index.html 62.2 KB
Newer Older
1
<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.0 Transitional//EN" "http://www.w3.org/TR/REC-html40/loose.dtd">
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
2 3 4 5
<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>
Nathalie Furmento's avatar
Nathalie Furmento committed
6
<link rel="stylesheet" type="text/css" href="/style.css" />
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
7 8 9 10
</HEAD>

<body>

11
<div class="title">
Nathalie Furmento's avatar
Nathalie Furmento committed
12
<h1><a href="/">StarPU</a></h1>
13 14
<h2>A Unified Runtime System for Heterogeneous Multicore Architectures</h2>
</div>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
15

16
<div class="menu">
17
<a href="https://team.inria.fr/storm/">STORM TEAM</a> |
18
&nbsp; &nbsp; &nbsp;
19
|
20
<a href="#overview">Overview</a> |
Nathalie Furmento's avatar
Nathalie Furmento committed
21
<a href="#news">News</a> |
22
<a href="#contact">Contact</a> |
Nathalie Furmento's avatar
Nathalie Furmento committed
23
<a href="people/">People</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> |
Nathalie Furmento's avatar
Nathalie Furmento committed
27
<a href="help/">Help</a> |
28
<a href="#publications">Publications</a> |
29
<a href="internships/">Jobs/Interns</a> |
30
<a href="files/">Download</a> |
THIBAULT Samuel's avatar
THIBAULT Samuel committed
31
<a href="market/">Market</a> |
Nathalie Furmento's avatar
Nathalie Furmento committed
32
<a href="tutorials/">Tutorials</a> |
Nathalie Furmento's avatar
Nathalie Furmento committed
33
<a href="/files/testing/morse/master/">Benchmarks</a> |
THIBAULT Samuel's avatar
THIBAULT Samuel committed
34
<a href="https://gitlab.inria.fr/starpu/starpu-intranet/-/wikis/home">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 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
<span class="note">The StarPU documentation is available in
THIBAULT Samuel's avatar
THIBAULT Samuel committed
65
<a href="./files/doc/starpu.pdf">PDF</a> and in <a href="./files/doc/html/">HTML</a>.</span>
66 67 68 69
Please note that these documents are up-to-date with the latest release of
StarPU.
</p>
<p>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
70 71
The latest documentation in <a href="/files/testing/master/doc/starpu.pdf">PDF</a>
and <a href="/files/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
<h3>News</h3>
78
<p>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
79 80
June 2020 <b>&raquo;&nbsp;</b><a href="http://gforge.inria.fr/frs/?group_id=1570"><b>The
      1.2.10 release of StarPU is now available!</b></a>.
81 82 83 84
	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>
85
<p>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
86 87 88
June
2020 <b>&raquo;&nbsp;</b><a href="http://gforge.inria.fr/frs/?group_id=1570"><b>The
      release 1.3.4 of StarPU is now
89 90 91 92 93 94 95 96
      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>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
97 98 99 100 101 102 103
<p>
  November
  2019 <b>&raquo;&nbsp;</b>.
  A <a href="/tutorials/2019-11-HPNS-Inria/">StarPU tutorial</a> will
  be given as part of the Inria automn school "High Performance
  Numerical Simulation".
</p>
104
<p>
105 106 107 108 109 110
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>
111
</div>
112 113

<div class="section emphasizebot" style="text-align: right; font-style: italic;">
114
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>.
115
See also the full <a href="news/">news</a>.
116 117
</div>

118 119 120 121 122 123 124 125 126 127 128
<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
129
<div class="section" id="tutorial">
130 131 132 133
<h3>Tutorial material</h3>
<p>
The latest tutorial material for StarPU is composed of two parts:
<ul>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
134 135
<li><a href="./tutorials/2016-06-PATC/slides/01_introducing_starpu.pdf">Introducing StarPU</a></li>
<li><a href="./tutorials/2016-06-PATC/slides/02_mastering_starpu.pdf">Mastering StarPU</a></li>
136 137 138 139
</ul>
</p>
</div>

THIBAULT Samuel's avatar
THIBAULT Samuel committed
140 141 142 143 144 145 146 147
<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>

148 149
<div class="section" id="contact">
<h3>Contact</h3>
150
<p>For any questions regarding StarPU, please contact the StarPU developers mailing list.</p>
151 152 153
<pre>
<a href="mailto:starpu-devel@lists.gforge.inria.fr?subject=StarPU">starpu-devel@lists.gforge.inria.fr</a>
</pre>
154
<p>Details of the <a href="people/">StarPU team people</a> are also available.</p>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
155 156
</div>

