Attention une mise à jour du serveur va être effectuée le lundi 17 mai entre 13h et 13h30. Cette mise à jour va générer une interruption du service de quelques minutes.

index.html 41.8 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="http://runtime.bordeaux.inria.fr/">RUNTIME 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> |
24
<a href="#features">Features</a> |
25
<a href="#software">Software</a> |
THIBAULT Samuel's avatar
THIBAULT Samuel committed
26
<a href="#tryit">Try it!</a> |
27
<a href="#publications">Publications</a> |
28
<a href="internships/">Jobs/Interns</a> |
29 30 31
<a href="files/">Download</a> |
<a href="tutorials">Tutorials</a> |
<a href="https://wiki.bordeaux.inria.fr/runtime/doku.php?id=starpu">Intranet</a>
Nathalie Furmento's avatar
Nathalie Furmento committed
32
</div>
33

34 35
<div class="section" id="overview">
<h3>Overview</h3>
36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60
  <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>
    <li>A graph of tasks, using either the StarPU's high level <b>GCC plugin</b> pragmas or StarPU's rich <b>C API</b></li>
    </ul>
<br>
</li>
<li><b>StarPU handles run-time concerns</b>
    <ul>
    <li>Task dependencies</li>
    <li>Optimized heterogeneous scheduling</li>
    <li>Optimized data transfers and replication between main memory and discrete memories</li>
    <li>Optimized cluster communications</li>
    </ul>
</li>
</ol>
</p>
<p>
<span class="important">Rather than handling low-level issues, <b>programmers can concentrate on algorithmic concerns!</b></span>
</p>

<p>
THIBAULT Samuel's avatar
fix URL  
THIBAULT Samuel committed
61
<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.
62 63 64 65
</p>
</div>

<div class="section emphasize newslist" id="news">
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
66 67
<h3>News</h3>
<p>
Nathalie Furmento's avatar
Nathalie Furmento committed
68 69 70 71 72 73 74
August 2016 <b>&raquo;&nbsp;</b><a href="http://gforge.inria.fr/frs/?group_id=1570"><b>The
      1.2.0 release of StarPU is now available!</b></a>.
      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>
75 76 77 78 79 80 81 82
August 2016 <b>&raquo;&nbsp;</b><a href="http://gforge.inria.fr/frs/?group_id=1570"><b>The
      sixth (and really hopefully last) release candidate of the v1.2.0 release of StarPU is now
      available!</b></a>.
      This release notably brings an out-of-core support, a MIC Xeon
      Phi support, an OpenMP runtime support, and a new internal
      communication system for MPI.
</p>
<p>
83 84
March 2016 <b>&raquo;&nbsp;</b> <b>Engineer job offer</b> at Inria: more
details on the job and on how to apply are available <a href="internships/hibox.html">here</a>
85 86
</p>
<p>
87 88 89 90 91 92 93 94
December 2015 <b>&raquo;&nbsp;</b><a href="http://gforge.inria.fr/frs/?group_id=1570"><b>The
      fifth (and hopefully last) release candidate of the v1.2.0 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>
Nathalie Furmento's avatar
Nathalie Furmento committed
95 96 97 98 99 100
September 2015 <b>&raquo;&nbsp;</b><a href="http://gforge.inria.fr/frs/?group_id=1570"><b>The
      v1.1.5 release of StarPU is now available!</b></a>. This release notably brings the concept of
      scheduling contexts which allows to separate computation
      resources.
</p>
<p>
101 102 103 104 105 106 107 108
August 2015 <b>&raquo;&nbsp;</b><a href="http://gforge.inria.fr/frs/?group_id=1570"><b>The
      fourth release candidate of the v1.2.0 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>
109 110 111 112 113 114 115 116
July 2015 <b>&raquo;&nbsp;</b><a href="http://gforge.inria.fr/frs/?group_id=1570"><b>The
      third release candidate of the v1.2.0 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>
Nathalie Furmento's avatar
Nathalie Furmento committed
117 118 119 120 121 122 123 124
May 2015 <b>&raquo;&nbsp;</b><a href="http://gforge.inria.fr/frs/?group_id=1570"><b>The
      second release candidate of the v1.2.0 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>
Nathalie Furmento's avatar
Nathalie Furmento committed
125
April 2015 <b>&raquo;&nbsp;</b>A <a href="https://events.prace-ri.eu/event/339/">tutorial</a> on runtime systems including
126 127 128
StarPU will be given at INRIA Bordeaux in June 2015.
</p>
<p>
129 130 131 132 133 134 135 136
March 2015 <b>&raquo;&nbsp;</b><a href="http://gforge.inria.fr/frs/?group_id=1570"><b>The
      first release candidate of the v1.2.0 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>
137 138 139 140 141
March 2015 <b>&raquo;&nbsp;</b><a href="http://gforge.inria.fr/frs/?group_id=1570"><b>The
      v1.1.4 release of StarPU is now available!</b></a>. This release notably brings the concept of
      scheduling contexts which allows to separate computation
      resources.
</p>
142
</div>
143 144

<div class="section emphasizebot" style="text-align: right; font-style: italic;">
145
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>.
146
See also the full <a href="news/">news</a>.
147 148
</div>

