<!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 `/Users/erlendbasso/.cargo/registry/src/github.com-1ecc6299db9ec823/nalgebra-0.32.1/src/base/blas_uninit.rs`."><meta name="keywords" content="rust, rustlang, rust-lang"><title>blas_uninit.rs - source</title><link rel="preload" as="font" type="font/woff2" crossorigin href="../../../static.files/SourceSerif4-Regular-1f7d512b176f0f72.ttf.woff2"><link rel="preload" as="font" type="font/woff2" crossorigin href="../../../static.files/FiraSans-Regular-018c141bf0843ffd.woff2"><link rel="preload" as="font" type="font/woff2" crossorigin href="../../../static.files/FiraSans-Medium-8f9a781e4970d388.woff2"><link rel="preload" as="font" type="font/woff2" crossorigin href="../../../static.files/SourceCodePro-Regular-562dcc5011b6de7d.ttf.woff2"><link rel="preload" as="font" type="font/woff2" crossorigin href="../../../static.files/SourceSerif4-Bold-124a1ca42af929b6.ttf.woff2"><link rel="preload" as="font" type="font/woff2" crossorigin href="../../../static.files/SourceCodePro-Semibold-d899c5a5c4aeb14a.ttf.woff2"><link rel="stylesheet" href="../../../static.files/normalize-76eba96aa4d2e634.css"><link rel="stylesheet" href="../../../static.files/rustdoc-6827029ac823cab7.css" id="mainThemeStyle"><link rel="stylesheet" id="themeStyle" href="../../../static.files/light-ebce58d0a40c3431.css"><link rel="stylesheet" disabled href="../../../static.files/dark-f23faae4a2daf9a6.css"><link rel="stylesheet" disabled href="../../../static.files/ayu-8af5e100b21cd173.css"><script id="default-settings" ></script><script src="../../../static.files/storage-d43fa987303ecbbb.js"></script><script defer src="../../../static.files/source-script-5cf2e01a42cc9858.js"></script><script defer src="../../../source-files.js"></script><script defer src="../../../static.files/main-c55e1eb52e1886b4.js"></script><noscript><link rel="stylesheet" href="../../../static.files/noscript-13285aec31fa243e.css"></noscript><link rel="icon" href="https://nalgebra.org/img/favicon.ico"></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"></nav><main><div class="width-limiter"><nav class="sub"><a class="sub-logo-container" href="../../../nalgebra/index.html"><img class="rust-logo" src="../../../static.files/rust-logo-151179464ae7ed46.svg" alt="logo"></a><form class="search-form"><span></span><input class="search-input" name="search" aria-label="Run search in the documentation" 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="../../../static.files/wheel-5ec35bf9ca753509.svg"></a></div></form></nav><section id="main-content" class="content"><div class="example-wrap"><pre class="src-line-numbers"><a href="#1" id="1">1</a>
<a href="#2" id="2">2</a>
<a href="#3" id="3">3</a>
<a href="#4" id="4">4</a>
<a href="#5" id="5">5</a>
<a href="#6" id="6">6</a>
<a href="#7" id="7">7</a>
<a href="#8" id="8">8</a>
<a href="#9" id="9">9</a>
<a href="#10" id="10">10</a>
<a href="#11" id="11">11</a>
<a href="#12" id="12">12</a>
<a href="#13" id="13">13</a>
<a href="#14" id="14">14</a>
<a href="#15" id="15">15</a>
<a href="#16" id="16">16</a>
<a href="#17" id="17">17</a>
<a href="#18" id="18">18</a>
<a href="#19" id="19">19</a>
<a href="#20" id="20">20</a>
<a href="#21" id="21">21</a>
<a href="#22" id="22">22</a>
<a href="#23" id="23">23</a>
<a href="#24" id="24">24</a>
<a href="#25" id="25">25</a>
<a href="#26" id="26">26</a>
<a href="#27" id="27">27</a>
<a href="#28" id="28">28</a>
<a href="#29" id="29">29</a>
<a href="#30" id="30">30</a>
<a href="#31" id="31">31</a>
<a href="#32" id="32">32</a>
