<div dir="ltr"><div class="gmail_extra"><div class="gmail_quote">On Thu, Nov 17, 2016 at 10:27 AM, Renato Golin <span dir="ltr"><<a href="mailto:renato.golin@linaro.org" target="_blank">renato.golin@linaro.org</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">On 17 November 2016 at 17:50, Mehdi Amini via llvm-dev<br>
<span><<a href="mailto:llvm-dev@lists.llvm.org" target="_blank">llvm-dev@lists.llvm.org</a>> wrote:<br>
> It depends on the use-case: with ThinLTO we scale linearly with the number<br>
> of physical cores. When you get over the number of physical cores you still<br>
> get some improvements, but that’s no longer linear.<br>
<br>
</span>Indeed, in HT, cores have two execution units on the same cache/bus<br>
line, so memory access is likely to be contrived. Linkers are memory<br>
hungry, which add to the I/O bottleneck which makes most of the gain<br>
disappear. :)<br>
<br>
Furthermore, the FP unit is also shared among the ALUs, so<br>
FP-intensive code does not make good use of HT. Not the case, here,<br>
though.<br></blockquote><div><br></div><div>We literally have no variables of type float or double in LLD. It would work fine on 486SX. :)</div></div></div></div>