149 150 151 152 153 154 155 156 157 158 159
<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>

160 161
<div class="section" id="contact">
<h3>Contact</h3>
162
<p>For any questions regarding StarPU, please contact the StarPU developers mailing list.</p>
163 164 165
<pre>
<a href="mailto:starpu-devel@lists.gforge.inria.fr?subject=StarPU">starpu-devel@lists.gforge.inria.fr</a>
</pre>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
166 167
</div>

168 169
<div class="section" id="features">
<h3>Features</h3>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
170

171
<h4>Portability</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
172
  <p>
173 174 175 176 177 178 179 180
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.
181 182 183 184
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
185 186
  </p>

187
<h4>Data transfers</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
188
  <p>
189
To relieve programmers from the burden of explicit data transfers, a high-level
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
190
data management library enforces memory coherency over the machine: before a
191 192
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
193 194
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
195
prefetching</b> data, which thus permits to <b>overlap data transfers with computations</b>
THIBAULT Samuel's avatar
update  
THIBAULT Samuel committed
196
(including <b>GPU-GPU direct transfers</b>) to achieve the most of the architecture.
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
197 198
  </p>

199
<h4>Dependencies</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
200
  <p>
201 202 203
Dependencies between tasks can be given several ways, to provide the
programmer with best flexibility:
  <ul>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
204 205
    <li><b>explicitly</b> between pairs of tasks,</li>
    <li>explicitly through <b>tags</b> which act as rendez-vous points between
206
    tasks (thus including tasks which have not been created yet),</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
207
    <li><b>implicitly</b> from RAW, WAW, and WAR data dependencies.</li>
208
  </ul>
209 210
  </p>
  <p>
211 212 213
  These dependencies are computed in a completely decentralized way.
  </p>
  <p>
214 215 216 217
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).
218 219 220 221 222
  </p>

<h4>Heterogeneous Scheduling</h4>
  <p>
StarPU obtains
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
223
portable performances by efficiently (and easily) using all computing resources
224
at the same time. StarPU also takes advantage of the <b>heterogeneous</b> nature of a
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
225
machine, for instance by using scheduling strategies based on auto-tuned
226 227 228
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>.
229
Various strategies and variants are available. Some of them are centralized, but
THIBAULT Samuel's avatar
THIBAULT Samuel committed
230
most of them are <b>completely distributed</b>. dmda (a data-aware MCT strategy,
231
thus similar to heft but starts executing tasks before the whole task graph is
232 233 234
submitted, thus allowing dynamic task submission and a decentralized scheduler),
eager (dumb centralized queue), decentralized locality-aware work-stealing, ...
The overhead per task is typically around the order of
235 236 237
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
238 239
  </p>

240 241
<h4>Clusters</h4>
  <p>
242
To deal with clusters, StarPU can nicely integrate with <a href="doc/html/MPISupport.html">MPI</a> through
243 244 245 246
explicit network communications, which will then be <b>automatically combined and
overlapped</b> with the intra-node data transfers and computation. 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
247 248 249 250
<b>generate all required MPI communications</b> accordingly (new in v0.9). We
have gotten excellent scaling on a 144-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
THIBAULT Samuel's avatar
update  
THIBAULT Samuel committed
251
pruning-tuned task submission, it should scale to about a <b>million nodes</b>.
252 253 254 255 256 257
  </p>

<h4>Out of core</h4>
  <p>
When memory is not big enough for the working set, one may have to resort to
using disks. StarPU makes this seamless thanks to its <a href="doc/html/OutOfCore.html">out of core support</a> (new in 1.2).
THIBAULT Samuel's avatar
update  
THIBAULT Samuel committed
258 259
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.
260 261
  </p>

262 263 264
<h4>Extensions to the C Language</h4>
<p>
  StarPU comes with a GCC plug-in
265
  that <a href="doc/html/cExtensions.html">extends the C programming
266 267
  language</a> with pragmas and attributes that make it easy
  to <b>annotate a sequential C program to turn it into a parallel
268 269 270
  StarPU program</b> (new in v1.0).
</p>

THIBAULT Samuel's avatar
THIBAULT Samuel committed
271 272
<h4>OpenMP 4 -compatible interface</h4>
<p>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
273
  <a href=http://kstar.gforge.inria.fr/>K'Star</a> provides an OpenMP
THIBAULT Samuel's avatar
THIBAULT Samuel committed
274 275 276 277 278 279 280 281 282
  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>

283 284 285 286
<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).
287 288
</p>

289 290 291 292
<h4>Simulation support</h4>
<p>
  StarPU can very accurately simulate an application execution
  and measure the resulting performance thanks to using the
293
  <a href="http://simgrid.gforge.inria.fr">SimGrid simulator</a> (new in v1.1).  This allows
294 295 296 297 298
  to quickly experiment with various scheduling heuristics, various application
  algorithms, and even various platforms (available GPUs and CPUs, available
  bandwidth)!
</p>

299 300
<h4>All in all</h4>
  <p>
301
All that means that, with the help
302
of <a href="doc/html/cExtensions.html">StarPU's extensions to the C
303 304
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
305
StarPU code, possibly running on all the CPUs and GPUs, and given a data
Nathalie Furmento's avatar
Nathalie Furmento committed
306
distribution over MPI nodes, it is even a distributed version!
307 308 309 310 311 312 313 314 315 316 317 318 319
  </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
320

