Gary Thomas: Code optimization bug in GCC-2.8 & EGCS

[David Edelsohn]

>>>>> Gary Thomas writes:

Gary> I spent many days trying to understand the "why" of this.  I believe that
Gary> in CSE, follow jumps only works properly if loops have been appropriately
Gary> "marked" (which is what fails to happen in JUMP in this case).  The CSE then
Gary> blindly optimizes away the first test (loop entry - above what should be
Gary> LOOP_VTOP).  I found that turning off CSE follow jumps kept the problem
Gary> from generating *icorrect* code at the expense of generating *non-optimal* code.

Gary> Correctly identified loops seem to work OK.  The failure here is that on the
Gary> PowerPC with signed characters, the loop exit code is "just a little too big."

	If CSE does not handle unmarked loops properly then that is a bug
in CSE.  For some sized loop, this always will fail. 

David