From owner-freebsd-amd64@FreeBSD.ORG  Tue Dec 30 02:16:32 2003
Return-Path: <owner-freebsd-amd64@FreeBSD.ORG>
Delivered-To: freebsd-amd64@freebsd.org
Received: from mx1.FreeBSD.org (mx1.freebsd.org [216.136.204.125])
	by hub.freebsd.org (Postfix) with ESMTP id 3073416A4CE
	for <freebsd-amd64@freebsd.org>; Tue, 30 Dec 2003 02:16:32 -0800 (PST)
Received: from bigtex.jrv.org (rrcs-sw-24-73-246-106.biz.rr.com
	[24.73.246.106])
	by mx1.FreeBSD.org (Postfix) with ESMTP id 218EA43D4C
	for <freebsd-amd64@freebsd.org>;
	Tue, 30 Dec 2003 02:16:31 -0800 (PST)
	(envelope-from james@bigtex.jrv.org)
Received: from bigtex.jrv.org (localhost [127.0.0.1])
	by bigtex.jrv.org (8.12.1/8.12.1) with ESMTP id hBUAGUo8085643
	(version=TLSv1/SSLv3 cipher=EDH-RSA-DES-CBC3-SHA bits=168 verify=NO)
	for <freebsd-amd64@freebsd.org>; Tue, 30 Dec 2003 04:16:30 -0600 (CST)
Received: (from james@localhost)
	by bigtex.jrv.org (8.12.1/8.12.1/Submit) id hBUAGT4Q085640;
	Tue, 30 Dec 2003 04:16:29 -0600 (CST)
Date: Tue, 30 Dec 2003 04:16:29 -0600 (CST)
Message-Id: <200312301016.hBUAGT4Q085640@bigtex.jrv.org>
From: James Van Artsdalen <james-freebsd-amd64@jrv.org>
To: freebsd-amd64@freebsd.org
Subject: Re: libc assembly optimizations?
X-BeenThere: freebsd-amd64@freebsd.org
X-Mailman-Version: 2.1.1
Precedence: list
List-Id: Porting FreeBSD to the AMD64 platform <freebsd-amd64.freebsd.org>
List-Unsubscribe: <http://lists.freebsd.org/mailman/listinfo/freebsd-amd64>,
	<mailto:freebsd-amd64-request@freebsd.org?subject=unsubscribe>
List-Archive: <http://lists.freebsd.org/pipermail/freebsd-amd64>
List-Post: <mailto:freebsd-amd64@freebsd.org>
List-Help: <mailto:freebsd-amd64-request@freebsd.org?subject=help>
List-Subscribe: <http://lists.freebsd.org/mailman/listinfo/freebsd-amd64>,
	<mailto:freebsd-amd64-request@freebsd.org?subject=subscribe>
X-List-Received-Date: Tue, 30 Dec 2003 10:16:32 -0000

Here's an alternative for fabs (3):

ENTRY(fabs)
	psllq	$1,%xmm0	/* 64-bit shifts lefts */
	psrlq	$1,%xmm0	/* logical shift right clears sign */
	ret

/usr/src/lib/libc/amd64/gen/fabs.S does the code below.
gcc generates essentially the same code as below.
The shifts above seem to work and look better to me.

The string ops can made be significantly improved if allowed to
read extra bytes around the string but within the same 16-byte
paragraph as the start or end of the string.  This seems safe in
userland.

Finally, can the SSE2 regs be safely used in kernel mode?
Page fill and aligned-bulk bcopy calls can be improved this way.

/*
 * Ok, this sucks. Is there really no way to push an xmm register onto
 * the FP stack directly?
 */

ENTRY(fabs)
	movsd	%xmm0, -8(%rsp)
	fldl	-8(%rsp)
	fabs
	fstpl	-8(%rsp)
	movsd	-8(%rsp),%xmm0
	ret