Date: Thu, 13 Oct 1994 08:21:58 +0100
From: terra AT diku DOT dk
To: dj AT stealth DOT ctron DOT com
Cc: babcock AT cfa DOT harvard DOT edu, buers AT dg1 DOT chemie DOT uni-konstanz DOT de,
        djgpp AT sun DOT soe DOT clarkson DOT edu
Subject: Intel's Manuals on Conditional Jumps Like Jcxz


Hi!

There seem to be some confusion about the movedata.s bug.  Let
me first say that it *is* a bug and has *nothing* to do with
recompilation of the libraries.

The Intel manuals say that the operation is

	if <condition> then
	  eip = eip + signextend(rel8/rel16/rel32)
          if <operandsize=16> then
            eip = eip & 0000FFFFh

modulo my memory.  Note that it is the *operand size* (not
the address size, not the size of the relative address) that
determines whether the high bits are masked.  This means that
"Jcxz" (not "Jecxz") is no good in 32-bit programs.

(And by the way, there is no Jcxz that has a 16-bit or 32-bit
relative address.)

In libc.a you will twice find the sequence

		66 E2? 02	Jcxz over
		?? ??		Rep movs?
	over:	...


Note that the operand size prefix (66) is present (to select cx
over ecx) and that the high bits are therefore masked.  That is
bad, and incidently: "jz over" would have worked just find.

I hope I have spread some light.

Morten Welinder
terra AT diku DOT dk