Path: utzoo!attcan!uunet!husc6!bloom-beacon!mit-eddie!rutgers!ucsd!ucbvax!hplabs!hpda!hpcuhb!hpindda!vandys
From: vandys@hpindda.HP.COM (Andy Valencia)
Newsgroups: comp.os.minix
Subject: Re: Re:  GNU c and c++ under MINIX (long)
Message-ID: <3580014@hpindda.HP.COM>
Date: 16 Sep 88 20:43:03 GMT
References: <2358@alliant.Alliant.COM>
Organization: HP Information Networks, Cupertino, CA
Lines: 32

/ hpindda:comp.os.minix / rana@hpqtdla.HP.COM (Rana Raychoudhury) /  5:40 am  Sep 15, 1988 /
>Now, porting gcc to *minix* may not be a problem per se, but running gcc
>on machines without certain features (above) will be. From the text above,
>it would appear that porting gcc to Intel 80X8X architectures MAY be difficult.

    This sounds like the argument against pcc working on 8086s.  No general
registers, no auto increment/decrement, mov instruction doesn't set the
flags, etc.  I managed to tweak over the pcc in a couple of days.  Wasn't
too bad, either.  Consider:

    AX, CX, DX - Perfectly good accumulators.  I made them AREGs
    BX, SI, DI - Perfectly good pointers.  BREGs.
    BP, SP, PC - Reserved.  BP keeps the stack frame.
    DS, ES, CS, SS - Sorry, I did small model.

    SI and DI could be used as register variables, otherwise they were in
the BREG pool.  The Sethi-Ullman numbers guided the compiler in either case.
It was interesting to watch the compiler adapt to various shortages of
registers....  Because the pcc code templates could specify FORCC or not,
I just made the obvious move template first, then had one which included
FORCC next.  The first one matched moves which only needed the effect; the
second one matched moves whose result was desired.  It was just like the
first, but would generate an extra instruction to set the flags after the
move.  Without memory-memory moves, the data had to be in a register anyway,
so it was pretty cheap.

    If any sites with V7 source would like to track my work, I'll be happy
to supply diffs.  It isn't easy to get a source license for a lowly PC, but
it makes an amusing cross compiler!  See the paper "A Tour through the
Portable C Compiler" to get a better idea of how this stuff works.

					Andy