157 158
<div class="section" id="features">
<h3>Features</h3>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
159

160
<h4>Portability</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
161
  <p>
162 163 164 165 166 167 168 169
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.
170 171 172 173
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).
174 175 176
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
177 178
  </p>

179 180 181
<h4>Genericity</h4>
  <p>
The StarPU programming interface is very generic. For intance, various data
182
structures are supported mainline (vectors, dense matrices, CSR/BCSR/COO sparse matrices, ...),
183 184 185 186 187 188
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>

189
<h4>Data transfers</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
190
  <p>
191
To relieve programmers from the burden of explicit data transfers, a high-level
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
192
data management library enforces memory coherency over the machine: before a
193 194
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
195 196
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
197
prefetching</b> data, which thus permits to <b>overlap data transfers with computations</b>
THIBAULT Samuel's avatar
update  
THIBAULT Samuel committed
198
(including <b>GPU-GPU direct transfers</b>) to achieve the most of the architecture.
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
199 200
  </p>

201
<h4>Dependencies</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
202
  <p>
203
Dependencies between tasks can be given either of several ways, to provide the
204 205
programmer with best flexibility:
  <ul>
206
    <li><b>implicitly</b> from RAW, WAW, and WAR data dependencies.</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
207
    <li>explicitly through <b>tags</b> which act as rendez-vous points between
208
    tasks (thus including tasks which have not been created yet),</li>
209
    <li><b>explicitly</b> between pairs of tasks,</li>
210
  </ul>
211 212
  </p>
  <p>
213 214 215
  These dependencies are computed in a completely decentralized way, and can be
  introduced completely dynamically as tasks get submitted by the application
  while tasks previously submitted are being executed.
216 217
  </p>
  <p>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
218
StarPU also supports an OpenMP-like <a href="files/doc/html/DataManagement.html#DataReduction">reduction</a> access mode (new in v0.9).
219 220
  </p>
  <p>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
221
It also supports a <a href="files/doc/html/DataManagement.html#DataCommute">commute</a> access mode to allow data access commutativity (new in v1.2).
222 223
  </p>

224 225 226 227 228
  <p>
It also supports transparent dependencies tracking between hierarchical subpieces of data
through asynchronous partitioning (new in v1.3).
  </p>

229 230 231
<h4>Heterogeneous Scheduling</h4>
  <p>
StarPU obtains
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
232
portable performances by efficiently (and easily) using all computing resources
233
at the same time. StarPU also takes advantage of the <b>heterogeneous</b> nature of a
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
234
machine, for instance by using scheduling strategies based on auto-tuned
235 236 237
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>.
238
Various strategies and variants are available. Some of them are centralized, but
239
most of them are <b>completely distributed</b>. <i>dmdas</i> (a data-locality-aware MCT strategy,
240
thus similar to heft but starts executing tasks before the whole task graph is
241 242 243
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), ...
244
The overhead per task is typically around the order of
245 246 247
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
248 249
  </p>

250 251
<h4>Clusters</h4>
  <p>
252
To deal with clusters, StarPU can nicely integrate with <a
THIBAULT Samuel's avatar
THIBAULT Samuel committed
253
	href="files/doc/html/MPISupport.html">MPI</a>, through explicit or implicit
254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273
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>
274 275 276 277 278
  </p>

<h4>Out of core</h4>
  <p>
When memory is not big enough for the working set, one may have to resort to
THIBAULT Samuel's avatar
THIBAULT Samuel committed
279
using disks. StarPU makes this seamless thanks to its <a href="files/doc/html/OutOfCore.html">out of core support</a> (new in v1.2).
THIBAULT Samuel's avatar
update  
THIBAULT Samuel committed
280 281
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.
282 283
  </p>

284
  <!--
285 286 287
<h4>Extensions to the C Language</h4>
<p>
  StarPU comes with a GCC plug-in
THIBAULT Samuel's avatar
THIBAULT Samuel committed
288
  that <a href="files/doc/html/cExtensions.html">extends the C programming
289 290
  language</a> with pragmas and attributes that make it easy
  to <b>annotate a sequential C program to turn it into a parallel
291 292
  StarPU program</b> (new in v1.0).
</p>
293 294 295 296 297 298
  -->

<h4>Fortran interface</h4>
<p>
  StarPU comes with native Fortran bindings and examples.
</p>
299

THIBAULT Samuel's avatar
THIBAULT Samuel committed
300 301
<h4>OpenMP 4 -compatible interface</h4>
<p>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
302
  <a href=http://kstar.gforge.inria.fr/>K'Star</a> provides an OpenMP
THIBAULT Samuel's avatar
THIBAULT Samuel committed
303 304 305 306 307 308 309 310 311
  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>

