RcppParallel/simd.html at feature/expanded-site · RcppCore/RcppParallel

History

421 lines (362 loc) · 15.8 KB

Raw

100

101

102

103

104

105

106

107

108

109

110

111

112

113

114

115

116

117

118

119

120

121

122

123

124

125

126

127

128

129

130

131

132

133

134

135

136

137

138

139

140

141

142

143

144

145

146

147

148

149

150

151

152

153

154

155

156

157

158

159

160

161

162

163

164

165

166

167

168

169

170

171

172

173

174

175

176

177

178

179

180

181

182

183

184

185

186

187

188

189

190

191

192

193

194

195

196

197

198

199

200

201

202

203

204

205

206

207

208

209

210

211

212

213

214

215

216

217

218

219

220

221

222

223

224

225

226

227

228

229

230

231

232

233

234

235

236

237

238

239

240

241

242

243

244

245

246

247

248

249

250

251

252

253

254

255

256

257

258

259

260

261

262

263

264

265

266

267

268

269

270

271

272

273

274

275

276

277

278

279

280

281

282

283

284

285

286

287

288

289

290

291

292

293

294

295

296

297

298

299

300

301

302

303

304

305

306

307

308

309

310

311

312

313

314

315

316

317

318

319

320

321

322

323

324

325

326

327

328

329

330

331

332

333

334

335

336

337

338

339

340

341

342

343

344

345

346

347

348

349

350

351

352

353

354

355

356

357

358

359

360

361

362

363

364

365

366

367

368

369

370

371

372

373

374

375

376

377

378

379

380

381

382

383

384

385

386

387

388

389

390

391

392

393

394

395

396

397

398

399

400

401

402

403

404

405

406

407

408

409

410

411

412

413

414

415

416

417

418

419

420

421

<!DOCTYPE html>

<head>

<title>Parallel Programming with Boost.SIMD</title>

href="libs/highlight/textmate.css"

type="text/css" />

pre:not([class]) {

background-color: white;

}

</style>

if (window.hljs && document.readyState && document.readyState === "complete") {

window.setTimeout(function() {

hljs.initHighlighting();

}, 0);

}

</script>

</head>

<body>

.main-container {

max-width: 940px;

margin-left: auto;

margin-right: auto;

}

code {

color: inherit;

background-color: rgba(0, 0, 0, 0.04);

}

img {

max-width:100%;

height: auto;

}

h1 {

font-size: 34px;

}

h1.title {

font-size: 38px;

}

h2 {

font-size: 30px;

}

h3 {

font-size: 24px;

}

h4 {

font-size: 18px;

}

h5 {

font-size: 16px;

}

h6 {

font-size: 12px;

}

</style>

$(function() {

// establish options

var options = {

selectors: "h1,h2,h3",

theme: "bootstrap3",

context: '.toc-content',

hashGenerator: function (text) {

return text.replace(/[.\/?&!#<>]/g, '').replace(/\s/g, '_').toLowerCase();

ignoreSelector: "h1.title",

scrollTo: 0

};

options.showAndHide = false;

options.smoothScroll = true;

// tocify

var toc = $("#TOC").tocify(options).data("toc-tocify");

});

</script>

#TOC {

margin: 25px 0px 20px 0px;

}

@media (max-width: 768px) {

#TOC {

position: relative;

width: 100%;

}

.toc-content {

padding-left: 30px;

padding-right: 40px;

}

div.main-container {

max-width: 1200px;

}

div.tocify {

width: 20%;

max-width: 260px;

max-height: 85%;

}

@media (min-width: 768px) and (max-width: 991px) {

div.tocify {

width: 25%;

}

.tocify ul, .tocify li {

line-height: 20px;

}

.tocify-subheader .tocify-item {

font-size: 0.9em;

padding-left: 5px;

}

.tocify .list-group-item {

border-radius: 0px;

}

.tocify-subheader {

display: inline;

