| <!DOCTYPE html><html lang="en"><head><meta charset="utf-8"><meta name="viewport" content="width=device-width, initial-scale=1.0"><meta name="generator" content="rustdoc"><meta name="description" content="Source of the Rust file `/root/.cargo/registry/src/github.com-1ecc6299db9ec823/memchr-2.5.0/src/memmem/prefilter/genericsimd.rs`."><meta name="keywords" content="rust, rustlang, rust-lang"><title>genericsimd.rs - source</title><link rel="preload" as="font" type="font/woff2" crossorigin href="../../../../SourceSerif4-Regular.ttf.woff2"><link rel="preload" as="font" type="font/woff2" crossorigin href="../../../../FiraSans-Regular.woff2"><link rel="preload" as="font" type="font/woff2" crossorigin href="../../../../FiraSans-Medium.woff2"><link rel="preload" as="font" type="font/woff2" crossorigin href="../../../../SourceCodePro-Regular.ttf.woff2"><link rel="preload" as="font" type="font/woff2" crossorigin href="../../../../SourceSerif4-Bold.ttf.woff2"><link rel="preload" as="font" type="font/woff2" crossorigin href="../../../../SourceCodePro-Semibold.ttf.woff2"><link rel="stylesheet" href="../../../../normalize.css"><link rel="stylesheet" href="../../../../rustdoc.css" id="mainThemeStyle"><link rel="stylesheet" href="../../../../ayu.css" disabled><link rel="stylesheet" href="../../../../dark.css" disabled><link rel="stylesheet" href="../../../../light.css" id="themeStyle"><script id="default-settings" ></script><script src="../../../../storage.js"></script><script defer src="../../../../source-script.js"></script><script defer src="../../../../source-files.js"></script><script defer src="../../../../main.js"></script><noscript><link rel="stylesheet" href="../../../../noscript.css"></noscript><link rel="alternate icon" type="image/png" href="../../../../favicon-16x16.png"><link rel="alternate icon" type="image/png" href="../../../../favicon-32x32.png"><link rel="icon" type="image/svg+xml" href="../../../../favicon.svg"></head><body class="rustdoc source"><!--[if lte IE 11]><div class="warning">This old browser is unsupported and will most likely display funky things.</div><![endif]--><nav class="sidebar"><a class="sidebar-logo" href="../../../../memchr/index.html"><div class="logo-container"><img class="rust-logo" src="../../../../rust-logo.svg" alt="logo"></div></a></nav><main><div class="width-limiter"><nav class="sub"><a class="sub-logo-container" href="../../../../memchr/index.html"><img class="rust-logo" src="../../../../rust-logo.svg" alt="logo"></a><form class="search-form"><div class="search-container"><span></span><input class="search-input" name="search" autocomplete="off" spellcheck="false" placeholder="Click or press ‘S’ to search, ‘?’ for more options…" type="search"><div id="help-button" title="help" tabindex="-1"><a href="../../../../help.html">?</a></div><div id="settings-menu" tabindex="-1"><a href="../../../../settings.html" title="settings"><img width="22" height="22" alt="Change settings" src="../../../../wheel.svg"></a></div></div></form></nav><section id="main-content" class="content"><div class="example-wrap"><pre class="src-line-numbers"><span id="1">1</span> |
| <span id="2">2</span> |
| <span id="3">3</span> |
| <span id="4">4</span> |
| <span id="5">5</span> |
| <span id="6">6</span> |
| <span id="7">7</span> |
| <span id="8">8</span> |
| <span id="9">9</span> |
| <span id="10">10</span> |
| <span id="11">11</span> |
| <span id="12">12</span> |
| <span id="13">13</span> |
| <span id="14">14</span> |
| <span id="15">15</span> |
| <span id="16">16</span> |
| <span id="17">17</span> |
| <span id="18">18</span> |
| <span id="19">19</span> |
| <span id="20">20</span> |
| <span id="21">21</span> |
| <span id="22">22</span> |
| <span id="23">23</span> |
| <span id="24">24</span> |
| <span id="25">25</span> |
| <span id="26">26</span> |
| <span id="27">27</span> |
| <span id="28">28</span> |
| <span id="29">29</span> |
| <span id="30">30</span> |
| <span id="31">31</span> |
| <span id="32">32</span> |
| <span id="33">33</span> |
| <span id="34">34</span> |
| <span id="35">35</span> |
