[PATCH] Fix for PR tree-optimization/27341

[Daniel Berlin]

This bug is twofold.

First, gimplify_val would create statements that nobody got a hold of
easily (IE you'd have to go walking back through the bb to see if
statements were added), but would never mark any of the new vars in them
for renaming.

*most* of the time, we were getting lucky, and other parts of
tree-complex.c would end up marking the variables for renaming as parts
of other statements.

However, in one of the attached testcases, this doesn't happen, and we
ICE as a result.

This bug is hidden by the second bug, which is that complex lowering
will sometimes change SMT usage when it creates IMAGEPART_EXPR or
REALPART_EXPR of an indirect ref, and that indirect ref is an incoming
Complex *, and a few other requirements.

This is fixed by just telling it to update smt usage.


Bootstrapped and regtested on i686-pc-linux-gnu.

Okay for mainline?

--Dan
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 27341.diff
Type: text/x-patch
Size: 2177 bytes
Desc: not available
URL: <http://gcc.gnu.org/pipermail/gcc-patches/attachments/20060616/f9979056/attachment.bin>