<a href="#33" id="33">33</a>
<a href="#34" id="34">34</a>
<a href="#35" id="35">35</a>
<a href="#36" id="36">36</a>
<a href="#37" id="37">37</a>
<a href="#38" id="38">38</a>
<a href="#39" id="39">39</a>
<a href="#40" id="40">40</a>
<a href="#41" id="41">41</a>
<a href="#42" id="42">42</a>
<a href="#43" id="43">43</a>
<a href="#44" id="44">44</a>
<a href="#45" id="45">45</a>
<a href="#46" id="46">46</a>
<a href="#47" id="47">47</a>
<a href="#48" id="48">48</a>
<a href="#49" id="49">49</a>
<a href="#50" id="50">50</a>
<a href="#51" id="51">51</a>
<a href="#52" id="52">52</a>
<a href="#53" id="53">53</a>
<a href="#54" id="54">54</a>
<a href="#55" id="55">55</a>
<a href="#56" id="56">56</a>
<a href="#57" id="57">57</a>
<a href="#58" id="58">58</a>
<a href="#59" id="59">59</a>
<a href="#60" id="60">60</a>
<a href="#61" id="61">61</a>
<a href="#62" id="62">62</a>
<a href="#63" id="63">63</a>
<a href="#64" id="64">64</a>
<a href="#65" id="65">65</a>
<a href="#66" id="66">66</a>
<a href="#67" id="67">67</a>
<a href="#68" id="68">68</a>
<a href="#69" id="69">69</a>
<a href="#70" id="70">70</a>
<a href="#71" id="71">71</a>
<a href="#72" id="72">72</a>
<a href="#73" id="73">73</a>
<a href="#74" id="74">74</a>
<a href="#75" id="75">75</a>
<a href="#76" id="76">76</a>
<a href="#77" id="77">77</a>
<a href="#78" id="78">78</a>
<a href="#79" id="79">79</a>
<a href="#80" id="80">80</a>
<a href="#81" id="81">81</a>
<a href="#82" id="82">82</a>
<a href="#83" id="83">83</a>
<a href="#84" id="84">84</a>
<a href="#85" id="85">85</a>
<a href="#86" id="86">86</a>
<a href="#87" id="87">87</a>
<a href="#88" id="88">88</a>
<a href="#89" id="89">89</a>
<a href="#90" id="90">90</a>
<a href="#91" id="91">91</a>
<a href="#92" id="92">92</a>
<a href="#93" id="93">93</a>
<a href="#94" id="94">94</a>
<a href="#95" id="95">95</a>
<a href="#96" id="96">96</a>
<a href="#97" id="97">97</a>
<a href="#98" id="98">98</a>
<a href="#99" id="99">99</a>
<a href="#100" id="100">100</a>
<a href="#101" id="101">101</a>
<a href="#102" id="102">102</a>
<a href="#103" id="103">103</a>
<a href="#104" id="104">104</a>
<a href="#105" id="105">105</a>
<a href="#106" id="106">106</a>
<a href="#107" id="107">107</a>
<a href="#108" id="108">108</a>
<a href="#109" id="109">109</a>
<a href="#110" id="110">110</a>
<a href="#111" id="111">111</a>
<a href="#112" id="112">112</a>
<a href="#113" id="113">113</a>
<a href="#114" id="114">114</a>
<a href="#115" id="115">115</a>
<a href="#116" id="116">116</a>
<a href="#117" id="117">117</a>
<a href="#118" id="118">118</a>
<a href="#119" id="119">119</a>
<a href="#120" id="120">120</a>
<a href="#121" id="121">121</a>
<a href="#122" id="122">122</a>
<a href="#123" id="123">123</a>
<a href="#124" id="124">124</a>
<a href="#125" id="125">125</a>
<a href="#126" id="126">126</a>
<a href="#127" id="127">127</a>
<a href="#128" id="128">128</a>
<a href="#129" id="129">129</a>
<a href="#130" id="130">130</a>
<a href="#131" id="131">131</a>
<a href="#132" id="132">132</a>
<a href="#133" id="133">133</a>
<a href="#134" id="134">134</a>
<a href="#135" id="135">135</a>
<a href="#136" id="136">136</a>
<a href="#137" id="137">137</a>
<a href="#138" id="138">138</a>
<a href="#139" id="139">139</a>
<a href="#140" id="140">140</a>
<a href="#141" id="141">141</a>
<a href="#142" id="142">142</a>
<a href="#143" id="143">143</a>
<a href="#144" id="144">144</a>
<a href="#145" id="145">145</a>
<a href="#146" id="146">146</a>
<a href="#147" id="147">147</a>
<a href="#148" id="148">148</a>
<a href="#149" id="149">149</a>