| <span id="36">36</span> |
| <span id="37">37</span> |
| <span id="38">38</span> |
| <span id="39">39</span> |
| <span id="40">40</span> |
| <span id="41">41</span> |
| <span id="42">42</span> |
| <span id="43">43</span> |
| <span id="44">44</span> |
| <span id="45">45</span> |
| <span id="46">46</span> |
| <span id="47">47</span> |
| <span id="48">48</span> |
| <span id="49">49</span> |
| <span id="50">50</span> |
| <span id="51">51</span> |
| <span id="52">52</span> |
| <span id="53">53</span> |
| <span id="54">54</span> |
| <span id="55">55</span> |
| <span id="56">56</span> |
| <span id="57">57</span> |
| <span id="58">58</span> |
| <span id="59">59</span> |
| <span id="60">60</span> |
| <span id="61">61</span> |
| <span id="62">62</span> |
| <span id="63">63</span> |
| <span id="64">64</span> |
| <span id="65">65</span> |
| <span id="66">66</span> |
| <span id="67">67</span> |
| <span id="68">68</span> |
| <span id="69">69</span> |
| <span id="70">70</span> |
| <span id="71">71</span> |
| <span id="72">72</span> |
| <span id="73">73</span> |
| <span id="74">74</span> |
| <span id="75">75</span> |
| <span id="76">76</span> |
| <span id="77">77</span> |
| <span id="78">78</span> |
| <span id="79">79</span> |
| <span id="80">80</span> |
| <span id="81">81</span> |
| <span id="82">82</span> |
| <span id="83">83</span> |
| <span id="84">84</span> |
| <span id="85">85</span> |
| <span id="86">86</span> |
| <span id="87">87</span> |
| <span id="88">88</span> |
| <span id="89">89</span> |
| <span id="90">90</span> |
| <span id="91">91</span> |
| <span id="92">92</span> |
| <span id="93">93</span> |
| <span id="94">94</span> |
| <span id="95">95</span> |
| <span id="96">96</span> |
| <span id="97">97</span> |
| <span id="98">98</span> |
| <span id="99">99</span> |
| <span id="100">100</span> |
| <span id="101">101</span> |
| <span id="102">102</span> |
| <span id="103">103</span> |
| <span id="104">104</span> |
| <span id="105">105</span> |
| <span id="106">106</span> |
| <span id="107">107</span> |
| <span id="108">108</span> |
| <span id="109">109</span> |
| <span id="110">110</span> |
| <span id="111">111</span> |
| <span id="112">112</span> |
| <span id="113">113</span> |
| <span id="114">114</span> |
| <span id="115">115</span> |
| <span id="116">116</span> |
| <span id="117">117</span> |
| <span id="118">118</span> |
| <span id="119">119</span> |
| <span id="120">120</span> |
| <span id="121">121</span> |
| <span id="122">122</span> |
| <span id="123">123</span> |
| <span id="124">124</span> |
| <span id="125">125</span> |
| <span id="126">126</span> |
| <span id="127">127</span> |
| <span id="128">128</span> |
| <span id="129">129</span> |
| <span id="130">130</span> |
| <span id="131">131</span> |
| <span id="132">132</span> |
| <span id="133">133</span> |
| <span id="134">134</span> |
| <span id="135">135</span> |
| <span id="136">136</span> |
| <span id="137">137</span> |
| <span id="138">138</span> |
| <span id="139">139</span> |
| <span id="140">140</span> |
| <span id="141">141</span> |
| <span id="142">142</span> |
| <span id="143">143</span> |
| <span id="144">144</span> |
| <span id="145">145</span> |
| <span id="146">146</span> |
| <span id="147">147</span> |
| <span id="148">148</span> |
| <span id="149">149</span> |
| <span id="150">150</span> |
| <span id="151">151</span> |
| <span id="152">152</span> |
| <span id="153">153</span> |
| <span id="154">154</span> |
| <span id="155">155</span> |
| <span id="156">156</span> |
| <span id="157">157</span> |
| <span id="158">158</span> |
| <span id="159">159</span> |
| <span id="160">160</span> |
| <span id="161">161</span> |
| <span id="162">162</span> |
| <span id="163">163</span> |
| <span id="164">164</span> |
| <span id="165">165</span> |
| <span id="166">166</span> |
| <span id="167">167</span> |
| <span id="168">168</span> |
| <span id="169">169</span> |
| <span id="170">170</span> |
| <span id="171">171</span> |
| <span id="172">172</span> |
| <span id="173">173</span> |
| <span id="174">174</span> |
