FT_DSP.gitlink.net/master/html/functionlib/dsplib/subgrad.html

217 lines
24 KiB
HTML
Raw Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

<!DOCTYPE html>
<html class="writer-html5" lang="zh-CN" data-content_root="../../">
<head>
<meta charset="utf-8" /><meta name="viewport" content="width=device-width, initial-scale=1" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
<title>Subgrad &mdash; MindSpore Signal+ 使用手册 alpha 文档</title>
<link rel="stylesheet" type="text/css" href="../../_static/pygments.css?v=03e43079" />
<link rel="stylesheet" type="text/css" href="../../_static/css/theme.css?v=e59714d7" />
<link rel="stylesheet" type="text/css" href="../../_static/table.css?v=72b86d3e" />
<script src="../../_static/jquery.js?v=5d32c60e"></script>
<script src="../../_static/_sphinx_javascript_frameworks_compat.js?v=2cd50e6c"></script>
<script src="../../_static/documentation_options.js?v=406e4f49"></script>
<script src="../../_static/doctools.js?v=9bcbadda"></script>
<script src="../../_static/sphinx_highlight.js?v=dc90522c"></script>
<script src="../../_static/translations.js?v=beaddf03"></script>
<script async="async" src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-mml-chtml.js"></script>
<script src="../../_static/table.js?v=99ed15ba"></script>
<script src="../../_static/js/theme.js"></script>
<link rel="index" title="索引" href="../../genindex.html" />
<link rel="search" title="搜索" href="../../search.html" />
<link rel="next" title="Switch" href="switch.html" />
<link rel="prev" title="Subfusion" href="subfusion.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">
MindSpore Signal+ 使用手册
</a>
<div role="search">
<form id="rtd-search-form" class="wy-form" action="../../search.html" method="get">
<input type="text" name="q" placeholder="搜索文档" aria-label="搜索文档" />
<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="导航菜单">
<ul>
<li class="toctree-l1"><a class="reference internal" href="../../quickstart/index.html">快速入门</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../appdevelop/index.html">应用开发示例</a></li>
<li class="toctree-l1"><a class="reference internal" href="../index.html">算子库支持</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../refdoc/index.html">参考资料</a></li>
</ul>
</div>
</div>
</nav>
<section data-toggle="wy-nav-shift" class="wy-nav-content-wrap"><nav class="wy-nav-top" aria-label="移动版导航菜单" >
<i data-toggle="wy-nav-top" class="fa fa-bars"></i>
<a href="../../index.html">MindSpore Signal+ 使用手册</a>
</nav>
<div class="wy-nav-content">
<div class="rst-content">
<div role="navigation" aria-label="页面导航">
<ul class="wy-breadcrumbs">
<li><a href="../../index.html" class="icon icon-home" aria-label="Home"></a></li>
<li class="breadcrumb-item"><a href="../index.html">算子库支持</a></li>
<li class="breadcrumb-item"><a href="dsplib_index.html">DSP Library C API Reference</a></li>
<li class="breadcrumb-item active">Subgrad</li>
<li class="wy-breadcrumbs-aside">
<a href="../../_sources/functionlib/dsplib/subgrad.rst.txt" rel="nofollow"> 查看页面源码</a>
</li>
</ul>
<hr/>
</div>
<div role="main" class="document" itemscope="itemscope" itemtype="http://schema.org/Article">
<div itemprop="articleBody">
<section id="subgrad">
<h1>Subgrad<a class="headerlink" href="#subgrad" title="Link to this heading"></a></h1>
<p>计算逐元素减法Sub操作的梯度。该算子是 Sub 算子的反向传播backward pass部分支持广播。</p>
<div class="math notranslate nohighlight">
\[\text{dx1} = \frac{\partial L}{\partial X1} = \frac{\partial L}{\partial Y} \times 1 = \frac{\partial L}{\partial Y}\]</div>
<div class="math notranslate nohighlight">
\[\text{dx2} = \frac{\partial L}{\partial X2} = \frac{\partial L}{\partial Y} \times (-1) = -\frac{\partial L}{\partial Y}\]</div>
<p>其中对于前向操作 <span class="math notranslate nohighlight">\(Y = X1 - X2\)</span><cite>dy</cite> 是来自后一层的上游梯度,<cite>dx1</cite><cite>dx2</cite> 分别是对 <cite>X1</cite><cite>X2</cite> 的梯度。</p>
<dl class="simple">
<dt>输入:</dt><dd><ul class="simple">
<li><p><strong>dy</strong> - 上游梯度数据地址(即 <span class="math notranslate nohighlight">\(\frac{\partial L}{\partial Y}\)</span>)。</p></li>
<li><p><strong>x1_dims</strong> - 前向传播时第一个输入 <cite>x1</cite> 的维度信息数组int*)。</p></li>
<li><p><strong>x2_dims</strong> - 前向传播时第二个输入 <cite>x2</cite> 的维度信息数组int*)。</p></li>
<li><p><strong>dy_dims</strong> - 上游梯度 <cite>dy</cite> 的维度信息数组int*)。</p></li>
<li><p><strong>num_dims</strong> - 维度数int</p></li>
<li><p><strong>core_mask</strong> - 核掩码int仅共享存储版本需要。</p></li>
</ul>
</dd>
<dt>输出:</dt><dd><ul class="simple">
<li><p><strong>dx1</strong> - 对 <cite>x1</cite> 的梯度数据地址。</p></li>
<li><p><strong>dx2</strong> - 对 <cite>x2</cite> 的梯度数据地址。</p></li>
