Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!samsung!uunet!lll-winken!decwrl!bacchus.pa.dec.com!shlump.nac.dec.com!shell.nac.dec.com!forecast
From: forecast@shell.nac.dec.com (John Forecast)
Newsgroups: comp.os.minix
Subject: C68 problems
Message-ID: <15219@shlump.nac.dec.com>
Date: 7 Sep 90 17:31:29 GMT
Sender: news@shlump.nac.dec.com
Reply-To: forecast@shell.enet.dec.com
Organization: Digital Equipment Corporation
Lines: 25

I tried to mail directly to Christoph but it doesn't look as though it made it.

1. Patch 2 seems to have broken some enum arithmetic. In particular,
   enum - enum now reports an error from g_addsub indicating that the compiler
   does not support the operation or it's a compiler bug.

2. The peephole optimizer for and (peep_and) does not check if a source
   addressing mode relies on the destination register before applying one of
   it's optimizations. This results in:

	move.w	0(a2,d0.l),d0
	and.l	#65535,d0.l		being changed to

	clr.l	d0.l
	move.w	0(a2,d0.l),d0

I added the following check to the last if statement in peep_and 

	|| (prev->oper1->mode == am_indx2 &&
	    prev->oper1->sreg == ip->oper2->preg)

If you use C68 to build your kernel, there is at least one of these in
mm/signal.s.

John.