Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <CAPM=9twNnV4zMCvrPkw3H-ajZOH-01JVh_kDrxdPYQErz8ZTdA@mail.gmail.com>
 <CAHk-=wi=eDN4Ub0qSN27ztBAvHSXCyiY2stu3_XbTpYpQX4x7w@mail.gmail.com>
 <CAHk-=wg9p+oJtzC0582vke8whcWPNSC9eq0Z6kdnSqWMEpah7g@mail.gmail.com> <CAKwvOd=dUWWSzQEWdguhe3SjYTw7a76skx370DbU2Y+5o=M8BA@mail.gmail.com>
In-Reply-To: <CAKwvOd=dUWWSzQEWdguhe3SjYTw7a76skx370DbU2Y+5o=M8BA@mail.gmail.com>
From:   Linus Torvalds <torvalds@linux-foundation.org>
Date:   Fri, 23 Jun 2023 12:33:58 -0700
Message-ID: <CAHk-=wgsK5H=ejAfFaABzK5PCb3=wqarq+L5frcTjRJkON9S+Q@mail.gmail.com>
Subject: Re: arm32 build warnings in workqueue.c
To:     Nick Desaulniers <ndesaulniers@google.com>
Cc:     Dave Airlie <airlied@gmail.com>,
        Nathan Chancellor <nathan@kernel.org>,
        Arnd Bergmann <arnd@arndb.de>, Tejun Heo <tj@kernel.org>,
        LKML <linux-kernel@vger.kernel.org>,
        clang-built-linux <llvm@lists.linux.dev>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Fri, 23 Jun 2023 at 12:16, Nick Desaulniers <ndesaulniers@google.com> wrote:
>
> >
> > -       movabsq $137438953440, %rcx             # imm = 0x1FFFFFFFE0
> > -       andq    %rax, %rcx
> > -       movabsq $68719476704, %rdx              # imm = 0xFFFFFFFE0
> > -       cmpq    %rdx, %rcx
> > +       shrq    $5, %rax
> > +       cmpl    $2147483647, %eax               # imm = 0x7FFFFFFF
> >
> > in several places.
>
> Sorry, are those equivalent?  Before looks to me like:
>
> if (0xFFFFFFFE0 - (0x1FFFFFFFE0 & rax))
>
> and after
>
> if (0x7FFFFFFF - (rax >> 5))

I cut off some of the other changes, with the old (confused) code also doing

-       shrq    $5, %rax

later.

And yes, as far as I can tell, the comparisons are 100% equivalent.
Look at what happens when you shift that big mask down by 5..

> > My guess is that clang keeps an enum as an enum as long as possible -
> > including past some (really) simple simplification phases of the
> > optimizer.
>
> I don't think so.
> https://godbolt.org/z/M8746c49z
> That's LLVM IR as soon as it leaves the front end.

Hmm. I have

     clang version 15.0.7 (Fedora 15.0.7-2.fc37)

and you can probably check the code generation of the current kernel
with and without that patch.

It might depend on exact config file - I'll send you mine in a
separate email to not pollute the mailing list.

                Linus