| <span id="175">175</span> |
| <span id="176">176</span> |
| <span id="177">177</span> |
| <span id="178">178</span> |
| <span id="179">179</span> |
| <span id="180">180</span> |
| <span id="181">181</span> |
| <span id="182">182</span> |
| <span id="183">183</span> |
| <span id="184">184</span> |
| <span id="185">185</span> |
| <span id="186">186</span> |
| <span id="187">187</span> |
| <span id="188">188</span> |
| <span id="189">189</span> |
| <span id="190">190</span> |
| <span id="191">191</span> |
| <span id="192">192</span> |
| <span id="193">193</span> |
| <span id="194">194</span> |
| <span id="195">195</span> |
| <span id="196">196</span> |
| <span id="197">197</span> |
| <span id="198">198</span> |
| <span id="199">199</span> |
| <span id="200">200</span> |
| <span id="201">201</span> |
| <span id="202">202</span> |
| <span id="203">203</span> |
| <span id="204">204</span> |
| <span id="205">205</span> |
| <span id="206">206</span> |
| <span id="207">207</span> |
| </pre><pre class="rust"><code><span class="kw">use </span>core::mem::size_of; |
| |
| <span class="kw">use </span><span class="kw">crate</span>::memmem::{ |
| prefilter::{PrefilterFnTy, PrefilterState}, |
| vector::Vector, |
| NeedleInfo, |
| }; |
| |
| <span class="doccomment">/// The implementation of the forward vector accelerated candidate finder. |
| /// |
| /// This is inspired by the "generic SIMD" algorithm described here: |
| /// http://0x80.pl/articles/simd-strfind.html#algorithm-1-generic-simd |
| /// |
| /// The main difference is that this is just a prefilter. That is, it reports |
| /// candidates once they are seen and doesn't attempt to confirm them. Also, |
| /// the bytes this routine uses to check for candidates are selected based on |
| /// an a priori background frequency distribution. This means that on most |
| /// haystacks, this will on average spend more time in vectorized code than you |
| /// would if you just selected the first and last bytes of the needle. |
| /// |
| /// Note that a non-prefilter variant of this algorithm can be found in the |
| /// parent module, but it only works on smaller needles. |
| /// |
| /// `prestate`, `ninfo`, `haystack` and `needle` are the four prefilter |
| /// function parameters. `fallback` is a prefilter that is used if the haystack |
| /// is too small to be handled with the given vector size. |
| /// |
| /// This routine is not safe because it is intended for callers to specialize |
| /// this with a particular vector (e.g., __m256i) and then call it with the |
| /// relevant target feature (e.g., avx2) enabled. |
| /// |
| /// # Panics |
| /// |
| /// If `needle.len() <= 1`, then this panics. |
| /// |
| /// # Safety |
| /// |
| /// Since this is meant to be used with vector functions, callers need to |
| /// specialize this inside of a function with a `target_feature` attribute. |
| /// Therefore, callers must ensure that whatever target feature is being used |
| /// supports the vector functions that this function is specialized for. (For |
| /// the specific vector functions used, see the Vector trait implementations.) |
| </span><span class="attribute">#[inline(always)] |
| </span><span class="kw">pub</span>(<span class="kw">crate</span>) <span class="kw">unsafe fn </span>find<V: Vector>( |
| prestate: <span class="kw-2">&mut </span>PrefilterState, |
| ninfo: <span class="kw-2">&</span>NeedleInfo, |
| haystack: <span class="kw-2">&</span>[u8], |
| needle: <span class="kw-2">&</span>[u8], |
| fallback: PrefilterFnTy, |
| ) -> <span class="prelude-ty">Option</span><usize> { |
| <span class="macro">assert!</span>(needle.len() >= <span class="number">2</span>, <span class="string">"needle must be at least 2 bytes"</span>); |
| <span class="kw">let </span>(rare1i, rare2i) = ninfo.rarebytes.as_rare_ordered_usize(); |
| <span class="kw">let </span>min_haystack_len = rare2i + size_of::<V>(); |
| <span class="kw">if </span>haystack.len() < min_haystack_len { |
| <span class="kw">return </span>fallback(prestate, ninfo, haystack, needle); |
| } |
| |
