<div><div dir="auto">Setting LLVM_PARALLEL_LINK_JOBS<br></div></div><div dir="auto">did not help a week or two weeks ago’s lld.</div><div dir="auto"><br></div><div dir="auto">But recent commits to lld might reflect the variable correctly.</div><div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Thu, Apr 2, 2020 at 22:52 Robinson, Paul <<a href="mailto:paul.robinson@sony.com">paul.robinson@sony.com</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><br>
<br>
> -----Original Message-----<br>
> From: llvm-dev <<a href="mailto:llvm-dev-bounces@lists.llvm.org" target="_blank">llvm-dev-bounces@lists.llvm.org</a>> On Behalf Of Tom Stellard<br>
> via llvm-dev<br>
> Sent: Wednesday, April 1, 2020 7:49 PM<br>
> To: Itaru Kitayama <<a href="mailto:itaru.kitayama@gmail.com" target="_blank">itaru.kitayama@gmail.com</a>><br>
> Cc: Nemanja Ivanovic via llvm-dev <<a href="mailto:llvm-dev@lists.llvm.org" target="_blank">llvm-dev@lists.llvm.org</a>><br>
> Subject: Re: [llvm-dev] LLD issue on a massively parallel build machine<br>
> <br>
> On 04/01/2020 04:12 PM, Itaru Kitayama wrote:<br>
> > On another login node which is 256 (GB)/48 (nodes) JURECA at JSC, I<br>
> never had an LLD issue without setting -j when executing ninja<br>
> > in the past few weeks.<br>
> ><br>
> > On Thu, Apr 2, 2020 at 7:17 AM Itaru Kitayama <<a href="mailto:itaru.kitayama@gmail.com" target="_blank">itaru.kitayama@gmail.com</a><br>
> <mailto:<a href="mailto:itaru.kitayama@gmail.com" target="_blank">itaru.kitayama@gmail.com</a>>> wrote:<br>
> ><br>
> > Tom,<br>
> > Then what ratio do you think it’s minimal?<br>
> ><br>
> <br>
> It really depends on your configuration, but I usually try to have at<br>
> least 2 GB<br>
> of memory per core. However, I usually do Release builds, so Debug builds<br>
> might<br>
> need more. If you aren't using LLVM_PARALLEL_LINK_JOBS it's pretty easy<br>
> to<br>
> run out of memory once ninja starts linking the tools and unittests.<br>
> <br>
> -Tom<br>
<br>
For Debug (or RelWithDebInfo) I usually figure on around 5GB per thread<br>
to avoid swapping. Compiling is never an issue, it's the linking phase<br>
that uses memory. LLVM_PARALLEL_LINK_JOBS works well for ninja builds,<br>
it has been a real help.<br>
--paulr<br>
</blockquote></div></div>