[PATCH]: Fix gfortran regressions

[Daniel Berlin]

I forgot that VIEW_CONVERT_EXPR was a tcc_reference expression, which
means it gets value numbered like the rest of our loads/stores.

The default case of value numbering references used to just happily
ignore the operation in question if it wasn't handled.

This means it would value number VIEW_CONVERT<int4>("A4") and
VIEW_CONVERT<int4>("A5") to the same.

To prevent this from happening in the future, I have modified the
default case to be gcc_unreachable, and made the things we were
*expecting* to fall into the default, explicit.

I will commit this once java finishes testing, it has already been
verified to fix the gfortran regressions.

Bootstrapped and regtested on i686-darwin and i686-linux


2007-07-01  Daniel Berlin  <dberlin@dberlin.org>

	* tree-ssa-sccvn.c (copy_reference_ops_from_ref): Handle constants
	and ohter expected operations explicitly, change default to
	gcc_unreachable.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: gfortran.diff
Type: text/x-patch
Size: 1236 bytes
Desc: not available
URL: <http://gcc.gnu.org/pipermail/fortran/attachments/20070701/d680d522/attachment.bin>