blob: 15c7becdd5f073082daeac4c61e17846af88cc68 [file] [log] [blame]
<!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/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>
<span id="208">208</span>
<span id="209">209</span>
<span id="210">210</span>
<span id="211">211</span>
<span id="212">212</span>
<span id="213">213</span>
<span id="214">214</span>
<span id="215">215</span>
<span id="216">216</span>
<span id="217">217</span>
<span id="218">218</span>
<span id="219">219</span>
<span id="220">220</span>
<span id="221">221</span>
<span id="222">222</span>
<span id="223">223</span>
<span id="224">224</span>
<span id="225">225</span>
<span id="226">226</span>
<span id="227">227</span>
<span id="228">228</span>
<span id="229">229</span>
<span id="230">230</span>
<span id="231">231</span>
<span id="232">232</span>
<span id="233">233</span>
<span id="234">234</span>
<span id="235">235</span>
<span id="236">236</span>
<span id="237">237</span>
<span id="238">238</span>
<span id="239">239</span>
<span id="240">240</span>
<span id="241">241</span>
<span id="242">242</span>
<span id="243">243</span>
<span id="244">244</span>
<span id="245">245</span>
<span id="246">246</span>
<span id="247">247</span>
<span id="248">248</span>
<span id="249">249</span>
<span id="250">250</span>
<span id="251">251</span>
<span id="252">252</span>
<span id="253">253</span>
<span id="254">254</span>
<span id="255">255</span>
<span id="256">256</span>
<span id="257">257</span>
<span id="258">258</span>
<span id="259">259</span>
<span id="260">260</span>
<span id="261">261</span>
<span id="262">262</span>
<span id="263">263</span>
<span id="264">264</span>
<span id="265">265</span>
<span id="266">266</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::{util::memcmp, vector::Vector, NeedleInfo};
<span class="doccomment">/// The minimum length of a needle required for this algorithm. The minimum
/// is 2 since a length of 1 should just use memchr and a length of 0 isn&#39;t
/// a case handled by this searcher.
</span><span class="kw">pub</span>(<span class="kw">crate</span>) <span class="kw">const </span>MIN_NEEDLE_LEN: usize = <span class="number">2</span>;
<span class="doccomment">/// The maximum length of a needle required for this algorithm.
///
/// In reality, there is no hard max here. The code below can handle any
/// length needle. (Perhaps that suggests there are missing optimizations.)
/// Instead, this is a heuristic and a bound guaranteeing our linear time
/// complexity.
///
/// It is a heuristic because when a candidate match is found, memcmp is run.
/// For very large needles with lots of false positives, memcmp can make the
/// code run quite slow.
///
/// It is a bound because the worst case behavior with memcmp is multiplicative
/// in the size of the needle and haystack, and we want to keep that additive.
/// This bound ensures we still meet that bound theoretically, since it&#39;s just
/// a constant. We aren&#39;t acting in bad faith here, memcmp on tiny needles
/// is so fast that even in pathological cases (see pathological vector
/// benchmarks), this is still just as fast or faster in practice.
///
/// This specific number was chosen by tweaking a bit and running benchmarks.
/// The rare-medium-needle, for example, gets about 5% faster by using this
/// algorithm instead of a prefilter-accelerated Two-Way. There&#39;s also a
/// theoretical desire to keep this number reasonably low, to mitigate the
/// impact of pathological cases. I did try 64, and some benchmarks got a
/// little better, and others (particularly the pathological ones), got a lot
/// worse. So... 32 it is?
</span><span class="kw">pub</span>(<span class="kw">crate</span>) <span class="kw">const </span>MAX_NEEDLE_LEN: usize = <span class="number">32</span>;
<span class="doccomment">/// The implementation of the forward vector accelerated substring search.
///
/// This is extremely similar to the prefilter vector module by the same name.
/// The key difference is that this is not a prefilter. Instead, it handles
/// confirming its own matches. The trade off is that this only works with
/// smaller needles. The speed up here is that an inlined memcmp on a tiny
/// needle is very quick, even on pathological inputs. This is much better than
/// combining a prefilter with Two-Way, where using Two-Way to confirm the
/// match has higher latency.
///
/// So why not use this for all needles? We could, and it would probably work
/// really well on most inputs. But its worst case is multiplicative and we
/// want to guarantee worst case additive time. Some of the benchmarks try to
/// justify this (see the pathological ones).
///
/// The prefilter variant of this has more comments. Also note that we only
/// implement this for forward searches for now. If you have a compelling use
/// case for accelerated reverse search, please file an issue.
</span><span class="attribute">#[derive(Clone, Copy, Debug)]
</span><span class="kw">pub</span>(<span class="kw">crate</span>) <span class="kw">struct </span>Forward {
rare1i: u8,
rare2i: u8,
}
<span class="kw">impl </span>Forward {
<span class="doccomment">/// Create a new &quot;generic simd&quot; forward searcher. If one could not be
/// created from the given inputs, then None is returned.