<a href="#150" id="150">150</a>
<a href="#151" id="151">151</a>
<a href="#152" id="152">152</a>
<a href="#153" id="153">153</a>
<a href="#154" id="154">154</a>
<a href="#155" id="155">155</a>
<a href="#156" id="156">156</a>
<a href="#157" id="157">157</a>
<a href="#158" id="158">158</a>
<a href="#159" id="159">159</a>
<a href="#160" id="160">160</a>
<a href="#161" id="161">161</a>
<a href="#162" id="162">162</a>
<a href="#163" id="163">163</a>
<a href="#164" id="164">164</a>
<a href="#165" id="165">165</a>
<a href="#166" id="166">166</a>
<a href="#167" id="167">167</a>
<a href="#168" id="168">168</a>
<a href="#169" id="169">169</a>
<a href="#170" id="170">170</a>
<a href="#171" id="171">171</a>
<a href="#172" id="172">172</a>
<a href="#173" id="173">173</a>
<a href="#174" id="174">174</a>
<a href="#175" id="175">175</a>
<a href="#176" id="176">176</a>
<a href="#177" id="177">177</a>
<a href="#178" id="178">178</a>
<a href="#179" id="179">179</a>
<a href="#180" id="180">180</a>
<a href="#181" id="181">181</a>
<a href="#182" id="182">182</a>
<a href="#183" id="183">183</a>
<a href="#184" id="184">184</a>
<a href="#185" id="185">185</a>
<a href="#186" id="186">186</a>
<a href="#187" id="187">187</a>
<a href="#188" id="188">188</a>
<a href="#189" id="189">189</a>
<a href="#190" id="190">190</a>
<a href="#191" id="191">191</a>
<a href="#192" id="192">192</a>
<a href="#193" id="193">193</a>
<a href="#194" id="194">194</a>
<a href="#195" id="195">195</a>
<a href="#196" id="196">196</a>
<a href="#197" id="197">197</a>
<a href="#198" id="198">198</a>
<a href="#199" id="199">199</a>
<a href="#200" id="200">200</a>
<a href="#201" id="201">201</a>
<a href="#202" id="202">202</a>
<a href="#203" id="203">203</a>
<a href="#204" id="204">204</a>
<a href="#205" id="205">205</a>
<a href="#206" id="206">206</a>
<a href="#207" id="207">207</a>
<a href="#208" id="208">208</a>
<a href="#209" id="209">209</a>
<a href="#210" id="210">210</a>
<a href="#211" id="211">211</a>
<a href="#212" id="212">212</a>
<a href="#213" id="213">213</a>
<a href="#214" id="214">214</a>
<a href="#215" id="215">215</a>
<a href="#216" id="216">216</a>
<a href="#217" id="217">217</a>
<a href="#218" id="218">218</a>
<a href="#219" id="219">219</a>
<a href="#220" id="220">220</a>
<a href="#221" id="221">221</a>
<a href="#222" id="222">222</a>
<a href="#223" id="223">223</a>
<a href="#224" id="224">224</a>
<a href="#225" id="225">225</a>
<a href="#226" id="226">226</a>
<a href="#227" id="227">227</a>
<a href="#228" id="228">228</a>
<a href="#229" id="229">229</a>
<a href="#230" id="230">230</a>
<a href="#231" id="231">231</a>
<a href="#232" id="232">232</a>
<a href="#233" id="233">233</a>
<a href="#234" id="234">234</a>
<a href="#235" id="235">235</a>
<a href="#236" id="236">236</a>
<a href="#237" id="237">237</a>
<a href="#238" id="238">238</a>
<a href="#239" id="239">239</a>
<a href="#240" id="240">240</a>
<a href="#241" id="241">241</a>
<a href="#242" id="242">242</a>
<a href="#243" id="243">243</a>
<a href="#244" id="244">244</a>
<a href="#245" id="245">245</a>
<a href="#246" id="246">246</a>
<a href="#247" id="247">247</a>
<a href="#248" id="248">248</a>
<a href="#249" id="249">249</a>
<a href="#250" id="250">250</a>
<a href="#251" id="251">251</a>
<a href="#252" id="252">252</a>
<a href="#253" id="253">253</a>
<a href="#254" id="254">254</a>
<a href="#255" id="255">255</a>
<a href="#256" id="256">256</a>
<a href="#257" id="257">257</a>
<a href="#258" id="258">258</a>
<a href="#259" id="259">259</a>
<a href="#260" id="260">260</a>
