gaspersic/triton
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
Files
5cefc81fceae86493d73b2d71b40055ef0e71207
triton/programming-guide/introduction.html

307 lines
17 KiB
HTML
Raw Normal View History

[GH-PAGES] Updated website
2021-03-29 11:59:18 -04:00
<!DOCTYPE html>
<html class="writer-html5" lang="en" >
<head>
<meta charset="utf-8" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
<title>Introduction &mdash; Triton documentation</title>
<link rel="stylesheet" href="../_static/css/theme.css" type="text/css" />
<link rel="stylesheet" href="../_static/pygments.css" type="text/css" />
<link rel="stylesheet" href="../_static/gallery.css" type="text/css" />
<link rel="stylesheet" href="../_static/gallery-binder.css" type="text/css" />
<link rel="stylesheet" href="../_static/gallery-dataframe.css" type="text/css" />
<link rel="stylesheet" href="../_static/gallery-rendered-html.css" type="text/css" />
<link rel="stylesheet" href="../_static/css/custom.css" type="text/css" />
<!--[if lt IE 9]>
<script src="../_static/js/html5shiv.min.js"></script>
<![endif]-->
<script type="text/javascript" id="documentation_options" data-url_root="../" src="../_static/documentation_options.js"></script>
<script src="../_static/jquery.js"></script>
<script src="../_static/underscore.js"></script>
<script src="../_static/doctools.js"></script>
<script type="text/javascript" src="../_static/js/theme.js"></script>
<link rel="index" title="Index" href="../genindex.html" />
<link rel="search" title="Search" href="../search.html" />
</head>
<body class="wy-body-for-nav">
<div class="wy-grid-for-nav">
<nav data-toggle="wy-nav-shift" class="wy-nav-side">
<div class="wy-side-scroll">
<div class="wy-side-nav-search" >
<a href="../index.html" class="icon icon-home"> Triton
</a>
<div role="search">
<form id="rtd-search-form" class="wy-form" action="../search.html" method="get">
<input type="text" name="q" placeholder="Search docs" />
<input type="hidden" name="check_keywords" value="yes" />
<input type="hidden" name="area" value="default" />
</form>
</div>
</div>
<div class="wy-menu wy-menu-vertical" data-spy="affix" role="navigation" aria-label="main navigation">
<p class="caption"><span class="caption-text">Getting Started</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../getting-started/installation.html">Installation</a></li>
<li class="toctree-l1"><a class="reference internal" href="../getting-started/tutorials/index.html">Tutorials</a></li>
</ul>
<p class="caption"><span class="caption-text">Programming Guide</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="chapter-1/introduction.html">Introduction</a></li>
<li class="toctree-l1"><a class="reference internal" href="chapter-2/related-work.html">Related Work</a></li>
<li class="toctree-l1"><a class="reference internal" href="chapter-3/triton-c.html">The Triton-C Language</a></li>
<li class="toctree-l1"><a class="reference internal" href="chapter-4/triton-ir.html">The Triton-IR Intermediate Representation</a></li>
</ul>
</div>
</div>
</nav>
<section data-toggle="wy-nav-shift" class="wy-nav-content-wrap">
<nav class="wy-nav-top" aria-label="top navigation">
<i data-toggle="wy-nav-top" class="fa fa-bars"></i>
<a href="../index.html">Triton</a>
</nav>
<div class="wy-nav-content">
<div class="rst-content">
<div role="navigation" aria-label="breadcrumbs navigation">
<ul class="wy-breadcrumbs">
<li><a href="../index.html" class="icon icon-home"></a> &raquo;</li>
<li>Introduction</li>
<li class="wy-breadcrumbs-aside">
<a href="../_sources/programming-guide/introduction.rst.txt" rel="nofollow"> View page source</a>
</li>
</ul>
<hr/>
</div>
<div role="main" class="document" itemscope="itemscope" itemtype="http://schema.org/Article">
<div itemprop="articleBody">
<div class="section" id="introduction">
<h1>Introduction<a class="headerlink" href="#introduction" title="Permalink to this headline"></a></h1>
<div class="section" id="motivations">
<h2>Motivations<a class="headerlink" href="#motivations" title="Permalink to this headline"></a></h2>