321
<h4>Supported Architectures</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
322 323
<ul>
<li>SMP/Multicore Processors (x86, PPC, ...) </li>
324
<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 1.1)</li>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
325 326 327
<li>OpenCL devices</li>
<li>Cell Processors (experimental)</li>
</ul>
328
and soon (in v1.2)
329 330
<ul>
<li>Intel SCC</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
331
<li>Intel MIC / Xeon Phi</li>
332
</ul>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
333

334
<h4>Supported Operating Systems</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
335
<ul>
Ludovic Courtès's avatar
Ludovic Courtès committed
336 337
<li>GNU/Linux</li>
<li>Mac OS X</li>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
338 339 340
<li>Windows</li>
</ul>

341
<h4>Performance analysis tools</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
342 343 344 345 346 347 348 349 350
  <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>

351
<p>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
352 353 354 355 356 357
<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>
358
<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>
359 360
</p>

Nathalie Furmento's avatar
website  
Nathalie Furmento committed
361 362 363 364 365 366
<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
367
GFlop/s</b>
368
<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
369 370
</p>

THIBAULT Samuel's avatar
THIBAULT Samuel committed
371 372
<p>
<a href="http://www.hlrs.de/temanejo">Temanejo</a> can be used to debug the task
373
graph, as shown below (new in v1.1).
THIBAULT Samuel's avatar
THIBAULT Samuel committed
374 375 376
</p>

<center>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
377
<a href="images/temanejo.png"><img src="images/temanejo.png" width="50%"/></a>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
378 379
</center>

Nathalie Furmento's avatar
website  
Nathalie Furmento committed
380 381
</div>

382 383 384 385 386
<div class="section" id="software">
<h3>Software using StarPU</h3>

<p>
Some software is known for being able to use StarPU to tackle heterogeneous
387 388
architectures, here is a non-exhaustive list (feel free to ask to be in the
list!):
389 390 391 392
</p>

<ul>
	<li><a href="http://icl.cs.utk.edu/magma/">MAGMA</a>, dense linear algebra library, starting from version 1.1</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
393
	<li><a href="https://project.inria.fr/chameleon/">Chameleon</a>, dense linear algebra library</li>
394 395
	<li><a href="http://www.ida.liu.se/~chrke/skepu/">SkePU</a>, a skeleton programming framework.</li>
	<li><a href="http://pastix.gforge.inria.fr/">PaStiX</a>, sparse linear algebra library, starting from version 5.2.1</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
396
	<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
397
	<li><a href="http://scalfmm-public.gforge.inria.fr/doc/">ScalFMM</a>, N-body interaction simulation using the Fast Multipole Method. </li>
398
	<li><a href="https://project.inria.fr/maphys/fr/">MaPHyS</a>, Massively Parallel Hybrid Solver</li>
399
	<li><a href="https://hal.archives-ouvertes.fr/hal-01086246">SignalPU</a>, a Dataflow-Graph-specific programming model. </li>
400 401
</ul>

402 403 404 405 406
<p>
You can find below the list of publications related to applications
using StarPU.
</p>

407 408
</div>

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

<tt><pre>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
421 422 423 424 425 426
$ 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
427
$ 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
428 429 430 431 432

<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
433 434 435 436 437 438 439 440 441 442 443 444 445 446
<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
447 448 449 450 451 452 453 454 455 456 457 458 459
<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
38400	10216	1847.6</pre></tt>
<p>(MPI simulation is not supported yet)</p>

460 461 462 463
<div class="section" id="publications">
<h3>Publications</h3>
<p>
All StarPU related publications are also
464
listed <a href="./publications">here</a>
465 466 467 468 469 470 471
with the corresponding Bibtex entries.
</p>