<a href="#261" id="261">261</a>
<a href="#262" id="262">262</a>
<a href="#263" id="263">263</a>
<a href="#264" id="264">264</a>
<a href="#265" id="265">265</a>
<a href="#266" id="266">266</a>
<a href="#267" id="267">267</a>
<a href="#268" id="268">268</a>
<a href="#269" id="269">269</a>
<a href="#270" id="270">270</a>
<a href="#271" id="271">271</a>
<a href="#272" id="272">272</a>
<a href="#273" id="273">273</a>
<a href="#274" id="274">274</a>
<a href="#275" id="275">275</a>
<a href="#276" id="276">276</a>
<a href="#277" id="277">277</a>
<a href="#278" id="278">278</a>
<a href="#279" id="279">279</a>
<a href="#280" id="280">280</a>
<a href="#281" id="281">281</a>
<a href="#282" id="282">282</a>
<a href="#283" id="283">283</a>
<a href="#284" id="284">284</a>
<a href="#285" id="285">285</a>
<a href="#286" id="286">286</a>
<a href="#287" id="287">287</a>
<a href="#288" id="288">288</a>
<a href="#289" id="289">289</a>
<a href="#290" id="290">290</a>
<a href="#291" id="291">291</a>
<a href="#292" id="292">292</a>
<a href="#293" id="293">293</a>
<a href="#294" id="294">294</a>
<a href="#295" id="295">295</a>
<a href="#296" id="296">296</a>
<a href="#297" id="297">297</a>
<a href="#298" id="298">298</a>
<a href="#299" id="299">299</a>
<a href="#300" id="300">300</a>
<a href="#301" id="301">301</a>
<a href="#302" id="302">302</a>
<a href="#303" id="303">303</a>
<a href="#304" id="304">304</a>
<a href="#305" id="305">305</a>
<a href="#306" id="306">306</a>
<a href="#307" id="307">307</a>
<a href="#308" id="308">308</a>
<a href="#309" id="309">309</a>
<a href="#310" id="310">310</a>
<a href="#311" id="311">311</a>
<a href="#312" id="312">312</a>
<a href="#313" id="313">313</a>
<a href="#314" id="314">314</a>
<a href="#315" id="315">315</a>
<a href="#316" id="316">316</a>
<a href="#317" id="317">317</a>
<a href="#318" id="318">318</a>
<a href="#319" id="319">319</a>
<a href="#320" id="320">320</a>
<a href="#321" id="321">321</a>
</pre><pre class="rust"><code><span class="comment">/*
* This file implements some BLAS operations in such a way that they work
* even if the first argument (the output parameter) is an uninitialized matrix.
*
* Because doing this makes the code harder to read, we only implemented the operations that we
* know would benefit from this performance-wise, namely, GEMM (which we use for our matrix
* multiplication code). If we identify other operations like that in the future, we could add
* them here.
*/
</span><span class="attr">#[cfg(feature = <span class="string">"std"</span>)]
</span><span class="kw">use </span>matrixmultiply;
<span class="kw">use </span>num::{One, Zero};
<span class="kw">use </span>simba::scalar::{ClosedAdd, ClosedMul};
<span class="attr">#[cfg(feature = <span class="string">"std"</span>)]
</span><span class="kw">use </span>std::mem;
<span class="kw">use </span><span class="kw">crate</span>::base::constraint::{
AreMultipliable, DimEq, SameNumberOfColumns, SameNumberOfRows, ShapeConstraint,
};
<span class="kw">use </span><span class="kw">crate</span>::base::dimension::{Dim, Dyn, U1};
<span class="kw">use </span><span class="kw">crate</span>::base::storage::{RawStorage, RawStorageMut};
<span class="kw">use </span><span class="kw">crate</span>::base::uninit::InitStatus;
<span class="kw">use </span><span class="kw">crate</span>::base::{Matrix, Scalar, Vector};
<span class="kw">use </span>std::any::TypeId;
<span class="comment">// # Safety
// The content of `y` must only contain values for which
// `Status::assume_init_mut` is sound.
