Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Subject: Re: [PATCH 4/4] MIPS: Avoid DIVU in `__div64_32' is result would be zero
Mime-Version: 1.0 (Mac OS X Mail 9.3 \(3124\))
Content-Type: text/plain; charset=us-ascii
From:   "H. Nikolaus Schaller" <hns@goldelico.com>
In-Reply-To: <alpine.DEB.2.21.2104200331110.44318@angie.orcam.me.uk>
Date:   Wed, 21 Apr 2021 18:05:37 +0200
Cc:     Arnd Bergmann <arnd@arndb.de>,
        Thomas Bogendoerfer <tsbogend@alpha.franken.de>,
        Huacai Chen <chenhuacai@kernel.org>,
        Huacai Chen <chenhuacai@loongson.cn>,
        Jiaxun Yang <jiaxun.yang@flygoat.com>,
        linux-arch@vger.kernel.org, linux-mips@vger.kernel.org,
        linux-kernel@vger.kernel.org
Content-Transfer-Encoding: quoted-printable
Message-Id: <284CBC37-0F4F-4077-A172-7E47C90B8B43@goldelico.com>
References: <alpine.DEB.2.21.2104200044060.44318@angie.orcam.me.uk> <alpine.DEB.2.21.2104200331110.44318@angie.orcam.me.uk>
To:     "Maciej W. Rozycki" <macro@orcam.me.uk>
Precedence: bulk


> Am 20.04.2021 um 04:50 schrieb Maciej W. Rozycki <macro@orcam.me.uk>:
>=20
> We already check the high part of the divident against zero to avoid =
the=20

nit-picking: s/divident/dividend/

(seems to come from from Latin "dividendum" =3D the number that is to be =
divided).

> costly DIVU instruction in that case, needed to reduce the high part =
of=20
> the divident, so we may well check against the divisor instead and set=20=

> the high part of the quotient to zero right away.  We need to treat =
the=20
> high part the divident in that case though as the remainder that would=20=

> be calculated by the DIVU instruction we avoided.
>=20
> This has passed correctness verification with test_div64 and reduced =
the
> module's average execution time down to 1.0445s and 0.2619s from =
1.0668s
> and 0.2629s respectively for an R3400 CPU @40MHz and a 5Kc CPU =
@160MHz.

Impressive.

>=20
> Signed-off-by: Maciej W. Rozycki <macro@orcam.me.uk>
> ---
> I have made an experimental change on top of this to put `__div64_32' =
out=20
> of line, and that increases the averages respectively up to 1.0785s =
and=20
> 0.2705s.  Not a terrible loss, especially compared to generic times =
quoted=20
> with 3/4, but still, so I think it would best be made where optimising =
for=20
> size, as noted in the cover letter.
> ---
> arch/mips/include/asm/div64.h |    6 ++++--
> 1 file changed, 4 insertions(+), 2 deletions(-)
>=20
> Index: linux-3maxp-div64/arch/mips/include/asm/div64.h
> =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
> --- linux-3maxp-div64.orig/arch/mips/include/asm/div64.h
> +++ linux-3maxp-div64/arch/mips/include/asm/div64.h
> @@ -68,9 +68,11 @@
> 									=
\
> 	__high =3D __div >> 32;						=
\
> 	__low =3D __div;							=
\
> -	__upper =3D __high;						=
\
> 									=
\
> -	if (__high) {							=
\
> +	if (__high < __radix) {						=
\
> +		__upper =3D __high;					=
\
> +		__high =3D 0;						=
\
> +	} else {							=
\
> 		__asm__("divu	$0, %z1, %z2"				=
\
> 		: "=3Dx" (__modquot)					=
\
> 		: "Jr" (__high), "Jr" (__radix));			=
\