------- Comment #8 from amodra at bigpond dot net dot au 2005-11-17 07:17
-------
In any case, I think the backporting 4.1 patches won't fix this problem
reliably. I can't see anything that really prevents a similar problem on
mainline, ie. of the first scheduling pass moving some instruction loading
constants to a place where r2 is invalid.
On the bright side, I've backported the patch you mention in comment #7, and
now my 4.0 tree bootstraps (regression test still in progress), and the
testcase here passes.
--
amodra at bigpond dot net dot au changed:
What |Removed |Added
----------------------------------------------------------------------------
Known to work|4.1.0 |
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24779