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

472
<h4>General Presentations</h4> 
473 474
<ol>
<li>
475 476 477 478 479 480 481
<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>
482 483 484
<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/>
485
[<a href="http://tel.archives-ouvertes.fr/tel-00777154">WWW</a>]
486 487
</li>
<li>
488 489 490
<a name="AugThiNamWac11CCPE"></a>Cédric Augonnet, Samuel Thibault, Raymond Namyst,  and Pierre-André Wacrenier<br/>
<strong>StarPU: A Unified Platform for Task Scheduling on Heterogeneous Multicore Architectures</strong><br/>
<em>Concurrency and Computation: Practice and Experience, Special Issue: Euro-Par 2009</em>, 23:187-198, February 2011<br/>
491
[<a href="http://hal.inria.fr/inria-00550877">WWW</a>]
492
[doi:<a href="http://dx.doi.org/10.1002/cpe.1631">10.1002/cpe.1631</a>]
493
</li>
494
<li>
495 496 497 498
<a name="AugThiNamWac10RR7240"></a>Cédric Augonnet, Samuel Thibault,  and Raymond Namyst<br/>
<strong>StarPU: a Runtime System for Scheduling Tasks over Accelerator-Based Multicore Machines</strong><br/>
Technical Report 7240, INRIA, March 2010<br/>
[<a href="http://hal.inria.fr/inria-00467677">WWW</a>]
499 500
</li>
<li>
501 502 503 504
<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/>
505
[<a href="http://hal.inria.fr/inria-00411581">WWW</a>]
506 507
</li>
<li>
508 509 510 511
<a name="AugThiNamWac09Europar"></a>Cédric Augonnet, Samuel Thibault, Raymond Namyst,  and Pierre-André Wacrenier<br/>
<strong>StarPU: A Unified Platform for Task Scheduling on Heterogeneous Multicore Architectures</strong><br/>
In <em>Proceedings of the 15th International Euro-Par Conference</em>, volume 5704 of <em>Lecture Notes in Computer Science</em>, Delft, The Netherlands, pages 863-874, August 2009<br/>
Springer<br/>
512
[<a href="http://hal.inria.fr/inria-00384363">WWW</a>]
513 514 515 516 517 518 519 520
[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/>
521
[<a href="http://hal.inria.fr/inria-00326917">WWW</a>]
522 523 524 525 526 527
[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/>
528
[<a href="http://hal.inria.fr/inria-00289361">WWW</a>]
529 530
</li>
</ol>
531
<h4>On Composability</h4> 
Nathalie Furmento's avatar
Nathalie Furmento committed
532 533
<ol>
<li>
534 535 536
<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/>
537
[<a href="http://hal.inria.fr/hal-00773610">WWW</a>]
Nathalie Furmento's avatar
Nathalie Furmento committed
538 539
</li>
<li>
540 541 542
<a name="hugo:hal-00824514"></a>Andra Hugo, Abdou Guermouche, Raymond Namyst,  and Pierre-André Wacrenier<br/>
<strong>Composing multiple StarPU applications over heterogeneous machines: a supervised approach</strong><br/>
In <em>Third International Workshop on Accelerators and Hybrid Exascale Systems</em>, Boston, USA, May 2013<br/>
543
[<a href="http://hal.inria.fr/hal-00824514">WWW</a>]
544 545 546 547 548
</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/>
549
[<a href="http://hal.inria.fr/inria-00619654/en/">WWW</a>]
Nathalie Furmento's avatar
Nathalie Furmento committed
550 551
</li>
</ol>
552
<h4>On Scheduling</h4> 
553
<ol>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
554
<li>
555 556 557 558 559 560 561 562
<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>
563 564 565
<a name="agullo:hal-01120507"></a>Emmanuel Agullo, Olivier Beaumont, Lionel Eyraud-Dubois, Julien Herrmann, Suraj Kumar, Loris Marchal,  and Samuel Thibault<br/>
<strong>Bridging the Gap between Performance and Bounds of Cholesky Factorization on Heterogeneous Platforms</strong><br/>
In <em>Heterogeneity in Computing Workshop 2015</em>, Hyderabad, India, May 2015<br/>
566
[<a href="https://hal.inria.fr/hal-01120507">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
567
</li>
568
<li>
569 570 571
<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/>
572 573
[<a href="http://hal.inria.fr/hal-00978364">WWW</a>]
[<a href="http://hal.inria.fr/hal-00978364/PDF/ordonnanceurs_modulaires.pdf">PDF</a>]
574 575
</li>
</ol>
576
<h4>On The C Extensions</h4> 
577
<ol>
578 579 580 581
<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/>
582 583
[<a href="http://hal.inria.fr/hal-00807033">WWW</a>]
[<a href="http://hal.inria.fr/hal-00807033/PDF/RR-8278.pdf">PDF</a>]
584 585
</li>
</ol>
586
<h4>On OpenMP Support on top of StarPU</h4> 
THIBAULT Samuel's avatar
THIBAULT Samuel committed
587
<ol>
