From owner-cvs-sys  Wed Sep  4 15:00:41 1996
Return-Path: owner-cvs-sys
Received: (from root@localhost)
          by freefall.freebsd.org (8.7.5/8.7.3) id PAA04961
          for cvs-sys-outgoing; Wed, 4 Sep 1996 15:00:41 -0700 (PDT)
Received: from godzilla.zeta.org.au (godzilla.zeta.org.au [203.2.228.19])
          by freefall.freebsd.org (8.7.5/8.7.3) with SMTP id PAA04954;
          Wed, 4 Sep 1996 15:00:35 -0700 (PDT)
Received: (from bde@localhost) by godzilla.zeta.org.au (8.6.12/8.6.9) id HAA19084; Thu, 5 Sep 1996 07:51:28 +1000
Date: Thu, 5 Sep 1996 07:51:28 +1000
From: Bruce Evans <bde@zeta.org.au>
Message-Id: <199609042151.HAA19084@godzilla.zeta.org.au>
To: bde@zeta.org.au, cvs-all@freefall.freebsd.org,
        CVS-committers@freefall.freebsd.org, cvs-sys@freefall.freebsd.org,
        julian@freefall.freebsd.org
Subject: Re: cvs commit:  src/sys/i386/boot/biosboot Makefile boot.c boot2.S
Sender: owner-cvs-sys@FreeBSD.ORG
X-Loop: FreeBSD.org
Precedence: bulk

I wrote:
>In the old generated code, howto is kept in a register and flags are
>ORed directly to memory.  This takes only 3 bytes per OR for flags
><= 0x80 (using an 8 bit constant).  It should take only 3 bytes for
    0xffu
>all (single) flags (using an 8 bit constant at an offset), but gcc
>generates poor code for this case (it generates a 32 but constant
>and 6 bytes per OR).  ...

The new code copies loadflags to a local varable, so the new generated
code is probably about as mediocre as the old generated code.  It can
be better iff the local variable can be allocated in %eax and most of
the constants are < 0x100u.  Then the ORing code can take 2 bytes:

	orb	$const,%al		// 2 bytes

For other registers it will be just as large as the indirect method:

	orb	$const,%dl		// 3 bytes
	orb	$const,0(%anyreg)	// 3 bytes

For large constants, best code is larger with a register than indirectly:

	orw	$bigconst,%ax		// 4 bytes (16 bit constant)
	orw	$bigconst,%dx		// 5 bytes (16 bit constant)
	orl	$bigconst,%eax		// 5 bytes (> 16-bit constant)
	orl	$bigconst,%edx		// 6 bytes (> 16-bit constant)
	orb	$bigconst>>8,1(%anyreg)	// 3 bytes
					// (0x100u <= bigconst < 0x10000u)

Bruce