</ul>
</dd>
<dt>支持平台:</dt><dd><p><code class="docutils literal notranslate"><span class="pre">FT78NE</span></code>
<code class="docutils literal notranslate"><span class="pre">MT7004</span></code></p>
</dd>
</dl>
<div class="admonition note">
<p class="admonition-title">备注</p>
<ul class="simple">
<li><p>MT7004 支持fp16, fp32</p></li>
<li><p>FT78NE 支持fp32</p></li>
<li><p>当输入张量被广播时,算子会自动处理广播维度的梯度累加</p></li>
</ul>
</div>
<p><strong>共享存储版本:</strong></p>
<dl class="c function">
<dt class="sig sig-object c" id="c.hp_subgrad_s">
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">hp_subgrad_s</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">dy</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">x1_dims</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">x2_dims</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">dy_dims</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">num_dims</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">dx1</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">dx2</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.hp_subgrad_s" title="Link to this definition"></a><br /></dt>
<dd></dd></dl>
<dl class="c function">
<dt class="sig sig-object c" id="c.fp_subgrad_s">
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">fp_subgrad_s</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">dy</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">x1_dims</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">x2_dims</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">dy_dims</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">num_dims</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">dx1</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">dx2</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">core_mask</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.fp_subgrad_s" title="Link to this definition"></a><br /></dt>
<dd></dd></dl>
<p><strong>C调用示例</strong></p>
<div class="highlight-c notranslate"><div class="highlight"><pre><span></span><span class="linenos"> 1</span><span class="c1">//MT7004示例</span>
<span class="linenos"> 2</span><span class="cp">#include</span><span class="w"> </span><span class="cpf">&lt;stdio.h&gt;</span>
<span class="linenos"> 3</span><span class="cp">#include</span><span class="w"> </span><span class="cpf">&lt;subgrad.h&gt;</span>
<span class="linenos"> 4</span>
<span class="linenos"> 5</span><span class="kt">int</span><span class="w"> </span><span class="nf">main</span><span class="p">(</span><span class="kt">int</span><span class="w"> </span><span class="n">argc</span><span class="p">,</span><span class="w"> </span><span class="kt">char</span><span class="o">*</span><span class="w"> </span><span class="n">argv</span><span class="p">[])</span><span class="w"> </span><span class="p">{</span>
<span class="linenos"> 6</span><span class="w"> </span><span class="kt">float</span><span class="w"> </span><span class="o">*</span><span class="n">dy</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">(</span><span class="kt">float</span><span class="w"> </span><span class="o">*</span><span class="p">)</span><span class="mh">0xA0000000</span><span class="p">;</span><span class="w"> </span><span class="c1">// 上游梯度在DDR空间</span>
<span class="linenos"> 7</span><span class="w"> </span><span class="kt">float</span><span class="w"> </span><span class="o">*</span><span class="n">dx1</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">(</span><span class="kt">float</span><span class="w"> </span><span class="o">*</span><span class="p">)</span><span class="mh">0xB0000000</span><span class="p">;</span><span class="w"> </span><span class="c1">// dx1输出</span>
<span class="linenos"> 8</span><span class="w"> </span><span class="kt">float</span><span class="w"> </span><span class="o">*</span><span class="n">dx2</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">(</span><span class="kt">float</span><span class="w"> </span><span class="o">*</span><span class="p">)</span><span class="mh">0xC0000000</span><span class="p">;</span><span class="w"> </span><span class="c1">// dx2输出</span>
<span class="linenos"> 9</span>
<span class="linenos">10</span><span class="w"> </span><span class="c1">// 示例x1形状 [3, 1]x2形状 [1, 4]dy形状 [3, 4]</span>
<span class="linenos">11</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">x1_dims</span><span class="p">[]</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">{</span><span class="mi">3</span><span class="p">,</span><span class="w"> </span><span class="mi">1</span><span class="p">};</span>