588 589 590 591 592
<li>
<a name="virouleau:hal-01081974"></a>Philippe Virouleau, Pierrick BRUNET, François Broquedis, Nathalie Furmento, Samuel Thibault, Olivier Aumage,  and Thierry Gautier<br/>
<strong>Evaluation of OpenMP Dependent Tasks with the KASTORS Benchmark Suite</strong><br/>
In <em>10th International Workshop on OpenMP, IWOMP2014</em>, 10th International Workshop on OpenMP, IWOMP2014, Salvador, Brazil, France, pages 16 - 29, September 2014<br/>
Springer<br/>
593
[<a href="https://hal.inria.fr/hal-01081974">WWW</a>]
594
[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
595 596
</li>
</ol>
597
<h4>On MPI Support</h4> 
598 599
<ol>
<li>
600 601 602 603 604 605 606
<a name="augonnet:hal-00992208"></a>Cédric Augonnet, Olivier Aumage, Nathalie Furmento, Samuel Thibault,  and Raymond Namyst<br/>
<strong>StarPU-MPI: Task Programming over Clusters of Machines Enhanced with Accelerators</strong><br/>
Rapport de recherche RR-8538, INRIA, May 2014<br/>
[<a href="http://hal.inria.fr/hal-00992208">WWW</a>]
[<a href="http://hal.inria.fr/hal-00992208/PDF/RR-8538.pdf">PDF</a>]
</li>
<li>
607 608 609 610 611
<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/>
612
[<a href="http://hal.inria.fr/hal-00725477">WWW</a>]
613
</li>
614
</ol>
Nathalie Furmento's avatar
Nathalie Furmento committed
615
<h4>On Memory Control</h4> 
THIBAULT Samuel's avatar
THIBAULT Samuel committed
616 617
<ol>
<li>
Nathalie Furmento's avatar
Nathalie Furmento committed
618
<a name="sergent:hal-01284004"></a>Marc Sergent, David Goudin, Samuel Thibault,  and Olivier Aumage<br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
619
<strong>Controlling the Memory Subscription of Distributed Applications with a Task-Based Runtime System</strong><br/>
Nathalie Furmento's avatar
Nathalie Furmento committed
620
In <em>21st International Workshop on High-Level Parallel Programming Models and Supportive Environments</em>, Chicago, United States, May 2016<br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
621 622 623 624
[<a href="https://hal.inria.fr/hal-01284004">WWW</a>]
[<a href="https://hal.inria.fr/hal-01284004/file/PID4127657.pdf">PDF</a>]
</li>
</ol>
625
<h4>On Data Transfer Management</h4> 
626 627
<ol>
<li>
628 629 630
<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/>
631
[<a href="http://hal.inria.fr/inria-00523937">WWW</a>]
632
[doi:<a href="http://dx.doi.org/10.1109/ICPADS.2010.129">10.1109/ICPADS.2010.129</a>]
633 634
</li>
</ol>
635
<h4>On Performance Model Tuning</h4> 
636 637
<ol>
<li>
638 639 640 641
<a name="AugThiNam09HPPC"></a>Cédric Augonnet, Samuel Thibault,  and Raymond Namyst<br/>
<strong>Automatic Calibration of Performance Models on Heterogeneous Multicore Architectures</strong><br/>
In <em>Proceedings of the International Euro-Par Workshops 2009, HPPC'09</em>, volume 6043 of <em>Lecture Notes in Computer Science</em>, Delft, The Netherlands, pages 56-65, August 2009<br/>
Springer<br/>
642
[<a href="http://hal.inria.fr/inria-00421333">WWW</a>]
643
[doi:<a href="http://dx.doi.org/10.1007/978-3-642-14122-5_9">10.1007/978-3-642-14122-5_9</a>]
644 645
</li>
</ol>
646
<h4>On The Simulation Support through SimGrid</h4> 
THIBAULT Samuel's avatar
THIBAULT Samuel committed
647
<ol>
THIBAULT Samuel's avatar
update  
THIBAULT Samuel committed
648
<li>
649 650 651
<a name="stanisic:hal-01147997"></a>Luka Stanisic, Samuel Thibault, Arnaud Legrand, Brice Videau,  and Jean-François Méhaut<br/>
<strong>Faithful Performance Prediction of a Dynamic Task-Based Runtime System for Heterogeneous Multi-Core Architectures</strong><br/>
<em>Concurrency and Computation: Practice and Experience</em>, pp 16, May 2015<br/>
652 653
[<a href="https://hal.inria.fr/hal-01147997">WWW</a>]
[<a href="https://hal.inria.fr/hal-01147997/file/CCPE14_article.pdf">PDF</a>]
654
[doi:<a href="http://dx.doi.org/10.1002/cpe">10.1002/cpe</a>]
THIBAULT Samuel's avatar
update  
THIBAULT Samuel committed
655
</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
656
<li>
657 658 659 660
<a name="stanisic:hal-01011633"></a>Luka Stanisic, Samuel Thibault, Arnaud Legrand, Brice Videau,  and Jean-François Méhaut<br/>
<strong>Modeling and Simulation of a Dynamic Task-Based Runtime System for Heterogeneous Multi-Core Architectures</strong><br/>
In <em>Euro-par - 20th International Conference on Parallel Processing</em>, Porto, Portugal, August 2014<br/>
Springer-Verlag<br/>