| <span class="kw">let </span>start_ptr = haystack.as_ptr(); |
| <span class="kw">let </span>end_ptr = start_ptr.add(haystack.len()); |
| <span class="kw">let </span>max_ptr = end_ptr.sub(min_haystack_len); |
| <span class="kw">let </span><span class="kw-2">mut </span>ptr = start_ptr; |
| |
| <span class="kw">let </span>rare1chunk = V::splat(needle[rare1i]); |
| <span class="kw">let </span>rare2chunk = V::splat(needle[rare2i]); |
| |
| <span class="comment">// N.B. I did experiment with unrolling the loop to deal with size(V) |
| // bytes at a time and 2*size(V) bytes at a time. The double unroll |
| // was marginally faster while the quadruple unroll was unambiguously |
| // slower. In the end, I decided the complexity from unrolling wasn't |
| // worth it. I used the memmem/krate/prebuilt/huge-en/ benchmarks to |
| // compare. |
| </span><span class="kw">while </span>ptr <= max_ptr { |
| <span class="kw">let </span>m = find_in_chunk2(ptr, rare1i, rare2i, rare1chunk, rare2chunk); |
| <span class="kw">if let </span><span class="prelude-val">Some</span>(chunki) = m { |
| <span class="kw">return </span><span class="prelude-val">Some</span>(matched(prestate, start_ptr, ptr, chunki)); |
| } |
| ptr = ptr.add(size_of::<V>()); |
| } |
| <span class="kw">if </span>ptr < end_ptr { |
| <span class="comment">// This routine immediately quits if a candidate match is found. |
| // That means that if we're here, no candidate matches have been |
| // found at or before 'ptr'. Thus, we don't need to mask anything |
| // out even though we might technically search part of the haystack |
| // that we've already searched (because we know it can't match). |
| </span>ptr = max_ptr; |
| <span class="kw">let </span>m = find_in_chunk2(ptr, rare1i, rare2i, rare1chunk, rare2chunk); |
| <span class="kw">if let </span><span class="prelude-val">Some</span>(chunki) = m { |
| <span class="kw">return </span><span class="prelude-val">Some</span>(matched(prestate, start_ptr, ptr, chunki)); |
| } |
| } |
| prestate.update(haystack.len()); |
| <span class="prelude-val">None |
| </span>} |
| |
| <span class="comment">// Below are two different techniques for checking whether a candidate |
| // match exists in a given chunk or not. find_in_chunk2 checks two bytes |
| // where as find_in_chunk3 checks three bytes. The idea behind checking |
| // three bytes is that while we do a bit more work per iteration, we |
| // decrease the chances of a false positive match being reported and thus |
| // make the search faster overall. This actually works out for the |
| // memmem/krate/prebuilt/huge-en/never-all-common-bytes benchmark, where |
| // using find_in_chunk3 is about 25% faster than find_in_chunk2. However, |
| // it turns out that find_in_chunk2 is faster for all other benchmarks, so |
| // perhaps the extra check isn't worth it in practice. |
| // |
| // For now, we go with find_in_chunk2, but we leave find_in_chunk3 around |
| // to make it easy to switch to and benchmark when possible. |
| |
| </span><span class="doccomment">/// Search for an occurrence of two rare bytes from the needle in the current |
| /// chunk pointed to by ptr. |
| /// |
| /// rare1chunk and rare2chunk correspond to vectors with the rare1 and rare2 |
| /// bytes repeated in each 8-bit lane, respectively. |
| /// |
| /// # Safety |
| /// |
| /// It must be safe to do an unaligned read of size(V) bytes starting at both |
| /// (ptr + rare1i) and (ptr + rare2i). |
| </span><span class="attribute">#[inline(always)] |
| </span><span class="kw">unsafe fn </span>find_in_chunk2<V: Vector>( |
| ptr: <span class="kw-2">*const </span>u8, |
| rare1i: usize, |
| rare2i: usize, |
| rare1chunk: V, |
| rare2chunk: V, |
| ) -> <span class="prelude-ty">Option</span><usize> { |
| <span class="kw">let </span>chunk0 = V::load_unaligned(ptr.add(rare1i)); |
| <span class="kw">let </span>chunk1 = V::load_unaligned(ptr.add(rare2i)); |
| |
| <span class="kw">let </span>eq0 = chunk0.cmpeq(rare1chunk); |
| <span class="kw">let </span>eq1 = chunk1.cmpeq(rare2chunk); |
| |
