bug in new ia32 backend

[Mike Stump]

> Date: Sat, 26 Jun 1999 21:53:22 +0200
> From: Toon Moene <toon@moene.indiv.nluug.nl>
> To: Mike Stump <mrs@wrs.com>

> Mike, I've tried to get a successful bootstrap with your patches (on
> i686-pc-linux-gnu) but failed: The stage1 compiler hangs while
> compiling regclass.c (compiling options -O2 -g).

As I have reported, I have fixed loop.c to version 1.162.  I suspect
recent checkins by Mark have fixed the problem I worked around, but
have not verified it.

> Although I do not agree with your "solution" to the null JUMP_LABEL
> field in the routine loop_iterations in unroll.c,

No, that wasn't a fix or a solution.  That was just a hack around to
allow me to continue testing, as I wanted complete test coverage, as
soon as possible.  I wanted to point out all the problems that might
exist as soon as I could.  I've done that.  Now I will go back, resync
my tree, and retest again, and see what work arounds are necessary,
and then go back and develop the fixes into real fixes.

> What I would like you to do is to send me the uuencoded, gzip'd tar file
> of the following

Let me resync, then we will have a common source tree to talk about.
Also, as I redive back into it, in this pass I will do much more
analysis than the first one.  I will provide a pointer to what routine
zeroed it, or what routine created it...

I will see about sending you what you requested...  I think you can
get it yourself by not bootstrapping.  bootstrapping isn't necessary
for development, or for verification of other bugs.  It is a way to
run additional tests and checks, or to work around bugs that may exist
with a native compiler.  If you native compiler is buggy, select a
different one.