661 662
[<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
663 664
</li>
</ol>
665
<h4>On The Cell Support</h4> 
666 667
<ol>
<li>
668 669 670
<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/>
671
[<a href="http://hal.inria.fr/inria-00378705">WWW</a>]
672
[doi:<a href="http://dx.doi.org/10.1007/978-3-642-03138-0_36">10.1007/978-3-642-03138-0_36</a>]
673 674
</li>
</ol>
675
<h4>On Applications</h4> 
676
<ol>
THIBAULT Samuel's avatar
update  
THIBAULT Samuel committed
677
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
678 679 680 681 682 683 684
<a name="agullo:hal-01316982"></a>E Agullo, L Giraud, A Guermouche, S Nakov,  and Jean Roman<br/>
<strong>Task-based Conjugate Gradient: from multi-GPU towards heterogeneous architectures</strong><br/>
Research Report 8912, Inria Bordeaux Sud-Ouest, May 2016<br/>
[<a href="https://hal.inria.fr/hal-01316982">WWW</a>]
[<a href="https://hal.inria.fr/hal-01316982/file/RR-8912.pdf">PDF</a>]
</li>
<li>
685 686 687
<a name="rossignon:tel-01230876"></a>Corentin Rossignon<br/>
<strong>A fine grain model programming for parallelization of sparse linear solver</strong><br/>
PhD thesis, Université de Bordeaux, July 2015<br/>
688 689
[<a href="https://tel.archives-ouvertes.fr/tel-01230876">WWW</a>]
[<a href="https://tel.archives-ouvertes.fr/tel-01230876/file/ROSSIGNON_CORENTIN_2015.pdf">PDF</a>]
THIBAULT Samuel's avatar
update  
THIBAULT Samuel committed
690
</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
691
<li>
692 693 694
<a name="MaMiDuAuThiAoNa15"></a>Vìctor Martìnez, David Michéa, Fabrice Dupros, Olivier Aumage, Samuel Thibault, Hideo Aochi,  and Philippe Olivier Alexandre Navaux<br/>
<strong>Towards seismic wave modeling on heterogeneous many-core architectures using task-based runtime system</strong><br/>
In <em>27th International Symposium on Computer Architecture and High Performance Computing (SBAC-PAD)</em>, Florianopolis, Brazil, October 2015<br/>
695 696
[<a href="https://hal.inria.fr/hal-01182746">WWW</a>]
[<a href="https://hal.inria.fr/hal-01182746/file/sbac2015_soumission.pdf">PDF</a>]
697 698
</li>
<li>
699 700 701 702
<a name="sylvain:hal-01005765"></a>Sylvain Henry, Alexandre Denis, Denis Barthou, Marie-Christine Counilh,  and Raymond Namyst<br/>
<strong>Toward OpenCL Automatic Multi-Device Support</strong><br/>
In Fernando Silva, Ines Dutra,  and Vitor Santos Costa, editors, <em>Euro-Par 2014</em>, Porto, Portugal, August 2014<br/>
Springer<br/>
703 704
[<a href="http://hal.inria.fr/hal-01005765">WWW</a>]
[<a href="http://hal.inria.fr/hal-01005765/PDF/final.pdf">PDF</a>]
705 706
</li>
<li>
707 708 709 710 711
<a name="lacoste:hal-00987094"></a>Xavier Lacoste, Mathieu Faverge, Pierre Ramet, Samuel Thibault,  and George Bosilca<br/>
<strong>Taking advantage of hybrid systems for sparse direct solvers via task-based runtimes</strong><br/>
In <em>HCW'2014 workshop of IPDPS</em>, Phoenix, États-Unis, May 2014<br/>
IEEE<br/>
Note: RR-8446 RR-8446<br/>
712 713
[<a href="http://hal.inria.fr/hal-00987094">WWW</a>]
[<a href="http://hal.inria.fr/hal-00987094/PDF/sparsegpus.pdf">PDF</a>]
714 715
</li>
<li>
716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740
<a name="lacoste:hal-00925017"></a>Xavier Lacoste, Mathieu Faverge, Pierre Ramet, Samuel Thibault,  and George Bosilca<br/>
<strong>Taking advantage of hybrid systems for sparse direct solvers via task-based runtimes</strong><br/>
Rapport de recherche RR-8446, INRIA, January 2014<br/>
[<a href="http://hal.inria.fr/hal-00925017">WWW</a>]
[<a href="http://hal.inria.fr/hal-00925017/PDF/RR-8446.pdf">PDF</a>]
</li>
<li>
<a name="sergent:hal-00978602"></a>Emmanuel Agullo, Olivier Aumage, Mathieu Faverge, Nathalie Furmento, Florent Pruvost, Marc Sergent,  and Samuel Thibault<br/>
<strong>Overview of Distributed Linear Algebra on Hybrid Nodes over the StarPU Runtime</strong><br/>
SIAM Conference on Parallel Processing for Scientific Computing, February 2014<br/>
[<a href="http://hal.inria.fr/hal-00978602">WWW</a>]
[<a href="http://hal.inria.fr/hal-00978602/PDF/siampp14.pdf">PDF</a>]
</li>
<li>
<a name="Bor13Thesis"></a>Cyril Bordage<br/>
<strong>Ordonnancement dynamique, adapté aux architectures hétérogènes, de la méthode multipôle pour les équations de Maxwell, en électromagnétisme</strong><br/>
PhD thesis, Université Bordeaux 1, 351 cours de la Libération --- 33405 TALENCE cedex, December 2013<br/>