</span><span class="attr">#[allow(clippy::too_many_arguments)]
</span><span class="kw">unsafe fn </span>array_axcpy<Status, T>(
<span class="kw">_</span>: Status,
y: <span class="kw-2">&mut </span>[Status::Value],
a: T,
x: <span class="kw-2">&</span>[T],
c: T,
beta: T,
stride1: usize,
stride2: usize,
len: usize,
) <span class="kw">where
</span>Status: InitStatus<T>,
T: Scalar + Zero + ClosedAdd + ClosedMul,
{
<span class="kw">for </span>i <span class="kw">in </span><span class="number">0</span>..len {
<span class="kw">let </span>y = Status::assume_init_mut(y.get_unchecked_mut(i * stride1));
<span class="kw-2">*</span>y =
a.clone() * x.get_unchecked(i * stride2).clone() * c.clone() + beta.clone() * y.clone();
}
}
<span class="kw">fn </span>array_axc<Status, T>(
<span class="kw">_</span>: Status,
y: <span class="kw-2">&mut </span>[Status::Value],
a: T,
x: <span class="kw-2">&</span>[T],
c: T,
stride1: usize,
stride2: usize,
len: usize,
) <span class="kw">where
</span>Status: InitStatus<T>,
T: Scalar + Zero + ClosedAdd + ClosedMul,
{
<span class="kw">for </span>i <span class="kw">in </span><span class="number">0</span>..len {
<span class="kw">unsafe </span>{
Status::init(
y.get_unchecked_mut(i * stride1),
a.clone() * x.get_unchecked(i * stride2).clone() * c.clone(),
);
}
}
}
<span class="doccomment">/// Computes `y = a * x * c + b * y`.
///
/// If `b` is zero, `y` is never read from and may be uninitialized.
///
/// # Safety
/// This is UB if b != 0 and any component of `y` is uninitialized.
</span><span class="attr">#[inline(always)]
#[allow(clippy::many_single_char_names)]
</span><span class="kw">pub unsafe fn </span>axcpy_uninit<Status, T, D1: Dim, D2: Dim, SA, SB>(
status: Status,
y: <span class="kw-2">&mut </span>Vector<Status::Value, D1, SA>,
a: T,
x: <span class="kw-2">&</span>Vector<T, D2, SB>,
c: T,
b: T,
) <span class="kw">where
</span>T: Scalar + Zero + ClosedAdd + ClosedMul,
SA: RawStorageMut<Status::Value, D1>,
SB: RawStorage<T, D2>,
ShapeConstraint: DimEq<D1, D2>,
Status: InitStatus<T>,
{
<span class="macro">assert_eq!</span>(y.nrows(), x.nrows(), <span class="string">"Axcpy: mismatched vector shapes."</span>);
<span class="kw">let </span>rstride1 = y.strides().<span class="number">0</span>;
<span class="kw">let </span>rstride2 = x.strides().<span class="number">0</span>;
<span class="comment">// SAFETY: the conversion to slices is OK because we access the
// elements taking the strides into account.
</span><span class="kw">let </span>y = y.data.as_mut_slice_unchecked();
<span class="kw">let </span>x = x.data.as_slice_unchecked();
<span class="kw">if </span>!b.is_zero() {
array_axcpy(status, y, a, x, c, b, rstride1, rstride2, x.len());
} <span class="kw">else </span>{
array_axc(status, y, a, x, c, rstride1, rstride2, x.len());
}
}
<span class="doccomment">/// Computes `y = alpha * a * x + beta * y`, where `a` is a matrix, `x` a vector, and
/// `alpha, beta` two scalars.
///
/// If `beta` is zero, `y` is never read from and may be uninitialized.
///
/// # Safety
/// This is UB if beta != 0 and any component of `y` is uninitialized.
</span><span class="attr">#[inline(always)]
</span><span class="kw">pub unsafe fn </span>gemv_uninit<Status, T, D1: Dim, R2: Dim, C2: Dim, D3: Dim, SA, SB, SC>(
status: Status,
y: <span class="kw-2">&mut </span>Vector<Status::Value, D1, SA>,
alpha: T,
a: <span class="kw-2">&</span>Matrix<T, R2, C2, SB>,
x: <span class="kw-2">&</span>Vector<T, D3, SC>,
beta: T,
) <span class="kw">where
</span>Status: InitStatus<T>,
T: Scalar + Zero + One + ClosedAdd + ClosedMul,
SA: RawStorageMut<Status::Value, D1>,
SB: RawStorage<T, R2, C2>,
SC: RawStorage<T, D3>,
ShapeConstraint: DimEq<D1, R2> + AreMultipliable<R2, C2, D3, U1>,
{
<span class="kw">let </span>dim1 = y.nrows();
<span class="kw">let </span>(nrows2, ncols2) = a.shape();
<span class="kw">let </span>dim3 = x.nrows();
<span class="macro">assert!</span>(
ncols2 == dim3 && dim1 == nrows2,
<span class="string">"Gemv: dimensions mismatch."