312 313
<h4>OpenCL-compatible interface</h4>
<p>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
314
  StarPU provides an <a href="files/doc/html/SOCLOpenclExtensions.html">OpenCL-compatible interface, SOCL</a>
315
  which allows to simply run OpenCL applications on top of StarPU (new in v1.0).
316 317
</p>

318 319 320 321
<h4>Simulation support</h4>
<p>
  StarPU can very accurately simulate an application execution
  and measure the resulting performance thanks to using the
322
  <a href="http://simgrid.gforge.inria.fr">SimGrid simulator</a> (new in v1.1).  This allows
323 324 325 326 327
  to quickly experiment with various scheduling heuristics, various application
  algorithms, and even various platforms (available GPUs and CPUs, available
  bandwidth)!
</p>

328 329
<h4>All in all</h4>
  <p>
330 331 332
All that means that the following sequential source code of a tiled version of
the classical Cholesky factorization algorithm using BLAS is also (a
almost) valid StarPU code, possibly running on all the CPUs and GPUs, and given a data
Nathalie Furmento's avatar
Nathalie Furmento committed
333
distribution over MPI nodes, it is even a distributed version!
334 335 336 337 338 339 340 341 342 343 344 345 346
  </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
347

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

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

THIBAULT Samuel's avatar
THIBAULT Samuel committed
364 365 366 367 368 369 370 371 372 373 374
<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>

375
<h4>Performance analysis tools</h4>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
376 377 378 379 380 381 382 383 384
  <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>

385
<p>
Nathalie Furmento's avatar
website  
Nathalie Furmento committed
386 387 388 389 390 391
<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>
392
<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>
393 394
</p>

Nathalie Furmento's avatar
website  
Nathalie Furmento committed
395 396 397 398 399 400
<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
401
GFlop/s</b>
402
<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
403 404
</p>

THIBAULT Samuel's avatar
THIBAULT Samuel committed
405 406
<p>
<a href="http://www.hlrs.de/temanejo">Temanejo</a> can be used to debug the task
407
graph, as shown below (new in v1.1).
THIBAULT Samuel's avatar
THIBAULT Samuel committed
408 409 410
</p>

<center>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
411
<a href="images/temanejo.png"><img src="images/temanejo.png" width="50%"/></a>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
412 413
</center>

Nathalie Furmento's avatar
website  
Nathalie Furmento committed
414 415
</div>

416 417 418 419 420
<div class="section" id="software">
<h3>Software using StarPU</h3>

<p>
Some software is known for being able to use StarPU to tackle heterogeneous
421
architectures, here is a non-exhaustive list (feel free to ask to be added to the
422
list!):
423 424 425
</p>

<ul>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
426
	<li><a href="https://github.com/ecrc/al4san">AL4SAN</a>, dense linear algebra library</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
427
	<li><a href="https://project.inria.fr/chameleon/">Chameleon</a>, dense linear algebra library</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
428
	<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
429
	<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
430
	<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
431
	<li><a href="http://github.com/ecrc/hicma">HiCMA</a>, Low-rank general linear algebra library</li>
432
	<li>hmat, hierarchical matrix C/C++ library</li>
433
        <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
434
	<li><a href="http://github.com/ecrc/ksvd">KSVD</a>, dense SVD on distributed-memory manycore systems</li>
435 436
	<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
437
	<li>MASA-StarPU, Parallel Sequence Comparison</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
438
	<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>
439
	<li><a href="http://pastix.gforge.inria.fr/">PaStiX</a>, sparse linear algebra library, starting from version 5.2.1</li>
440
	<li>PEPPHER, Performance Portability and Programmability for Heterogeneous Many-core Architectures</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
441
	<li><a href="http://github.com/ecrc/qdwh">QDWH</a>, QR-based Dynamically Weighted Halley</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
442
	<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
443
	<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
444
	<li><a href="https://tel.archives-ouvertes.fr/tel-01410049/">SCHNAPS</a>, Solver for Conservative Hypebolic Non-linear systems Applied to PlasmaS. </li>
445 446
	<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
447
	<li><a href="https://github.com/NLAFET/StarNEig/">StarNEig</a>, a dense nonsymmetric (generalized) eigenvalue solving library.</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
448
	<li><a href="http://github.com/ecrc/stars-h">STARS-H</a>, HPC low-rank matrix market</li>
449
	<li><a href="http://www.xcalablemp.org/">XcalableMP</a>, Directive-based language eXtension for Scalable and performance-aware Parallel Programming</li>
450 451
</ul>