<p>Over the past decade, Deep Neural Networks (DNNs) have emerged as an important class of Machine Learning (ML) models, capable of achieving state-of-the-art performance across many domains ranging from natural language processing <a class="footnote-reference brackets" href="#id10" id="id1">1</a> to computer vision <a class="footnote-reference brackets" href="#id11" id="id2">2</a> to computational neuroscience <a class="footnote-reference brackets" href="#id12" id="id3">3</a>. The strength of these models lies in their hierarchical structure, composed of a sequence of parametric (e.g., convolutional) and non-parametric (e.g., rectified linearity) <em>layers</em>. This pattern, though notoriously computationally expensive, also generates a large amount of highly parallelizable work particularly well suited for multi- and many- core processors.</p>
<p>As a consequence, Graphics Processing Units (GPUs) have become a cheap and accessible resource for exploring and/or deploying novel research ideas in the field. This trend has been accelerated by the release of several frameworks for General-Purpose GPU (GPGPU) computing, such as CUDA and OpenCL, which have made the development of high-performance programs easier. Yet, GPUs remain incredibly challenging to optimize for locality and parallelism, especially for computations that cannot be efficiently implemented using a combination of pre-existing optimized primitives. To make matters worse, GPU architectures are also rapidly evolving and specializing, as evidenced by the addition of tensor cores to NVIDIA (and more recently AMD) micro-architectures.</p>
<p>This tension between the computational opportunities offered by DNNs and the practical difficulty of GPU programming has created substantial academic and industrial interest for Domain-Specific Languages (DSLs) and compilers. Regrettably, these systems – whether they be based on polyhedral machinery (<em>e.g.</em>, Tiramisu <a class="footnote-reference brackets" href="#id13" id="id4">4</a>, Tensor Comprehensions <a class="footnote-reference brackets" href="#id14" id="id5">5</a>) or scheduling languages (<em>e.g.</em>, Halide <a class="footnote-reference brackets" href="#id15" id="id6">6</a>, TVM <a class="footnote-reference brackets" href="#id16" id="id7">7</a>) – remain less flexible and (for the same algorithm) markedly slower than the best handwritten compute kernels available in libraries like <a class="reference external" href="https://docs.nvidia.com/cuda/cublas/index.html">cuBLAS</a>, <a class="reference external" href="https://docs.nvidia.com/deeplearning/cudnn/api/index.html">cuDNN</a> or <a class="reference external" href="https://docs.nvidia.com/deeplearning/tensorrt/developer-guide/index.html">TensorRT</a>.</p>
<p>The main premise of this project is the following: programming paradigms based on blocked algorithms <a class="footnote-reference brackets" href="#id17" id="id8">8</a> can facilitate the construction of high-performance compute kernels for neural networks. We specifically revisit traditional “Single Program, Multiple Data” (SPMD <a class="footnote-reference brackets" href="#id18" id="id9">9</a>) execution models for GPUs, and propose a variant in which programs – rather than threads – are blocked. For example, in the case of matrix multiplication, CUDA and Triton differ as follows:</p>
<table class="colwidths-given docutils align-default">
<colgroup>
<col style="width: 50%" />
<col style="width: 50%" />
</colgroup>
<thead>
<tr class="row-odd"><th class="head"><p>CUDA Programming Model</p>
<p>(Scalar Program, Blocked Threads)</p>
</th>
<th class="head"><p>Triton Programming Model</p>
<p>(Blocked Program, Scalar Threads)</p>
</th>
</tr>
</thead>
<tbody>
<tr class="row-even"><td><div class="highlight-C notranslate"><div class="highlight"><pre><span></span><span class="cp">#pragma parallel</span>
<span class="k">for</span><span class="p">(</span><span class="kt">int</span> <span class="n">m</span> <span class="o">=</span> <span class="mi">0</span><span class="p">;</span> <span class="n">i</span> <span class="o">&lt;</span> <span class="n">M</span><span class="p">;</span> <span class="n">m</span><span class="o">++</span><span class="p">)</span>
<span class="cp">#pragma parallel</span>