</span>);
<span class="kw">if </span>ncols2 == <span class="number">0 </span>{
<span class="kw">if </span>beta.is_zero() {
y.apply(|e| Status::init(e, T::zero()));
} <span class="kw">else </span>{
<span class="comment">// SAFETY: this is UB if y is uninitialized.
</span>y.apply(|e| <span class="kw-2">*</span>Status::assume_init_mut(e) <span class="kw-2">*</span>= beta.clone());
}
<span class="kw">return</span>;
}
<span class="comment">// TODO: avoid bound checks.
</span><span class="kw">let </span>col2 = a.column(<span class="number">0</span>);
<span class="kw">let </span>val = x.vget_unchecked(<span class="number">0</span>).clone();
<span class="comment">// SAFETY: this is the call that makes this method unsafe: it is UB if Status = Uninit and beta != 0.
</span>axcpy_uninit(status, y, alpha.clone(), <span class="kw-2">&</span>col2, val, beta);
<span class="kw">for </span>j <span class="kw">in </span><span class="number">1</span>..ncols2 {
<span class="kw">let </span>col2 = a.column(j);
<span class="kw">let </span>val = x.vget_unchecked(j).clone();
<span class="comment">// SAFETY: safe because y was initialized above.
</span>axcpy_uninit(status, y, alpha.clone(), <span class="kw-2">&</span>col2, val, T::one());
}
}
<span class="doccomment">/// Computes `y = alpha * a * b + beta * y`, where `a, b, y` are matrices.
/// `alpha` and `beta` are scalar.
///
/// If `beta` is zero, `y` is never read from and may be uninitialized.
///
/// # Safety
/// This is UB if beta != 0 and any component of `y` is uninitialized.
</span><span class="attr">#[inline(always)]
</span><span class="kw">pub unsafe fn </span>gemm_uninit<
Status,
T,
R1: Dim,
C1: Dim,
R2: Dim,
C2: Dim,
R3: Dim,
C3: Dim,
SA,
SB,
SC,
>(
status: Status,
y: <span class="kw-2">&mut </span>Matrix<Status::Value, R1, C1, SA>,
alpha: T,
a: <span class="kw-2">&</span>Matrix<T, R2, C2, SB>,
b: <span class="kw-2">&</span>Matrix<T, R3, C3, SC>,
beta: T,
) <span class="kw">where
</span>Status: InitStatus<T>,
T: Scalar + Zero + One + ClosedAdd + ClosedMul,
SA: RawStorageMut<Status::Value, R1, C1>,
SB: RawStorage<T, R2, C2>,
SC: RawStorage<T, R3, C3>,
ShapeConstraint:
SameNumberOfRows<R1, R2> + SameNumberOfColumns<C1, C3> + AreMultipliable<R2, C2, R3, C3>,
{
<span class="kw">let </span>ncols1 = y.ncols();
<span class="attr">#[cfg(feature = <span class="string">"std"</span>)]
</span>{
<span class="comment">// We assume large matrices will be Dyn but small matrices static.
// We could use matrixmultiply for large statically-sized matrices but the performance
// threshold to activate it would be different from SMALL_DIM because our code optimizes
// better for statically-sized matrices.
</span><span class="kw">if </span>R1::is::<Dyn>()
|| C1::is::<Dyn>()
|| R2::is::<Dyn>()
|| C2::is::<Dyn>()
|| R3::is::<Dyn>()
|| C3::is::<Dyn>()
{
<span class="comment">// matrixmultiply can be used only if the std feature is available.
</span><span class="kw">let </span>nrows1 = y.nrows();
<span class="kw">let </span>(nrows2, ncols2) = a.shape();
<span class="kw">let </span>(nrows3, ncols3) = b.shape();
<span class="comment">// Threshold determined empirically.