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

456 457
</div>

THIBAULT Samuel's avatar
THIBAULT Samuel committed
458 459 460 461
<div class="section" id="tryit">
<h3>Give it a try!</h3>
<p>
You can easily try the performance on the Cholesky factorization for
THIBAULT Samuel's avatar
THIBAULT Samuel committed
462 463 464
instance. Make sure to have the pkg-config and
<a href="http://www.open-mpi.org/projects/hwloc/">hwloc</a>
software installed for
THIBAULT Samuel's avatar
THIBAULT Samuel committed
465 466 467
proper CPU control and BLAS kernels for your computation units and configured in
your environment (e.g. MKL for CPUs and CUBLAS for GPUs).
</p>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
468 469

<tt><pre>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
470 471 472 473 474 475
$ wget http://starpu.gforge.inria.fr/files/starpu-someversion.tar.gz
$ tar xf starpu-someversion.tar.gz
$ cd starpu-someversion
$ ./configure
$ make -j 12
$ STARPU_SCHED=dmdas ./examples/cholesky/cholesky_implicit -size $((960*40)) -nblocks 40
THIBAULT Samuel's avatar
THIBAULT Samuel committed
476
$ STARPU_SCHED=dmdas mpirun -np 4 -machinefile mymachines ./mpi/examples/matrix_decomposition/mpi_cholesky_distributed -size $((960*40*4)) -nblocks $((40*4))</pre></tt>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
477 478 479 480 481

<p>Note that the dmdas scheduler uses performance models, and thus needs
calibration execution before exhibiting optimized performance (until the "model
something is not calibrated enough" messages go away).</p>

THIBAULT Samuel's avatar
THIBAULT Samuel committed
482 483 484 485 486 487 488 489 490 491 492 493 494 495
<p>To get a glimpse at what happened, you can get an execution trace by
installing
<a href="http://savannah.nongnu.org/projects/fkt">FxT</a>
and <a href="http://vite.gforge.inria.fr/">ViTE</a>, and enabling traces:
</p>

<tt><pre>
$ ./configure --with-fxt
$ make -j 12
$ STARPU_SCHED=dmdas ./examples/cholesky/cholesky_implicit -size $((960*40)) -nblocks 40
$ ./tools/starpu_fxt_tool -i /tmp/prof_file_${USER}_0
$ vite paje.trace
</pre></tt>

THIBAULT Samuel's avatar
THIBAULT Samuel committed
496 497 498 499 500 501 502 503 504 505
<p>
Starting with StarPU 1.1, it is also possible to reproduce the performance that
we show in our articles on our machines, by installing simgrid, and then using
the simulation mode of StarPU using the performance models of our machines:
</p>
  <tt><pre>
$ ./configure --enable-simgrid
$ make -j 12
$ STARPU_PERF_MODEL_DIR=$PWD/tools/perfmodels/sampling STARPU_HOSTNAME=mirage STARPU_SCHED=dmdas ./examples/cholesky/cholesky_implicit -size $((960*40)) -nblocks 40
# size	ms	GFlops
THIBAULT Samuel's avatar
THIBAULT Samuel committed
506
38400	9915	1903.7</pre></tt>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
507 508
<p>(MPI simulation is not supported yet)</p>

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

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

522 523
<p>
If you need to cite StarPU, please
524
reference <a href="publications/Year/2011.html#AugThiNamWac11CCPE">[StarPU: A Unified Platform
525 526 527 528 529
    for Task Scheduling on Heterogeneous Multicore Architectures]</a>
for a general presentation. Other sub-sections below will give you
references for more specific aspects of StarPU.
</p>