<span class="linenos">12</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">x2_dims</span><span class="p">[]</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">{</span><span class="mi">1</span><span class="p">,</span><span class="w"> </span><span class="mi">4</span><span class="p">};</span>
<span class="linenos">13</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">dy_dims</span><span class="p">[]</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">{</span><span class="mi">3</span><span class="p">,</span><span class="w"> </span><span class="mi">4</span><span class="p">};</span>
<span class="linenos">14</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">num_dims</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="mi">2</span><span class="p">;</span>
<span class="linenos">15</span>
<span class="linenos">16</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">core_mask</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="mh">0xff</span><span class="p">;</span>
<span class="linenos">17</span>
<span class="hll"><span class="linenos">18</span><span class="w"> </span><span class="n">fp_subgrad_s</span><span class="p">(</span><span class="n">dy</span><span class="p">,</span><span class="w"> </span><span class="n">x1_dims</span><span class="p">,</span><span class="w"> </span><span class="n">x2_dims</span><span class="p">,</span><span class="w"> </span><span class="n">dy_dims</span><span class="p">,</span><span class="w"> </span><span class="n">num_dims</span><span class="p">,</span><span class="w"> </span><span class="n">dx1</span><span class="p">,</span><span class="w"> </span><span class="n">dx2</span><span class="p">,</span><span class="w"> </span><span class="n">core_mask</span><span class="p">);</span>
</span><span class="linenos">19</span>
<span class="linenos">20</span><span class="w"> </span><span class="k">return</span><span class="w"> </span><span class="mi">0</span><span class="p">;</span>
<span class="linenos">21</span><span class="p">}</span>
</pre></div>
</div>
<p><strong>私有存储版本:</strong></p>
<dl class="c function">
<dt class="sig sig-object c" id="c.hp_subgrad_p">
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">hp_subgrad_p</span></span></span><span class="sig-paren">(</span><span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">dy</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">x1_dims</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">x2_dims</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">dy_dims</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">num_dims</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">dx1</span></span>, <span class="n"><span class="pre">half</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">dx2</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.hp_subgrad_p" title="Link to this definition"></a><br /></dt>
<dd></dd></dl>
<dl class="c function">
<dt class="sig sig-object c" id="c.fp_subgrad_p">
<span class="kt"><span class="pre">void</span></span><span class="w"> </span><span class="sig-name descname"><span class="n"><span class="pre">fp_subgrad_p</span></span></span><span class="sig-paren">(</span><span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">dy</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">x1_dims</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">x2_dims</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">dy_dims</span></span>, <span class="kt"><span class="pre">int</span></span><span class="w"> </span><span class="n"><span class="pre">num_dims</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">dx1</span></span>, <span class="kt"><span class="pre">float</span></span><span class="w"> </span><span class="p"><span class="pre">*</span></span><span class="n"><span class="pre">dx2</span></span><span class="sig-paren">)</span><a class="headerlink" href="#c.fp_subgrad_p" title="Link to this definition"></a><br /></dt>
<dd></dd></dl>
<p><strong>C调用示例</strong></p>
<div class="highlight-c notranslate"><div class="highlight"><pre><span></span><span class="linenos"> 1</span><span class="c1">//MT7004示例</span>
<span class="linenos"> 2</span><span class="cp">#include</span><span class="w"> </span><span class="cpf">&lt;stdio.h&gt;</span>
<span class="linenos"> 3</span><span class="cp">#include</span><span class="w"> </span><span class="cpf">&lt;subgrad.h&gt;</span>
<span class="linenos"> 4</span>
