From: "M. Schulter" <mschulter AT value DOT net>
Newsgroups: comp.os.msdos.djgpp
Subject: fsdb question: cltd shown as cwd
Date: 9 Oct 1997 16:45:19 GMT
Organization: Value Net Internetwork Services Inc.
Lines: 77
Message-ID: <61j1mv$fo6$1@vnetnews.value.net>
NNTP-Posting-Host: value.net
To: djgpp AT delorie DOT com
DJ-Gateway: from newsgroup comp.os.msdos.djgpp
Precedence: bulk

Hi, there.

This message is to ask a question about the way that fsdb represents
the `cltd' instruction in an AT&T assembler program for GAS as `cdw'
instead of `cdq' in its Intel (or Intel-like?) disassembly display.

According to

info as machine i386 i386-opcodes

the AT&T `cltd' instruction is equivalent to Intel `cdq': a 32-bit
integer gets sign-extended to 64 bits. So far, I've used this before a
64-bit division, as in the attached brief program.

For some reason, however, fsdb shows the instruction as `cwd' -- which
sign-extends a word to a dword (in other words, a word to a long or
int in GAS terminology).

This happens for me with DJGPP 2.01, using a version of fsdb
(identified on the System Info screen as v. 1.00) with a date for the
executable of 31 October 1996.

Since I'm just learning assembly, I wonder if this is indeed
unexpected behavior in fsdb, a great debugger which I have found
indispensable both in routine learning and bug-detecting.

By the way, my apologies for any infelicities or nonobvious bugs in
the program which follows; when I compile it, fsdb runs it fine, but
displays my `cltd' as `cdw' rather than `cdq'.

Most appreciatively,

Margo Schulter
mschulter AT value DOT net 

/*
 *	cltdtst.s
 *
 *	This file is designed to document an interesting result with
 *	fsdb version 1.00 distributed with DJGPP v.2.01: the AT&T
 *	assembly opcode 'cltd' displays as 'cwd' ("sign-extend word
 *	to double") instead of the expected 'cdq' ("sign-extend dword
 *	to quad").
 *
 *	To compile, please use the following DJGPP command line:
 *
 *	gcc -o cltdtst.exe cltdtst.s
 *
 *      Margo Schulter       mschulter AT value DOT net       8 October 1997
 *
 * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */

.file	"cltdtst.s"
.data
string0:	
	.ascii "\n%d/%d = %d, and %d mod %d = %d\n\0"
.text
.global	_main
_main:
	pushl	%ebp		# save stack pointers
	movl	%esp, %ebp	# for `leave' before exiting
	movl	$-1500000001, %eax	# a 32-bit signed dividend
	movl	%eax, %esi	# move a copy of %eax into %esi
	movl	$3, %ebx	# 3 is our divisor
	cltd			# sign-extend %eax to %edx:%eax
	idivl	%ebx		# and do 64-bit division
	pushl	%edx		# %edx is remainder
	pushl	%ebx		# %ebx is divisor
	pushl	%esi		# %esi has copy of dividend
	pushl	%eax		# %eax is quotient
	pushl	%ebx		# %ebx again has divisor
	pushl	%esi		# %esi again has dividend
	pushl	$string0	# string includes six format specifiers
	call	_printf		# for our six integers from stack
	leave			# after _printf, we restore stack frame
	ret