807033">WWW</a>]
[<a href="http://hal.inria.fr/hal-00807033/PDF/RR-8278.pdf">PDF</a>]
787 788
</li>
</ol>
789
<h4>On OpenMP Support on top of StarPU</h4> 
790
<a name="PublicationsOnOpenMPSupportontopofStarPU"></a>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
791
<ol>
792
<li>
Nathalie Furmento's avatar
Nathalie Furmento committed
793 794 795 796 797 798 799 800
<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>
801
<a name="agullo:hal-01372022"></a>Emmanuel Agullo, Olivier Aumage, Berenger Bramas, Olivier Coulaud,  and Samuel Pitoiset<br/>
802
<strong>Bridging the gap between OpenMP 4.0 and native runtime systems for the fast multipole method</strong><br/>
803 804 805 806 807
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
808
<a name="virouleau:hal-01081974"></a>Philippe Virouleau, Pierrick Brunet, François Broquedis, Nathalie Furmento, Samuel Thibault, Olivier Aumage,  and Thierry Gautier<br/>
809
<strong>Evaluation of OpenMP Dependent Tasks with the KASTORS Benchmark Suite</strong><br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
810
In <em>IWOMP2014 - 10th International Workshop on OpenMP</em>, 10th International Workshop on OpenMP, IWOMP2014, Salvador, Brazil, France, pages 16 - 29, September 2014<br/>
811
Springer<br/>
812
[<a href="https://hal.inria.fr/hal-01081974">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
813
[<a href="https://hal.inria.fr/hal-01081974/document">PDF</a>]
814
[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
815 816
</li>
</ol>
817
<h4>On MPI Support</h4> 
818
<a name="PublicationsOnMPISupport"></a>
819 820
<ol>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
821
<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
822
<strong>Achieving High Performance on Supercomputers with a Sequential Task-based Programming Model</strong><br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
823
<em>TPDS - IEEE Transactions on Parallel and Distributed Systems</em>, December 2017<br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
824 825
[<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
826
[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
827 828 829 830
</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
831
PhD thesis, Université de Bordeaux, December 2016<br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
832 833
[<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
834 835
</li>
<li>
836 837 838 839 840 841 842
<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>
843 844
<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
845
Research Report RR-8538, INRIA, May 2014<br/>
846 847 848 849
[<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>
850 851 852 853 854
<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/>
855
[<a href="http://hal.inria.fr/hal-00725477">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
856
[<a href="http://hal.inria.fr/hal-00725477/document">PDF</a>]
857
</li>
858
</ol>
Nathalie Furmento's avatar
Nathalie Furmento committed
859
<h4>On Memory Control</h4> 
860
<a name="PublicationsOnMemoryControl"></a>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
861 862
<ol>
<li>
863 864 865 866 867 868 869
<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
870
<a name="sergent:hal-01284004"></a>Marc Sergent, David Goudin, Samuel Thibault,  and Olivier Aumage<br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
871
<strong>Controlling the Memory Subscription of Distributed Applications with a Task-Based Runtime System</strong><br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
872
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
873 874
[<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
875
[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
876 877
</li>
</ol>
878
<h4>On Performance Model Tuning</h4> 
879
<a name="PublicationsOnPerformanceModelTuning"></a>
880 881
<ol>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
882 883 884 885 886 887 888
<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>
889 890
<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
891
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/>
892
Springer<br/>
893
[<a href="http://hal.inria.fr/inria-00421333">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
894
[<a href="http://hal.inria.fr/inria-00421333/document">PDF</a>]
895
[doi:<a href="http://dx.doi.org/10.1007/978-3-642-14122-5_9">10.1007/978-3-642-14122-5_9</a>]
896 897
</li>
</ol>
898
<h4>On The Simulation Support through SimGrid</h4> 
899
<a name="PublicationsOnTheSimulationSupportthroughSimGrid"></a>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
900
<ol>
THIBAULT Samuel's avatar
update  
THIBAULT Samuel committed
901
<li>
902 903
<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
904
<em>CCPE - Concurrency and Computation: Practice and Experience</em>, pp 16, May 2015<br/>
905 906
[<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
907
[doi:<a href="http://dx.doi.org/10.1002/cpe.3555">10.1002/cpe.3555</a>]
THIBAULT Samuel's avatar
update  
THIBAULT Samuel committed
908
</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
909
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
910 911 912 913 914 915 916
<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>
917 918
<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
919
In <em>Euro-Par - 20th International Conference on Parallel Processing</em>, Porto, Portugal, August 2014<br/>
920
Springer-Verlag<br/>
921 922
[<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
923
[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
924 925
</li>
</ol>
926
<h4>On The Cell Support</h4> 
927
<a name="PublicationsOnTheCellSupport"></a>
928 929
<ol>
<li>
930 931 932
<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/>
933
[<a href="http://hal.inria.fr/inria-00378705">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
934
[<a href="http://hal.inria.fr/inria-00378705/document">PDF</a>]
935
[doi:<a href="http://dx.doi.org/10.1007/978-3-642-03138-0_36">10.1007/978-3-642-03138-0_36</a>]
936 937
</li>
</ol>
938
<h4>On Applications</h4>