<span class="linenos"> 5</span><span class="kt">int</span><span class="w"> </span><span class="nf">main</span><span class="p">(</span><span class="kt">int</span><span class="w"> </span><span class="n">argc</span><span class="p">,</span><span class="w"> </span><span class="kt">char</span><span class="o">*</span><span class="w"> </span><span class="n">argv</span><span class="p">[])</span><span class="w"> </span><span class="p">{</span>
<span class="linenos"> 6</span><span class="w"> </span><span class="kt">float</span><span class="w"> </span><span class="o">*</span><span class="n">dy</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">(</span><span class="kt">float</span><span class="w"> </span><span class="o">*</span><span class="p">)</span><span class="mh">0x10000000</span><span class="p">;</span><span class="w"> </span><span class="c1">// 上游梯度在L2空间</span>
<span class="linenos"> 7</span><span class="w"> </span><span class="kt">float</span><span class="w"> </span><span class="o">*</span><span class="n">dx1</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">(</span><span class="kt">float</span><span class="w"> </span><span class="o">*</span><span class="p">)</span><span class="mh">0x10001000</span><span class="p">;</span><span class="w"> </span><span class="c1">// dx1输出</span>
<span class="linenos"> 8</span><span class="w"> </span><span class="kt">float</span><span class="w"> </span><span class="o">*</span><span class="n">dx2</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">(</span><span class="kt">float</span><span class="w"> </span><span class="o">*</span><span class="p">)</span><span class="mh">0x10002000</span><span class="p">;</span><span class="w"> </span><span class="c1">// dx2输出</span>
<span class="linenos"> 9</span>
<span class="linenos">10</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">x1_dims</span><span class="p">[]</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">{</span><span class="mi">3</span><span class="p">,</span><span class="w"> </span><span class="mi">1</span><span class="p">};</span>
<span class="linenos">11</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">x2_dims</span><span class="p">[]</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">{</span><span class="mi">1</span><span class="p">,</span><span class="w"> </span><span class="mi">4</span><span class="p">};</span>
<span class="linenos">12</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">dy_dims</span><span class="p">[]</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">{</span><span class="mi">3</span><span class="p">,</span><span class="w"> </span><span class="mi">4</span><span class="p">};</span>
<span class="linenos">13</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">num_dims</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="mi">2</span><span class="p">;</span>
<span class="linenos">14</span>
<span class="hll"><span class="linenos">15</span><span class="w"> </span><span class="n">fp_subgrad_p</span><span class="p">(</span><span class="n">dy</span><span class="p">,</span><span class="w"> </span><span class="n">x1_dims</span><span class="p">,</span><span class="w"> </span><span class="n">x2_dims</span><span class="p">,</span><span class="w"> </span><span class="n">dy_dims</span><span class="p">,</span><span class="w"> </span><span class="n">num_dims</span><span class="p">,</span><span class="w"> </span><span class="n">dx1</span><span class="p">,</span><span class="w"> </span><span class="n">dx2</span><span class="p">);</span>
</span><span class="linenos">16</span>
<span class="linenos">17</span><span class="w"> </span><span class="k">return</span><span class="w"> </span><span class="mi">0</span><span class="p">;</span>
<span class="linenos">18</span><span class="p">}</span>
</pre></div>
</div>
</section>
</div>
</div>
<footer><div class="rst-footer-buttons" role="navigation" aria-label="页脚">
<a href="subfusion.html" class="btn btn-neutral float-left" title="Subfusion" accesskey="p" rel="prev"><span class="fa fa-arrow-circle-left" aria-hidden="true"></span> 上一页</a>
<a href="switch.html" class="btn btn-neutral float-right" title="Switch" accesskey="n" rel="next">下一页 <span class="fa fa-arrow-circle-right" aria-hidden="true"></span></a>
</div>
<hr/>
<div role="contentinfo">
<p>&#169; 版权所有 2025 - 2026, NUDT-674。</p>
</div>
利用 <a href="https://www.sphinx-doc.org/">Sphinx</a> 构建,使用的
<a href="https://github.com/readthedocs/sphinx_rtd_theme">主题</a>
<a href="https://readthedocs.org">Read the Docs</a> 开发.
</footer>
</div>
</div>
</section>
</div>
<script>
jQuery(function () {
SphinxRtdTheme.Navigation.enable(true);
});
</script>
</body>
</html>