From:	SMTP%"MACRO32%WKUVX1.BITNET@uu7.psi.com"  7-AUG-1993 12:17:59.55
To:	EVERHART
CC:	
Subj:	Re: AMAC vs. computed JMP

Message-Id: <9308040742.AA08370@uu7.psi.com>
X-Listname: "VMS Internals, MACRO, & BLISS Discussions" <MACRO32@WKUVX1.BITNET>
Warnings-To: <>
Errors-To: MacroMan%WKUVX1.BITNET@uu7.psi.com
Sender: MacroMan%WKUVX1.BITNET@uu7.psi.com
Date: Wed, 04 Aug 1993 09:37:23 +0200
From: "GWDGV1::MOELLER" <moeller@gwdgv1.dnet.gwdg.de>
Reply-To: MACRO32%WKUVX1.BITNET@uu7.psi.com
To: MACRO32@WKUVX1.BITNET
Subject: Re: AMAC vs. computed JMP

<BJJ@ECL.PSU.EDU> writes [in response to my question and Arne's answer]:
>[...]
>            Explanation: When an instruction such as MOVAx or PUSHAx
>            is used to store the address of a code label which has not
>            been declared as an entry point, the compiler reports the
>            STORCODLAB error. The reason for this is that any legal
>            reference to the stored address will result in a JSB or CALL
>            to that address. For example, JMP 4(R0) will be treated as a
>            JSB (followed immediately by an RSB in the current routine).
>
>
> You'd probably use .JSB32_ENTRY instead of .JSB_ENTRY so
> the compiler doesn't go trashing all your registers.

I think this still means that you cannot 'randomly' transfer control
(all within the same module, of course) with JMP and have the same
semantics as a branch (with regard to the stack contents).

In the meantime, I converted my program to use some big CASEL intructions
(by storing 'case labels' instead of the actual adresses).

The 'stored code label' problem also affects the 'NEWPC' argument
to the $UNWIND system service: In my undestanding, in the following
example, the stack pointer changes when the 'alternate return' [x1]
is taken. This won't hurt if the main program returns soon (like in the
example), but if the main program would want to continue, I wouldn't know
how to limit the stack expansion ...

Example (just to test the machine code):

        .title  z
;
        .psect  _rodata,quad,nowrt,noexe
k_0:    .long   0
;
        .psect  _code,quad,nowrt,exe
;
handler::       .call_entry
        $unwind_s       depadr=k_0,newpc=x1
        ret
;
main::  .call_entry
        movab   handler,(fp)
        clrl    r1
        divl3   r1,#1,r0
        ret
;
x1:     .jsb32_entry
;;                              ;; something pushed on stack here ???
        movl    #4,r0
        ret
;
        .end    main

Wolfgang J. Moeller, GWDG, D-37018 Goettingen, F.R.Germany | Disclaimer ...
PSI%(0262)45050352008::MOELLER       Phone: +49 551 201516 | No claim intended!
Internet: moeller@gwdgv1.dnet.gwdg.de    | This space intentionally left blank.