</span><span class="kw">pub</span>(<span class="kw">crate</span>) <span class="kw">fn </span>new(ninfo: <span class="kw-2">&amp;</span>NeedleInfo, needle: <span class="kw-2">&amp;</span>[u8]) -&gt; <span class="prelude-ty">Option</span>&lt;Forward&gt; {
<span class="kw">let </span>(rare1i, rare2i) = ninfo.rarebytes.as_rare_ordered_u8();
<span class="comment">// If the needle is too short or too long, give up. Also, give up
// if the rare bytes detected are at the same position. (It likely
// suggests a degenerate case, although it should technically not be
// possible.)
</span><span class="kw">if </span>needle.len() &lt; MIN_NEEDLE_LEN
|| needle.len() &gt; MAX_NEEDLE_LEN
|| rare1i == rare2i
{
<span class="kw">return </span><span class="prelude-val">None</span>;
}
<span class="prelude-val">Some</span>(Forward { rare1i, rare2i })
}
<span class="doccomment">/// Returns the minimum length of haystack that is needed for this searcher
/// to work for a particular vector. Passing a haystack with a length
/// smaller than this will cause `fwd_find` to panic.
</span><span class="attribute">#[inline(always)]
</span><span class="kw">pub</span>(<span class="kw">crate</span>) <span class="kw">fn </span>min_haystack_len&lt;V: Vector&gt;(<span class="kw-2">&amp;</span><span class="self">self</span>) -&gt; usize {
<span class="self">self</span>.rare2i <span class="kw">as </span>usize + size_of::&lt;V&gt;()
}
}
<span class="doccomment">/// Searches the given haystack for the given needle. The needle given should
/// be the same as the needle that this searcher was initialized with.
///
/// # Panics
///
/// When the given haystack has a length smaller than `min_haystack_len`.
///
/// # 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>fwd_find&lt;V: Vector&gt;(
fwd: <span class="kw-2">&amp;</span>Forward,
haystack: <span class="kw-2">&amp;</span>[u8],
needle: <span class="kw-2">&amp;</span>[u8],
) -&gt; <span class="prelude-ty">Option</span>&lt;usize&gt; {
<span class="comment">// It would be nice if we didn&#39;t have this check here, since the meta
// searcher should handle it for us. But without this, I don&#39;t think we
// guarantee that end_ptr.sub(needle.len()) won&#39;t result in UB. We could
// put it as part of the safety contract, but it makes it more complicated
// than necessary.
</span><span class="kw">if </span>haystack.len() &lt; needle.len() {
<span class="kw">return </span><span class="prelude-val">None</span>;
}
<span class="kw">let </span>min_haystack_len = fwd.min_haystack_len::&lt;V&gt;();
<span class="macro">assert!</span>(haystack.len() &gt;= min_haystack_len, <span class="string">&quot;haystack too small&quot;</span>);
<span class="macro">debug_assert!</span>(needle.len() &lt;= haystack.len());
<span class="macro">debug_assert!</span>(
needle.len() &gt;= MIN_NEEDLE_LEN,
<span class="string">&quot;needle must be at least {} bytes&quot;</span>,
MIN_NEEDLE_LEN,
);
<span class="macro">debug_assert!</span>(
needle.len() &lt;= MAX_NEEDLE_LEN,
<span class="string">&quot;needle must be at most {} bytes&quot;</span>,
MAX_NEEDLE_LEN,
);
<span class="kw">let </span>(rare1i, rare2i) = (fwd.rare1i <span class="kw">as </span>usize, fwd.rare2i <span class="kw">as </span>usize);
<span class="kw">let </span>rare1chunk = V::splat(needle[rare1i]);
<span class="kw">let </span>rare2chunk = V::splat(needle[rare2i]);
<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="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&#39;t worth it. I
// used the memmem/krate/prebuilt/huge-en/ benchmarks to compare.
</span><span class="kw">while </span>ptr &lt;= max_ptr {
<span class="kw">let </span>m = fwd_find_in_chunk(
fwd, needle, ptr, end_ptr, rare1chunk, rare2chunk, !<span class="number">0</span>,
);
<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(start_ptr, ptr, chunki));
}
ptr = ptr.add(size_of::&lt;V&gt;());
}
<span class="kw">if </span>ptr &lt; end_ptr {
<span class="kw">let </span>remaining = diff(end_ptr, ptr);
<span class="macro">debug_assert!</span>(
remaining &lt; min_haystack_len,
<span class="string">&quot;remaining bytes should be smaller than the minimum haystack \
length of {}, but there are {} bytes remaining&quot;</span>,
min_haystack_len,
remaining,
);
<span class="kw">if </span>remaining &lt; needle.len() {
<span class="kw">return </span><span class="prelude-val">None</span>;
}
<span class="macro">debug_assert!</span>(
max_ptr &lt; ptr,
<span class="string">&quot;after main loop, ptr should have exceeded max_ptr&quot;</span>,
);
<span class="kw">let </span>overlap = diff(ptr, max_ptr);
<span class="macro">debug_assert!</span>(
overlap &gt; <span class="number">0</span>,
<span class="string">&quot;overlap ({}) must always be non-zero&quot;</span>,
overlap,
);
<span class="macro">debug_assert!</span>(
overlap &lt; size_of::&lt;V&gt;(),
<span class="string">&quot;overlap ({}) cannot possibly be &gt;= than a vector ({})&quot;</span>,
overlap,
size_of::&lt;V&gt;(),
);
<span class="comment">// The mask has all of its bits set except for the first N least
// significant bits, where N=overlap. This way, any matches that
// occur in find_in_chunk within the overlap are automatically
// ignored.