</li>
<li>
<a name="Hen13Thesis"></a>Sylvain Henry<br/>
<strong>Modèles de programmation et supports exécutifs pour architectures hétérogènes</strong><br/>
PhD thesis, Université Bordeaux 1, 351 cours de la Libération --- 33405 TALENCE cedex, November 2013<br/>
[<a href="http://tel.archives-ouvertes.fr/tel-00948309">WWW</a>]
</li>
<li>
741 742 743
<a name="hen13fhpc"></a>Sylvain Henry<br/>
<strong>ViperVM: a Runtime System for Parallel Functional High-Performance Computing on Heterogeneous Architectures</strong><br/>
In <em>2nd Workshop on Functional High-Performance Computing (FHPC'13)</em>, Boston, États-Unis, September 2013<br/>
744 745
[<a href="http://hal.inria.fr/hal-00851122">WWW</a>]
[<a href="http://hal.inria.fr/hal-00851122/PDF/fhpc13.pdf">PDF</a>]
746 747
</li>
<li>
748 749 750
<a name="odajima:hal-00920915"></a>Tetsuya Odajima, Taisuke Boku, Mitsuhisa Sato, Toshihiro Hanawa, Yuetsu Kodama, Raymond Namyst, Samuel Thibault,  and Olivier Aumage<br/>
<strong>Adaptive Task Size Control on High Level Programming for GPU/CPU Work Sharing</strong><br/>
In <em>The 2013 International Symposium on Advances of Distributed and Parallel Computing (ADPC 2013)</em>, Vietri sul Mare, Italie, December 2013<br/>
751 752
[<a href="http://hal.inria.fr/hal-00920915">WWW</a>]
[<a href="http://hal.inria.fr/hal-00920915/PDF/ADPC2013-117.pdf">PDF</a>]
753 754 755 756 757 758
</li>
<li>
<a name="ohshima:hal-00926144"></a>Satoshi Ohshima, Satoshi Katagiri, Kengo Nakajima, Samuel Thibault,  and Raymond Namyst<br/>
<strong>Implementation of FEM Application on GPU with StarPU</strong><br/>
In <em>SIAM CSE13 - SIAM Conference on Computational Science and Engineering 2013</em>, Boston, États-Unis, February 2013<br/>
SIAM<br/>
759
[<a href="http://hal.inria.fr/hal-00926144">WWW</a>]
760 761 762 763 764
</li>
<li>
<a name="Ros13Renpar"></a>Corentin Rossignon<br/>
<strong>Optimisation du produit matrice-vecteur creux sur architecture GPU pour un simulateur de reservoir</strong><br/>
In <em>21èmes Rencontres Francophones du Parallélisme (RenPar'21)</em>, Grenoble, France, January 2013<br/>
765
[<a href="http://hal.inria.fr/hal-00773571">WWW</a>]
766 767
</li>
<li>
768 769 770
<a name="rossignon:hal-00858350"></a>Corentin Rossignon, Pascal Hénon, Olivier Aumage,  and Samuel Thibault<br/>
<strong>A NUMA-aware fine grain parallelization framework for multi-core architecture</strong><br/>
In <em>PDSEC - 14th IEEE International Workshop on Parallel and Distributed Scientific and Engineering Computing - 2013</em>, Boston, États-Unis, May 2013<br/>
771 772
[<a href="http://hal.inria.fr/hal-00858350">WWW</a>]
[<a href="http://hal.inria.fr/hal-00858350/PDF/taggre_pdsec_2013.pdf">PDF</a>]
773
</li>
774
<li>
775 776 777 778 779 780 781 782 783 784 785 786
<a name="HenDenBar2012TSI"></a>Sylvain Henry, Alexandre Denis,  and Denis Barthou<br/>
<strong>Programmation unifiée multi-accélérateur OpenCL</strong><br/>
<em>Techniques et Sciences Informatiques</em>, (8-9-10):1233-1249, 2012<br/>
[<a href="http://hal.inria.fr/hal-00772742">WWW</a>]
</li>
<li>
<a name="MahManAugThi12TSI"></a>Sidi Ahmed Mahmoudi, Pierre Manneback, Cédric Augonnet,  and Samuel Thibault<br/>
<strong>Traitements d'Images sur Architectures Parallèles et Hétérogènes</strong><br/>
<em>Technique et Science Informatiques</em>, 2012<br/>
[<a href="http://hal.inria.fr/hal-00714858/">WWW</a>]
</li>
<li>
787 788 789
<a name="BenkBajMarSanNamThiEuroPar2012"></a>Siegfried Benkner, Enes Bajrovic, Erich Marth, Martin Sandrieser, Raymond Namyst,  and Samuel Thibault<br/>
<strong>High-Level Support for Pipeline Parallelism on Many-Core Architectures</strong><br/>
In <em>Europar - International European Conference on Parallel and Distributed Computing - 2012</em>, Rhodes Island, Grèce, August 2012<br/>
790 791
[<a href="http://hal.inria.fr/hal-00697020">WWW</a>]
[<a href="http://hal.inria.fr/hal-00697020/PDF/europar2012-submitted.pdf">PDF</a>]
792 793
</li>
<li>
794 795 796
<a name="kessler:hal-00776610"></a>Christoph Kessler, Usman Dastgeer, Samuel Thibault, Raymond Namyst, Andrew Richards, Uwe Dolinsky, Siegfried Benkner, Jesper Larsson Träff,  and Sabri Pllana<br/>
<strong>Programmability and Performance Portability Aspects of Heterogeneous Multi-/Manycore Systems</strong><br/>
In <em>Design, Automation and Test in Europe (DATE)</em>, Dresden, Allemagne, March 2012<br/>
797 798