530
<h4>General Presentations</h4> 
531
<a name="PublicationsGeneralPresentations"></a>
532 533
<ol>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
534 535 536 537 538 539 540
<a name="thibault:tel-01959127"></a>Samuel Thibault<br/>
<strong>On Runtime Systems for Task-based Programming on Heterogeneous Platforms</strong><br/>
Habilitation à diriger des recherches, Université de Bordeaux, December 2018<br/>
[<a href="https://hal.inria.fr/tel-01959127">WWW</a>]
[<a href="https://hal.inria.fr/tel-01959127/file/hdr.pdf">PDF</a>]
</li>
<li>
541 542 543
<a name="Aug11Thesis"></a>Cédric Augonnet<br/>
<strong>Scheduling Tasks over Multicore machines enhanced with Accelerators: a Runtime System's Perspective</strong><br/>
PhD thesis, Université Bordeaux 1, 351 cours de la Libération --- 33405 TALENCE cedex, December 2011<br/>
544
[<a href="http://tel.archives-ouvertes.fr/tel-00777154">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
545
[<a href="http://tel.archives-ouvertes.fr/tel-00777154/document">PDF</a>]
546 547
</li>
<li>
548 549
<a name="AugThiNamWac11CCPE"></a>Cédric Augonnet, Samuel Thibault, Raymond Namyst,  and Pierre-André Wacrenier<br/>
<strong>StarPU: A Unified Platform for Task Scheduling on Heterogeneous Multicore Architectures</strong><br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
550
<em>CCPE - Concurrency and Computation: Practice and Experience, Special Issue: Euro-Par 2009</em>, 23:187-198, February 2011<br/>
551
[<a href="http://hal.inria.fr/inria-00550877">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
552
[<a href="http://hal.inria.fr/inria-00550877/document">PDF</a>]
553
[doi:<a href="http://dx.doi.org/10.1002/cpe.1631">10.1002/cpe.1631</a>]
554
</li>
555
<li>
556 557
<a name="AugThiNamWac10RR7240"></a>Cédric Augonnet, Samuel Thibault,  and Raymond Namyst<br/>
<strong>StarPU: a Runtime System for Scheduling Tasks over Accelerator-Based Multicore Machines</strong><br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
558
Research Report RR-7240, INRIA, March 2010<br/>
559
[<a href="http://hal.inria.fr/inria-00467677">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
560
[<a href="http://hal.inria.fr/inria-00467677/document">PDF</a>]
561 562
</li>
<li>
563 564 565 566
<a name="Aug09Renpar19"></a>Cédric Augonnet<br/>
<strong>StarPU: un support exécutif unifié pour les architectures multicoeurs hétérogènes</strong><br/>
In <em>19èmes Rencontres Francophones du Parallélisme</em>, Toulouse / France, September 2009<br/>
Note: Best Paper Award<br/>
567
[<a href="http://hal.inria.fr/inria-00411581">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
568
[<a href="http://hal.inria.fr/inria-00411581/document">PDF</a>]
569 570
</li>
<li>
571 572
<a name="AugThiNamWac09Europar"></a>Cédric Augonnet, Samuel Thibault, Raymond Namyst,  and Pierre-André Wacrenier<br/>
<strong>StarPU: A Unified Platform for Task Scheduling on Heterogeneous Multicore Architectures</strong><br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
573
In <em>Euro-Par - 15th International Conference on Parallel Processing</em>, volume 5704 of <em>Lecture Notes in Computer Science</em>, Delft, The Netherlands, pages 863-874, August 2009<br/>
574
Springer<br/>
575
[<a href="http://hal.inria.fr/inria-00384363">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
576
[<a href="http://hal.inria.fr/inria-00384363/document">PDF</a>]
577 578 579 580 581 582 583 584
[doi:<a href="http://dx.doi.org/10.1007/978-3-642-03869-3_80">10.1007/978-3-642-03869-3_80</a>]
</li>
<li>
<a name="AugNam08HPPC"></a>Cédric Augonnet and Raymond Namyst<br/>
<strong>A unified runtime system for heterogeneous multicore architectures</strong><br/>
In <em>Proceedings of the International Euro-Par Workshops 2008, HPPC'08</em>, volume 5415 of <em>Lecture Notes in Computer Science</em>, Las Palmas de Gran Canaria, Spain, pages 174-183, August 2008<br/>
Springer<br/>
<strong>ISBN:</strong> 978-3-642-00954-9<br/>
585
[<a href="http://hal.inria.fr/inria-00326917">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
586
[<a href="http://hal.inria.fr/inria-00326917/document">PDF</a>]
587 588 589 590 591 592
[doi:<a href="http://dx.doi.org/10.1007/978-3-642-00955-6_22">10.1007/978-3-642-00955-6_22</a>]
</li>
<li>
<a name="Aug08Master"></a>Cédric Augonnet<br/>
<strong>Vers des supports d'exécution capables d'exploiter les machines multicoeurs hétérogènes</strong><br/>
Mémoire de DEA, Université Bordeaux 1, June 2008<br/>
593
[<a href="http://hal.inria.fr/inria-00289361">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
594
[<a href="http://hal.inria.fr/inria-00289361/document">PDF</a>]
595 596
</li>
</ol>
597
<h4>On Composability</h4> 
598
<a name="PublicationsOnComposability"></a>
Nathalie Furmento's avatar
Nathalie Furmento committed
599 600
<ol>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
