<div dir="ltr">Have you considered using OpenSSL's implementation? (just wondering, I don't know much about OpenSSL, but I guess their code is optimized.)</div><div class="gmail_extra"><br><div class="gmail_quote">On Wed, Nov 16, 2016 at 4:16 PM, Mehdi Amini <span dir="ltr"><<a href="mailto:mehdi.amini@apple.com" target="_blank">mehdi.amini@apple.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div style="word-wrap:break-word">No, otherwise I’d have pick it instead of this one ;-)<div><br><div>The alternative plan was either to:</div><div>1) reach out to someone who has written an optimized one and convince him to contribute it to LLVM</div><div>2) “optimize” the one in tree.</div><div><br></div><div>But not high enough in my priority list.</div><div><br></div><div>— </div><span class="HOEnZb"><font color="#888888"><div>Mehdi</div></font></span><div><div class="h5"><div><br><div><blockquote type="cite"><div>On Nov 16, 2016, at 4:13 PM, Rui Ueyama <<a href="mailto:ruiu@google.com" target="_blank">ruiu@google.com</a>> wrote:</div><br class="m_5123718511511992452Apple-interchange-newline"><div><div dir="ltr">I should've said that do you know if there's an optimized SHA1 implementation that we can use?</div><div class="gmail_extra"><br><div class="gmail_quote">On Wed, Nov 16, 2016 at 4:11 PM, Mehdi Amini <span dir="ltr"><<a href="mailto:mehdi.amini@apple.com" target="_blank">mehdi.amini@apple.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div style="word-wrap:break-word">The current implementation was “copy/pasted” from somewhere (it was explicitly public domain).<div><div class="m_5123718511511992452h5"><div><br><div><div><blockquote type="cite"><div>On Nov 16, 2016, at 4:05 PM, Rui Ueyama <<a href="mailto:ruiu@google.com" target="_blank">ruiu@google.com</a>> wrote:</div><br class="m_5123718511511992452m_838021890767585268Apple-interchange-newline"><div><div dir="ltr">Can we just copy-and-paste optimized code from somewhere?</div><div class="gmail_extra"><br><div class="gmail_quote">On Wed, Nov 16, 2016 at 4:03 PM, Mehdi Amini <span dir="ltr"><<a href="mailto:mehdi.amini@apple.com" target="_blank">mehdi.amini@apple.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div style="word-wrap:break-word">SHA1 in LLVM is *very* naive, any improvement is welcome there!<div>It think Amaury pointed it originally and he had an alternative implementation IIRC.</div><div><br></div><div>— </div><div>Mehdi</div><div><br><div><blockquote type="cite"><div><div class="m_5123718511511992452m_838021890767585268h5"><div>On Nov 16, 2016, at 3:58 PM, Rui Ueyama via llvm-dev <<a href="mailto:llvm-dev@lists.llvm.org" target="_blank">llvm-dev@lists.llvm.org</a>> wrote:</div><br class="m_5123718511511992452m_838021890767585268m_-254240913159878184Apple-interchange-newline"></div></div><div><div><div class="m_5123718511511992452m_838021890767585268h5"><div dir="ltr">By the way, while running benchmark, I found that our SHA1 function seems much slower than the one in gold. gold slowed down by only 1.3 seconds to compute a SHA1 of output, but we spent 6.0 seconds to do the same thing (I believe). Something doesn't seem right.<div><br></div><div>Here is a table to link the same binary with -no-threads and -build-id={none,md5,sha1}. The numbers are in seconds.</div><div><br></div><div><div><font face="monospace, monospace"> LLD gold</font></div><div><font face="monospace, monospace">none 7.82 13.78</font></div><div><font face="monospace, monospace">MD5 9.68 14.56</font></div><div><font face="monospace, monospace">SHA1 13.85 15.05</font></div></div><div><br></div></div><div class="gmail_extra"><br><div class="gmail_quote">On Wed, Nov 16, 2016 at 1:46 PM, Rafael Espíndola <span dir="ltr"><<a href="mailto:rafael.espindola@gmail.com" target="_blank">rafael.espindola@gmail.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">On 16 November 2016 at 15:52, Rafael Espíndola<br>