[<a href="http://hal.inria.fr/hal-00776610">WWW</a>]
[<a href="http://hal.inria.fr/hal-00776610/PDF/date12-paper.pdf">PDF</a>]
799 800
</li>
<li>
801 802 803 804 805 806 807 808
<a name="BenPllTraTsiDolAugBacKesMolOsi11IEEEMicro"></a>Siegfried Benkner, Sabri Pllana, Jesper Larsson Träff, Philippas Tsigas, Uwe Dolinsky, Cédric Augonnet, Beverly Bachmayer, Christoph Kessler, David Moloney,  and Vitaly Osipov<br/>
<strong>PEPPHER: Efficient and Productive Usage of Hybrid Computing Systems</strong><br/>
<em>IEEE Micro</em>, 31(5):28-41, September 2011<br/>
<strong>ISSN:</strong> 0272-1732<br/>
[<a href="http://hal.inria.fr/hal-00648480">WWW</a>]
[doi:<a href="http://dx.doi.org/10.1109/MM.2011.67">10.1109/MM.2011.67</a>]
</li>
<li>
809 810 811
<a name="AguAugDonFavLanLtaTomAICCSA11"></a>Emmanuel Agullo, Cédric Augonnet, Jack Dongarra, Mathieu Faverge, Julien Langou, Hatem Ltaief,  and Stanimire Tomov<br/>
<strong>LU factorization for accelerator-based systems</strong><br/>
In <em>9th ACS/IEEE International Conference on Computer Systems and Applications (AICCSA 11)</em>, Sharm El-Sheikh, Egypt, June 2011<br/>
812
[<a href="http://hal.inria.fr/hal-00654193">WWW</a>]
813 814 815 816 817
</li>
<li>
<a name="AguAugDonFavLtaThiTom11IPDPS"></a>Emmanuel Agullo, Cédric Augonnet, Jack Dongarra, Mathieu Faverge, Hatem Ltaief, Samuel Thibault,  and Stanimire Tomov<br/>
<strong>QR Factorization on a Multicore Node Enhanced with Multiple GPU Accelerators</strong><br/>
In <em>25th IEEE International Parallel & Distributed Processing Symposium (IEEE IPDPS 2011)</em>, Anchorage, Alaska, USA, May 2011<br/>
818
[<a href="http://hal.inria.fr/inria-00547614">WWW</a>]
819 820 821 822 823 824
[doi:<a href="http://dx.doi.org/10.1109/IPDPS.2011.90">10.1109/IPDPS.2011.90</a>]
</li>
<li>
<a name="DasKesThi11ParCo"></a>Usman Dastgeer, Christoph Kessler,  and Samuel Thibault<br/>
<strong>Flexible runtime support for efficient skeleton programming on hybrid systems</strong><br/>
In <em>Proceedings of the International Conference on Parallel Computing (ParCo), Applications, Tools and Techniques on the Road to Exascale Computing</em>, volume 22 of <em>Advances of Parallel Computing</em>, Gent, Belgium, pages 159-166, August 2011<br/>
825
[<a href="http://hal.inria.fr/inria-00606200/">WWW</a>]
826 827 828 829 830
</li>
<li>
<a name="Hen11Renpar"></a>Sylvain Henry<br/>
<strong>Programmation multi-accélérateurs unifiée en OpenCL</strong><br/>
In <em>20èmes Rencontres Francophones du Parallélisme (RenPar'20)</em>, Saint Malo, France, May 2011<br/>
831
[<a href="http://hal.archives-ouvertes.fr/hal-00643257">WWW</a>]
832 833 834 835 836
</li>
<li>
<a name="MahManAugThi11Renpar20"></a>Sidi Ahmed Mahmoudi, Pierre Manneback, Cédric Augonnet,  and Samuel Thibault<br/>
<strong>Détection optimale des coins et contours dans des bases d'images volumineuses sur architectures multicoeurs hétérogènes</strong><br/>
In <em>20èmes Rencontres Francophones du Parallélisme</em>, Saint-Malo / France, May 2011<br/>
837
[<a href="http://hal.inria.fr/inria-00606195">WWW</a>]
838 839
</li>
<li>
840 841 842 843 844 845 846
<a name="AguAugDonLtaNamThiTomGPUgems"></a>Emmanuel Agullo, Cédric Augonnet, Jack Dongarra, Hatem Ltaief, Raymond Namyst, Samuel Thibault,  and Stanimire Tomov<br/>
<strong>A Hybridization Methodology for High-Performance Linear Algebra Software for GPUs</strong><br/>
In Wen-mei W<br/> Hwu, editor, <em>GPU Computing Gems</em>, volume 2<br/>
Morgan Kaufmann, September 2010<br/>
[<a href="http://hal.inria.fr/inria-00547847">WWW</a>]
</li>
<li>
847 848 849
<a name="AguAugDonLtaNamRomThiTom10SAAHPC"></a>Emmanuel Agullo, Cédric Augonnet, Jack Dongarra, Hatem Ltaief, Raymond Namyst, Jean Roman, Samuel Thibault,  and Stanimire Tomov<br/>
<strong>Dynamically scheduled Cholesky factorization on multicore architectures with GPU accelerators</strong><br/>
In <em>Symposium on Application Accelerators in High Performance Computing (SAAHPC)</em>, Knoxville, USA, July 2010<br/>
850
[<a href="http://hal.inria.fr/inria-00547616">WWW</a>]
851
</li>
852
</ol>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
853

854
<div class="section bot">
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
855
<p class="updated">
856
  Last updated on 2016/04/13.
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
857
</p>
858
</div>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
859 860 861

</body>
</html>