Your change causing regressions

[David S. Miller]

   From: law@redhat.com
   Date: Tue, 25 Jun 2002 12:53:45 -0600

   I've verified that your change is definitely responsible for the 
   f90-intrinsic-bit.f failures for ppc-eabisim as well as x86 native.
   I did not investigate the 20020108-1.c failure.
   
   f90-intrinsic-bit.f is failing with -Os because compilation of
   the test times out.  

The 20020108-1.c failure occurs on Sparc too and fails for the
same reason, compilation timeout with -Os.

It is hanging during the code hoisting pass of 'main', the GDB
backtrace is:

et_forest_common_ancestor
nearest_common_dominator
dominated_by_p
hoist_code
one_code_hoisting_pass
gcse_main
rest_of_compilation
c_expand_body
finish_function
yyparse
c_common_parse_file
compile_file
do_compile
toplev_main

There is another similar regression (nonterminating algorithm) still
in the mainline which has been there for quite some time and it is
"well known", see PR middle-end/6566 which is assigned to Jan.