llvm.org GIT mirror llvm / release_31 docs / BranchWeightMetadata.html
release_31

Tree @release_31 (Download .tar.gz)

BranchWeightMetadata.html @release_31raw · history · blame

<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
                      "http://www.w3.org/TR/html4/strict.dtd">
<html>
<head>
  <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
  <title>LLVM Branch Weight Metadata</title>
  <link rel="stylesheet" href="llvm.css" type="text/css">
</head>
<body>

<h1>
  LLVM Branch Weight Metadata
</h1>

<ol>
  <li><a href="#introduction">Introduction</a></li>
  <li><a href="#supported_instructions">Supported Instructions</a></li>
  <li><a href="#builtin_expect">Built-in "expect" Instruction </a></li>
  <li><a href="#cfg_modifications">CFG Modifications</a></li>
</ol>

<div class="doc_author">
  <p>Written by <a href="mailto:jstaszak@apple.com">Jakub Staszak</a></p>
</div>

<h2>
  <a name="introduction">Introduction</a>
</h2>
<div>
<p>Branch Weight Metadata represents branch weights as its likeliness to
be taken. Metadata is assigned to the <tt>TerminatorInst</tt> as a
<tt>MDNode</tt> of the <tt>MD_prof</tt> kind. The first operator is always a
<tt>MDString</tt> node with the string "branch_weights". Number of operators
depends on the terminator type.</p>

<p>Branch weights might be fetch from the profiling file, or generated based on
<a href="#builtin_expect"><tt>__builtin_expect</tt></a> instruction.
</p>

<p>All weights are represented as an unsigned 32-bit values, where higher value
indicates greater chance to be taken.</p>
</div>

<h2>
  <a name="supported_instructions">Supported Instructions</a>
</h2>

<div>
  <h4>BranchInst</h4>
  <div>
    <p>Metadata is only assign to the conditional branches. There are two extra
    operarands, for the true and the false branch.</p>
  </div>
  <div class="doc_code">
  <pre>
!0 = metadata !{
  metadata !"branch_weights",
  i32 &lt;TRUE_BRANCH_WEIGHT&gt;,
  i32 &lt;FALSE_BRANCH_WEIGHT&gt;
}
  </pre>
  </div>

  <h4>SwitchInst</h4>
  <div>
  <p>Branch weights are assign to every case (including <tt>default</tt> case
  which is always case #0).</p>
  </div>
  <div class="doc_code">
  <pre>
!0 = metadata !{
  metadata !"branch_weights",
  i32 &lt;DEFAULT_BRANCH_WEIGHT&gt;
  [ , i32 &lt;CASE_BRANCH_WEIGHT&gt; ... ]
}
  </pre>
  </div>

  <h4>IndirectBrInst</h4>
  <div>
  <p>Branch weights are assign to every destination.</p>
  </div>
  <div class="doc_code">
  <pre>
!0 = metadata !{
  metadata !"branch_weights",
  i32 &lt;LABEL_BRANCH_WEIGHT&gt;
  [ , i32 &lt;LABEL_BRANCH_WEIGHT&gt; ... ]
}
  </pre>
  </div>

  <h4>Other</h4>
  <div>
  <p>Other terminator instructions are not allowed to contain Branch Weight
  Metadata.</p>
  </div>
</div>

<h2>
  <a name="builtin_expect">Built-in "expect" Instructions</a>
</h2>
<div>
  <p><tt>__builtin_expect(long exp, long c)</tt> instruction provides branch
  prediction information. The return value is the value of <tt>exp</tt>.</p>

  <p>It is especially useful in conditional statements. Currently Clang supports
  two conditional statements:
  </p>
  <h4><tt>if</tt> statement</h4>
  <div>
  <p>The <tt>exp</tt> parameter is the condition. The <tt>c</tt> parameter is
  the expected comparision value. If it is equal to 1 (true), the condition is
  likely to be true, in other case condition is likely to be false. For example:
  </p>
  </div>
  <div class="doc_code">
  <pre>
  if (__builtin_expect(x &gt; 0, 1)) {
    // This block is likely to be taken.
  }
  </pre>
  </div>

  <h4><tt>switch</tt> statement</h4>
  <div>
  <p>The <tt>exp</tt> parameter is the value. The <tt>c</tt> parameter is the
  expected value. If the expected value doesn't show on the cases list, the
  <tt>default</tt> case is assumed to be likely taken.</p>
  </div>
  <div class="doc_code">
  <pre>
  switch (__builtin_expect(x, 5)) {
  default: break;
  case 0:  // ...
  case 3:  // ...
  case 5:  // This case is likely to be taken.
  }
  </pre>
  </div>
</div>

<h2>
  <a name="cfg_modifications">CFG Modifications</a>
</h2>
<div>
<p>Branch Weight Metatada is not proof against CFG changes. If terminator
operands' are changed some action should be taken. In other case some
misoptimizations may occur due to incorrent branch prediction information.</p>
</div>

<hr>
<address>
  <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
  src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
  <a href="http://validator.w3.org/check/referer"><img
  src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>

  <a href="mailto:jstaszak@apple.com">Jakub Staszak</a><br>
  <a href="http://llvm.org/">LLVM Compiler Infrastructure</a><br>
</address>

</body>
</html>