Re: llvm10 build failure on Rpi3

From: Denis Ovsienko <denis_at_ovsienko.info>
Date: Wed, 23 Jun 2021 19:19:20 +0100
On Wed, 23 Jun 2021 10:43:38 -0700
bob prohaska <fbsd_at_www.zefox.net> wrote:

> A Pi4 doing a bulk build of chromium, lxqt and apache has gone far
> past that point building llvm10, suggesting the fault lies somewhere
> in my setup.

In case this information helps, I occasionally build Clang/LLVM on
NetBSD/AArch64. When trying to do that directly on the RPI3B, the main
difficulty was that cc1plus process typically grows 500-700MB big before
it produces an .o file, so instead of -j4 it had to run -j1 to avoid
swapping (1GB RAM). What's worse, for some files the process size
reached 1500MB anyway, and the incurred swapping was effectively
blocking any progress.

After waiting a week for the build to finish I spawned another NetBSD
in a Cortex-A53 QEMU VM. The emulated CPU core performance measured at
40% the RPI3 CPU core performance in my [rather crude] test, but this
way you can emulate as many cores and as much RAM as the host system
can provide. In my case 4 cores and 8GB of emulated RAM were enough to
build both LLVM and Clang in acceptable time. The resulting packages
install and work on RPI3B as expected.

As far as I understand, much of this logic should apply to FreeBSD as
well. Could you please confirm if any resources were maxing out during
the failed build on RPI3B?

-- 
    Denis Ovsienko
Received on Wed Jun 23 2021 - 18:19:20 UTC

Original text of this message