Date: Tue, 11 Oct 94 20:42:47 +0100
From: buers AT dg1 DOT chemie DOT uni-konstanz DOT de (Dieter Buerssner)
To: babcock AT cfa DOT harvard DOT edu
Cc: djgpp AT sun DOT soe DOT clarkson DOT edu
Subject: Re: bug in movedata (or in gas), with fix


Dear Bob Babcock, 

 > In the file dosmem.s in libsrc/c/dos gas produces for the jcxz zero
 > instructions code for a jump with 16 bit displacements. This will mask
 > the high word of EIP, so (when cx is zero) it will jump from say
 > 0x00012345 to 0x00002347 instead of the correct 0x00012347.  In the
 > gas manual, there is mentioned, that using jcxz may be risky, but I
 > really don't understand what they mean. The code should get assembled
 > correctly (with 8-bit displacement, "short" jump).  The following
 > patch seems to fix the problem (using jz instead of jcxz).

 I just looked at some of my code with the debugger.  The two jcxz
 instructions in dosmem.s seem to be assembled with 8-bit displacements in the
 standard distribution, not 16-bit.  Did you rebuild the library, and if so,
 did you use a different version of gas than DJ uses?

You are right, I actually recompiled the library. I thought I recompiled
with the newest tools from djgpp1.12 (I have no space for old versions),
but I will check the version of gas I used.

You might have seen a similar message Sven Palme
<palme AT irz DOT hu-berlin DOT de>, sent to the djgpp mailing list on 10th Oct.,
who noted the 0x66 Prefix for the jcxz instruction as well. 

Regards,

Dieter