601 602 603 604 605 606 607
<a name="hugo:tel-01162975"></a>Andra-Ecaterina Hugo<br/>
<strong>Composability of parallel codes on heterogeneous architectures</strong><br/>
Theses, Université de Bordeaux, December 2014<br/>
[<a href="https://tel.archives-ouvertes.fr/tel-01162975">WWW</a>]
[<a href="https://tel.archives-ouvertes.fr/tel-01162975/file/HUGO_ANDRA_2014.pdf">PDF</a>]
</li>
<li>
608 609 610
<a name="AH13Renpar"></a>Andra Hugo<br/>
<strong>Le problème de la composition parallèle : une approche supervisée</strong><br/>
In <em>21èmes Rencontres Francophones du Parallélisme (RenPar'21)</em>, Grenoble, France, January 2013<br/>
611
[<a href="http://hal.inria.fr/hal-00773610">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
612
[<a href="http://hal.inria.fr/hal-00773610/document">PDF</a>]
Nathalie Furmento's avatar
Nathalie Furmento committed
613 614
</li>
<li>
615 616 617
<a name="hugo:hal-00824514"></a>Andra Hugo, Abdou Guermouche, Raymond Namyst,  and Pierre-André Wacrenier<br/>
<strong>Composing multiple StarPU applications over heterogeneous machines: a supervised approach</strong><br/>
In <em>Third International Workshop on Accelerators and Hybrid Exascale Systems</em>, Boston, USA, May 2013<br/>
618
[<a href="http://hal.inria.fr/hal-00824514">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
619
[<a href="http://hal.inria.fr/hal-00824514/document">PDF</a>]
620 621 622 623 624
</li>
<li>
<a name="AH11Master"></a>Andra Hugo<br/>
<strong>Composabilité de codes parallèles sur architectures hétérogènes</strong><br/>
Mémoire de Master, Université Bordeaux 1, June 2011<br/>
625
[<a href="http://hal.inria.fr/inria-00619654/en/">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
626
[<a href="http://hal.inria.fr/inria-00619654/document">PDF</a>]
Nathalie Furmento's avatar
Nathalie Furmento committed
627 628
</li>
</ol>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
629 630 631 632 633 634 635 636 637 638 639
<h4>On Parallel Tasks</h4> 
<a name="PublicationsOnParallelTasks"></a>
<ol>
<li>
<a name="cojean:tel-01816341"></a>Terry Cojean<br/>
<strong>Programmation of heterogeneous architectures using moldable tasks</strong><br/>
Theses, Université de Bordeaux, March 2018<br/>
[<a href="https://tel.archives-ouvertes.fr/tel-01816341">WWW</a>]
[<a href="https://tel.archives-ouvertes.fr/tel-01816341/file/COJEAN_TERRY_2018.pdf">PDF</a>]
</li>
</ol>
640
<h4>On Scheduling</h4> 
641
<a name="PublicationsOnScheduling"></a>
642
<ol>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
643
<li>
644 645 646 647 648 649 650 651
<a name="bramas:hal-02120736"></a>Bérenger Bramas<br/>
<strong>Impact study of data locality on task-based applications through the Heteroprio scheduler</strong><br/>
<em>PeerJ Computer Science</em>, May 2019<br/>
[<a href="https://hal.inria.fr/hal-02120736">WWW</a>]
[<a href="https://hal.inria.fr/hal-02120736/file/peerj-cs-190.pdf">PDF</a>]
[doi:<a href="http://dx.doi.org/10.7717/peerj-cs.190">10.7717/peerj-cs.190</a>]
</li>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
652 653 654 655 656 657 658 659 660
<a name="leandronesi:hal-02275363"></a>Lucas Leandro Nesi, Samuel Thibault, Luka Stanisic,  and Lucas Mello Schnorr<br/>
<strong>Visual Performance Analysis of Memory Behavior in a Task-Based Runtime on Hybrid Platforms</strong><br/>
In <em>2019 19th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (CCGRID)</em>, Larnaca, Cyprus, pages 142-151, May 2019<br/>
IEEE<br/>
[<a href="https://hal.inria.fr/hal-02275363">WWW</a>]
[<a href="https://hal.inria.fr/hal-02275363/file/CCGRID_camera_ready.pdf">PDF</a>]
[doi:<a href="http://dx.doi.org/10.1109/CCGRID.2019.00025">10.1109/CCGRID.2019.00025</a>]
</li>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
661 662 663 664 665 666 667
<a name="alias:hal-02421327"></a>Christophe Alias, Samuel Thibault,  and Laure Gonnord<br/>
<strong>A Compiler Algorithm to Guide Runtime Scheduling</strong><br/>
Research Report RR-9315, INRIA Grenoble ; INRIA Bordeaux, December 2019<br/>
[<a href="https://hal.inria.fr/hal-02421327">WWW</a>]
[<a href="https://hal.inria.fr/hal-02421327/file/RR-9315.pdf">PDF</a>]
</li>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
668 669
<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
670
<em>CCPE - Concurrency and Computation: Practice and Experience</em>, 30, April 2018<br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
671 672 673 674 675
[<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>
676
<a name="pinto:hal-01842038"></a>Vinicius Garcia Pinto, Lucas Mello Schnorr, Arnaud Legrand, Samuel Thibault, Luka Stanisic,  and Vincent Danjean<br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
677
<strong>Detecção de Anomalias de Desempenho em Aplicações de Alto Desempenho baseadas em Tarefas em Clusters Hìbridos</strong><br/>