}

.tocify-subheader .tocify-item {

font-size: 0.95em;

padding-left: 10px;

}

</style>

</div>

</button>

<a class="navbar-brand" href="/">Rcpp Parallel</a>

</div>

<li><a href="/tbb.html">Intel TBB</a></li>

<li><a href="/simd.html">Boost.SIMD</a></li>

</ul>

<li><a href="https://github.com/RcppCore/RcppParallel">GitHub</a></li>

</ul>

</div>

</div>

</div>

// manage active state of menu based on current page

$(document).ready(function () {

// active menu

href = window.location.pathname

href = href.substr(href.lastIndexOf('/'))

$('a[href="' + href + '"]').parent().addClass('active');

});

</script>

<h1 class="title">Parallel Programming with Boost.SIMD</h1>

</div>

<p><strong>IMPORTANT NOTE</strong>: Support for Boost.SIMD is currently only available in the development version of RcppParallel. You can install the development version as follows:</p>

<pre class="r"><code>devtools::install_github("RcppCore/RcppParallel")</code></pre>

<h2>Introduction</h2>

<p>Modern CPU processors are built with new, extended instruction sets that optimize for certain operations. A class of these allow for vectorized operations, called Single Instruction / Multiple Data (SIMD) instructions. Although modern compilers will use these instructions when possible, they are often unable to reason about whether or not a particular block of code can be executed using SIMD instructions.</p>

<p><code>Boost.SIMD</code> [<a href="https://meetingcpp.com/tl_files/mcpp/slides/12/simd.pdf">PDF</a>] is a C++ header-only library that makes it possible to explicitly request the use of SIMD instructions when possible, while falling back to regular scalar operations when not. <a href="http://rcppcore.github.io/RcppParallel/"><code>RcppParallel</code></a> wraps and exposes this library for use with <code>R</code> vectors.</p>

<p>The primary abstraction that <code>Boost.SIMD</code> uses under the hood is the <code>boost::simd::pack<></code> data structure. This item represents a small, contiguous, pack of integral objects (e.g. <code>double</code>s), and comes with a host of functions that facilitate the use of SIMD operations on those objects when possible. Although you don’t need to know the details to use the high-level functionality provided by <code>Boost.SIMD</code>, it’s useful for understanding what happens behind the scenes.</p>

<p>Here’s a quick example of how we might compute the sum of elements in a vector, using <code>Boost.SIMD</code>.</p>

<pre class="cpp"><code>// [[Rcpp::depends(RcppParallel)]]

#define RCPP_PARALLEL_USE_SIMD

#include <RcppParallel.h>

using namespace RcppParallel;

#include <Rcpp.h>

using namespace Rcpp;

// Define a functor -- a C++ class which defines a templated

// 'function call' operator -- to perform the addition of

// two pieces of data.

struct add_two {

template <typename T>

T operator()(const T& lhs, const T& rhs) {

return lhs + rhs;

}

};

// [[Rcpp::export]]

double simd_sum(NumericVector x) {

// Pass the functor to 'simdReduce()'.

return simdReduce(x.begin(), x.end(), 0.0, add_two());

}</code></pre>

<p>Behind the scenes, <code>simdReduce()</code> takes care of iteration over our sequence, and ensures that we use optimized SIMD instructions over packs of numbers when possible, and scalar instructions when not. By passing a templated functor, <code>simdReduce()</code> can automagically choose the correct template specialization depending on whether it’s working with a pack or not. In other words, two template specializations will be generated in this case: one with <code>T = double</code>, and another with <code>T = boost::simd::pack<double></code>.</p>

<p>Let’s confirm that this produces the correct output, and run a small benchmark.</p>

<pre class="r"><code># helper function for printing microbenchmark output

printBm <- function(bm) {

summary <- summary(bm)

print(summary[, 1:7], row.names = FALSE)

}

# generate some data

data <- rnorm(1024 * 1000)

