On Wed, Jul 10, 2019 at 10:38:27PM +0100, Ken Moffat via blfs-dev wrote:
>
> > Now I'm trying to do a Python-3.7.4 build with LTO (its configure script now
> > supports --with-lto to do a LTO build).
>
> A quick test after one of my early builds had completed confirmed it
> (3.7.3, I think) used PGO (Profile Guided Optimization, for those not
> familiar with that TLA) and the build takes a _lot_ longer.
>
> I build Python3 twice: once in LFS, and then fairly early in BLFS
> after I've built extra dependencies. Was thinking about using LTO
> for the second build, but at the moment I don't use Python3 much and
> I doubt that the benefits when running things like Meson will be more
> than a few milliseconds per package.
>
I started by looking at 2.7.16, where 'expensive optimizations' are
described as PGO and possibly LTO. In 3.7.3 (sic, no real point
moving on in the test builds, I'll be careful :) it doesn't mention
LTO for that. Anyway, gave that a go with -O3. Confirmed that LTO
is NOT used unless --with-lto is supplied, so this is only adding
PGO.
Times in seconds for configure ; make -j8 ; make install
2.7.16 3.7.3
------ -----
march=native -O2 cheap hardening, no PGO 50.371s 76.844s
march=native -O3 cheap hardening, PGO 668.128s 1810.117s
Somehow, I don't think I'll be using PGO again for the pythons.
A lot of the PGO part seems to only use one core.
ĸen
--
This is magic for grown-ups; it has to be hard because we know there's
no such thing as a free goblin.
-- Pratchett, Stewart & Cohen - The Science of Discworld II
--
http://lists.linuxfromscratch.org/listinfo/blfs-dev
FAQ: http://www.linuxfromscratch.org/blfs/faq.html
Unsubscribe: See the above information page