THIBAULT Samuel's avatar
typo  
THIBAULT Samuel committed
678
In <em>WPerformance - 17o Workshop em Desempenho de Sistemas Computacionais e de Comunicação</em>, Natal, Brazil, July 2018<br/>
679 680 681 682
[<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
683 684
<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
685
PhD thesis, Université de Bordeaux, April 2017<br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
686 687 688 689
[<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
690 691 692 693 694 695 696 697
<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>
698 699 700 701 702 703 704 705
<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
706 707 708 709
<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
710
[<a href="https://hal.inria.fr/hal-01361992v2/document">PDF</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
711 712
</li>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
713 714
<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/>
715
In <em>HeteroPar'2016 workshop of Euro-Par</em>, Grenoble, France, August 2016<br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
716 717
[<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
718 719
</li>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
720 721
<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
722
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
723 724
Note: Held in conjunction with SC16<br/>
[<a href="https://hal.inria.fr/hal-01353962">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
725 726
[<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
727
</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
728
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
729 730 731
<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
732
[<a href="http://ieeexplore.ieee.org/abstract/document/7789322/">WWW</a>]
733 734 735
[doi:<a href="http://dx.doi.org/10.1109/SBAC-PAD.2016.14">10.1109/SBAC-PAD.2016.14</a>]
</li>
<li>
736 737 738 739 740 741 742
<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>
743 744
<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
745
In <em>HCW'2015 - Heterogeneity in Computing Workshop of IPDPS</em>, Hyderabad, India, May 2015<br/>
746
[<a href="https://hal.inria.fr/hal-01120507">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
747
[<a href="https://hal.inria.fr/hal-01120507/document">PDF</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
748
[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
749
</li>
750
<li>
751 752 753
<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/>
754 755
[<a href="http://hal.inria.fr/hal-00978364">WWW</a>]
[<a href="http://hal.inria.fr/hal-00978364/PDF/ordonnanceurs_modulaires.pdf">PDF</a>]
756
</li>
757 758 759 760 761 762 763 764
<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>
765
</ol>
766
<h4>On The C Extensions</h4> 
767
<a name="PublicationsOnTheCExtensions"></a>
768
<ol>
769 770 771 772
<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/>
773 774
[<a href="http://hal.inria.fr/hal-00807033">WWW</a>]
[<a href="http://hal.inria.fr/hal-00807033/PDF/RR-8278.pdf">PDF</a>]
775 776
</li>
</ol>
777
<h4>On OpenMP Support on top of StarPU</h4> 
778
<a name="PublicationsOnOpenMPSupportontopofStarPU"></a>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
779
<ol>
780
<li>
Nathalie Furmento's avatar
Nathalie Furmento committed
781 782 783 784 785 786 787 788
<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>
789
<a name="agullo:hal-01372022"></a>Emmanuel Agullo, Olivier Aumage, Berenger Bramas, Olivier Coulaud,  and Samuel Pitoiset<br/>
790
<strong>Bridging the gap between OpenMP 4.0 and native runtime systems for the fast multipole method</strong><br/>
791 792 793 794 795
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
796
<a name="virouleau:hal-01081974"></a>Philippe Virouleau, Pierrick Brunet, François Broquedis, Nathalie Furmento, Samuel Thibault, Olivier Aumage,  and Thierry Gautier<br/>
797
<strong>Evaluation of OpenMP Dependent Tasks with the KASTORS Benchmark Suite</strong><br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
798
In <em>IWOMP2014 - 10th International Workshop on OpenMP</em>, 10th International Workshop on OpenMP, IWOMP2014, Salvador, Brazil, France, pages 16 - 29, September 2014<br/>
799
Springer<br/>
800
[<a href="https://hal.inria.fr/hal-01081974">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
801
[<a href="https://hal.inria.fr/hal-01081974/document">PDF</a>]
802
[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
803 804
</li>
</ol>
805
<h4>On MPI Support</h4> 
806
<a name="PublicationsOnMPISupport"></a>
807 808
<ol>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
809 810 811 812 813 814 815 816
<a name="denis:hal-02872765"></a>Alexandre Denis, Emmanuel Jeannot, Philippe Swartvagher,  and Samuel Thibault<br/>
