Path: utzoo!utgpu!jarvis.csri.toronto.edu!cs.utexas.edu!usc!snorkelwacker!bloom-beacon!eru!luth!sunic!tut!santra!tuura!risto
From: risto@tuura.UUCP (Risto Lankinen)
Newsgroups: comp.sys.intel
Subject: Undoc'd immediate MUL in 80X86 (incl.8086)
Message-ID: <622@tuura.UUCP>
Date: 14 Feb 90 09:26:35 GMT
Distribution: comp
Organization: Nokia Data Systems Oy
Lines: 53

A little experimenting with Microsoft Macro Assembler version 5.10 and
the instructions AAM and AAD revealed a neat way to do immediate 8-bit
MUL and DIV on 80X86:s.  The 'feature' is *ABSOLUTELY* undocumented by
either Intel or Microsoft or any other organization associated.  Who-
ever uses it must consider the possibility of the instructions not
working in a particular computer.  I took the chance to use them in
an application for personal use only, since the only computer they are
used is my ow only, in which they work.

I ran into this using the SYMDEB, with which I accidentally used 'u' to
"disassemble" random address.  The SYMDEB listed instructions like
AAM 2B and AAD 2A .  A little browsin of the Microsoft Press' 386-book
revealed the op-codes for the AAM and AAD are D40Ah and D50Ah .  The 
specified function for AAM is ( AL <- AL mod 10d , AH <- AL div 10d )
and for AAD ( AL <- 10d * AH + AL , AH <- 0 ) .  Now, the 2nd byte of
both opcodes is 10d .  I tried replacing this byte by different values
instead of 0Ah (=10d), and it still worked in a similar way.  Then I
made two macros for MS-MASM 5.10:

bmul MACRO immed
   db 0D5h,immed
ENDM
bdiv MACRO immed
   db 0D4h,immed
ENDM
   ...
   mov ah,octave
   mov al,note_number
   bmul 12
;  AL would now contain 12*octave+note_number for a MIDI application...
;  the SYMDEB lists the previous instruction as
;  AAD 0C

The bdiv does essentially the same as "div <immed>","xchg al,ah", so
the advantage would be the immediacy of the division + different 
register output which is good for some purposes (nt to mention small
space saving for perfectionists).  The speed difference is not very
significant.
The bmul does a bit more: "mov temp,al","xchg ah,al","mul <immed>",
"add al,temp" and "sub ah,ah" .  For the purpose in the example above
this suits well, as it is used in an interrupt service routine and
is rather quick and small compared to the 'complete' way.

Once more, this habit is absolutely discouraged for programs that
should work in different processors, present or future.  As an added
curiosity, and because I don't have such in my possession, I would
like to know what happens, if you run this on a NEC Vprocessor. I
have run it in 8086, 286 and 386 (full ver.), and it works in those
all.

terveisin: Risto Lankinen

--