Received-SPF: softfail (google.com: domain of transitioning linux-kernel-owner@vger.kernel.org does not designate 23.128.96.19 as permitted sender) client-ip=23.128.96.19;
MIME-Version: 1.0
References: <23e0ba7863d51ab629498762a97d477645aeafea.1653123744.git.christophe.jaillet@wanadoo.fr>
 <CAK7LNASHGyzKS-d+S1GYz_L0pBxxSwk05YfvV+LK4ghZ5xx=cA@mail.gmail.com> <20220523123436.b5gwj55xacc2hxcv@mail.google.com>
In-Reply-To: <20220523123436.b5gwj55xacc2hxcv@mail.google.com>
From:   Arnd Bergmann <arnd@arndb.de>
Date:   Mon, 23 May 2022 14:50:28 +0200
Message-ID: <CAK8P3a2N=azrGfJUShkBkkQerMZ1TZKKbdd_7ZArLrUXNsEPDQ@mail.gmail.com>
Subject: Re: [RFC PATCH] kbuild: Add an option to enable -O1 and speed-up
 compilation time
To:     Changbin Du <changbin.du@gmail.com>
Cc:     Masahiro Yamada <masahiroy@kernel.org>,
        Christophe JAILLET <christophe.jaillet@wanadoo.fr>,
        Dan Carpenter <dan.carpenter@oracle.com>,
        Michal Marek <michal.lkml@markovi.net>,
        Nick Desaulniers <ndesaulniers@google.com>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        kernel-janitors <kernel-janitors@vger.kernel.org>,
        Linux Kbuild mailing list <linux-kbuild@vger.kernel.org>,
        Arnd Bergmann <arnd@arndb.de>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Mon, May 23, 2022 at 2:34 PM Changbin Du <changbin.du@gmail.com> wrote:
> On Sun, May 22, 2022 at 12:38:17AM +0900, Masahiro Yamada wrote:
> > (+CC: Arnd, Changbin Du)
> >
> >
> > If -O1 support does not require additional effort,
> > I have no objection to this patch.
> >
> > (but I do not have enough insight about
> > the compiler's inlining heuristic)
> >
> >
> >
> > BTW, when we attempted to add the -Og support,
> > we fixed various parts, and Linus rejected it.
> >
> > https://lore.kernel.org/linux-kbuild/CAK7LNARQggM3aKEPRKJqa4tunFAfmfErMZuS-rrnRv6UB1VpPQ@mail.gmail.com/
> >
> >
> >
> >
> >
> >
> >
> I am afraid that '-O1' has the same situation with '-Og'. As described in GCC
> mannual:
>
> Like -O0, -Og completely disables a number of optimization passes so that
> individual options controlling them have no effect. Otherwise -Og enables all
> -O1 optimization flags except for those that may interfere with debugging:
>  -fbranch-count-reg  -fdelayed-branch
>  -fdse  -fif-conversion  -fif-conversion2
>  -finline-functions-called-once
>  -fmove-loop-invariants  -fmove-loop-stores  -fssa-phiopt
>  -ftree-bit-ccp  -ftree-dse  -ftree-pta  -ftree-sra

Correct, I'm fairly sure this never worked. It may be possible to narrow down
the individual optimization flags to have something that works inbetween
-O1 and -O2, but that requires a lot of testing work, and it is questionable
whether there are any practical upsides.

Note that we already disable some optimizations for certain configurations,
e.g. CONFIG_READABLE_ASM, CONFIG_FRAME_POINTER or
CONFIG_DEBUG_SECTION_MISMATCH. The gcc manual suggests
listing all optimization flags with

$ gcc -c -Q -O1 --help=optimizers > O1-opts
$ gcc -c -Q -O2 --help=optimizers > O2-opts
$ diff -u O1-opts O2-opts

If the goal is to speed up compilation, there may be value in trying
which of the various options make the most difference here and
still produce a working kernel when disabled. I know we need the
inlining to happen or things go badly wrong, but I don't know if e.g.
-fno-expensive-optimizations would work.

       Arnd