| <span class="kw">let </span>match_offsets = eq0.and(eq1).movemask(); |
| <span class="kw">if </span>match_offsets == <span class="number">0 </span>{ |
| <span class="kw">return </span><span class="prelude-val">None</span>; |
| } |
| <span class="prelude-val">Some</span>(match_offsets.trailing_zeros() <span class="kw">as </span>usize) |
| } |
| |
| <span class="doccomment">/// Search for an occurrence of two rare bytes and the first byte (even if one |
| /// of the rare bytes is equivalent to the first byte) from the needle in the |
| /// current chunk pointed to by ptr. |
| /// |
| /// firstchunk, rare1chunk and rare2chunk correspond to vectors with the first, |
| /// rare1 and rare2 bytes repeated in each 8-bit lane, respectively. |
| /// |
| /// # Safety |
| /// |
| /// It must be safe to do an unaligned read of size(V) bytes starting at ptr, |
| /// (ptr + rare1i) and (ptr + rare2i). |
| </span><span class="attribute">#[allow(dead_code)] |
| #[inline(always)] |
| </span><span class="kw">unsafe fn </span>find_in_chunk3<V: Vector>( |
| ptr: <span class="kw-2">*const </span>u8, |
| rare1i: usize, |
| rare2i: usize, |
| firstchunk: V, |
| rare1chunk: V, |
| rare2chunk: V, |
| ) -> <span class="prelude-ty">Option</span><usize> { |
| <span class="kw">let </span>chunk0 = V::load_unaligned(ptr); |
| <span class="kw">let </span>chunk1 = V::load_unaligned(ptr.add(rare1i)); |
| <span class="kw">let </span>chunk2 = V::load_unaligned(ptr.add(rare2i)); |
| |
| <span class="kw">let </span>eq0 = chunk0.cmpeq(firstchunk); |
| <span class="kw">let </span>eq1 = chunk1.cmpeq(rare1chunk); |
| <span class="kw">let </span>eq2 = chunk2.cmpeq(rare2chunk); |
| |
| <span class="kw">let </span>match_offsets = eq0.and(eq1).and(eq2).movemask(); |
| <span class="kw">if </span>match_offsets == <span class="number">0 </span>{ |
| <span class="kw">return </span><span class="prelude-val">None</span>; |
| } |
| <span class="prelude-val">Some</span>(match_offsets.trailing_zeros() <span class="kw">as </span>usize) |
| } |
| |
| <span class="doccomment">/// Accepts a chunk-relative offset and returns a haystack relative offset |
| /// after updating the prefilter state. |
| /// |
| /// Why do we use this unlineable function when a search completes? Well, |
| /// I don't know. Really. Obviously this function was not here initially. |
| /// When doing profiling, the codegen for the inner loop here looked bad and |
| /// I didn't know why. There were a couple extra 'add' instructions and an |
| /// extra 'lea' instruction that I couldn't explain. I hypothesized that the |
| /// optimizer was having trouble untangling the hot code in the loop from the |
| /// code that deals with a candidate match. By putting the latter into an |
| /// unlineable function, it kind of forces the issue and it had the intended |
| /// effect: codegen improved measurably. It's good for a ~10% improvement |
| /// across the board on the memmem/krate/prebuilt/huge-en/ benchmarks. |
| </span><span class="attribute">#[cold] |
| #[inline(never)] |
| </span><span class="kw">fn </span>matched( |
| prestate: <span class="kw-2">&mut </span>PrefilterState, |
| start_ptr: <span class="kw-2">*const </span>u8, |
| ptr: <span class="kw-2">*const </span>u8, |
| chunki: usize, |
| ) -> usize { |
| <span class="kw">let </span>found = diff(ptr, start_ptr) + chunki; |
| prestate.update(found); |
| found |
| } |
| |
| <span class="doccomment">/// Subtract `b` from `a` and return the difference. `a` must be greater than |
| /// or equal to `b`. |
| </span><span class="kw">fn </span>diff(a: <span class="kw-2">*const </span>u8, b: <span class="kw-2">*const </span>u8) -> usize { |
| <span class="macro">debug_assert!</span>(a >= b); |
| (a <span class="kw">as </span>usize) - (b <span class="kw">as </span>usize) |
| } |
| </code></pre></div> |
| </section></div></main><div id="rustdoc-vars" data-root-path="../../../../" data-current-crate="memchr" data-themes="ayu,dark,light" data-resource-suffix="" data-rustdoc-version="1.66.0-nightly (5c8bff74b 2022-10-21)" ></div></body></html> |