From mboxrd@z Thu Jan  1 00:00:00 1970
X-Spam-Checker-Version: SpamAssassin 3.4.4 (2020-01-24) on inbox.vuxu.org
X-Spam-Level: 
X-Spam-Status: No, score=-3.3 required=5.0 tests=MAILING_LIST_MULTI,
	RCVD_IN_DNSWL_MED,RCVD_IN_MSPIKE_H3,RCVD_IN_MSPIKE_WL autolearn=ham
	autolearn_force=no version=3.4.4
Received: (qmail 14497 invoked from network); 3 Aug 2021 20:27:50 -0000
Received: from mother.openwall.net (195.42.179.200)
  by inbox.vuxu.org with ESMTPUTF8; 3 Aug 2021 20:27:50 -0000
Received: (qmail 9256 invoked by uid 550); 3 Aug 2021 20:27:48 -0000
Mailing-List: contact musl-help@lists.openwall.com; run by ezmlm
Precedence: bulk
List-Post: <mailto:musl@lists.openwall.com>
List-Help: <mailto:musl-help@lists.openwall.com>
List-Unsubscribe: <mailto:musl-unsubscribe@lists.openwall.com>
List-Subscribe: <mailto:musl-subscribe@lists.openwall.com>
List-ID: <musl.lists.openwall.com>
Reply-To: musl@lists.openwall.com
Received: (qmail 9235 invoked from network); 3 Aug 2021 20:27:47 -0000
Date: Tue, 3 Aug 2021 22:27:35 +0200
From: Szabolcs Nagy <nsz@port70.net>
To: Stefan Kanthak <stefan.kanthak@nexgo.de>
Cc: musl@lists.openwall.com
Message-ID: <20210803202735.GA37904@port70.net>
Mail-Followup-To: Stefan Kanthak <stefan.kanthak@nexgo.de>,
	musl@lists.openwall.com
References: <04BD4026EE364FF7AFBAF8C593E9A2E7@H270>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <04BD4026EE364FF7AFBAF8C593E9A2E7@H270>
Subject: Re: [musl] [Patch] src/math/i386/remquo.s: remove conditional
 branch, shorter bit twiddling

* Stefan Kanthak <stefan.kanthak@nexgo.de> [2021-08-01 17:59:52 +0200]:
> Halve the number of instructions (from 12 to 6) to fetch the
> (3-bit partial) quotient from the FPU flags C0:C3:C1, and
> perform its negation without conditional branch.

i haven't tested it but it looks good.

i think we should not tweak x87 asm code too much though.
it can introduce bugs and there are not many users of it.
i think only the size saving can justify keeping any i386
math code at all.

but i'm not against committing this.
thanks for the patch.

> --- -/math/i386/remquo.s
> +++ +/math/i386/remquo.s
> @@ -2,49 +2,44 @@
>  .type remquof,@function
>  remquof:
>         mov 12(%esp),%ecx
> +       mov 8(%esp),%eax
> +       xor 4(%esp),%eax
>         flds 8(%esp)
>         flds 4(%esp)
> -       mov 11(%esp),%dh
> -       xor 7(%esp),%dh
> -       jmp 1f
> +       jmp 0f
> 
>  .global remquol
>  .type remquol,@function
>  remquol:
>         mov 28(%esp),%ecx
> +       mov 24(%esp),%eax
> +       xor 12(%esp),%eax
> +       cwtl
>         fldt 16(%esp)
>         fldt 4(%esp)
> -       mov 25(%esp),%dh
> -       xor 13(%esp),%dh
> -       jmp 1f
> +       jmp 0f
> 
>  .global remquo
>  .type remquo,@function
>  remquo:
>         mov 20(%esp),%ecx
> +       mov 16(%esp),%eax
> +       xor 8(%esp),%eax
>         fldl 12(%esp)
>         fldl 4(%esp)
> -       mov 19(%esp),%dh
> -       xor 11(%esp),%dh
> +0:     cltd
>  1:     fprem1
>         fnstsw %ax
>         sahf
>         jp 1b
>         fstp %st(1)
> -       mov %ah,%dl
> -       shr %dl
> -       and $1,%dl
> -       mov %ah,%al
> -       shr $5,%al
> -       and $2,%al
> -       or %al,%dl
> -       mov %ah,%al
> -       shl $2,%al
> -       and $4,%al
> -       or %al,%dl
> -       test %dh,%dh
> -       jns 1f
> -       neg %dl
> -1:     movsbl %dl,%edx
> -       mov %edx,(%ecx)
> +       adc %al,%al
> +       shl $2,%ah
> +       adc %al,%al
> +       shl $5,%ah
> +       adc %al,%al
> +       and $7,%eax
> +       xor %edx,%eax
> +       sub %edx,%eax
> +       mov %eax,(%ecx)
>         ret