------- Additional Comments From para at cfl dot rr dot com 2005-07-05 14:40 ------- The analysis is slightly flawed. For example:
UINT32 r0045025C = opt_and(ic, r004501D4); // N = and (_, M) :00022108 UINT32 r00450994 = opt_not(r0045025C); // b = not (N) :fffddef7 Since you are using 1s to represent bits that are known to be 0 and 0s to represent bits whose value are completely unknown, b should be 00000000. All known 0s flip to 1s, and you can say nothing about the other bits because their values are all unknown. However, I ran my own analysis and came to the same conclusion. It seems that the mistake was mine as I was using and (&) instead of add (+) for the round key. My appologies. Interestingly, GCC 4.0 does *not* catch this optimization opportunity. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21970