<span class="k">for</span><span class="p">(</span><span class="kt">int</span> <span class="n">n</span> <span class="o">=</span> <span class="mi">0</span><span class="p">;</span> <span class="n">j</span> <span class="o">&lt;</span> <span class="n">N</span><span class="p">;</span> <span class="n">n</span><span class="o">++</span><span class="p">){</span>
<span class="kt">float</span> <span class="n">acc</span> <span class="o">=</span> <span class="mi">0</span><span class="p">;</span>
<span class="k">for</span><span class="p">(</span><span class="kt">int</span> <span class="n">k</span> <span class="o">=</span> <span class="mi">0</span><span class="p">;</span> <span class="n">k</span> <span class="o">&lt;</span> <span class="n">K</span><span class="p">;</span><span class="n">k</span> <span class="o">++</span><span class="p">)</span>
<span class="n">acc</span> <span class="o">+=</span> <span class="n">A</span><span class="p">[</span><span class="n">i</span><span class="p">,</span> <span class="n">k</span><span class="p">]</span><span class="o">*</span> <span class="n">B</span><span class="p">[</span><span class="n">k</span><span class="p">,</span> <span class="n">j</span><span class="p">];</span>
<span class="n">C</span><span class="p">[</span><span class="n">i</span><span class="p">,</span> <span class="n">j</span><span class="p">]</span> <span class="o">=</span> <span class="n">acc</span><span class="p">;</span>
<span class="p">}</span>
</pre></div>
</div>
</td>
<td><div class="highlight-C notranslate"><div class="highlight"><pre><span></span><span class="cp">#pragma parallel</span>
<span class="k">for</span><span class="p">(</span><span class="kt">int</span> <span class="n">m</span> <span class="o">=</span> <span class="mi">0</span><span class="p">;</span> <span class="n">m</span> <span class="o">&lt;</span> <span class="n">M</span><span class="p">;</span> <span class="n">m</span> <span class="o">+=</span> <span class="n">MB</span><span class="p">)</span>
<span class="cp">#pragma parallel</span>
<span class="k">for</span><span class="p">(</span><span class="kt">int</span> <span class="n">n</span> <span class="o">=</span> <span class="mi">0</span><span class="p">;</span> <span class="n">n</span> <span class="o">&lt;</span> <span class="n">N</span><span class="p">;</span> <span class="n">n</span> <span class="o">+=</span> <span class="n">NB</span><span class="p">){</span>
<span class="kt">float</span> <span class="n">acc</span><span class="p">[</span><span class="n">MB</span><span class="p">,</span> <span class="n">NB</span><span class="p">]</span> <span class="o">=</span> <span class="mi">0</span><span class="p">;</span>
<span class="k">for</span><span class="p">(</span><span class="kt">int</span> <span class="n">k</span> <span class="o">=</span> <span class="mi">0</span><span class="p">;</span> <span class="n">k</span> <span class="o">&lt;</span> <span class="n">K</span><span class="p">;</span> <span class="n">k</span> <span class="o">+=</span> <span class="n">KB</span><span class="p">)</span>
<span class="n">acc</span> <span class="o">+=</span> <span class="n">A</span><span class="p">[</span><span class="nl">m</span><span class="p">:</span><span class="n">m</span><span class="o">+</span><span class="n">MB</span><span class="p">,</span> <span class="nl">k</span><span class="p">:</span><span class="n">k</span><span class="o">+</span><span class="n">KB</span><span class="p">]</span>
<span class="err">@</span> <span class="n">B</span><span class="p">[</span><span class="nl">k</span><span class="p">:</span><span class="n">k</span><span class="o">+</span><span class="n">KB</span><span class="p">,</span> <span class="nl">n</span><span class="p">:</span><span class="n">n</span><span class="o">+</span><span class="n">NB</span><span class="p">];</span>
<span class="n">C</span><span class="p">[</span><span class="nl">m</span><span class="p">:</span><span class="n">m</span><span class="o">+</span><span class="n">MB</span><span class="p">,</span> <span class="nl">n</span><span class="p">:</span><span class="n">n</span><span class="o">+</span><span class="n">NB</span><span class="p">]</span> <span class="o">=</span> <span class="n">acc</span><span class="p">;</span>
<span class="p">}</span>
</pre></div>
</div>
</td>
</tr>
