Fix handling of CFG_ALTERED in main DOM loop.

[Jeffrey A Law <law at redhat dot com>]

This is something I noticed a long long time ago, but didn't install it
into the tree because it results in a small compile-time regression.

Basically in two places we should be OR-ing in a value to CFG_ALTERED
rather than just assigning a new value to CFG_ALTERED.  The net result
was that DOM was not iterating in some cases where is should have
iterated.  Redundancies exposed by jump threading were left to be
detected during a later DOM pass (or the RTL optimizers).

The long term hope is that other changes will supercede the main
part of DOM and/or the iteration step can be drastically revamped
to reduce its cost (by realizing that most of the time only a small
number of statements actually need to be revisited and re-optimized
after jump threading).

Bootstrapped and regression tested on i686-pc-linux-gnu.  Expect a
1% compile-time slowdown.  Instealled into the mainline sources only.

Attachment: PP
Description: Text document