</span><span class="kw">let </span>mask = !((<span class="number">1 </span>&lt;&lt; overlap) - <span class="number">1</span>);
ptr = max_ptr;
<span class="kw">let </span>m = fwd_find_in_chunk(
fwd, needle, ptr, end_ptr, rare1chunk, rare2chunk, mask,
);
<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(start_ptr, ptr, chunki));
}
}
<span class="prelude-val">None
</span>}
<span class="doccomment">/// Search for an occurrence of two rare bytes from the needle in the chunk
/// pointed to by ptr, with the end of the haystack pointed to by end_ptr. When
/// an occurrence is found, memcmp is run to check if a match occurs at the
/// corresponding position.
///
/// rare1chunk and rare2chunk correspond to vectors with the rare1 and rare2
/// bytes repeated in each 8-bit lane, respectively.
///
/// mask should have bits set corresponding the positions in the chunk in which
/// matches are considered. This is only used for the last vector load where
/// the beginning of the vector might have overlapped with the last load in
/// the main loop. The mask lets us avoid visiting positions that have already
/// been discarded as matches.
///
/// # Safety
///
/// It must be safe to do an unaligned read of size(V) bytes starting at both
/// (ptr + rare1i) and (ptr + rare2i). It must also be safe to do unaligned
/// loads on ptr up to (end_ptr - needle.len()).
</span><span class="attribute">#[inline(always)]
</span><span class="kw">unsafe fn </span>fwd_find_in_chunk&lt;V: Vector&gt;(
fwd: <span class="kw-2">&amp;</span>Forward,
needle: <span class="kw-2">&amp;</span>[u8],
ptr: <span class="kw-2">*const </span>u8,
end_ptr: <span class="kw-2">*const </span>u8,
rare1chunk: V,
rare2chunk: V,
mask: u32,
) -&gt; <span class="prelude-ty">Option</span>&lt;usize&gt; {
<span class="kw">let </span>chunk0 = V::load_unaligned(ptr.add(fwd.rare1i <span class="kw">as </span>usize));
<span class="kw">let </span>chunk1 = V::load_unaligned(ptr.add(fwd.rare2i <span class="kw">as </span>usize));
<span class="kw">let </span>eq0 = chunk0.cmpeq(rare1chunk);
<span class="kw">let </span>eq1 = chunk1.cmpeq(rare2chunk);
<span class="kw">let </span><span class="kw-2">mut </span>match_offsets = eq0.and(eq1).movemask() &amp; mask;
<span class="kw">while </span>match_offsets != <span class="number">0 </span>{
<span class="kw">let </span>offset = match_offsets.trailing_zeros() <span class="kw">as </span>usize;
<span class="kw">let </span>ptr = ptr.add(offset);
<span class="kw">if </span>end_ptr.sub(needle.len()) &lt; ptr {
<span class="kw">return </span><span class="prelude-val">None</span>;
}
<span class="kw">let </span>chunk = core::slice::from_raw_parts(ptr, needle.len());
<span class="kw">if </span>memcmp(needle, chunk) {
<span class="kw">return </span><span class="prelude-val">Some</span>(offset);
}
match_offsets &amp;= match_offsets - <span class="number">1</span>;
}
<span class="prelude-val">None
</span>}
<span class="doccomment">/// Accepts a chunk-relative offset and returns a haystack relative offset
/// after updating the prefilter state.
///
/// See the same function with the same name in the prefilter variant of this
/// algorithm to learned why it&#39;s tagged with inline(never). Even here, where
/// the function is simpler, inlining it leads to poorer codegen. (Although
/// it does improve some benchmarks, like prebuiltiter/huge-en/common-you.)
</span><span class="attribute">#[cold]
#[inline(never)]
</span><span class="kw">fn </span>matched(start_ptr: <span class="kw-2">*const </span>u8, ptr: <span class="kw-2">*const </span>u8, chunki: usize) -&gt; usize {
diff(ptr, start_ptr) + chunki
}
<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) -&gt; usize {
<span class="macro">debug_assert!</span>(a &gt;= 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>