PATCH: Small combine glitch, testcase, and patch.

[Jeffrey A Law]

  In message < Pine.SOL.3.91.1010221184804.24150A-100000@cse.cygnus.com >you writ
e:
  > 
  > In the testcase below, combine was erroneously deleting an smax operation
  > because it concluded that the value of the operation under the if_then_else
  > it fed (with the same operands) was always zero. This is due to incorrect
  > handling of NE/EQ by known_cond.
  > 
  > Bootstrapped on i686-linux (the testcase originally failed on hppa, but 
  > bootstrapping there is agonizingly slow).
  > 
  > 
  > 2001-02-21  Jason Eckhardt  <jle@redhat.com>
  > 
  > 	* combine.c (known_cond): Do not reverse the condition when
  > 	SMAX/UMAX is being considered and the condition is for equality 
  > 	or inequality.
  > 
  > 	* testsuite/gcc.c-torture/execute/20010221-1.c: New test.
Approved.  Please install.

Thanks,
jeff