<span><<a href="mailto:rafael.espindola@gmail.com" target="_blank">rafael.espindola@gmail.com</a>> wrote:<br>
> I will do a quick benchmark run.<br>
<br>
<br>
</span>On a mac pro (running linux) the results I got with all cores available:<br>
<br>
firefox<br>
master <a href="tel:7.146418217" value="+17146418217" target="_blank">7.146418217</a><br>
patch 5.<a href="tel:304271767%201" value="+13042717671" target="_blank">304271767 1</a>.34729488437x faster<br>
firefox-gc<br>
master <a href="tel:7.316743822" value="+17316743822" target="_blank">7.316743822</a><br>
patch 5.46436812 1.33899174824x faster<br>
chromium<br>
master 4.265597914<br>
patch 3.972218527 1.07385781648x faster<br>
chromium fast<br>
master 1.823614026<br>
patch 1.686059427 1.08158348205x faster<br>
the gold plugin<br>
master 0.340167513<br>
patch 0.318601465 1.06768973269x faster<br>
clang<br>
master 0.579914119<br>
patch 0.520784947 1.11353855817x faster<br>
llvm-as<br>
master 0.03323043<br>
patch 0.041571719 1.251013574x slower<br>
the gold plugin fsds<br>
master 0.36675887<br>
patch 0.350970944 1.04498356992x faster<br>
clang fsds<br>
master 0.656180056<br>
patch 0.591607603 1.10914743602x faster<br>
llvm-as fsds<br>
master 0.030324313<br>
patch 0.040045353 1.32056917497x slower<br>
scylla<br>
master 3.23378908<br>
patch 2.019191831 1.60152642773x faster<br>
<br>
With only 2 cores:<br>
<br>
firefox<br>
master 7.174839911<br>
patch 6.319808477 1.13529388384x faster<br>
firefox-gc<br>
master 7.345525844<br>
patch 6.493005841 1.13129820362x faster<br>
chromium<br>
master 4.180752414<br>
patch 4.129515199 1.01240756179x faster<br>
chromium fast<br>
master 1.847296843<br>
patch 1.78837299 1.0329483018x faster<br>
the gold plugin<br>
master 0.341725451<br>
patch 0.339943222 1.0052427255x faster<br>
clang<br>
master 0.581901114<br>
patch 0.566932481 1.02640284955x faster<br>
llvm-as<br>
master 0.03381059<br>
patch 0.036671392 1.08461260215x slower<br>
the gold plugin fsds<br>
master 0.369184003<br>
patch 0.368774353 1.00111084189x faster<br>
clang fsds<br>
master 0.660120583<br>
patch 0.641040511 1.02976422187x faster<br>
llvm-as fsds<br>
master 0.031074029<br>
patch 0.035421531 1.13990789543x slower<br>
scylla<br>
master 3.243011681<br>
patch 2.630991522 1.23261958615x faster<br>
<br>
<br>
With only 1 core:<br>
<br>
firefox<br>
master 7.174323116<br>
patch 7.301968002 1.01779190649x slower<br>
firefox-gc<br>
master 7.339104117<br>
patch 7.466171668 1.01731376868x slower<br>
chromium<br>
master 4.176958448<br>
patch 4.188387233 1.00273615003x slower<br>
chromium fast<br>
master 1.848922713<br>
patch 1.858714219 1.00529578978x slower<br>
the gold plugin<br>
master 0.342383846<br>
patch 0.347106743 1.01379415838x slower<br>
clang<br>
master 0.582476955<br>
patch 0.600524655 1.03098440178x slower<br>
llvm-as<br>
master 0.033248459<br>
patch 0.035622988 1.07141771593x slower<br>
the gold plugin fsds<br>
master 0.369510236<br>
patch 0.376390506 1.01861997133x slower<br>
clang fsds<br>
master 0.661267753<br>
patch 0.683417482 1.03349585535x slower<br>
llvm-as fsds<br>
master 0.030574688<br>
patch 0.033052779 1.08105041006x slower<br>
scylla<br>
master 3.236604638<br>
patch 3.325831407 1.02756801617x slower<br>
<br>
Given that we have an improvement even with just two cores available, LGTM.<br>
<br>
Cheers,<br>
Rafael<br>
</blockquote></div><br></div></div></div>
______________________________<wbr>_________________<br>LLVM Developers mailing list<br><a href="mailto:llvm-dev@lists.llvm.org" target="_blank">llvm-dev@lists.llvm.org</a><br><a href="http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev" target="_blank">http://lists.llvm.org/cgi-bin/<wbr>mailman/listinfo/llvm-dev</a><br></div></blockquote></div><br></div></div></blockquote></div><br></div>
</div></blockquote></div><br></div></div></div></div></div></blockquote></div><br></div>
</div></blockquote></div><br></div></div></div></div></div></blockquote></div><br></div>