<tr class="row-odd"><td><p><img alt="pic1" src="../_images/cuda-parallel-matmul1.png" /></p></td>
<td><p><img alt="pic2" src="../_images/triton-parallel-matmul1.png" /></p></td>
</tr>
</tbody>
</table>
<p>A key benefit of this approach is that it leads to block-structured iteration spaces that offer programmers more flexibility than existing DSLs when implementing sparse operations, all while allowing compilers to aggressively optimize programs for data locality and parallelism.</p>
</div>
<div class="section" id="challenges">
<h2>Challenges<a class="headerlink" href="#challenges" title="Permalink to this headline"></a></h2>
<p>The main challenge posed by our proposed paradigm is that of work scheduling, i.e., how the work done by each program instance should be partitioned for efficient execution on modern GPUs. To address this issue, the Triton compiler makes heavy use of <em>block-level data-flow analysis</em>, a technique for scheduling iteration blocks statically based on the control- and data-flow structure of the target program. The resulting system actually works surprisingly well: our compiler manages to apply a broad range of interesting optimization automatically (e.g., automatic coalescing, thread swizzling, pre-fetching, automatic vectorization, tensor core-aware instruction selection, shared memory allocation/synchronization, asynchronous copy scheduling). Of course doing all this is not trivial; one of the purposes of this guide is to give you a sense of how it works.</p>
</div>
<div class="section" id="references">
<h2>References<a class="headerlink" href="#references" title="Permalink to this headline"></a></h2>
<dl class="footnote brackets">
<dt class="label" id="id10"><span class="brackets"><a class="fn-backref" href="#id1">1</a></span></dt>
<dd><p>Sutskever et al., “Sequence to Sequence Learning with Neural Networks”, NIPS 2014</p>
</dd>
<dt class="label" id="id11"><span class="brackets"><a class="fn-backref" href="#id2">2</a></span></dt>
<dd><p>Redmon et al., “You Only Look Once: Unified, Real-Time Object Detection”, CVPR 2016</p>
</dd>
<dt class="label" id="id12"><span class="brackets"><a class="fn-backref" href="#id3">3</a></span></dt>
<dd><p>Lee et al., “Superhuman Accuracy on the SNEMI3D Connectomics Challenge”, ArXiV 2017</p>
</dd>
<dt class="label" id="id13"><span class="brackets"><a class="fn-backref" href="#id4">4</a></span></dt>
<dd><p>Baghdadi et al., “Tiramisu: A Polyhedral Compiler for Expressing Fast and Portable Code”, CGO 2021</p>
</dd>
<dt class="label" id="id14"><span class="brackets"><a class="fn-backref" href="#id5">5</a></span></dt>
<dd><p>Vasilache et al., “Tensor Comprehensions: Framework-Agnostic High-Performance Machine Learning Abstractions”, ArXiV 2018</p>
</dd>
<dt class="label" id="id15"><span class="brackets"><a class="fn-backref" href="#id6">6</a></span></dt>
<dd><p>Ragan-Kelley et al., “Halide: A Language and Compiler for Optimizing Parallelism, Locality, and Recomputation in Image Processing Pipelines”, PLDI 2013</p>
</dd>
<dt class="label" id="id16"><span class="brackets"><a class="fn-backref" href="#id7">7</a></span></dt>
<dd><p>Chen et al., “TVM: An Automated End-to-End Optimizing Compiler for Deep Learning”, OSDI 2018</p>
</dd>
<dt class="label" id="id17"><span class="brackets"><a class="fn-backref" href="#id8">8</a></span></dt>
<dd><p>Lam et al., “The Cache Performance and Optimizations of Blocked Algorithms”, ASPLOS 1991</p>
</dd>
<dt class="label" id="id18"><span class="brackets"><a class="fn-backref" href="#id9">9</a></span></dt>
<dd><p>Auguin et al., “Opsila: an advanced SIMD for numerical analysis and signal processing”, EUROMICRO 1983</p>
</dd>
</dl>
</div>
</div>
</div>
</div>
<footer>
<hr/>
<div role="contentinfo">
<p>
&#169; Copyright 2020, Philippe Tillet.
</p>
</div>
Built with <a href="https://www.sphinx-doc.org/">Sphinx</a> using a
<a href="https://github.com/readthedocs/sphinx_rtd_theme">theme</a>
provided by <a href="https://readthedocs.org">Read the Docs</a>.
</footer>
</div>
</div>
</section>
</div>
<script type="text/javascript">
jQuery(function () {
SphinxRtdTheme.Navigation.enable(true);
});
</script>
</body>
</html>
Reference in New Issue Copy Permalink