From mboxrd@z Thu Jan  1 00:00:00 1970
Message-ID: <d84403de694e4cb39d58012aad2f033f@quanstro.net>
From: erik quanstrom <quanstro@quanstro.net>
Date: Tue, 15 Sep 2009 17:20:36 -0400
To: 9fans@9fans.net
MIME-Version: 1.0
Content-Type: text/plain; charset="US-ASCII"
Content-Transfer-Encoding: 7bit
Subject: [9fans] stopgap for vlong bug
Topicbox-Message-UUID: 6ec26d6a-ead5-11e9-9d60-3106f5b1d025

reported here http://9fans.net/narchive/2009/08/1042

i haven't had much time to even look at this problem.  this
will at least abort the compile, rather than miscompile.  so
it's marginally better than nothing.

the warning is just documentation of the problem.  i recompiled
everything with this patch and didn't see it trip (except in
my lucky code), so i wonder if there is always a problem if
t->right->reg != D_DX holds.

/n/sources/plan9//sys/src/cmd/8c/cgen64.c:2293,2298 - cgen64.c:2293,2300
  					evacaxdx(l);
  				}
  			}
+ 			if(l->op == OSUB && l->right != Z && l->right->op == OCAST)
+ 				warn(n, "likely vlong bug");
  			t = regpair(Z, n);
  			sugen(r, t, 8);
  			r = t;
/n/sources/plan9//sys/src/cmd/8c/cgen64.c:2336,2341 - cgen64.c:2338,2344
  			zapreg(t->left);
  		}
  		if(t->right->reg != D_DX) {
+ 			fatal(Z, "vlong double-spill botch");
  			d = &nod4;
  			regsalloc(d, t->right);
  			gmove(&nod2, d);

- erik