<strong>Using Dynamic Broadcasts to improve Task-Based Runtime Performances</strong><br/>
In <em>Euro-Par - 26th International European Conference on Parallel and Distributed Computing</em>, Euro-Par 2020, Warsaw, Poland, August 2020<br/>
Rzadca and Malawski, Springer<br/>
[<a href="https://hal.inria.fr/hal-02872765">WWW</a>]
[<a href="https://hal.inria.fr/hal-02872765/file/dynamic_broadcasts.pdf">PDF</a>]
</li>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
817 818 819 820 821 822 823
<a name="lion:hal-02296118"></a>Romain Lion<br/>
<strong>Tolérance aux pannes dans l'exécution distribuée de graphes de tâches</strong><br/>
In <em>Conférence d'informatique en Parallélisme, Architecture et Système</em>, Anglet, France, June 2019<br/>
[<a href="https://hal.inria.fr/hal-02296118">WWW</a>]
[<a href="https://hal.inria.fr/hal-02296118/file/Compas_Romain_LION_submitted_final.pdf">PDF</a>]
</li>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
824
<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
825
<strong>Achieving High Performance on Supercomputers with a Sequential Task-based Programming Model</strong><br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
826
<em>TPDS - IEEE Transactions on Parallel and Distributed Systems</em>, December 2017<br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
827 828
[<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
829
[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
830 831 832 833
</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
834
PhD thesis, Université de Bordeaux, December 2016<br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
835 836
[<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
837 838
</li>
<li>
839 840 841 842 843 844 845
<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>
846 847
<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
848
Research Report RR-8538, INRIA, May 2014<br/>
849 850 851 852
[<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>
853 854 855 856 857
<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/>
858
[<a href="http://hal.inria.fr/hal-00725477">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
859
[<a href="http://hal.inria.fr/hal-00725477/document">PDF</a>]
860
</li>
861
</ol>
Nathalie Furmento's avatar
Nathalie Furmento committed
862
<h4>On Memory Control</h4> 
863
<a name="PublicationsOnMemoryControl"></a>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
864 865
<ol>
<li>
866 867 868 869 870 871 872
<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
873
<a name="sergent:hal-01284004"></a>Marc Sergent, David Goudin, Samuel Thibault,  and Olivier Aumage<br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
874
<strong>Controlling the Memory Subscription of Distributed Applications with a Task-Based Runtime System</strong><br/>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
875
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
876 877
[<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
878
[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
879 880
</li>
</ol>
881
<h4>On Performance Model Tuning</h4> 
882
<a name="PublicationsOnPerformanceModelTuning"></a>
883 884
<ol>
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
885 886 887 888 889 890 891
<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>
892 893
<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
894
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/>
895
Springer<br/>
896
[<a href="http://hal.inria.fr/inria-00421333">WWW</a>]
THIBAULT Samuel's avatar
THIBAULT Samuel committed
897
[<a href="http://hal.inria.fr/inria-00421333/document">PDF</a>]
898
[doi:<a href="http://dx.doi.org/10.1007/978-3-642-14122-5_9">10.1007/978-3-642-14122-5_9</a>]
899 900
</li>
</ol>
901
<h4>On The Simulation Support through SimGrid</h4> 
902
<a name="PublicationsOnTheSimulationSupportthroughSimGrid"></a>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
903
<ol>
THIBAULT Samuel's avatar
update  
THIBAULT Samuel committed
904
<li>
905 906
<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
907
<em>CCPE - Concurrency and Computation: Practice and Experience</em>, pp 16, May 2015<br/>
908 909
[<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
910
[doi:<a href="http://dx.doi.org/10.1002/cpe.3555">10.1002/cpe.3555</a>]
THIBAULT Samuel's avatar
update  
THIBAULT Samuel committed
911
</li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
912
<li>
THIBAULT Samuel's avatar
THIBAULT Samuel committed
913 914 915 916 917 918 919
<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>
920 921
<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
922
In <em>Euro-Par - 20th International Conference on Parallel Processing</em>, Porto, Portugal, August 2014<br/>
923
Springer-Verlag<br/>