<div dir="ltr"><br><div class="gmail_extra"><br><div class="gmail_quote">On Fri, Apr 29, 2016 at 2:21 PM, Sean Silva <span dir="ltr"><<a href="mailto:chisophugis@gmail.com" target="_blank">chisophugis@gmail.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div dir="ltr"><br><div class="gmail_extra"><br><div class="gmail_quote"><span class="">On Thu, Apr 28, 2016 at 4:23 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">This is the idea we chatted today to try to avoid mallocs os windows.<br>
I was surprised to find this a small speedup up on linux</blockquote><div><br></div></span><div>The numbers you show below indicate a 0.07% difference in performance. It is somewhat spurious to call that "small". More like "negligible".</div><div><br></div><div>To put some numbers on this, CPU's will clock differently depending on the number of cores available</div></div></div></div></blockquote><div><br></div><div>this should say "cores in use"</div><div><br></div><div>-- Sean Silva</div><div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div dir="ltr"><div class="gmail_extra"><div class="gmail_quote"><div>. With only one core being used, they may be at their fastest clock, while if two cores are being used they will clock down by, say, 20% (and even more if more CPU's are being used).</div><div><br></div><div>Assuming that LTO speed is linearly proportional to the clock frequency, then 0.07% performance difference corresponds to a difference of 0.4% in time spent at the highest clock frequency (in absolute terms, this is about 100ms out of a 30s LTO run). E.g. you may have touched the mouse during one run and not the other or you refreshed gmail in one run but not the other.</div><span class="HOEnZb"><font color="#888888"><div><br></div><div>-- Sean Silva</div></font></span><span class=""><div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> (with /tmp in<br>
tmpfs), so I wonder if we are doing something funny in SmallVector.<br>
<br>
Linking llvm-as I got<br>
<br>
master: 29.774786873 seconds<br>
patch 29.752283093 seconds<br>
<br>
Could one of you benchmark this on windows? If it is a performance win<br>
there too I will clean the patch up and send for proper review.<br>
<br>
Cheers,<br>
Rafael<br>
</blockquote></span></div><br></div></div>
</blockquote></div><br></div></div>