# verify that it produces the correct sum

all.equal(simd_sum(data), sum(data))</code></pre>

<pre class="r"><code># compare results

library(microbenchmark)

bm <- microbenchmark(sum(data), simd_sum(data))

printBm(bm)</code></pre>

<pre><code>## expr min lq mean median uq max

## sum(data) 824.013 836.9370 880.5446 870.0565 909.2475 1300.552

## simd_sum(data) 416.062 421.2825 456.2859 432.6560 481.2070 595.670</code></pre>

<p>We get a noticable gain by taking advantage of SIMD instructions here, although it’s worth noting that we don’t handle <code>NA</code> and <code>NaN</code> with the same granularity as <code>R</code>.</p>

</div>

<h2>SIMD Algorithms</h2>

<h3>Built-In Algorithms</h3>

<p><code>Boost.SIMD</code> provides two primary abstractions for the implementation of SIMD algorithms:</p>

<table>

<thead>

<th align="left">Algorithm</th>

<th align="left">Transformation</th>

</tr>

</thead>

<tbody>

<td align="left"><code>boost::simd::transform()</code></td>

<td align="left"><code>vector</code> -> <code>vector</code></td>

</tr>

<td align="left"><code>boost::simd::accumulate()</code></td>

<td align="left"><code>vector</code> -> <code>scalar</code></td>

</tr>

</tbody>

</table>

<p>These functions operate like their <code>std::</code> counterparts, but expect a functor with a templated call operator. By making the call operator templated, <code>Boost.SIMD</code> can generate code using its own optimized SIMD functions when appropriate, and fall back to a default implementation (based on the types provided) when not.</p>

<p><code>RcppParallel</code> augments this with its own algorithms as well, for consistency with <code>parallelFor()</code> and <code>parallelReduce()</code>:</p>

<table>

<thead>

<th align="left">Algorithm</th>

<th align="left">Transformation</th>

</tr>

</thead>

<tbody>

<td align="left"><code>RcppParallel::simdTransform()</code></td>

<td align="left"><code>vector</code> -> <code>vector</code></td>

</tr>

<td align="left"><code>RcppParallel::simdReduce()</code></td>

<td align="left"><code>vector</code> -> <code>scalar</code></td>

</tr>

<td align="left"><code>RcppParallel::simdFor()</code></td>

<td align="left"><code>vector</code> -> <code>any</code></td>

</tr>

</tbody>

</table>

<p><code>simdFor()</code> is useful in particular when neither <code>transform()</code> nor <code>accumulate()</code> seem to be a good fit.</p>

</div>

<h3>Custom Algorithms</h3>

<p>To take advantage of <code>Boost.SIMD</code>, you should try to perform the following steps:</p>

<li>Decompose your problem into separate, vectorizable pieces,</li>

<li>Select an appropriate algorithm provided by <code>RcppParallel</code>,</li>

<li>Write templated functors in a <code>Boost.SIMD</code>-aware way.</li>

</ol>

<p><code>Boost.SIMD</code> provides a large number of functions that have optimized specializations for packed data structures, while falling back to regular operations for scalar data structures. These functions can typically be accessed within the <code>boost::simd</code> namespace. To illustrate, here’s an example of a functor that computes the square for a set of data, using the <code>boost::simd::sqr()</code> function:</p>

<pre class="cpp"><code>class simd_square {

template <typename T>

void operator()(const T& data) {

return boost::simd::sqr(data);

}

};</code></pre>

<p>A reference guide for other functions provided is available <a href="http://nt2.metascale.fr/doc/html/boost_simd_functions_and_operators/reference.html">here</a>.</p>

</div>

<h2>Using SIMD in an R Package</h2>

<p><strong>IMPORTANT NOTE</strong>: Support for Boost.SIMD is currently only available in the development version of RcppParallel. Therefore, packages using Boost.SIMD should not yet be submitted to CRAN.</p>

