Date: Wed, 1 Feb 95 19:52:23 PST
From: paktor@hal.com (David Paktor)
Subject: Item #226: "Split" and "Join" words for 64-Bit Extension


P1275 Open Firmware Working Group Proposal -- Proposal #:226 Ver 1

Title:		"Split" and "Join" words for 64-Bit Extension

Author:		David L. Paktor

Date:		Wed 01 Feb 95

Ed/Tech:	Editorial

Synopsis:	Clarify the phrasing, in the 64-Bit Extension, of the descrip-
		    tions of  bxjoin  lxjoin  wxjoin  xbsplit  xlsplit  and
		     xwsplit  along the lines of the clarifications amended to
		    the Core document.

Doc & Version:	1275.6 D4

Problem:	The descriptions of  bxjoin  lxjoin  wxjoin  xbsplit  xlsplit
		    and  xwsplit , if taken literally, would be inaccurate.

Proposal:  Change the text on the following page:lines as shown:

    pg:ln#	Word Described
		Changed Text

     2:39	bxjoin
		The eight least significant bits of each operand are combined
		    to form an octlet.  Other operand bits are ignored.

     2:51	lxjoin
		The thirty-two least significant bits of each operand are com-
		   bined to form an octlet.  Other operand bits are ignored.

     3:35	wxjoin
		The sixteen least significant bits of each operand are com-
		    bined to form an octlet.  Other operand bits are ignored.

     4:06	xbsplit
		The bits of greater significance than the eight least
		    significant bits of each of the eight resulting values
		    will be zero.

     4:29	xlsplit
		The bits of greater significance than the thirty-two least
		    significant bits of each of the two resulting values
		    will be zero.

     4:52	xwsplit
		The bits of greater significance than the sixteen least
		    significant bits of each of the four resulting values
		    will be zero.


Discussion:  Perhaps we should highlight the subtle but potentially important
	shift between the "join" words in the 64-bit supplement and those in
	the Core:  The Core specifies that the other operand bits "must be
	zero", i.e., it is the responsibility of the user to make sure they
	are zero before applying an XXjoin word; in the 64-bit supplement, on
	the other hand, the implication is that the iplementation will take
	care of that step.

    In point of fact, I have just tested an implementation in which the other
	operand bits are not ignored.  Should we modify the phrases that say
	"are ignored" to read "must be zero"?

    I would prefer to place the burden of trimming out the "other operand
	bits" on the implementation, rather than on the application, but
	it might be more important to maintain a consistent policy.


David

[ P1275 Item #226 -- Received: Wed Feb  1 19:54:24 PST 1995 ]