</span><span class="kw">const </span>SMALL_DIM: usize = <span class="number">5</span>;
<span class="kw">if </span>nrows1 > SMALL_DIM && ncols1 > SMALL_DIM && nrows2 > SMALL_DIM && ncols2 > SMALL_DIM
{
<span class="macro">assert_eq!</span>(
ncols2, nrows3,
<span class="string">"gemm: dimensions mismatch for multiplication."
</span>);
<span class="macro">assert_eq!</span>(
(nrows1, ncols1),
(nrows2, ncols3),
<span class="string">"gemm: dimensions mismatch for addition."
</span>);
<span class="comment">// NOTE: this case should never happen because we enter this
// codepath only when ncols2 > SMALL_DIM. Though we keep this
// here just in case if in the future we change the conditions to
// enter this codepath.
</span><span class="kw">if </span>ncols2 == <span class="number">0 </span>{
<span class="comment">// NOTE: we can't just always multiply by beta
// because we documented the guaranty that `self` is
// never read if `beta` is zero.
</span><span class="kw">if </span>beta.is_zero() {
y.apply(|e| Status::init(e, T::zero()));
} <span class="kw">else </span>{
<span class="comment">// SAFETY: this is UB if Status = Uninit
</span>y.apply(|e| <span class="kw-2">*</span>Status::assume_init_mut(e) <span class="kw-2">*</span>= beta.clone());
}
<span class="kw">return</span>;
}
<span class="kw">if </span>TypeId::of::<T>() == TypeId::of::<f32>() {
<span class="kw">let </span>(rsa, csa) = a.strides();
<span class="kw">let </span>(rsb, csb) = b.strides();
<span class="kw">let </span>(rsc, csc) = y.strides();
matrixmultiply::sgemm(
nrows2,
ncols2,
ncols3,
mem::transmute_copy(<span class="kw-2">&</span>alpha),
a.data.ptr() <span class="kw">as </span><span class="kw-2">*const </span>f32,
rsa <span class="kw">as </span>isize,
csa <span class="kw">as </span>isize,
b.data.ptr() <span class="kw">as </span><span class="kw-2">*const </span>f32,
rsb <span class="kw">as </span>isize,
csb <span class="kw">as </span>isize,
mem::transmute_copy(<span class="kw-2">&</span>beta),
y.data.ptr_mut() <span class="kw">as </span><span class="kw-2">*mut </span>f32,
rsc <span class="kw">as </span>isize,
csc <span class="kw">as </span>isize,
);
<span class="kw">return</span>;
} <span class="kw">else if </span>TypeId::of::<T>() == TypeId::of::<f64>() {
<span class="kw">let </span>(rsa, csa) = a.strides();
<span class="kw">let </span>(rsb, csb) = b.strides();
<span class="kw">let </span>(rsc, csc) = y.strides();
matrixmultiply::dgemm(
nrows2,
ncols2,
ncols3,
mem::transmute_copy(<span class="kw-2">&</span>alpha),
a.data.ptr() <span class="kw">as </span><span class="kw-2">*const </span>f64,
rsa <span class="kw">as </span>isize,
csa <span class="kw">as </span>isize,
b.data.ptr() <span class="kw">as </span><span class="kw-2">*const </span>f64,
rsb <span class="kw">as </span>isize,
csb <span class="kw">as </span>isize,
mem::transmute_copy(<span class="kw-2">&</span>beta),
y.data.ptr_mut() <span class="kw">as </span><span class="kw-2">*mut </span>f64,
rsc <span class="kw">as </span>isize,
csc <span class="kw">as </span>isize,
);
<span class="kw">return</span>;
}
}
}
}
<span class="kw">for </span>j1 <span class="kw">in </span><span class="number">0</span>..ncols1 {
<span class="comment">// TODO: avoid bound checks.
// SAFETY: this is UB if Status = Uninit && beta != 0
</span>gemv_uninit(
status,
<span class="kw-2">&mut </span>y.column_mut(j1),
alpha.clone(),
a,
<span class="kw-2">&</span>b.column(j1),
beta.clone(),
);
}
}
</code></pre></div>
</section></div></main><div id="rustdoc-vars" data-root-path="../../../" data-static-root-path="../../../static.files/" data-current-crate="nalgebra" data-themes="" data-resource-suffix="" data-rustdoc-version="1.67.1 (d5a82bbd2 2023-02-07)" data-search-js="search-444266647c4dba98.js" data-settings-js="settings-bebeae96e00e4617.js" data-settings-css="settings-af96d9e2fc13e081.css" ></div></body></html>