Re: R65C02 PI

[nparker@cie-2.uoregon.edu (Neil Parker)]

In article <500k7k$mf9@nnrp1.news.primenet.com> johnlb@primenet.com (John
Bowling) writes:
>In article <4vv3gg$an7@europa.frii.com>,
>Randy Shackelford <shack@deimos.frii.com> wrote:
>>All the old games which I have tried with the exception of One on One work
>>on a //c and IIgs for that matter.
>
>Europe Ablaze, Think Tank, and Evelyn Wood Speed Reading don't work on the
>gs.  I haven't bothered to look into why, but it has to be related to
>either use of jumps into RAM or Dos/ProDOS which are undocumented, or use
>of 6502 undocumented opcodes.

Actually, there is at least one more reason why non-GS software can fail on
a GS:  absolute,X and absolute,Y addressing.  This is probably a once-in-a-
lifetime pitfall, but I have a program that fell into it.

Machine-language programmers might want to take note of this.

A few years ago I inherited a copy of MicroMotion Forth-79 from a former
Apple II user.  It worked just fine on an Apple IIe, but I couldn't get it
to boot on my IIGS.  For a long time I couldn't figure out why--it makes
only minimal use of the ROMs, and only through published entry points, and
it doesn't seem to use any undocumented 6502 opcodes.

Finally a few months ago I found the cause of the problem.  There are
several places in its machine code where it uses instructions like this:

    LDA $FFFF,X
    ORA $FFFE,X

The intent here was for the indexing to wrap around to zero page, and
access the zero-page memory location one or two bytes before the location
pointed to by the X register.  This works just fine on the 6502 and 65C02,
but on the 65816, it indexes into bank 1 and picks up the wrong data.

I doubt the programmer who wrote that code had any idea it was happening.
The original assembler source code probably looked something like this:

     LABEL     EQU     0
               LDA     LABEL-1,X
               ORA     LABEL-2,X

Since countless nearby references to "LABEL+1,X", "LABEL+2,X", etc., all
assemble to zero-page,X addressing, it's not immediately obvious that this
is any different.  But even the dumbest 8-bit assembler knows that
0 - 1 = $FFFF, which has a non-zero high byte, so it dutifully assembles
absolute,X addressing.

Had zero-page,X addressing been used for those instructions, there wouldn't
have been any problem, since zero-page,X addressing wraps within zero page
on the 65816 as long as the direct page register is page-aligned (as it
almost always is in emulation mode).  It would have been a couple of cycles
faster too.

Moral:  If you're storing data in a table on page zero, check the output of
your assembler carefully.  You may be generating GS-incompatible code
without even realizing it.


I was able to get MicroMotion Forth-79 to work on my IIGS by hunting down
all the $FFFF,X and $FFFE,X references, and replacing them like this:

     LDA $FF,X
     NOP
     ORA $FE,X
     NOP

               - Neil Parker

P.S.  Is there anybody out there besides me who actually owns a copy of
MicroMotion Forth-79, and can't get to work on your GS?  If so, send
e-mail, and I'll send you the patch info.
-- 
Neil Parker, nparker@{cie-2,cie}.uoregon.edu, http://cie-2.uoregon.edu/~nparker

     "Evolution is vastly overrated."  -- Ambassador Delenn, _Babylon_5_