Re: gcc-7.1.0 fails to build with flto profiled bootstrap build (with gold linker) + graphite optimisations

[Branko <brankob at s5tehnika dot net>]

I thought that no one answered because I asked something that was documented
somewhere and I simply overlooked it.

Can't find the link now, but the gist of it is that one has to inform 
configure about what he wants
and for lto use the right --with-build-config option, which chooses 
desired config/***.mk file.
Which sets the right xxFLAGS for each stage.

One shouldn't just forcefeed the make his own LTO combinations.

Well, I cheated a bit there. I forced -O3 instead of -O2, but it seems 
to be working just fine.

Branko

On 16. 07. 2017 11:23, Brian Drummond wrote:
> On Sun, 2017-07-16 at 09:47 +0000, Branko wrote:
> > Eh, nevermind. I've found the link with hints how to do it properly. It
> > seems to be building OK now. With graphite, LTO and PGO.
> When following up, it would be useful to add the link you found so that
> it becomes part of the public record, for anyone else with a similar
> problem, who finds this thread through a search.
>
> -- Brian
>
> > On 14. 07. 2017 05:15, Branko wrote:
> > > I have adapted existing gcc-7.1.0 ebuild and flag-o-matic eclass so
> > > that it allows me to build gcc with my flags without filtering them.
> > >
> > > It works fine, when built with prefiltered flags, but I wanted to
> > > compile it with flto and pgo steps ( and later do same with binutils)
> > > to see if that would shorten long compile times.
> ...
> > > Has anyone tried to compile gcc-7 with such combination ?
> > >
> > > I used gcc-7.1.0 for compilation, along with binutils-2.28