<h3>Package Configuration</h3>

<p>To build an R package that uses <code>Boost.SIMD</code>, you need to make some modifications to the standard <code>RcppParallel</code> configuration. Within the <code>DESCRIPTION</code> file of your package, you need to:</p>

<li>Add the <a href="https://cran.r-project.org/package=BH"><strong>BH</strong></a> package as a <code>LinkingTo</code> dependency, and</li>

<li>Add <code>C++11</code> as a <code>SystemRequirement</code>.</li>

</ol>

<p>For example:</p>

<pre class="yaml"><code>Imports: RcppParallel

LinkingTo: RcppParallel, BH

SystemRequirements: GNU make, C++11</code></pre>

</div>

<h3>Platform Compatibility</h3>

<p><code>Boost.SIMD</code> requires a C++11 conformant compiler. This means that packages making use of SIMD features may not compile on platforms with older compilers, including Windows and RedHat/CentOS Linux. You can however create a package that takes advantage of <code>Boost.SIMD</code> where available and falls back to a non-SIMD implementation otherwise.</p>

<p>You can opt-in to the use of <code>Boost.SIMD</code> by defining the <code>RCPP_PARALLEL_USE_SIMD</code> macro before including <code><RcppParallel.h></code>, e.g.</p>

<pre><code>#define RCPP_PARALLEL_USE_SIMD

#include <RcppParallel.h></code></pre>

<p>You can test for the availability of <code>Boost.SIMD</code> on a given platform using the <code>RCPP_PARALLEL_USE_SIMD</code> preprocessor variable. If the current compiler doesn’t support C++11 (as determined by <code>__cplusplus <= 199711L</code>) the variable will be undefined (even if you defined it explicitly). This allows you to write code like this:</p>

<pre class="cpp"><code>#define RCPP_PARALLEL_USE_SIMD

#include <RcppParallel.h>

#if RCPP_PARALLEL_USE_SIMD

IntegerVector transformDataImpl(IntegerVector x) {

// Implement with Boost.SIMD

}

#else

IntegerVector transformDataImpl(IntegerVector x) {

// Implement without Boost.SIMD

}

#endif

// [[Rcpp::export]]

IntegerVector transformData(IntegerVector x) {

return transformDataImpl(x);

}</code></pre>

<p>The two <code>transformDataImpl</code> functions have the same name, but only one will be compiled and linked based on whether the target platform supports <code>Boost.SIMD</code>.</p>

<p>Note that if you conditionally compile all uses of <code>Boost.SIMD</code> within your package, then you can drop the <code>C++11</code> from <code>SystemRequirements</code> (it’s no longer required as a result of your fallback implementation).</p>

</div>

<h2>Learning More</h2>

<p>If you want to dive deeper into <code>Boost.SIMD</code>, you can <a href="http://nt2.metascale.fr/doc/html/boost_simd.html">read the online documentation</a>, and also browse the examples <a href="https://github.com/RcppCore/RcppParallel/tree/master/inst/examples/boost-simd">here</a>.</p>

<hr />

<p>If you want to try out <code>Boost.SIMD</code> yourself, please install the development version of <a href="http://rcppcore.github.io/RcppParallel/"><code>RcppParallel</code></a> with <code>devtools::install_github("RcppCore/RcppParallel")</code>.</p>

</div>

// add bootstrap table styles to pandoc tables

$(document).ready(function () {

$('tr.header').parent('thead').parent('table').addClass('table table-condensed');

});

</script>

(function () {

var script = document.createElement("script");

script.type = "text/javascript";

script.src = "https://cdn.mathjax.org/mathjax/latest/MathJax.js?config=TeX-AMS-MML_HTMLorMML";

document.getElementsByTagName("head")[0].appendChild(script);

})();

</script>

</body>

</html>

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

FilesExpand file tree

simd.html

Latest commit

History

simd.html

File metadata and controls